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) | |
3d48ae45 | 27 | * mapping->i_mmap_mutex |
2b575eb6 | 28 | * anon_vma->mutex |
82591e6e NP |
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) | |
250df6ed | 34 | * inode->i_lock (in set_page_dirty's __mark_inode_dirty) |
a66979ab | 35 | * inode_wb_list_lock (in set_page_dirty's __mark_inode_dirty) |
82591e6e NP |
36 | * sb_lock (within inode_lock in fs/fs-writeback.c) |
37 | * mapping->tree_lock (widely used, in set_page_dirty, | |
38 | * in arch-dependent flush_dcache_mmap_lock, | |
a66979ab | 39 | * within inode_wb_list_lock in __sync_single_inode) |
6a46079c AK |
40 | * |
41 | * (code doesn't rely on that order so it could be switched around) | |
42 | * ->tasklist_lock | |
2b575eb6 | 43 | * anon_vma->mutex (memory_failure, collect_procs_anon) |
6a46079c | 44 | * pte map lock |
1da177e4 LT |
45 | */ |
46 | ||
47 | #include <linux/mm.h> | |
48 | #include <linux/pagemap.h> | |
49 | #include <linux/swap.h> | |
50 | #include <linux/swapops.h> | |
51 | #include <linux/slab.h> | |
52 | #include <linux/init.h> | |
5ad64688 | 53 | #include <linux/ksm.h> |
1da177e4 LT |
54 | #include <linux/rmap.h> |
55 | #include <linux/rcupdate.h> | |
a48d07af | 56 | #include <linux/module.h> |
8a9f3ccd | 57 | #include <linux/memcontrol.h> |
cddb8a5c | 58 | #include <linux/mmu_notifier.h> |
64cdd548 | 59 | #include <linux/migrate.h> |
0fe6e20b | 60 | #include <linux/hugetlb.h> |
1da177e4 LT |
61 | |
62 | #include <asm/tlbflush.h> | |
63 | ||
b291f000 NP |
64 | #include "internal.h" |
65 | ||
fdd2e5f8 | 66 | static struct kmem_cache *anon_vma_cachep; |
5beb4930 | 67 | static struct kmem_cache *anon_vma_chain_cachep; |
fdd2e5f8 AB |
68 | |
69 | static inline struct anon_vma *anon_vma_alloc(void) | |
70 | { | |
01d8b20d PZ |
71 | struct anon_vma *anon_vma; |
72 | ||
73 | anon_vma = kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL); | |
74 | if (anon_vma) { | |
75 | atomic_set(&anon_vma->refcount, 1); | |
76 | /* | |
77 | * Initialise the anon_vma root to point to itself. If called | |
78 | * from fork, the root will be reset to the parents anon_vma. | |
79 | */ | |
80 | anon_vma->root = anon_vma; | |
81 | } | |
82 | ||
83 | return anon_vma; | |
fdd2e5f8 AB |
84 | } |
85 | ||
01d8b20d | 86 | static inline void anon_vma_free(struct anon_vma *anon_vma) |
fdd2e5f8 | 87 | { |
01d8b20d | 88 | VM_BUG_ON(atomic_read(&anon_vma->refcount)); |
fdd2e5f8 AB |
89 | kmem_cache_free(anon_vma_cachep, anon_vma); |
90 | } | |
1da177e4 | 91 | |
5beb4930 RR |
92 | static inline struct anon_vma_chain *anon_vma_chain_alloc(void) |
93 | { | |
94 | return kmem_cache_alloc(anon_vma_chain_cachep, GFP_KERNEL); | |
95 | } | |
96 | ||
e574b5fd | 97 | static void anon_vma_chain_free(struct anon_vma_chain *anon_vma_chain) |
5beb4930 RR |
98 | { |
99 | kmem_cache_free(anon_vma_chain_cachep, anon_vma_chain); | |
100 | } | |
101 | ||
d9d332e0 LT |
102 | /** |
103 | * anon_vma_prepare - attach an anon_vma to a memory region | |
104 | * @vma: the memory region in question | |
105 | * | |
106 | * This makes sure the memory mapping described by 'vma' has | |
107 | * an 'anon_vma' attached to it, so that we can associate the | |
108 | * anonymous pages mapped into it with that anon_vma. | |
109 | * | |
110 | * The common case will be that we already have one, but if | |
23a0790a | 111 | * not we either need to find an adjacent mapping that we |
d9d332e0 LT |
112 | * can re-use the anon_vma from (very common when the only |
113 | * reason for splitting a vma has been mprotect()), or we | |
114 | * allocate a new one. | |
115 | * | |
116 | * Anon-vma allocations are very subtle, because we may have | |
117 | * optimistically looked up an anon_vma in page_lock_anon_vma() | |
118 | * and that may actually touch the spinlock even in the newly | |
119 | * allocated vma (it depends on RCU to make sure that the | |
120 | * anon_vma isn't actually destroyed). | |
121 | * | |
122 | * As a result, we need to do proper anon_vma locking even | |
123 | * for the new allocation. At the same time, we do not want | |
124 | * to do any locking for the common case of already having | |
125 | * an anon_vma. | |
126 | * | |
127 | * This must be called with the mmap_sem held for reading. | |
128 | */ | |
1da177e4 LT |
129 | int anon_vma_prepare(struct vm_area_struct *vma) |
130 | { | |
131 | struct anon_vma *anon_vma = vma->anon_vma; | |
5beb4930 | 132 | struct anon_vma_chain *avc; |
1da177e4 LT |
133 | |
134 | might_sleep(); | |
135 | if (unlikely(!anon_vma)) { | |
136 | struct mm_struct *mm = vma->vm_mm; | |
d9d332e0 | 137 | struct anon_vma *allocated; |
1da177e4 | 138 | |
5beb4930 RR |
139 | avc = anon_vma_chain_alloc(); |
140 | if (!avc) | |
141 | goto out_enomem; | |
142 | ||
1da177e4 | 143 | anon_vma = find_mergeable_anon_vma(vma); |
d9d332e0 LT |
144 | allocated = NULL; |
145 | if (!anon_vma) { | |
1da177e4 LT |
146 | anon_vma = anon_vma_alloc(); |
147 | if (unlikely(!anon_vma)) | |
5beb4930 | 148 | goto out_enomem_free_avc; |
1da177e4 | 149 | allocated = anon_vma; |
1da177e4 LT |
150 | } |
151 | ||
cba48b98 | 152 | anon_vma_lock(anon_vma); |
1da177e4 LT |
153 | /* page_table_lock to protect against threads */ |
154 | spin_lock(&mm->page_table_lock); | |
155 | if (likely(!vma->anon_vma)) { | |
156 | vma->anon_vma = anon_vma; | |
5beb4930 RR |
157 | avc->anon_vma = anon_vma; |
158 | avc->vma = vma; | |
159 | list_add(&avc->same_vma, &vma->anon_vma_chain); | |
26ba0cb6 | 160 | list_add_tail(&avc->same_anon_vma, &anon_vma->head); |
1da177e4 | 161 | allocated = NULL; |
31f2b0eb | 162 | avc = NULL; |
1da177e4 LT |
163 | } |
164 | spin_unlock(&mm->page_table_lock); | |
cba48b98 | 165 | anon_vma_unlock(anon_vma); |
31f2b0eb ON |
166 | |
167 | if (unlikely(allocated)) | |
01d8b20d | 168 | put_anon_vma(allocated); |
31f2b0eb | 169 | if (unlikely(avc)) |
5beb4930 | 170 | anon_vma_chain_free(avc); |
1da177e4 LT |
171 | } |
172 | return 0; | |
5beb4930 RR |
173 | |
174 | out_enomem_free_avc: | |
175 | anon_vma_chain_free(avc); | |
176 | out_enomem: | |
177 | return -ENOMEM; | |
1da177e4 LT |
178 | } |
179 | ||
5beb4930 RR |
180 | static void anon_vma_chain_link(struct vm_area_struct *vma, |
181 | struct anon_vma_chain *avc, | |
182 | struct anon_vma *anon_vma) | |
1da177e4 | 183 | { |
5beb4930 RR |
184 | avc->vma = vma; |
185 | avc->anon_vma = anon_vma; | |
186 | list_add(&avc->same_vma, &vma->anon_vma_chain); | |
187 | ||
cba48b98 | 188 | anon_vma_lock(anon_vma); |
05759d38 AA |
189 | /* |
190 | * It's critical to add new vmas to the tail of the anon_vma, | |
191 | * see comment in huge_memory.c:__split_huge_page(). | |
192 | */ | |
5beb4930 | 193 | list_add_tail(&avc->same_anon_vma, &anon_vma->head); |
cba48b98 | 194 | anon_vma_unlock(anon_vma); |
1da177e4 LT |
195 | } |
196 | ||
5beb4930 RR |
197 | /* |
198 | * Attach the anon_vmas from src to dst. | |
199 | * Returns 0 on success, -ENOMEM on failure. | |
200 | */ | |
201 | int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src) | |
1da177e4 | 202 | { |
5beb4930 RR |
203 | struct anon_vma_chain *avc, *pavc; |
204 | ||
646d87b4 | 205 | list_for_each_entry_reverse(pavc, &src->anon_vma_chain, same_vma) { |
5beb4930 RR |
206 | avc = anon_vma_chain_alloc(); |
207 | if (!avc) | |
208 | goto enomem_failure; | |
209 | anon_vma_chain_link(dst, avc, pavc->anon_vma); | |
210 | } | |
211 | return 0; | |
1da177e4 | 212 | |
5beb4930 RR |
213 | enomem_failure: |
214 | unlink_anon_vmas(dst); | |
215 | return -ENOMEM; | |
1da177e4 LT |
216 | } |
217 | ||
5beb4930 RR |
218 | /* |
219 | * Attach vma to its own anon_vma, as well as to the anon_vmas that | |
220 | * the corresponding VMA in the parent process is attached to. | |
221 | * Returns 0 on success, non-zero on failure. | |
222 | */ | |
223 | int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma) | |
1da177e4 | 224 | { |
5beb4930 RR |
225 | struct anon_vma_chain *avc; |
226 | struct anon_vma *anon_vma; | |
1da177e4 | 227 | |
5beb4930 RR |
228 | /* Don't bother if the parent process has no anon_vma here. */ |
229 | if (!pvma->anon_vma) | |
230 | return 0; | |
231 | ||
232 | /* | |
233 | * First, attach the new VMA to the parent VMA's anon_vmas, | |
234 | * so rmap can find non-COWed pages in child processes. | |
235 | */ | |
