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 | * mm->mmap_sem |
25 | * page->flags PG_locked (lock_page) | |
3d48ae45 | 26 | * mapping->i_mmap_mutex |
5a505085 | 27 | * anon_vma->rwsem |
82591e6e NP |
28 | * mm->page_table_lock or pte_lock |
29 | * zone->lru_lock (in mark_page_accessed, isolate_lru_page) | |
30 | * swap_lock (in swap_duplicate, swap_info_get) | |
31 | * mmlist_lock (in mmput, drain_mmlist and others) | |
32 | * mapping->private_lock (in __set_page_dirty_buffers) | |
250df6ed | 33 | * inode->i_lock (in set_page_dirty's __mark_inode_dirty) |
f758eeab | 34 | * bdi.wb->list_lock (in set_page_dirty's __mark_inode_dirty) |
82591e6e NP |
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, | |
f758eeab | 38 | * within bdi.wb->list_lock in __sync_single_inode) |
6a46079c | 39 | * |
5a505085 | 40 | * anon_vma->rwsem,mapping->i_mutex (memory_failure, collect_procs_anon) |
9b679320 | 41 | * ->tasklist_lock |
6a46079c | 42 | * pte map lock |
1da177e4 LT |
43 | */ |
44 | ||
45 | #include <linux/mm.h> | |
46 | #include <linux/pagemap.h> | |
47 | #include <linux/swap.h> | |
48 | #include <linux/swapops.h> | |
49 | #include <linux/slab.h> | |
50 | #include <linux/init.h> | |
5ad64688 | 51 | #include <linux/ksm.h> |
1da177e4 LT |
52 | #include <linux/rmap.h> |
53 | #include <linux/rcupdate.h> | |
b95f1b31 | 54 | #include <linux/export.h> |
8a9f3ccd | 55 | #include <linux/memcontrol.h> |
cddb8a5c | 56 | #include <linux/mmu_notifier.h> |
64cdd548 | 57 | #include <linux/migrate.h> |
0fe6e20b | 58 | #include <linux/hugetlb.h> |
ef5d437f | 59 | #include <linux/backing-dev.h> |
1da177e4 LT |
60 | |
61 | #include <asm/tlbflush.h> | |
62 | ||
b291f000 NP |
63 | #include "internal.h" |
64 | ||
6fa3eb70 S |
65 | #define MUTEX_RETRY_COUNT (65536) |
66 | #define MUTEX_RETRY_RESCHED (1024) | |
67 | ||
68 | ||
69 | ||
fdd2e5f8 | 70 | static struct kmem_cache *anon_vma_cachep; |
5beb4930 | 71 | static struct kmem_cache *anon_vma_chain_cachep; |
fdd2e5f8 AB |
72 | |
73 | static inline struct anon_vma *anon_vma_alloc(void) | |
74 | { | |
01d8b20d PZ |
75 | struct anon_vma *anon_vma; |
76 | ||
77 | anon_vma = kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL); | |
78 | if (anon_vma) { | |
79 | atomic_set(&anon_vma->refcount, 1); | |
80 | /* | |
81 | * Initialise the anon_vma root to point to itself. If called | |
82 | * from fork, the root will be reset to the parents anon_vma. | |
83 | */ | |
84 | anon_vma->root = anon_vma; | |
85 | } | |
86 | ||
87 | return anon_vma; | |
fdd2e5f8 AB |
88 | } |
89 | ||
01d8b20d | 90 | static inline void anon_vma_free(struct anon_vma *anon_vma) |
fdd2e5f8 | 91 | { |
01d8b20d | 92 | VM_BUG_ON(atomic_read(&anon_vma->refcount)); |
88c22088 PZ |
93 | |
94 | /* | |
4fc3f1d6 | 95 | * Synchronize against page_lock_anon_vma_read() such that |
88c22088 PZ |
96 | * we can safely hold the lock without the anon_vma getting |
97 | * freed. | |
98 | * | |
99 | * Relies on the full mb implied by the atomic_dec_and_test() from | |
100 | * put_anon_vma() against the acquire barrier implied by | |
4fc3f1d6 | 101 | * down_read_trylock() from page_lock_anon_vma_read(). This orders: |
88c22088 | 102 | * |
4fc3f1d6 IM |