236 | if (anon_vma_clone(vma, pvma)) | |
237 | return -ENOMEM; | |
238 | ||
239 | /* Then add our own anon_vma. */ | |
240 | anon_vma = anon_vma_alloc(); | |
241 | if (!anon_vma) | |
242 | goto out_error; | |
243 | avc = anon_vma_chain_alloc(); | |
244 | if (!avc) | |
245 | goto out_error_free_anon_vma; | |
5c341ee1 RR |
246 | |
247 | /* | |
248 | * The root anon_vma's spinlock is the lock actually used when we | |
249 | * lock any of the anon_vmas in this anon_vma tree. | |
250 | */ | |
251 | anon_vma->root = pvma->anon_vma->root; | |
76545066 | 252 | /* |
01d8b20d PZ |
253 | * With refcounts, an anon_vma can stay around longer than the |
254 | * process it belongs to. The root anon_vma needs to be pinned until | |
255 | * this anon_vma is freed, because the lock lives in the root. | |
76545066 RR |
256 | */ |
257 | get_anon_vma(anon_vma->root); | |
5beb4930 RR |
258 | /* Mark this anon_vma as the one where our new (COWed) pages go. */ |
259 | vma->anon_vma = anon_vma; | |
5c341ee1 | 260 | anon_vma_chain_link(vma, avc, anon_vma); |
5beb4930 RR |
261 | |
262 | return 0; | |
263 | ||
264 | out_error_free_anon_vma: | |
01d8b20d | 265 | put_anon_vma(anon_vma); |
5beb4930 | 266 | out_error: |
4946d54c | 267 | unlink_anon_vmas(vma); |
5beb4930 | 268 | return -ENOMEM; |
1da177e4 LT |
269 | } |
270 | ||
5beb4930 | 271 | static void anon_vma_unlink(struct anon_vma_chain *anon_vma_chain) |
1da177e4 | 272 | { |
5beb4930 | 273 | struct anon_vma *anon_vma = anon_vma_chain->anon_vma; |
1da177e4 LT |
274 | int empty; |
275 | ||
5beb4930 | 276 | /* If anon_vma_fork fails, we can get an empty anon_vma_chain. */ |
1da177e4 LT |
277 | if (!anon_vma) |
278 | return; | |
279 | ||
cba48b98 | 280 | anon_vma_lock(anon_vma); |
5beb4930 | 281 | list_del(&anon_vma_chain->same_anon_vma); |
1da177e4 LT |
282 | |
283 | /* We must garbage collect the anon_vma if it's empty */ | |
01d8b20d | 284 | empty = list_empty(&anon_vma->head); |
cba48b98 | 285 | anon_vma_unlock(anon_vma); |
1da177e4 | 286 | |
01d8b20d PZ |
287 | if (empty) |
288 | put_anon_vma(anon_vma); | |
1da177e4 LT |
289 | } |
290 | ||
5beb4930 RR |
291 | void unlink_anon_vmas(struct vm_area_struct *vma) |
292 | { | |
293 | struct anon_vma_chain *avc, *next; | |
294 | ||
5c341ee1 RR |
295 | /* |
296 | * Unlink each anon_vma chained to the VMA. This list is ordered | |
297 | * from newest to oldest, ensuring the root anon_vma gets freed last. | |
298 | */ | |
5beb4930 RR |
299 | list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) { |
300 | anon_vma_unlink(avc); | |
301 | list_del(&avc->same_vma); | |
302 | anon_vma_chain_free(avc); | |
303 | } | |
304 | } | |
305 | ||
51cc5068 | 306 | static void anon_vma_ctor(void *data) |
1da177e4 | 307 | { |
a35afb83 | 308 | struct anon_vma *anon_vma = data; |
1da177e4 | 309 | |
2b575eb6 | 310 | mutex_init(&anon_vma->mutex); |
83813267 | 311 | atomic_set(&anon_vma->refcount, 0); |
a35afb83 | 312 | INIT_LIST_HEAD(&anon_vma->head); |
1da177e4 LT |
313 | } |
314 | ||
315 | void __init anon_vma_init(void) | |
316 | { | |
317 | anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma), | |
20c2df83 | 318 | 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor); |
5beb4930 | 319 | anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain, SLAB_PANIC); |
1da177e4 LT |
320 | } |
321 | ||
322 | /* | |
6111e4ca PZ |
323 | * Getting a lock on a stable anon_vma from a page off the LRU is tricky! |
324 | * | |
325 | * Since there is no serialization what so ever against page_remove_rmap() | |
326 | * the best this function can do is return a locked anon_vma that might | |
327 | * have been relevant to this page. | |
328 | * | |
329 | * The page might have been remapped to a different anon_vma or the anon_vma | |
330 | * returned may already be freed (and even reused). | |
331 | * | |
332 | * All users of this function must be very careful when walking the anon_vma | |
333 | * chain and verify that the page in question is indeed mapped in it | |
334 | * [ something equivalent to page_mapped_in_vma() ]. | |
335 | * | |
336 | * Since anon_vma's slab is DESTROY_BY_RCU and we know from page_remove_rmap() | |
337 | * that the anon_vma pointer from page->mapping is valid if there is a | |
338 | * mapcount, we can dereference the anon_vma after observing those. | |
1da177e4 | 339 | */ |
746b18d4 | 340 | struct anon_vma *page_get_anon_vma(struct page *page) |
1da177e4 | 341 | { |
746b18d4 | 342 | struct anon_vma *anon_vma = NULL; |
1da177e4 LT |
343 | unsigned long anon_mapping; |
344 | ||
345 | rcu_read_lock(); | |
80e14822 | 346 | anon_mapping = (unsigned long) ACCESS_ONCE(page->mapping); |
3ca7b3c5 | 347 | if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON) |
1da177e4 LT |
348 | goto out; |
349 | if (!page_mapped(page)) | |
350 | goto out; | |
351 | ||
352 | anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON); | |
746b18d4 PZ |
353 | if (!atomic_inc_not_zero(&anon_vma->refcount)) { |
354 | anon_vma = NULL; | |
355 | goto out; | |
356 | } | |
f1819427 HD |
357 | |
358 | /* | |
359 | * If this page is still mapped, then its anon_vma cannot have been | |
746b18d4 PZ |
360 | * freed. But if it has been unmapped, we have no security against the |
361 | * anon_vma structure being freed and reused (for another anon_vma: | |
362 | * SLAB_DESTROY_BY_RCU guarantees that - so the atomic_inc_not_zero() | |
363 | * above cannot corrupt). | |
f1819427 | 364 | */ |
746b18d4 PZ |
365 | if (!page_mapped(page)) { |
366 | put_anon_vma(anon_vma); | |
367 | anon_vma = NULL; | |
368 | } | |
1da177e4 LT |
369 | out: |
370 | rcu_read_unlock(); | |
746b18d4 PZ |
371 | |
372 | return anon_vma; | |
373 | } | |
374 | ||
375 | struct anon_vma *page_lock_anon_vma(struct page *page) | |
376 | { | |
377 | struct anon_vma *anon_vma = page_get_anon_vma(page); | |
378 | ||
379 | if (anon_vma) | |
380 | anon_vma_lock(anon_vma); | |
381 | ||
382 | return anon_vma; | |
34bbd704 ON |
383 | } |
384 | ||
10be22df | 385 | void page_unlock_anon_vma(struct anon_vma *anon_vma) |
34bbd704 | 386 | { |
cba48b98 | 387 | anon_vma_unlock(anon_vma); |
746b18d4 | 388 | put_anon_vma(anon_vma); |
1da177e4 LT |
389 | } |
390 | ||
391 | /* | |
3ad33b24 LS |
392 | * At what user virtual address is page expected in @vma? |
393 | * Returns virtual address or -EFAULT if page's index/offset is not | |
394 | * within the range mapped the @vma. | |
1da177e4 | 395 | */ |
71e3aac0 | 396 | inline unsigned long |
1da177e4 LT |
397 | vma_address(struct page *page, struct vm_area_struct *vma) |
398 | { | |
399 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
400 | unsigned long address; | |
401 | ||
0fe6e20b NH |
402 | if (unlikely(is_vm_hugetlb_page(vma))) |
403 | pgoff = page->index << huge_page_order(page_hstate(page)); | |
1da177e4 LT |
404 | address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); |
405 | if (unlikely(address < vma->vm_start || address >= vma->vm_end)) { | |
3ad33b24 | 406 | /* page should be within @vma mapping range */ |
1da177e4 LT |
407 | return -EFAULT; |
408 | } | |
409 | return address; | |
410 | } | |
411 | ||
412 | /* | |
bf89c8c8 | 413 | * At what user virtual address is page expected in vma? |
ab941e0f | 414 | * Caller should check the page is actually part of the vma. |
1da177e4 LT |
415 | */ |
416 | unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma) | |
417 | { | |
21d0d443 | 418 | if (PageAnon(page)) { |
4829b906 HD |
419 | struct anon_vma *page__anon_vma = page_anon_vma(page); |
420 | /* | |
421 | * Note: swapoff's unuse_vma() is more efficient with this | |
422 | * check, and needs it to match anon_vma when KSM is active. | |
423 | */ | |
424 | if (!vma->anon_vma || !page__anon_vma || | |
425 | vma->anon_vma->root != page__anon_vma->root) | |
21d0d443 AA |
426 | return -EFAULT; |
427 | } else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) { | |
ee498ed7 HD |
428 | if (!vma->vm_file || |
429 | vma->vm_file->f_mapping != page->mapping) | |
1da177e4 LT |
430 | return -EFAULT; |
431 | } else | |
432 | return -EFAULT; | |
433 | return vma_address(page, vma); | |
434 | } | |
435 | ||
81b4082d ND |
436 | /* |
437 | * Check that @page is mapped at @address into @mm. | |
438 | * | |
479db0bf NP |
439 | * If @sync is false, page_check_address may perform a racy check to avoid |