103 | * page_lock_anon_vma_read() VS put_anon_vma() |
104 | * down_read_trylock() atomic_dec_and_test() | |
88c22088 | 105 | * LOCK MB |
4fc3f1d6 | 106 | * atomic_read() rwsem_is_locked() |
88c22088 PZ |
107 | * |
108 | * LOCK should suffice since the actual taking of the lock must | |
109 | * happen _before_ what follows. | |
110 | */ | |
ef151648 | 111 | might_sleep(); |
6fa3eb70 | 112 | if (anon_vma->root && rwsem_is_locked(&anon_vma->root->rwsem)) { |
4fc3f1d6 | 113 | anon_vma_lock_write(anon_vma); |
08b52706 | 114 | anon_vma_unlock_write(anon_vma); |
88c22088 PZ |
115 | } |
116 | ||
fdd2e5f8 AB |
117 | kmem_cache_free(anon_vma_cachep, anon_vma); |
118 | } | |
1da177e4 | 119 | |
dd34739c | 120 | static inline struct anon_vma_chain *anon_vma_chain_alloc(gfp_t gfp) |
5beb4930 | 121 | { |
dd34739c | 122 | return kmem_cache_alloc(anon_vma_chain_cachep, gfp); |
5beb4930 RR |
123 | } |
124 | ||
e574b5fd | 125 | static void anon_vma_chain_free(struct anon_vma_chain *anon_vma_chain) |
5beb4930 RR |
126 | { |
127 | kmem_cache_free(anon_vma_chain_cachep, anon_vma_chain); | |
128 | } | |
129 | ||
6583a843 KC |
130 | static void anon_vma_chain_link(struct vm_area_struct *vma, |
131 | struct anon_vma_chain *avc, | |
132 | struct anon_vma *anon_vma) | |
133 | { | |
134 | avc->vma = vma; | |
135 | avc->anon_vma = anon_vma; | |
136 | list_add(&avc->same_vma, &vma->anon_vma_chain); | |
bf181b9f | 137 | anon_vma_interval_tree_insert(avc, &anon_vma->rb_root); |
6583a843 KC |
138 | } |
139 | ||
d9d332e0 LT |
140 | /** |
141 | * anon_vma_prepare - attach an anon_vma to a memory region | |
142 | * @vma: the memory region in question | |
143 | * | |
144 | * This makes sure the memory mapping described by 'vma' has | |
145 | * an 'anon_vma' attached to it, so that we can associate the | |
146 | * anonymous pages mapped into it with that anon_vma. | |
147 | * | |
148 | * The common case will be that we already have one, but if | |
23a0790a | 149 | * not we either need to find an adjacent mapping that we |
d9d332e0 LT |
150 | * can re-use the anon_vma from (very common when the only |
151 | * reason for splitting a vma has been mprotect()), or we | |
152 | * allocate a new one. | |
153 | * | |
154 | * Anon-vma allocations are very subtle, because we may have | |
4fc3f1d6 | 155 | * optimistically looked up an anon_vma in page_lock_anon_vma_read() |
d9d332e0 LT |
156 | * and that may actually touch the spinlock even in the newly |
157 | * allocated vma (it depends on RCU to make sure that the | |
158 | * anon_vma isn't actually destroyed). | |
159 | * | |
160 | * As a result, we need to do proper anon_vma locking even | |
161 | * for the new allocation. At the same time, we do not want | |
162 | * to do any locking for the common case of already having | |
163 | * an anon_vma. | |
164 | * | |
165 | * This must be called with the mmap_sem held for reading. | |
166 | */ | |
1da177e4 LT |
167 | int anon_vma_prepare(struct vm_area_struct *vma) |
168 | { | |
169 | struct anon_vma *anon_vma = vma->anon_vma; | |
5beb4930 | 170 | struct anon_vma_chain *avc; |
1da177e4 LT |
171 | |
172 | might_sleep(); | |
173 | if (unlikely(!anon_vma)) { | |
174 | struct mm_struct *mm = vma->vm_mm; | |
d9d332e0 | 175 | struct anon_vma *allocated; |
1da177e4 | 176 | |
dd34739c | 177 | avc = anon_vma_chain_alloc(GFP_KERNEL); |
5beb4930 RR |
178 | if (!avc) |
179 | goto out_enomem; | |
180 | ||
1da177e4 | 181 | anon_vma = find_mergeable_anon_vma(vma); |
d9d332e0 LT |
182 | allocated = NULL; |
183 | if (!anon_vma) { | |
1da177e4 LT |
184 | anon_vma = anon_vma_alloc(); |
185 | if (unlikely(!anon_vma)) | |
5beb4930 | 186 | goto out_enomem_free_avc; |
1da177e4 | 187 | allocated = anon_vma; |
1da177e4 LT |
188 | } |
189 | ||
4fc3f1d6 | 190 | anon_vma_lock_write(anon_vma); |
1da177e4 LT |
191 | /* page_table_lock to protect against threads */ |
192 | spin_lock(&mm->page_table_lock); | |
193 | if (likely(!vma->anon_vma)) { | |
194 | vma->anon_vma = anon_vma; | |
6583a843 | 195 | anon_vma_chain_link(vma, avc, anon_vma); |
1da177e4 | 196 | allocated = NULL; |
31f2b0eb | 197 | avc = NULL; |
1da177e4 LT |
198 | } |
199 | spin_unlock(&mm->page_table_lock); | |
08b52706 | 200 | anon_vma_unlock_write(anon_vma); |
31f2b0eb ON |
201 | |
202 | if (unlikely(allocated)) | |
01d8b20d | 203 | put_anon_vma(allocated); |
31f2b0eb | 204 | if (unlikely(avc)) |
5beb4930 | 205 | anon_vma_chain_free(avc); |
1da177e4 LT |
206 | } |
207 | return 0; | |
5beb4930 RR |
208 | |
209 | out_enomem_free_avc: | |
210 | anon_vma_chain_free(avc); | |
211 | out_enomem: | |
212 | return -ENOMEM; | |
1da177e4 LT |
213 | } |
214 | ||
bb4aa396 LT |
215 | /* |
216 | * This is a useful helper function for locking the anon_vma root as | |
217 | * we traverse the vma->anon_vma_chain, looping over anon_vma's that | |
218 | * have the same vma. | |
219 | * | |
220 | * Such anon_vma's should have the same root, so you'd expect to see | |
221 | * just a single mutex_lock for the whole traversal. | |
222 | */ | |
223 | static inline struct anon_vma *lock_anon_vma_root(struct anon_vma *root, struct anon_vma *anon_vma) | |
224 | { | |
225 | struct anon_vma *new_root = anon_vma->root; | |
226 | if (new_root != root) { | |
227 | if (WARN_ON_ONCE(root)) | |
5a505085 | 228 | up_write(&root->rwsem); |
bb4aa396 | 229 | root = new_root; |
5a505085 | 230 | down_write(&root->rwsem); |
bb4aa396 LT |
231 | } |
232 | return root; | |
233 | } | |
234 | ||
235 | static inline void unlock_anon_vma_root(struct anon_vma *root) | |
236 | { | |
237 | if (root) | |
5a505085 | 238 | up_write(&root->rwsem); |
bb4aa396 LT |
239 | } |
240 | ||
5beb4930 RR |
241 | /* |
242 | * Attach the anon_vmas from src to dst. | |
243 | * Returns 0 on success, -ENOMEM on failure. | |
244 | */ | |
245 | int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src) | |
1da177e4 | 246 | { |
5beb4930 | 247 | struct anon_vma_chain *avc, *pavc; |
bb4aa396 | 248 | struct anon_vma *root = NULL; |
5beb4930 | 249 | |
646d87b4 | 250 | list_for_each_entry_reverse(pavc, &src->anon_vma_chain, same_vma) { |
bb4aa396 LT |
251 | struct anon_vma *anon_vma; |
252 | ||
dd34739c LT |
253 | avc = anon_vma_chain_alloc(GFP_NOWAIT | __GFP_NOWARN); |
254 | if (unlikely(!avc)) { | |
255 | unlock_anon_vma_root(root); | |
256 | root = NULL; | |
257 | avc = anon_vma_chain_alloc(GFP_KERNEL); | |
258 | if (!avc) | |