440 | * the page table lock when the pte is not present (helpful when reclaiming | |
441 | * highly shared pages). | |
442 | * | |
b8072f09 | 443 | * On success returns with pte mapped and locked. |
81b4082d | 444 | */ |
e9a81a82 | 445 | pte_t *__page_check_address(struct page *page, struct mm_struct *mm, |
479db0bf | 446 | unsigned long address, spinlock_t **ptlp, int sync) |
81b4082d ND |
447 | { |
448 | pgd_t *pgd; | |
449 | pud_t *pud; | |
450 | pmd_t *pmd; | |
451 | pte_t *pte; | |
c0718806 | 452 | spinlock_t *ptl; |
81b4082d | 453 | |
0fe6e20b NH |
454 | if (unlikely(PageHuge(page))) { |
455 | pte = huge_pte_offset(mm, address); | |
456 | ptl = &mm->page_table_lock; | |
457 | goto check; | |
458 | } | |
459 | ||
81b4082d | 460 | pgd = pgd_offset(mm, address); |
c0718806 HD |
461 | if (!pgd_present(*pgd)) |
462 | return NULL; | |
463 | ||
464 | pud = pud_offset(pgd, address); | |
465 | if (!pud_present(*pud)) | |
466 | return NULL; | |
467 | ||
468 | pmd = pmd_offset(pud, address); | |
469 | if (!pmd_present(*pmd)) | |
470 | return NULL; | |
71e3aac0 AA |
471 | if (pmd_trans_huge(*pmd)) |
472 | return NULL; | |
c0718806 HD |
473 | |
474 | pte = pte_offset_map(pmd, address); | |
475 | /* Make a quick check before getting the lock */ | |
479db0bf | 476 | if (!sync && !pte_present(*pte)) { |
c0718806 HD |
477 | pte_unmap(pte); |
478 | return NULL; | |
479 | } | |
480 | ||
4c21e2f2 | 481 | ptl = pte_lockptr(mm, pmd); |
0fe6e20b | 482 | check: |
c0718806 HD |
483 | spin_lock(ptl); |
484 | if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) { | |
485 | *ptlp = ptl; | |
486 | return pte; | |
81b4082d | 487 | } |
c0718806 HD |
488 | pte_unmap_unlock(pte, ptl); |
489 | return NULL; | |
81b4082d ND |
490 | } |
491 | ||
b291f000 NP |
492 | /** |
493 | * page_mapped_in_vma - check whether a page is really mapped in a VMA | |
494 | * @page: the page to test | |
495 | * @vma: the VMA to test | |
496 | * | |
497 | * Returns 1 if the page is mapped into the page tables of the VMA, 0 | |
498 | * if the page is not mapped into the page tables of this VMA. Only | |
499 | * valid for normal file or anonymous VMAs. | |
500 | */ | |
6a46079c | 501 | int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma) |
b291f000 NP |
502 | { |
503 | unsigned long address; | |
504 | pte_t *pte; | |
505 | spinlock_t *ptl; | |
506 | ||
507 | address = vma_address(page, vma); | |
508 | if (address == -EFAULT) /* out of vma range */ | |
509 | return 0; | |
510 | pte = page_check_address(page, vma->vm_mm, address, &ptl, 1); | |
511 | if (!pte) /* the page is not in this mm */ | |
512 | return 0; | |
513 | pte_unmap_unlock(pte, ptl); | |
514 | ||
515 | return 1; | |
516 | } | |
517 | ||
1da177e4 LT |
518 | /* |
519 | * Subfunctions of page_referenced: page_referenced_one called | |
520 | * repeatedly from either page_referenced_anon or page_referenced_file. | |
521 | */ | |
5ad64688 HD |
522 | int page_referenced_one(struct page *page, struct vm_area_struct *vma, |
523 | unsigned long address, unsigned int *mapcount, | |
524 | unsigned long *vm_flags) | |
1da177e4 LT |
525 | { |
526 | struct mm_struct *mm = vma->vm_mm; | |
1da177e4 LT |
527 | int referenced = 0; |
528 | ||
71e3aac0 AA |
529 | if (unlikely(PageTransHuge(page))) { |
530 | pmd_t *pmd; | |
531 | ||
532 | spin_lock(&mm->page_table_lock); | |
2da28bfd AA |
533 | /* |
534 | * rmap might return false positives; we must filter | |
535 | * these out using page_check_address_pmd(). | |
536 | */ | |
71e3aac0 AA |
537 | pmd = page_check_address_pmd(page, mm, address, |
538 | PAGE_CHECK_ADDRESS_PMD_FLAG); | |
2da28bfd AA |
539 | if (!pmd) { |
540 | spin_unlock(&mm->page_table_lock); | |
541 | goto out; | |
542 | } | |
543 | ||
544 | if (vma->vm_flags & VM_LOCKED) { | |
545 | spin_unlock(&mm->page_table_lock); | |
546 | *mapcount = 0; /* break early from loop */ | |
547 | *vm_flags |= VM_LOCKED; | |
548 | goto out; | |
549 | } | |
550 | ||
551 | /* go ahead even if the pmd is pmd_trans_splitting() */ | |
552 | if (pmdp_clear_flush_young_notify(vma, address, pmd)) | |
71e3aac0 AA |
553 | referenced++; |
554 | spin_unlock(&mm->page_table_lock); | |
555 | } else { | |
556 | pte_t *pte; | |
557 | spinlock_t *ptl; | |
558 | ||
2da28bfd AA |
559 | /* |
560 | * rmap might return false positives; we must filter | |
561 | * these out using page_check_address(). | |
562 | */ | |
71e3aac0 AA |
563 | pte = page_check_address(page, mm, address, &ptl, 0); |
564 | if (!pte) | |
565 | goto out; | |
566 | ||
2da28bfd AA |
567 | if (vma->vm_flags & VM_LOCKED) { |
568 | pte_unmap_unlock(pte, ptl); | |
569 | *mapcount = 0; /* break early from loop */ | |
570 | *vm_flags |= VM_LOCKED; | |
571 | goto out; | |
572 | } | |
573 | ||
71e3aac0 AA |
574 | if (ptep_clear_flush_young_notify(vma, address, pte)) { |
575 | /* | |
576 | * Don't treat a reference through a sequentially read | |
577 | * mapping as such. If the page has been used in | |
578 | * another mapping, we will catch it; if this other | |
579 | * mapping is already gone, the unmap path will have | |
580 | * set PG_referenced or activated the page. | |
581 | */ | |
582 | if (likely(!VM_SequentialReadHint(vma))) | |
583 | referenced++; | |
584 | } | |
585 | pte_unmap_unlock(pte, ptl); | |
586 | } | |
587 | ||
2da28bfd AA |
588 | /* Pretend the page is referenced if the task has the |
589 | swap token and is in the middle of a page fault. */ | |
590 | if (mm != current->mm && has_swap_token(mm) && | |
591 | rwsem_is_locked(&mm->mmap_sem)) | |
592 | referenced++; | |
593 | ||
c0718806 | 594 | (*mapcount)--; |
273f047e | 595 | |
6fe6b7e3 WF |
596 | if (referenced) |
597 | *vm_flags |= vma->vm_flags; | |
273f047e | 598 | out: |
1da177e4 LT |
599 | return referenced; |
600 | } | |
601 | ||
bed7161a | 602 | static int page_referenced_anon(struct page *page, |
6fe6b7e3 WF |
603 | struct mem_cgroup *mem_cont, |
604 | unsigned long *vm_flags) | |
1da177e4 LT |
605 | { |
606 | unsigned int mapcount; | |
607 | struct anon_vma *anon_vma; | |
5beb4930 | 608 | struct anon_vma_chain *avc; |
1da177e4 LT |
609 | int referenced = 0; |
610 | ||
611 | anon_vma = page_lock_anon_vma(page); | |
612 | if (!anon_vma) | |
613 | return referenced; | |
614 | ||
615 | mapcount = page_mapcount(page); | |
5beb4930 RR |
616 | list_for_each_entry(avc, &anon_vma->head, same_anon_vma) { |
617 | struct vm_area_struct *vma = avc->vma; | |
1cb1729b HD |
618 | unsigned long address = vma_address(page, vma); |
619 | if (address == -EFAULT) | |
620 | continue; | |
bed7161a BS |
621 | /* |
622 | * If we are reclaiming on behalf of a cgroup, skip | |
623 | * counting on behalf of references from different | |
624 | * cgroups | |
625 | */ | |
bd845e38 | 626 | if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont)) |
bed7161a | 627 | continue; |
1cb1729b | 628 | referenced += page_referenced_one(page, vma, address, |
6fe6b7e3 | 629 | &mapcount, vm_flags); |
1da177e4 LT |
630 | if (!mapcount) |
631 | break; | |
632 | } | |
34bbd704 ON |
633 | |
634 | page_unlock_anon_vma(anon_vma); | |
1da177e4 LT |
635 | return referenced; |
636 | } | |
637 | ||
638 | /** | |
639 | * page_referenced_file - referenced check for object-based rmap | |
640 | * @page: the page we're checking references on. | |
43d8eac4 | 641 | * @mem_cont: target memory controller |
6fe6b7e3 | 642 | * @vm_flags: collect encountered vma->vm_flags who actually referenced the page |
1da177e4 LT |
643 | * |
644 | * For an object-based mapped page, find all the places it is mapped and | |
645 | * check/clear the referenced flag. This is done by following the page->mapping | |
646 | * pointer, then walking the chain of vmas it holds. It returns the number | |
647 | * of references it found. | |
648 | * | |
649 | * This function is only called from page_referenced for object-based pages. | |
650 | */ | |
bed7161a | 651 | static int page_referenced_file(struct page *page, |
6fe6b7e3 WF |
652 | struct mem_cgroup *mem_cont, |
653 | unsigned long *vm_flags) | |
1da177e4 LT |
654 | { |
655 | unsigned int mapcount; | |
656 | struct address_space *mapping = page->mapping; | |
657 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
658 | struct vm_area_struct *vma; | |
659 | struct prio_tree_iter iter; | |
660 | int referenced = 0; | |
661 | ||
662 | /* | |