259 | goto enomem_failure; | |
260 | } | |
bb4aa396 LT |
261 | anon_vma = pavc->anon_vma; |
262 | root = lock_anon_vma_root(root, anon_vma); | |
263 | anon_vma_chain_link(dst, avc, anon_vma); | |
5beb4930 | 264 | } |
bb4aa396 | 265 | unlock_anon_vma_root(root); |
5beb4930 | 266 | return 0; |
1da177e4 | 267 | |
5beb4930 RR |
268 | enomem_failure: |
269 | unlink_anon_vmas(dst); | |
270 | return -ENOMEM; | |
1da177e4 LT |
271 | } |
272 | ||
5beb4930 RR |
273 | /* |
274 | * Attach vma to its own anon_vma, as well as to the anon_vmas that | |
275 | * the corresponding VMA in the parent process is attached to. | |
276 | * Returns 0 on success, non-zero on failure. | |
277 | */ | |
278 | int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma) | |
1da177e4 | 279 | { |
5beb4930 RR |
280 | struct anon_vma_chain *avc; |
281 | struct anon_vma *anon_vma; | |
1da177e4 | 282 | |
5beb4930 RR |
283 | /* Don't bother if the parent process has no anon_vma here. */ |
284 | if (!pvma->anon_vma) | |
285 | return 0; | |
286 | ||
287 | /* | |
288 | * First, attach the new VMA to the parent VMA's anon_vmas, | |
289 | * so rmap can find non-COWed pages in child processes. | |
290 | */ | |
291 | if (anon_vma_clone(vma, pvma)) | |
292 | return -ENOMEM; | |
293 | ||
294 | /* Then add our own anon_vma. */ | |
295 | anon_vma = anon_vma_alloc(); | |
296 | if (!anon_vma) | |
297 | goto out_error; | |
dd34739c | 298 | avc = anon_vma_chain_alloc(GFP_KERNEL); |
5beb4930 RR |
299 | if (!avc) |
300 | goto out_error_free_anon_vma; | |
5c341ee1 RR |
301 | |
302 | /* | |
303 | * The root anon_vma's spinlock is the lock actually used when we | |
304 | * lock any of the anon_vmas in this anon_vma tree. | |
305 | */ | |
306 | anon_vma->root = pvma->anon_vma->root; | |
76545066 | 307 | /* |
01d8b20d PZ |
308 | * With refcounts, an anon_vma can stay around longer than the |
309 | * process it belongs to. The root anon_vma needs to be pinned until | |
310 | * this anon_vma is freed, because the lock lives in the root. | |
76545066 RR |
311 | */ |
312 | get_anon_vma(anon_vma->root); | |
5beb4930 RR |
313 | /* Mark this anon_vma as the one where our new (COWed) pages go. */ |
314 | vma->anon_vma = anon_vma; | |
4fc3f1d6 | 315 | anon_vma_lock_write(anon_vma); |
5c341ee1 | 316 | anon_vma_chain_link(vma, avc, anon_vma); |
08b52706 | 317 | anon_vma_unlock_write(anon_vma); |
5beb4930 RR |
318 | |
319 | return 0; | |
320 | ||
321 | out_error_free_anon_vma: | |
01d8b20d | 322 | put_anon_vma(anon_vma); |
5beb4930 | 323 | out_error: |
4946d54c | 324 | unlink_anon_vmas(vma); |
5beb4930 | 325 | return -ENOMEM; |
1da177e4 LT |
326 | } |
327 | ||
5beb4930 RR |
328 | void unlink_anon_vmas(struct vm_area_struct *vma) |
329 | { | |
330 | struct anon_vma_chain *avc, *next; | |
eee2acba | 331 | struct anon_vma *root = NULL; |
5beb4930 | 332 | |
5c341ee1 RR |
333 | /* |
334 | * Unlink each anon_vma chained to the VMA. This list is ordered | |
335 | * from newest to oldest, ensuring the root anon_vma gets freed last. | |
336 | */ | |
5beb4930 | 337 | list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) { |
eee2acba PZ |
338 | struct anon_vma *anon_vma = avc->anon_vma; |
339 | ||