663 | * The caller's checks on page->mapping and !PageAnon have made | |
664 | * sure that this is a file page: the check for page->mapping | |
665 | * excludes the case just before it gets set on an anon page. | |
666 | */ | |
667 | BUG_ON(PageAnon(page)); | |
668 | ||
669 | /* | |
670 | * The page lock not only makes sure that page->mapping cannot | |
671 | * suddenly be NULLified by truncation, it makes sure that the | |
672 | * structure at mapping cannot be freed and reused yet, | |
3d48ae45 | 673 | * so we can safely take mapping->i_mmap_mutex. |
1da177e4 LT |
674 | */ |
675 | BUG_ON(!PageLocked(page)); | |
676 | ||
3d48ae45 | 677 | mutex_lock(&mapping->i_mmap_mutex); |
1da177e4 LT |
678 | |
679 | /* | |
3d48ae45 | 680 | * i_mmap_mutex does not stabilize mapcount at all, but mapcount |
1da177e4 LT |
681 | * is more likely to be accurate if we note it after spinning. |
682 | */ | |
683 | mapcount = page_mapcount(page); | |
684 | ||
685 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { | |
1cb1729b HD |
686 | unsigned long address = vma_address(page, vma); |
687 | if (address == -EFAULT) | |
688 | continue; | |
bed7161a BS |
689 | /* |
690 | * If we are reclaiming on behalf of a cgroup, skip | |
691 | * counting on behalf of references from different | |
692 | * cgroups | |
693 | */ | |
bd845e38 | 694 | if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont)) |
bed7161a | 695 | continue; |
1cb1729b | 696 | referenced += page_referenced_one(page, vma, address, |
6fe6b7e3 | 697 | &mapcount, vm_flags); |
1da177e4 LT |
698 | if (!mapcount) |
699 | break; | |
700 | } | |
701 | ||
3d48ae45 | 702 | mutex_unlock(&mapping->i_mmap_mutex); |
1da177e4 LT |
703 | return referenced; |
704 | } | |
705 | ||
706 | /** | |
707 | * page_referenced - test if the page was referenced | |
708 | * @page: the page to test | |
709 | * @is_locked: caller holds lock on the page | |
43d8eac4 | 710 | * @mem_cont: target memory controller |
6fe6b7e3 | 711 | * @vm_flags: collect encountered vma->vm_flags who actually referenced the page |
1da177e4 LT |
712 | * |
713 | * Quick test_and_clear_referenced for all mappings to a page, | |
714 | * returns the number of ptes which referenced the page. | |
715 | */ | |
6fe6b7e3 WF |
716 | int page_referenced(struct page *page, |
717 | int is_locked, | |
718 | struct mem_cgroup *mem_cont, | |
719 | unsigned long *vm_flags) | |
1da177e4 LT |
720 | { |
721 | int referenced = 0; | |
5ad64688 | 722 | int we_locked = 0; |
1da177e4 | 723 | |
6fe6b7e3 | 724 | *vm_flags = 0; |
3ca7b3c5 | 725 | if (page_mapped(page) && page_rmapping(page)) { |
5ad64688 HD |
726 | if (!is_locked && (!PageAnon(page) || PageKsm(page))) { |
727 | we_locked = trylock_page(page); | |
728 | if (!we_locked) { | |
729 | referenced++; | |
730 | goto out; | |
731 | } | |
732 | } | |
733 | if (unlikely(PageKsm(page))) | |
734 | referenced += page_referenced_ksm(page, mem_cont, | |
735 | vm_flags); | |
736 | else if (PageAnon(page)) | |
6fe6b7e3 WF |
737 | referenced += page_referenced_anon(page, mem_cont, |
738 | vm_flags); | |
5ad64688 | 739 | else if (page->mapping) |
6fe6b7e3 WF |
740 | referenced += page_referenced_file(page, mem_cont, |
741 | vm_flags); | |
5ad64688 | 742 | if (we_locked) |
1da177e4 | 743 | unlock_page(page); |
1da177e4 | 744 | } |
5ad64688 | 745 | out: |
2d42552d | 746 | if (page_test_and_clear_young(page_to_pfn(page))) |
5b7baf05 CB |
747 | referenced++; |
748 | ||
1da177e4 LT |
749 | return referenced; |
750 | } | |
751 | ||
1cb1729b HD |
752 | static int page_mkclean_one(struct page *page, struct vm_area_struct *vma, |
753 | unsigned long address) | |
d08b3851 PZ |
754 | { |
755 | struct mm_struct *mm = vma->vm_mm; | |
c2fda5fe | 756 | pte_t *pte; |
d08b3851 PZ |
757 | spinlock_t *ptl; |
758 | int ret = 0; | |
759 | ||
479db0bf | 760 | pte = page_check_address(page, mm, address, &ptl, 1); |
d08b3851 PZ |
761 | if (!pte) |
762 | goto out; | |
763 | ||
c2fda5fe PZ |
764 | if (pte_dirty(*pte) || pte_write(*pte)) { |
765 | pte_t entry; | |
d08b3851 | 766 | |
c2fda5fe | 767 | flush_cache_page(vma, address, pte_pfn(*pte)); |
cddb8a5c | 768 | entry = ptep_clear_flush_notify(vma, address, pte); |
c2fda5fe PZ |
769 | entry = pte_wrprotect(entry); |
770 | entry = pte_mkclean(entry); | |
d6e88e67 | 771 | set_pte_at(mm, address, pte, entry); |
c2fda5fe PZ |
772 | ret = 1; |
773 | } | |
d08b3851 | 774 | |
d08b3851 PZ |
775 | pte_unmap_unlock(pte, ptl); |
776 | out: | |
777 | return ret; | |
778 | } | |
779 | ||
780 | static int page_mkclean_file(struct address_space *mapping, struct page *page) | |
781 | { | |
782 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
783 | struct vm_area_struct *vma; | |
784 | struct prio_tree_iter iter; | |
785 | int ret = 0; | |
786 | ||
787 | BUG_ON(PageAnon(page)); | |
788 | ||
3d48ae45 | 789 | mutex_lock(&mapping->i_mmap_mutex); |
d08b3851 | 790 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { |
1cb1729b HD |
791 | if (vma->vm_flags & VM_SHARED) { |
792 | unsigned long address = vma_address(page, vma); | |
793 | if (address == -EFAULT) | |
794 | continue; | |
795 | ret += page_mkclean_one(page, vma, address); | |
796 | } | |
d08b3851 | 797 | } |
3d48ae45 | 798 | mutex_unlock(&mapping->i_mmap_mutex); |
d08b3851 PZ |
799 | return ret; |
800 | } | |
801 | ||
802 | int page_mkclean(struct page *page) | |
803 | { | |
804 | int ret = 0; | |
805 | ||
806 | BUG_ON(!PageLocked(page)); | |
807 | ||
808 | if (page_mapped(page)) { | |
809 | struct address_space *mapping = page_mapping(page); | |
ce7e9fae | 810 | if (mapping) { |
d08b3851 | 811 | ret = page_mkclean_file(mapping, page); |
2d42552d | 812 | if (page_test_and_clear_dirty(page_to_pfn(page), 1)) |
ce7e9fae | 813 | ret = 1; |
6c210482 | 814 | } |
d08b3851 PZ |
815 | } |
816 | ||
817 | return ret; | |
818 | } | |
60b59bea | 819 | EXPORT_SYMBOL_GPL(page_mkclean); |
d08b3851 | 820 | |
c44b6743 RR |
821 | /** |
822 | * page_move_anon_rmap - move a page to our anon_vma | |
823 | * @page: the page to move to our anon_vma | |
824 | * @vma: the vma the page belongs to | |
825 | * @address: the user virtual address mapped | |
826 | * | |
827 | * When a page belongs exclusively to one process after a COW event, | |
828 | * that page can be moved into the anon_vma that belongs to just that | |
829 | * process, so the rmap code will not search the parent or sibling | |
830 | * processes. | |
831 | */ | |
832 | void page_move_anon_rmap(struct page *page, | |
833 | struct vm_area_struct *vma, unsigned long address) | |
834 | { | |
835 | struct anon_vma *anon_vma = vma->anon_vma; | |
836 | ||
837 | VM_BUG_ON(!PageLocked(page)); | |
838 | VM_BUG_ON(!anon_vma); | |
839 | VM_BUG_ON(page->index != linear_page_index(vma, address)); | |
840 | ||
841 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; | |
842 | page->mapping = (struct address_space *) anon_vma; | |
843 | } | |
844 | ||
9617d95e | 845 | /** |
4e1c1975 AK |
846 | * __page_set_anon_rmap - set up new anonymous rmap |
847 | * @page: Page to add to rmap | |
848 | * @vma: VM area to add page to. | |
849 | * @address: User virtual address of the mapping | |
e8a03feb | 850 | * @exclusive: the page is exclusively owned by the current process |
9617d95e NP |
851 | */ |
852 | static void __page_set_anon_rmap(struct page *page, | |
e8a03feb | 853 | struct vm_area_struct *vma, unsigned long address, int exclusive) |
9617d95e | 854 | { |
e8a03feb | 855 | struct anon_vma *anon_vma = vma->anon_vma; |
ea90002b | 856 | |
e8a03feb | 857 | BUG_ON(!anon_vma); |
ea90002b | 858 | |
4e1c1975 AK |
859 | if (PageAnon(page)) |
860 | return; | |
861 | ||
ea90002b | 862 | /* |
e8a03feb RR |
863 | * If the page isn't exclusively mapped into this vma, |
864 | * we must use the _oldest_ possible anon_vma for the | |
865 | * page mapping! | |
ea90002b | 866 | */ |
4e1c1975 | 867 | if (!exclusive) |
288468c3 | 868 | anon_vma = anon_vma->root; |
9617d95e | 869 | |
9617d95e NP |
870 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; |
871 | page->mapping = (struct address_space *) anon_vma; | |
9617d95e | 872 | page->index = linear_page_index(vma, address); |
9617d95e NP |
873 | } |
874 | ||
c97a9e10 | 875 | /** |
43d8eac4 | 876 | * __page_check_anon_rmap - sanity check anonymous rmap addition |
c97a9e10 NP |