340 | root = lock_anon_vma_root(root, anon_vma); | |
bf181b9f | 341 | anon_vma_interval_tree_remove(avc, &anon_vma->rb_root); |
eee2acba PZ |
342 | |
343 | /* | |
344 | * Leave empty anon_vmas on the list - we'll need | |
345 | * to free them outside the lock. | |
346 | */ | |
bf181b9f | 347 | if (RB_EMPTY_ROOT(&anon_vma->rb_root)) |
eee2acba PZ |
348 | continue; |
349 | ||
350 | list_del(&avc->same_vma); | |
351 | anon_vma_chain_free(avc); | |
352 | } | |
353 | unlock_anon_vma_root(root); | |
354 | ||
355 | /* | |
356 | * Iterate the list once more, it now only contains empty and unlinked | |
357 | * anon_vmas, destroy them. Could not do before due to __put_anon_vma() | |
5a505085 | 358 | * needing to write-acquire the anon_vma->root->rwsem. |
eee2acba PZ |
359 | */ |
360 | list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) { | |
361 | struct anon_vma *anon_vma = avc->anon_vma; | |
362 | ||
363 | put_anon_vma(anon_vma); | |
364 | ||
5beb4930 RR |
365 | list_del(&avc->same_vma); |
366 | anon_vma_chain_free(avc); | |
367 | } | |
368 | } | |
369 | ||
51cc5068 | 370 | static void anon_vma_ctor(void *data) |
1da177e4 | 371 | { |
a35afb83 | 372 | struct anon_vma *anon_vma = data; |
1da177e4 | 373 | |
5a505085 | 374 | init_rwsem(&anon_vma->rwsem); |
83813267 | 375 | atomic_set(&anon_vma->refcount, 0); |
bf181b9f | 376 | anon_vma->rb_root = RB_ROOT; |
1da177e4 LT |
377 | } |
378 | ||
379 | void __init anon_vma_init(void) | |
380 | { | |
381 | anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma), | |
20c2df83 | 382 | 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor); |
5beb4930 | 383 | anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain, SLAB_PANIC); |
1da177e4 LT |
384 | } |
385 | ||
386 | /* | |
6111e4ca PZ |
387 | * Getting a lock on a stable anon_vma from a page off the LRU is tricky! |
388 | * | |
389 | * Since there is no serialization what so ever against page_remove_rmap() | |
390 | * the best this function can do is return a locked anon_vma that might | |
391 | * have been relevant to this page. | |
392 | * | |
393 | * The page might have been remapped to a different anon_vma or the anon_vma | |
394 | * returned may already be freed (and even reused). | |
395 | * | |
bc658c96 PZ |
396 | * In case it was remapped to a different anon_vma, the new anon_vma will be a |
397 | * child of the old anon_vma, and the anon_vma lifetime rules will therefore | |
398 | * ensure that any anon_vma obtained from the page will still be valid for as | |
399 | * long as we observe page_mapped() [ hence all those page_mapped() tests ]. | |
400 | * | |
6111e4ca PZ |
401 | * All users of this function must be very careful when walking the anon_vma |
402 | * chain and verify that the page in question is indeed mapped in it | |
403 | * [ something equivalent to page_mapped_in_vma() ]. | |
404 | * | |
405 | * Since anon_vma's slab is DESTROY_BY_RCU and we know from page_remove_rmap() | |
406 | * that the anon_vma pointer from page->mapping is valid if there is a | |
407 | * mapcount, we can dereference the anon_vma after observing those. | |
1da177e4 | 408 | */ |
746b18d4 | 409 | struct anon_vma *page_get_anon_vma(struct page *page) |
1da177e4 | 410 | { |
746b18d4 | 411 | struct anon_vma *anon_vma = NULL; |
1da177e4 LT |
412 | unsigned long anon_mapping; |
413 | ||
414 | rcu_read_lock(); | |
80e14822 | 415 | anon_mapping = (unsigned long) ACCESS_ONCE(page->mapping); |
3ca7b3c5 | 416 | if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON) |
1da177e4 LT |
417 | goto out; |
418 | if (!page_mapped(page)) | |
419 | goto out; | |
420 | ||
421 | anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON); | |
746b18d4 PZ |
422 | if (!atomic_inc_not_zero(&anon_vma->refcount)) { |
423 | anon_vma = NULL; | |
424 | goto out; | |
425 | } | |
f1819427 HD |
426 | |
427 | /* | |
428 | * If this page is still mapped, then its anon_vma cannot have been | |
746b18d4 PZ |
429 | * freed. But if it has been unmapped, we have no security against the |
430 | * anon_vma structure being freed and reused (for another anon_vma: | |
431 | * SLAB_DESTROY_BY_RCU guarantees that - so the atomic_inc_not_zero() | |
432 | * above cannot corrupt). | |
f1819427 | 433 | */ |
746b18d4 | 434 | if (!page_mapped(page)) { |
ef151648 | 435 | rcu_read_unlock(); |
746b18d4 | 436 | put_anon_vma(anon_vma); |
ef151648 | 437 | return NULL; |
746b18d4 | 438 | } |
1da177e4 LT |
439 | out: |
440 | rcu_read_unlock(); | |
746b18d4 PZ |
441 | |
442 | return anon_vma; | |
443 | } | |
444 | ||
88c22088 PZ |
445 | /* |
446 | * Similar to page_get_anon_vma() except it locks the anon_vma. | |
447 | * | |
448 | * Its a little more complex as it tries to keep the fast path to a single | |
449 | * atomic op -- the trylock. If we fail the trylock, we fall back to getting a | |
450 | * reference like with page_get_anon_vma() and then block on the mutex. | |
451 | */ | |
4fc3f1d6 | 452 | struct anon_vma *page_lock_anon_vma_read(struct page *page) |
746b18d4 | 453 | { |
88c22088 | 454 | struct anon_vma *anon_vma = NULL; |
eee0f252 | 455 | struct anon_vma *root_anon_vma; |
88c22088 | 456 | unsigned long anon_mapping; |
746b18d4 | 457 | |
88c22088 PZ |
458 | rcu_read_lock(); |
459 | anon_mapping = (unsigned long) ACCESS_ONCE(page->mapping); | |
460 | if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON) | |
461 | goto out; | |
462 | if (!page_mapped(page)) | |
463 | goto out; | |
464 | ||
465 | anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON); | |
eee0f252 | 466 | root_anon_vma = ACCESS_ONCE(anon_vma->root); |
4fc3f1d6 | 467 | if (down_read_trylock(&root_anon_vma->rwsem)) { |
88c22088 | 468 | /* |
eee0f252 HD |
469 | * If the page is still mapped, then this anon_vma is still |
470 | * its anon_vma, and holding the mutex ensures that it will | |
bc658c96 | 471 | * not go away, see anon_vma_free(). |
88c22088 | 472 | */ |
eee0f252 | 473 | if (!page_mapped(page)) { |
4fc3f1d6 | 474 | up_read(&root_anon_vma->rwsem); |
88c22088 PZ |
475 | anon_vma = NULL; |
476 | } | |
477 | goto out; | |
478 | } | |
746b18d4 | 479 | |
88c22088 PZ |
480 | /* trylock failed, we got to sleep */ |
481 | if (!atomic_inc_not_zero(&anon_vma->refcount)) { | |
482 | anon_vma = NULL; | |
483 | goto out; | |
484 | } | |
485 | ||
486 | if (!page_mapped(page)) { | |
ef151648 | 487 | rcu_read_unlock(); |
88c22088 | 488 | put_anon_vma(anon_vma); |