877 | * @page: the page to add the mapping to |
878 | * @vma: the vm area in which the mapping is added | |
879 | * @address: the user virtual address mapped | |
880 | */ | |
881 | static void __page_check_anon_rmap(struct page *page, | |
882 | struct vm_area_struct *vma, unsigned long address) | |
883 | { | |
884 | #ifdef CONFIG_DEBUG_VM | |
885 | /* | |
886 | * The page's anon-rmap details (mapping and index) are guaranteed to | |
887 | * be set up correctly at this point. | |
888 | * | |
889 | * We have exclusion against page_add_anon_rmap because the caller | |
890 | * always holds the page locked, except if called from page_dup_rmap, | |
891 | * in which case the page is already known to be setup. | |
892 | * | |
893 | * We have exclusion against page_add_new_anon_rmap because those pages | |
894 | * are initially only visible via the pagetables, and the pte is locked | |
895 | * over the call to page_add_new_anon_rmap. | |
896 | */ | |
44ab57a0 | 897 | BUG_ON(page_anon_vma(page)->root != vma->anon_vma->root); |
c97a9e10 NP |
898 | BUG_ON(page->index != linear_page_index(vma, address)); |
899 | #endif | |
900 | } | |
901 | ||
1da177e4 LT |
902 | /** |
903 | * page_add_anon_rmap - add pte mapping to an anonymous page | |
904 | * @page: the page to add the mapping to | |
905 | * @vma: the vm area in which the mapping is added | |
906 | * @address: the user virtual address mapped | |
907 | * | |
5ad64688 | 908 | * The caller needs to hold the pte lock, and the page must be locked in |
80e14822 HD |
909 | * the anon_vma case: to serialize mapping,index checking after setting, |
910 | * and to ensure that PageAnon is not being upgraded racily to PageKsm | |
911 | * (but PageKsm is never downgraded to PageAnon). | |
1da177e4 LT |
912 | */ |
913 | void page_add_anon_rmap(struct page *page, | |
914 | struct vm_area_struct *vma, unsigned long address) | |
ad8c2ee8 RR |
915 | { |
916 | do_page_add_anon_rmap(page, vma, address, 0); | |
917 | } | |
918 | ||
919 | /* | |
920 | * Special version of the above for do_swap_page, which often runs | |
921 | * into pages that are exclusively owned by the current process. | |
922 | * Everybody else should continue to use page_add_anon_rmap above. | |
923 | */ | |
924 | void do_page_add_anon_rmap(struct page *page, | |
925 | struct vm_area_struct *vma, unsigned long address, int exclusive) | |
1da177e4 | 926 | { |
5ad64688 | 927 | int first = atomic_inc_and_test(&page->_mapcount); |
79134171 AA |
928 | if (first) { |
929 | if (!PageTransHuge(page)) | |
930 | __inc_zone_page_state(page, NR_ANON_PAGES); | |
931 | else | |
932 | __inc_zone_page_state(page, | |
933 | NR_ANON_TRANSPARENT_HUGEPAGES); | |
934 | } | |
5ad64688 HD |
935 | if (unlikely(PageKsm(page))) |
936 | return; | |
937 | ||
c97a9e10 NP |
938 | VM_BUG_ON(!PageLocked(page)); |
939 | VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); | |
5ad64688 | 940 | if (first) |
ad8c2ee8 | 941 | __page_set_anon_rmap(page, vma, address, exclusive); |
69029cd5 | 942 | else |
c97a9e10 | 943 | __page_check_anon_rmap(page, vma, address); |
1da177e4 LT |
944 | } |
945 | ||
43d8eac4 | 946 | /** |
9617d95e NP |
947 | * page_add_new_anon_rmap - add pte mapping to a new anonymous page |
948 | * @page: the page to add the mapping to | |
949 | * @vma: the vm area in which the mapping is added | |
950 | * @address: the user virtual address mapped | |
951 | * | |
952 | * Same as page_add_anon_rmap but must only be called on *new* pages. | |
953 | * This means the inc-and-test can be bypassed. | |
c97a9e10 | 954 | * Page does not have to be locked. |
9617d95e NP |
955 | */ |
956 | void page_add_new_anon_rmap(struct page *page, | |
957 | struct vm_area_struct *vma, unsigned long address) | |
958 | { | |
b5934c53 | 959 | VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); |
cbf84b7a HD |
960 | SetPageSwapBacked(page); |
961 | atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */ | |
79134171 AA |
962 | if (!PageTransHuge(page)) |
963 | __inc_zone_page_state(page, NR_ANON_PAGES); | |
964 | else | |
965 | __inc_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES); | |
e8a03feb | 966 | __page_set_anon_rmap(page, vma, address, 1); |
b5934c53 | 967 | if (page_evictable(page, vma)) |
cbf84b7a | 968 | lru_cache_add_lru(page, LRU_ACTIVE_ANON); |
b5934c53 HD |
969 | else |
970 | add_page_to_unevictable_list(page); | |
9617d95e NP |
971 | } |
972 | ||
1da177e4 LT |
973 | /** |
974 | * page_add_file_rmap - add pte mapping to a file page | |
975 | * @page: the page to add the mapping to | |
976 | * | |
b8072f09 | 977 | * The caller needs to hold the pte lock. |
1da177e4 LT |
978 | */ |
979 | void page_add_file_rmap(struct page *page) | |
980 | { | |
d69b042f | 981 | if (atomic_inc_and_test(&page->_mapcount)) { |
65ba55f5 | 982 | __inc_zone_page_state(page, NR_FILE_MAPPED); |
2a7106f2 | 983 | mem_cgroup_inc_page_stat(page, MEMCG_NR_FILE_MAPPED); |
d69b042f | 984 | } |
1da177e4 LT |
985 | } |
986 | ||
987 | /** | |
988 | * page_remove_rmap - take down pte mapping from a page | |
989 | * @page: page to remove mapping from | |
990 | * | |
b8072f09 | 991 | * The caller needs to hold the pte lock. |
1da177e4 | 992 | */ |
edc315fd | 993 | void page_remove_rmap(struct page *page) |
1da177e4 | 994 | { |
b904dcfe KM |
995 | /* page still mapped by someone else? */ |
996 | if (!atomic_add_negative(-1, &page->_mapcount)) | |
997 | return; | |
998 | ||
999 | /* | |
1000 | * Now that the last pte has gone, s390 must transfer dirty | |
1001 | * flag from storage key to struct page. We can usually skip | |
1002 | * this if the page is anon, so about to be freed; but perhaps | |
1003 | * not if it's in swapcache - there might be another pte slot | |
1004 | * containing the swap entry, but page not yet written to swap. | |
1005 | */ | |
2d42552d MS |
1006 | if ((!PageAnon(page) || PageSwapCache(page)) && |
1007 | page_test_and_clear_dirty(page_to_pfn(page), 1)) | |
b904dcfe | 1008 | set_page_dirty(page); |
0fe6e20b NH |
1009 | /* |
1010 | * Hugepages are not counted in NR_ANON_PAGES nor NR_FILE_MAPPED | |
1011 | * and not charged by memcg for now. | |
1012 | */ | |
1013 | if (unlikely(PageHuge(page))) | |
1014 | return; | |
b904dcfe KM |
1015 | if (PageAnon(page)) { |
1016 | mem_cgroup_uncharge_page(page); | |
79134171 AA |
1017 | if (!PageTransHuge(page)) |
1018 | __dec_zone_page_state(page, NR_ANON_PAGES); | |
1019 | else | |
1020 | __dec_zone_page_state(page, | |
1021 | NR_ANON_TRANSPARENT_HUGEPAGES); | |
b904dcfe KM |
1022 | } else { |
1023 | __dec_zone_page_state(page, NR_FILE_MAPPED); | |
2a7106f2 | 1024 | mem_cgroup_dec_page_stat(page, MEMCG_NR_FILE_MAPPED); |
b904dcfe | 1025 | } |
b904dcfe KM |
1026 | /* |
1027 | * It would be tidy to reset the PageAnon mapping here, | |
1028 | * but that might overwrite a racing page_add_anon_rmap | |
1029 | * which increments mapcount after us but sets mapping | |
1030 | * before us: so leave the reset to free_hot_cold_page, | |
1031 | * and remember that it's only reliable while mapped. | |
1032 | * Leaving it set also helps swapoff to reinstate ptes | |
1033 | * faster for those pages still in swapcache. | |
1034 | */ | |
1da177e4 LT |
1035 | } |
1036 | ||
1037 | /* | |
1038 | * Subfunctions of try_to_unmap: try_to_unmap_one called | |
1039 | * repeatedly from either try_to_unmap_anon or try_to_unmap_file. | |
1040 | */ | |
5ad64688 HD |
1041 | int try_to_unmap_one(struct page *page, struct vm_area_struct *vma, |
1042 | unsigned long address, enum ttu_flags flags) | |
1da177e4 LT |
1043 | { |
1044 | struct mm_struct *mm = vma->vm_mm; | |
1da177e4 LT |
1045 | pte_t *pte; |
1046 | pte_t pteval; | |
c0718806 | 1047 | spinlock_t *ptl; |
1da177e4 LT |
1048 | int ret = SWAP_AGAIN; |
1049 | ||
479db0bf | 1050 | pte = page_check_address(page, mm, address, &ptl, 0); |
c0718806 | 1051 | if (!pte) |
81b4082d | 1052 | goto out; |
1da177e4 LT |
1053 | |
1054 | /* | |
1055 | * If the page is mlock()d, we cannot swap it out. | |
1056 | * If it's recently referenced (perhaps page_referenced | |
1057 | * skipped over this mm) then we should reactivate it. | |
1058 | */ | |
14fa31b8 | 1059 | if (!(flags & TTU_IGNORE_MLOCK)) { |
caed0f48 KM |
1060 | if (vma->vm_flags & VM_LOCKED) |
1061 | goto out_mlock; | |
1062 | ||
af8e3354 | 1063 | if (TTU_ACTION(flags) == TTU_MUNLOCK) |