ef151648 | 489 | return NULL; |
88c22088 PZ |
490 | } |
491 | ||
492 | /* we pinned the anon_vma, its safe to sleep */ | |
493 | rcu_read_unlock(); | |
4fc3f1d6 | 494 | anon_vma_lock_read(anon_vma); |
88c22088 PZ |
495 | |
496 | if (atomic_dec_and_test(&anon_vma->refcount)) { | |
497 | /* | |
498 | * Oops, we held the last refcount, release the lock | |
499 | * and bail -- can't simply use put_anon_vma() because | |
4fc3f1d6 | 500 | * we'll deadlock on the anon_vma_lock_write() recursion. |
88c22088 | 501 | */ |
4fc3f1d6 | 502 | anon_vma_unlock_read(anon_vma); |
88c22088 PZ |
503 | __put_anon_vma(anon_vma); |
504 | anon_vma = NULL; | |
505 | } | |
506 | ||
507 | return anon_vma; | |
508 | ||
509 | out: | |
510 | rcu_read_unlock(); | |
746b18d4 | 511 | return anon_vma; |
34bbd704 ON |
512 | } |
513 | ||
4fc3f1d6 | 514 | void page_unlock_anon_vma_read(struct anon_vma *anon_vma) |
34bbd704 | 515 | { |
4fc3f1d6 | 516 | anon_vma_unlock_read(anon_vma); |
1da177e4 LT |
517 | } |
518 | ||
519 | /* | |
3ad33b24 | 520 | * At what user virtual address is page expected in @vma? |
1da177e4 | 521 | */ |
86c2ad19 ML |
522 | static inline unsigned long |
523 | __vma_address(struct page *page, struct vm_area_struct *vma) | |
1da177e4 LT |
524 | { |
525 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
1da177e4 | 526 | |
0fe6e20b NH |
527 | if (unlikely(is_vm_hugetlb_page(vma))) |
528 | pgoff = page->index << huge_page_order(page_hstate(page)); | |
86c2ad19 ML |
529 | |
530 | return vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); | |
531 | } | |
532 | ||
533 | inline unsigned long | |
534 | vma_address(struct page *page, struct vm_area_struct *vma) | |
535 | { | |
536 | unsigned long address = __vma_address(page, vma); | |
537 | ||
538 | /* page should be within @vma mapping range */ | |
539 | VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); | |
540 | ||
1da177e4 LT |
541 | return address; |
542 | } | |
543 | ||
544 | /* | |
bf89c8c8 | 545 | * At what user virtual address is page expected in vma? |
ab941e0f | 546 | * Caller should check the page is actually part of the vma. |
1da177e4 LT |
547 | */ |
548 | unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma) | |
549 | { | |
86c2ad19 | 550 | unsigned long address; |
21d0d443 | 551 | if (PageAnon(page)) { |
4829b906 HD |
552 | struct anon_vma *page__anon_vma = page_anon_vma(page); |
553 | /* | |
554 | * Note: swapoff's unuse_vma() is more efficient with this | |
555 | * check, and needs it to match anon_vma when KSM is active. | |
556 | */ | |
557 | if (!vma->anon_vma || !page__anon_vma || | |
558 | vma->anon_vma->root != page__anon_vma->root) | |
21d0d443 AA |
559 | return -EFAULT; |
560 | } else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) { | |
ee498ed7 HD |
561 | if (!vma->vm_file || |
562 | vma->vm_file->f_mapping != page->mapping) | |
1da177e4 LT |
563 | return -EFAULT; |
564 | } else | |
565 | return -EFAULT; | |
86c2ad19 ML |
566 | address = __vma_address(page, vma); |
567 | if (unlikely(address < vma->vm_start || address >= vma->vm_end)) | |
568 | return -EFAULT; | |
569 | return address; | |
1da177e4 LT |
570 | } |
571 | ||
6219049a BL |