53f79acb | 1064 | goto out_unmap; |
14fa31b8 AK |
1065 | } |
1066 | if (!(flags & TTU_IGNORE_ACCESS)) { | |
b291f000 NP |
1067 | if (ptep_clear_flush_young_notify(vma, address, pte)) { |
1068 | ret = SWAP_FAIL; | |
1069 | goto out_unmap; | |
1070 | } | |
1071 | } | |
1da177e4 | 1072 | |
1da177e4 LT |
1073 | /* Nuke the page table entry. */ |
1074 | flush_cache_page(vma, address, page_to_pfn(page)); | |
cddb8a5c | 1075 | pteval = ptep_clear_flush_notify(vma, address, pte); |
1da177e4 LT |
1076 | |
1077 | /* Move the dirty bit to the physical page now the pte is gone. */ | |
1078 | if (pte_dirty(pteval)) | |
1079 | set_page_dirty(page); | |
1080 | ||
365e9c87 HD |
1081 | /* Update high watermark before we lower rss */ |
1082 | update_hiwater_rss(mm); | |
1083 | ||
888b9f7c AK |
1084 | if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) { |
1085 | if (PageAnon(page)) | |
d559db08 | 1086 | dec_mm_counter(mm, MM_ANONPAGES); |
888b9f7c | 1087 | else |
d559db08 | 1088 | dec_mm_counter(mm, MM_FILEPAGES); |
888b9f7c AK |
1089 | set_pte_at(mm, address, pte, |
1090 | swp_entry_to_pte(make_hwpoison_entry(page))); | |
1091 | } else if (PageAnon(page)) { | |
4c21e2f2 | 1092 | swp_entry_t entry = { .val = page_private(page) }; |
0697212a CL |
1093 | |
1094 | if (PageSwapCache(page)) { | |
1095 | /* | |
1096 | * Store the swap location in the pte. | |
1097 | * See handle_pte_fault() ... | |
1098 | */ | |
570a335b HD |
1099 | if (swap_duplicate(entry) < 0) { |
1100 | set_pte_at(mm, address, pte, pteval); | |
1101 | ret = SWAP_FAIL; | |
1102 | goto out_unmap; | |
1103 | } | |
0697212a CL |
1104 | if (list_empty(&mm->mmlist)) { |
1105 | spin_lock(&mmlist_lock); | |
1106 | if (list_empty(&mm->mmlist)) | |
1107 | list_add(&mm->mmlist, &init_mm.mmlist); | |
1108 | spin_unlock(&mmlist_lock); | |
1109 | } | |
d559db08 | 1110 | dec_mm_counter(mm, MM_ANONPAGES); |
b084d435 | 1111 | inc_mm_counter(mm, MM_SWAPENTS); |
64cdd548 | 1112 | } else if (PAGE_MIGRATION) { |
0697212a CL |
1113 | /* |
1114 | * Store the pfn of the page in a special migration | |
1115 | * pte. do_swap_page() will wait until the migration | |
1116 | * pte is removed and then restart fault handling. | |
1117 | */ | |
14fa31b8 | 1118 | BUG_ON(TTU_ACTION(flags) != TTU_MIGRATION); |
0697212a | 1119 | entry = make_migration_entry(page, pte_write(pteval)); |
1da177e4 LT |
1120 | } |
1121 | set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); | |
1122 | BUG_ON(pte_file(*pte)); | |
14fa31b8 | 1123 | } else if (PAGE_MIGRATION && (TTU_ACTION(flags) == TTU_MIGRATION)) { |
04e62a29 CL |
1124 | /* Establish migration entry for a file page */ |
1125 | swp_entry_t entry; | |
1126 | entry = make_migration_entry(page, pte_write(pteval)); | |
1127 | set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); | |
1128 | } else | |
d559db08 | 1129 | dec_mm_counter(mm, MM_FILEPAGES); |
1da177e4 | 1130 | |
edc315fd | 1131 | page_remove_rmap(page); |
1da177e4 LT |
1132 | page_cache_release(page); |
1133 | ||
1134 | out_unmap: | |
c0718806 | 1135 | pte_unmap_unlock(pte, ptl); |
caed0f48 KM |
1136 | out: |
1137 | return ret; | |
53f79acb | 1138 | |
caed0f48 KM |
1139 | out_mlock: |
1140 | pte_unmap_unlock(pte, ptl); | |
1141 | ||
1142 | ||
1143 | /* | |
1144 | * We need mmap_sem locking, Otherwise VM_LOCKED check makes | |
1145 | * unstable result and race. Plus, We can't wait here because | |
2b575eb6 | 1146 | * we now hold anon_vma->mutex or mapping->i_mmap_mutex. |
caed0f48 KM |
1147 | * if trylock failed, the page remain in evictable lru and later |
1148 | * vmscan could retry to move the page to unevictable lru if the | |
1149 | * page is actually mlocked. | |
1150 | */ | |
1151 | if (down_read_trylock(&vma->vm_mm->mmap_sem)) { | |
1152 | if (vma->vm_flags & VM_LOCKED) { | |
1153 | mlock_vma_page(page); | |
1154 | ret = SWAP_MLOCK; | |
53f79acb | 1155 | } |
caed0f48 | 1156 | up_read(&vma->vm_mm->mmap_sem); |
53f79acb | 1157 | } |
1da177e4 LT |
1158 | return ret; |
1159 | } | |
1160 | ||
1161 | /* | |
1162 | * objrmap doesn't work for nonlinear VMAs because the assumption that | |
1163 | * offset-into-file correlates with offset-into-virtual-addresses does not hold. | |
1164 | * Consequently, given a particular page and its ->index, we cannot locate the | |
1165 | * ptes which are mapping that page without an exhaustive linear search. | |
1166 | * | |
1167 | * So what this code does is a mini "virtual scan" of each nonlinear VMA which | |
1168 | * maps the file to which the target page belongs. The ->vm_private_data field | |
1169 | * holds the current cursor into that scan. Successive searches will circulate | |
1170 | * around the vma's virtual address space. | |
1171 | * | |
1172 | * So as more replacement pressure is applied to the pages in a nonlinear VMA, | |
1173 | * more scanning pressure is placed against them as well. Eventually pages | |
1174 | * will become fully unmapped and are eligible for eviction. | |
1175 | * | |
1176 | * For very sparsely populated VMAs this is a little inefficient - chances are | |
1177 | * there there won't be many ptes located within the scan cluster. In this case | |
1178 | * maybe we could scan further - to the end of the pte page, perhaps. | |
b291f000 NP |
1179 | * |
1180 | * Mlocked pages: check VM_LOCKED under mmap_sem held for read, if we can | |
1181 | * acquire it without blocking. If vma locked, mlock the pages in the cluster, | |
1182 | * rather than unmapping them. If we encounter the "check_page" that vmscan is | |
1183 | * trying to unmap, return SWAP_MLOCK, else default SWAP_AGAIN. | |
1da177e4 LT |
1184 | */ |
1185 | #define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE) | |
1186 | #define CLUSTER_MASK (~(CLUSTER_SIZE - 1)) | |
1187 | ||
b291f000 NP |
1188 | static int try_to_unmap_cluster(unsigned long cursor, unsigned int *mapcount, |
1189 | struct vm_area_struct *vma, struct page *check_page) | |
1da177e4 LT |
1190 | { |
1191 | struct mm_struct *mm = vma->vm_mm; | |
1192 | pgd_t *pgd; | |
1193 | pud_t *pud; | |
1194 | pmd_t *pmd; | |
c0718806 | 1195 | pte_t *pte; |
1da177e4 | 1196 | pte_t pteval; |
c0718806 | 1197 | spinlock_t *ptl; |
1da177e4 LT |
1198 | struct page *page; |
1199 | unsigned long address; | |
1200 | unsigned long end; | |
b291f000 NP |
1201 | int ret = SWAP_AGAIN; |
1202 | int locked_vma = 0; | |
1da177e4 | 1203 | |
1da177e4 LT |
1204 | address = (vma->vm_start + cursor) & CLUSTER_MASK; |
1205 | end = address + CLUSTER_SIZE; | |
1206 | if (address < vma->vm_start) | |
1207 | address = vma->vm_start; | |
1208 | if (end > vma->vm_end) | |
1209 | end = vma->vm_end; | |
1210 | ||
1211 | pgd = pgd_offset(mm, address); | |
1212 | if (!pgd_present(*pgd)) | |
b291f000 | 1213 | return ret; |
1da177e4 LT |
1214 | |
1215 | pud = pud_offset(pgd, address); | |
1216 | if (!pud_present(*pud)) | |
b291f000 | 1217 | return ret; |
1da177e4 LT |
1218 | |
1219 | pmd = pmd_offset(pud, address); | |
1220 | if (!pmd_present(*pmd)) | |
b291f000 NP |
1221 | return ret; |
1222 | ||
1223 | /* | |
af8e3354 | 1224 | * If we can acquire the mmap_sem for read, and vma is VM_LOCKED, |
b291f000 NP |
1225 | * keep the sem while scanning the cluster for mlocking pages. |
1226 | */ | |
af8e3354 | 1227 | if (down_read_trylock(&vma->vm_mm->mmap_sem)) { |
b291f000 NP |
1228 | locked_vma = (vma->vm_flags & VM_LOCKED); |
1229 | if (!locked_vma) | |
1230 | up_read(&vma->vm_mm->mmap_sem); /* don't need it */ | |
1231 | } | |
c0718806 HD |
1232 | |
1233 | pte = pte_offset_map_lock(mm, pmd, address, &ptl); | |
1da177e4 | 1234 | |
365e9c87 HD |
1235 | /* Update high watermark before we lower rss */ |
1236 | update_hiwater_rss(mm); | |
1237 | ||
c0718806 | 1238 | for (; address < end; pte++, address += PAGE_SIZE) { |
1da177e4 LT |
1239 | if (!pte_present(*pte)) |
1240 | continue; | |
6aab341e LT |
1241 | page = vm_normal_page(vma, address, *pte); |
1242 | BUG_ON(!page || PageAnon(page)); | |
1da177e4 | 1243 | |
b291f000 NP |
1244 | if (locked_vma) { |
1245 | mlock_vma_page(page); /* no-op if already mlocked */ | |
1246 | if (page == check_page) | |
1247 | ret = SWAP_MLOCK; | |
1248 | continue; /* don't unmap */ | |
1249 | } | |
1250 | ||
cddb8a5c | 1251 | if (ptep_clear_flush_young_notify(vma, address, pte)) |
1da177e4 LT |
1252 | continue; |
1253 | ||
1254 | /* Nuke the page table entry. */ | |
eca35133 | 1255 | flush_cache_page(vma, address, pte_pfn(*pte)); |
cddb8a5c | 1256 | pteval = ptep_clear_flush_notify(vma, address, pte); |
1da177e4 LT |
1257 | |
1258 | /* If nonlinear, store the file page offset in the pte. */ | |
1259 | if (page->index != linear_page_index(vma, address)) | |
1260 | set_pte_at(mm, address, pte, pgoff_to_pte(page->index)); | |
1261 | ||
1262 | /* Move the dirty bit to the physical page now the pte is gone. */ | |
1263 | if (pte_dirty(pteval)) | |
1264 | set_page_dirty(page); | |
1265 | ||
edc315fd | 1266 | page_remove_rmap(page); |
1da177e4 | 1267 | page_cache_release(page); |
d559db08 | 1268 | dec_mm_counter(mm, MM_FILEPAGES); |
1da177e4 LT |
1269 | (*mapcount)--; |
1270 | } | |
c0718806 | 1271 | pte_unmap_unlock(pte - 1, ptl); |
b291f000 NP |
1272 | if (locked_vma) |
1273 | up_read(&vma->vm_mm->mmap_sem); | |
1274 | return ret; | |
1da177e4 LT |
1275 | } |
1276 | ||
71e3aac0 | 1277 | bool is_vma_temporary_stack(struct vm_area_struct *vma) |
a8bef8ff MG |
1278 | { |
1279 | int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP); | |
1280 | ||
1281 | if (!maybe_stack) | |
1282 | return false; | |
1283 | ||
1284 | if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) == | |
1285 | VM_STACK_INCOMPLETE_SETUP) | |
1286 | return true; | |
1287 | ||
1288 | return false; | |
1289 | } | |
1290 | ||
b291f000 NP |
1291 | /** |
1292 | * try_to_unmap_anon - unmap or unlock anonymous page using the object-based | |
1293 | * rmap method | |
1294 | * @page: the page to unmap/unlock | |
8051be5e | 1295 | * @flags: action and flags |
b291f000 NP |
1296 | * |
1297 | * Find all the mappings of a page using the mapping pointer and the vma chains | |
1298 | * contained in the anon_vma struct it points to. | |
1299 | * | |
1300 | * This function is only called from try_to_unmap/try_to_munlock for | |
1301 | * anonymous pages. | |
1302 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma | |
1303 | * where the page was found will be held for write. So, we won't recheck | |
1304 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be | |
1305 | * 'LOCKED. | |
1306 | */ | |
14fa31b8 | 1307 | static int try_to_unmap_anon(struct page *page, enum ttu_flags flags) |
1da177e4 LT |
1308 | { |
1309 | struct anon_vma *anon_vma; | |
5beb4930 | 1310 | struct anon_vma_chain *avc; |
1da177e4 | 1311 | int ret = SWAP_AGAIN; |
b291f000 | 1312 | |
1da177e4 LT |
1313 | anon_vma = page_lock_anon_vma(page); |
1314 | if (!anon_vma) | |
1315 | return ret; | |
1316 | ||
5beb4930 RR |
1317 | list_for_each_entry(avc, &anon_vma->head, same_anon_vma) { |
1318 | struct vm_area_struct *vma = avc->vma; | |
a8bef8ff MG |
1319 | unsigned long address; |
1320 | ||
1321 | /* | |
1322 | * During exec, a temporary VMA is setup and later moved. | |
1323 | * The VMA is moved under the anon_vma lock but not the | |
1324 | * page tables leading to a race where migration cannot | |
1325 | * find the migration ptes. Rather than increasing the | |
1326 | * locking requirements of exec(), migration skips | |
1327 | * temporary VMAs until after exec() completes. | |
1328 | */ | |
1329 | if (PAGE_MIGRATION && (flags & TTU_MIGRATION) && | |
1330 | is_vma_temporary_stack(vma)) | |
1331 | continue; | |
1332 | ||
1333 | address = vma_address(page, vma); | |
1cb1729b HD |
1334 | if (address == -EFAULT) |
1335 | continue; | |
1336 | ret = try_to_unmap_one(page, vma, address, flags); | |
53f79acb HD |
1337 | if (ret != SWAP_AGAIN || !page_mapped(page)) |
1338 | break; | |
1da177e4 | 1339 | } |
34bbd704 ON |
1340 | |
1341 | page_unlock_anon_vma(anon_vma); | |
1da177e4 LT |
1342 | return ret; |
1343 | } | |
1344 | ||
1345 | /** | |
b291f000 NP |
1346 | * try_to_unmap_file - unmap/unlock file page using the object-based rmap method |
1347 | * @page: the page to unmap/unlock | |
14fa31b8 | 1348 | * @flags: action and flags |
1da177e4 LT |
1349 | * |
1350 | * Find all the mappings of a page using the mapping pointer and the vma chains | |
1351 | * contained in the address_space struct it points to. | |
1352 | * | |
b291f000 NP |
1353 | * This function is only called from try_to_unmap/try_to_munlock for |
1354 | * object-based pages. | |
1355 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma | |
1356 | * where the page was found will be held for write. So, we won't recheck | |
1357 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be | |
1358 | * 'LOCKED. | |
1da177e4 | 1359 | */ |
14fa31b8 | 1360 | static int try_to_unmap_file(struct page *page, enum ttu_flags flags) |
1da177e4 LT |
1361 | { |
1362 | struct address_space *mapping = page->mapping; | |
1363 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
1364 | struct vm_area_struct *vma; | |
1365 | struct prio_tree_iter iter; | |
1366 | int ret = SWAP_AGAIN; | |
1367 | unsigned long cursor; | |
1368 | unsigned long max_nl_cursor = 0; | |
1369 | unsigned long max_nl_size = 0; | |
1370 | unsigned int mapcount; | |
1371 | ||
3d48ae45 | 1372 | mutex_lock(&mapping->i_mmap_mutex); |
1da177e4 | 1373 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { |
1cb1729b HD |
1374 | unsigned long address = vma_address(page, vma); |
1375 | if (address == -EFAULT) | |
1376 | continue; | |
1377 | ret = try_to_unmap_one(page, vma, address, flags); | |
53f79acb HD |
1378 | if (ret != SWAP_AGAIN || !page_mapped(page)) |
1379 | goto out; | |
1da177e4 LT |
1380 | } |
1381 | ||
1382 | if (list_empty(&mapping->i_mmap_nonlinear)) | |
1383 | goto out; | |
1384 | ||
53f79acb HD |
1385 | /* |
1386 | * We don't bother to try to find the munlocked page in nonlinears. | |
1387 | * It's costly. Instead, later, page reclaim logic may call | |
1388 | * try_to_unmap(TTU_MUNLOCK) and recover PG_mlocked lazily. | |
1389 | */ | |
1390 | if (TTU_ACTION(flags) == TTU_MUNLOCK) | |
1391 | goto out; | |
1392 | ||
1da177e4 LT |
1393 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, |
1394 | shared.vm_set.list) { | |
1da177e4 LT |
1395 | cursor = (unsigned long) vma->vm_private_data; |
1396 | if (cursor > max_nl_cursor) | |
1397 | max_nl_cursor = cursor; | |
1398 | cursor = vma->vm_end - vma->vm_start; | |
1399 | if (cursor > max_nl_size) | |
1400 | max_nl_size = cursor; | |
1401 | } | |
1402 | ||
b291f000 | 1403 | if (max_nl_size == 0) { /* all nonlinears locked or reserved ? */ |
1da177e4 LT |
1404 | ret = SWAP_FAIL; |
1405 | goto out; | |
1406 | } | |
1407 | ||
1408 | /* | |
1409 | * We don't try to search for this page in the nonlinear vmas, | |
1410 | * and page_referenced wouldn't have found it anyway. Instead | |
1411 | * just walk the nonlinear vmas trying to age and unmap some. | |
1412 | * The mapcount of the page we came in with is irrelevant, | |
1413 | * but even so use it as a guide to how hard we should try? | |
1414 | */ | |
1415 | mapcount = page_mapcount(page); | |
1416 | if (!mapcount) | |
1417 | goto out; | |
3d48ae45 | 1418 | cond_resched(); |
1da177e4 LT |
1419 | |
1420 | max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK; | |
1421 | if (max_nl_cursor == 0) | |
1422 | max_nl_cursor = CLUSTER_SIZE; | |
1423 | ||
1424 | do { | |
1425 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, | |
1426 | shared.vm_set.list) { | |
1da177e4 | 1427 | cursor = (unsigned long) vma->vm_private_data; |
839b9685 | 1428 | while ( cursor < max_nl_cursor && |
1da177e4 | 1429 | cursor < vma->vm_end - vma->vm_start) { |
53f79acb HD |
1430 | if (try_to_unmap_cluster(cursor, &mapcount, |
1431 | vma, page) == SWAP_MLOCK) | |
1432 | ret = SWAP_MLOCK; | |
1da177e4 LT |
1433 | cursor += CLUSTER_SIZE; |
1434 | vma->vm_private_data = (void *) cursor; | |
1435 | if ((int)mapcount <= 0) | |
1436 | goto out; | |
1437 | } | |
1438 | vma->vm_private_data = (void *) max_nl_cursor; | |
1439 | } | |
3d48ae45 | 1440 | cond_resched(); |
1da177e4 LT |
1441 | max_nl_cursor += CLUSTER_SIZE; |
1442 | } while (max_nl_cursor <= max_nl_size); | |
1443 | ||
1444 | /* | |
1445 | * Don't loop forever (perhaps all the remaining pages are | |
1446 | * in locked vmas). Reset cursor on all unreserved nonlinear | |
1447 | * vmas, now forgetting on which ones it had fallen behind. | |
1448 | */ | |
101d2be7 HD |
1449 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list) |
1450 | vma->vm_private_data = NULL; | |
1da177e4 | 1451 | out: |
3d48ae45 | 1452 | mutex_unlock(&mapping->i_mmap_mutex); |
1da177e4 LT |
1453 | return ret; |
1454 | } | |
1455 | ||
1456 | /** | |
1457 | * try_to_unmap - try to remove all page table mappings to a page | |
1458 | * @page: the page to get unmapped | |
14fa31b8 | 1459 | * @flags: action and flags |
1da177e4 LT |
1460 | * |
1461 | * Tries to remove all the page table entries which are mapping this | |
1462 | * page, used in the pageout path. Caller must hold the page lock. | |
1463 | * Return values are: | |
1464 | * | |
1465 | * SWAP_SUCCESS - we succeeded in removing all mappings | |
1466 | * SWAP_AGAIN - we missed a mapping, try again later | |
1467 | * SWAP_FAIL - the page is unswappable | |
b291f000 | 1468 | * SWAP_MLOCK - page is mlocked. |
1da177e4 | 1469 | */ |
14fa31b8 | 1470 | int try_to_unmap(struct page *page, enum ttu_flags flags) |
1da177e4 LT |
1471 | { |
1472 | int ret; | |
1473 | ||
1da177e4 | 1474 | BUG_ON(!PageLocked(page)); |
91600e9e | 1475 | VM_BUG_ON(!PageHuge(page) && PageTransHuge(page)); |
1da177e4 | 1476 | |
5ad64688 HD |
1477 | if (unlikely(PageKsm(page))) |
1478 | ret = try_to_unmap_ksm(page, flags); | |
1479 | else if (PageAnon(page)) | |
14fa31b8 | 1480 | ret = try_to_unmap_anon(page, flags); |
1da177e4 | 1481 | else |
14fa31b8 | 1482 | ret = try_to_unmap_file(page, flags); |
b291f000 | 1483 | if (ret != SWAP_MLOCK && !page_mapped(page)) |
1da177e4 LT |
1484 | ret = SWAP_SUCCESS; |
1485 | return ret; | |
1486 | } | |
81b4082d | 1487 | |
b291f000 NP |
1488 | /** |
1489 | * try_to_munlock - try to munlock a page | |
1490 | * @page: the page to be munlocked | |
1491 | * | |
1492 | * Called from munlock code. Checks all of the VMAs mapping the page | |
1493 | * to make sure nobody else has this page mlocked. The page will be | |
1494 | * returned with PG_mlocked cleared if no other vmas have it mlocked. | |
1495 | * | |
1496 | * Return values are: | |
1497 | * | |
53f79acb | 1498 | * SWAP_AGAIN - no vma is holding page mlocked, or, |
b291f000 | 1499 | * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem |
5ad64688 | 1500 | * SWAP_FAIL - page cannot be located at present |
b291f000 NP |
1501 | * SWAP_MLOCK - page is now mlocked. |
1502 | */ | |
1503 | int try_to_munlock(struct page *page) | |
1504 | { | |
1505 | VM_BUG_ON(!PageLocked(page) || PageLRU(page)); | |
1506 | ||
5ad64688 HD |
1507 | if (unlikely(PageKsm(page))) |
1508 | return try_to_unmap_ksm(page, TTU_MUNLOCK); | |
1509 | else if (PageAnon(page)) | |
14fa31b8 | 1510 | return try_to_unmap_anon(page, TTU_MUNLOCK); |
b291f000 | 1511 | else |
14fa31b8 | 1512 | return try_to_unmap_file(page, TTU_MUNLOCK); |
b291f000 | 1513 | } |
e9995ef9 | 1514 | |
01d8b20d | 1515 | void __put_anon_vma(struct anon_vma *anon_vma) |
76545066 | 1516 | { |
01d8b20d | 1517 | struct anon_vma *root = anon_vma->root; |
76545066 | 1518 | |
01d8b20d PZ |
1519 | if (root != anon_vma && atomic_dec_and_test(&root->refcount)) |
1520 | anon_vma_free(root); | |
76545066 | 1521 | |
01d8b20d | 1522 | anon_vma_free(anon_vma); |
76545066 | 1523 | } |
76545066 | 1524 | |
e9995ef9 HD |
1525 | #ifdef CONFIG_MIGRATION |
1526 | /* | |
1527 | * rmap_walk() and its helpers rmap_walk_anon() and rmap_walk_file(): | |
1528 | * Called by migrate.c to remove migration ptes, but might be used more later. | |
1529 | */ | |
1530 | static int rmap_walk_anon(struct page *page, int (*rmap_one)(struct page *, | |
1531 | struct vm_area_struct *, unsigned long, void *), void *arg) | |
1532 | { | |
1533 | struct anon_vma *anon_vma; | |
5beb4930 | 1534 | struct anon_vma_chain *avc; |
e9995ef9 HD |
1535 | int ret = SWAP_AGAIN; |
1536 | ||
1537 | /* | |
1538 | * Note: remove_migration_ptes() cannot use page_lock_anon_vma() | |
1539 | * because that depends on page_mapped(); but not all its usages | |
3f6c8272 MG |
1540 | * are holding mmap_sem. Users without mmap_sem are required to |
1541 | * take a reference count to prevent the anon_vma disappearing | |
e9995ef9 HD |
1542 | */ |
1543 | anon_vma = page_anon_vma(page); | |
1544 | if (!anon_vma) | |
1545 | return ret; | |
cba48b98 | 1546 | anon_vma_lock(anon_vma); |
5beb4930 RR |
1547 | list_for_each_entry(avc, &anon_vma->head, same_anon_vma) { |
1548 | struct vm_area_struct *vma = avc->vma; | |
e9995ef9 HD |
1549 | unsigned long address = vma_address(page, vma); |
1550 | if (address == -EFAULT) | |
1551 | continue; | |
1552 | ret = rmap_one(page, vma, address, arg); | |
1553 | if (ret != SWAP_AGAIN) | |
1554 | break; | |
1555 | } | |
cba48b98 | 1556 | anon_vma_unlock(anon_vma); |
e9995ef9 HD |
1557 | return ret; |
1558 | } | |
1559 | ||
1560 | static int rmap_walk_file(struct page *page, int (*rmap_one)(struct page *, | |
1561 | struct vm_area_struct *, unsigned long, void *), void *arg) | |
1562 | { | |
1563 | struct address_space *mapping = page->mapping; | |
1564 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
1565 | struct vm_area_struct *vma; | |
1566 | struct prio_tree_iter iter; | |
1567 | int ret = SWAP_AGAIN; | |
1568 | ||
1569 | if (!mapping) | |
1570 | return ret; | |
3d48ae45 | 1571 | mutex_lock(&mapping->i_mmap_mutex); |
e9995ef9 HD |
1572 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { |
1573 | unsigned long address = vma_address(page, vma); | |
1574 | if (address == -EFAULT) | |
1575 | continue; | |
1576 | ret = rmap_one(page, vma, address, arg); | |
1577 | if (ret != SWAP_AGAIN) | |
1578 | break; | |
1579 | } | |
1580 | /* | |
1581 | * No nonlinear handling: being always shared, nonlinear vmas | |
1582 | * never contain migration ptes. Decide what to do about this | |
1583 | * limitation to linear when we need rmap_walk() on nonlinear. | |
1584 | */ | |
3d48ae45 | 1585 | mutex_unlock(&mapping->i_mmap_mutex); |
e9995ef9 HD |
1586 | return ret; |
1587 | } | |
1588 | ||
1589 | int rmap_walk(struct page *page, int (*rmap_one)(struct page *, | |
1590 | struct vm_area_struct *, unsigned long, void *), void *arg) | |
1591 | { | |
1592 | VM_BUG_ON(!PageLocked(page)); | |
1593 | ||
1594 | if (unlikely(PageKsm(page))) | |
1595 | return rmap_walk_ksm(page, rmap_one, arg); | |
1596 | else if (PageAnon(page)) | |
1597 | return rmap_walk_anon(page, rmap_one, arg); | |
1598 | else | |
1599 | return rmap_walk_file(page, rmap_one, arg); | |
1600 | } | |
1601 | #endif /* CONFIG_MIGRATION */ | |
0fe6e20b | 1602 | |
e3390f67 | 1603 | #ifdef CONFIG_HUGETLB_PAGE |
0fe6e20b NH |
1604 | /* |
1605 | * The following three functions are for anonymous (private mapped) hugepages. | |
1606 | * Unlike common anonymous pages, anonymous hugepages have no accounting code | |
1607 | * and no lru code, because we handle hugepages differently from common pages. | |
1608 | */ | |
1609 | static void __hugepage_set_anon_rmap(struct page *page, | |
1610 | struct vm_area_struct *vma, unsigned long address, int exclusive) | |
1611 | { | |
1612 | struct anon_vma *anon_vma = vma->anon_vma; | |
433abed6 | 1613 | |
0fe6e20b | 1614 | BUG_ON(!anon_vma); |
433abed6 NH |
1615 | |
1616 | if (PageAnon(page)) | |
1617 | return; | |
1618 | if (!exclusive) | |
1619 | anon_vma = anon_vma->root; | |
1620 | ||
0fe6e20b NH |
1621 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; |
1622 | page->mapping = (struct address_space *) anon_vma; | |
1623 | page->index = linear_page_index(vma, address); | |
1624 | } | |
1625 | ||
1626 | void hugepage_add_anon_rmap(struct page *page, | |
1627 | struct vm_area_struct *vma, unsigned long address) | |
1628 | { | |
1629 | struct anon_vma *anon_vma = vma->anon_vma; | |
1630 | int first; | |
a850ea30 NH |
1631 | |
1632 | BUG_ON(!PageLocked(page)); | |
0fe6e20b NH |
1633 | BUG_ON(!anon_vma); |
1634 | BUG_ON(address < vma->vm_start || address >= vma->vm_end); | |
1635 | first = atomic_inc_and_test(&page->_mapcount); | |
1636 | if (first) | |
1637 | __hugepage_set_anon_rmap(page, vma, address, 0); | |
1638 | } | |
1639 | ||
1640 | void hugepage_add_new_anon_rmap(struct page *page, | |
1641 | struct vm_area_struct *vma, unsigned long address) | |
1642 | { | |
1643 | BUG_ON(address < vma->vm_start || address >= vma->vm_end); | |
1644 | atomic_set(&page->_mapcount, 0); | |
1645 | __hugepage_set_anon_rmap(page, vma, address, 1); | |
1646 | } | |
e3390f67 | 1647 | #endif /* CONFIG_HUGETLB_PAGE */ |