572 | pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address) |
573 | { | |
574 | pgd_t *pgd; | |
575 | pud_t *pud; | |
576 | pmd_t *pmd = NULL; | |
577 | ||
578 | pgd = pgd_offset(mm, address); | |
579 | if (!pgd_present(*pgd)) | |
580 | goto out; | |
581 | ||
582 | pud = pud_offset(pgd, address); | |
583 | if (!pud_present(*pud)) | |
584 | goto out; | |
585 | ||
586 | pmd = pmd_offset(pud, address); | |
587 | if (!pmd_present(*pmd)) | |
588 | pmd = NULL; | |
589 | out: | |
590 | return pmd; | |
591 | } | |
592 | ||
81b4082d ND |
593 | /* |
594 | * Check that @page is mapped at @address into @mm. | |
595 | * | |
479db0bf NP |
596 | * If @sync is false, page_check_address may perform a racy check to avoid |
597 | * the page table lock when the pte is not present (helpful when reclaiming | |
598 | * highly shared pages). | |
599 | * | |
b8072f09 | 600 | * On success returns with pte mapped and locked. |
81b4082d | 601 | */ |
e9a81a82 | 602 | pte_t *__page_check_address(struct page *page, struct mm_struct *mm, |
479db0bf | 603 | unsigned long address, spinlock_t **ptlp, int sync) |
81b4082d | 604 | { |
81b4082d ND |
605 | pmd_t *pmd; |
606 | pte_t *pte; | |
c0718806 | 607 | spinlock_t *ptl; |
81b4082d | 608 | |
0fe6e20b | 609 | if (unlikely(PageHuge(page))) { |
398bbc57 | 610 | /* when pud is not present, pte will be NULL */ |
0fe6e20b | 611 | pte = huge_pte_offset(mm, address); |
398bbc57 JW |
612 | if (!pte) |
613 | return NULL; | |
614 | ||
0fe6e20b NH |
615 | ptl = &mm->page_table_lock; |
616 | goto check; | |
617 | } | |
618 | ||
6219049a BL |
619 | pmd = mm_find_pmd(mm, address); |
620 | if (!pmd) | |
c0718806 HD |
621 | return NULL; |
622 | ||
71e3aac0 AA |
623 | if (pmd_trans_huge(*pmd)) |
624 | return NULL; | |
c0718806 HD |
625 | |
626 | pte = pte_offset_map(pmd, address); | |
627 | /* Make a quick check before getting the lock */ | |
479db0bf | 628 | if (!sync && !pte_present(*pte)) { |
c0718806 HD |
629 | pte_unmap(pte); |
630 | return NULL; | |
631 | } | |
632 | ||
4c21e2f2 | 633 | ptl = pte_lockptr(mm, pmd); |
0fe6e20b | 634 | check: |
c0718806 HD |
635 | spin_lock(ptl); |
636 | if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) { | |
637 | *ptlp = ptl; | |
638 | return pte; | |
81b4082d | 639 | } |
c0718806 HD |
640 | pte_unmap_unlock(pte, ptl); |
641 | return NULL; | |
81b4082d ND |
642 | } |
643 | ||
b291f000 NP |
644 | /** |
645 | * page_mapped_in_vma - check whether a page is really mapped in a VMA | |
646 | * @page: the page to test | |
647 | * @vma: the VMA to test | |
648 | * | |
649 | * Returns 1 if the page is mapped into the page tables of the VMA, 0 | |
650 | * if the page is not mapped into the page tables of this VMA. Only | |
651 | * valid for normal file or anonymous VMAs. | |
652 | */ | |
6a46079c | 653 | int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma) |
b291f000 NP |
654 | { |
655 | unsigned long address; | |
656 | pte_t *pte; | |
657 | spinlock_t *ptl; | |
658 | ||
86c2ad19 ML |
659 | address = __vma_address(page, vma); |
660 | if (unlikely(address < vma->vm_start || address >= vma->vm_end)) | |
b291f000 NP |
661 | return 0; |
662 | pte = page_check_address(page, vma->vm_mm, address, &ptl, 1); | |
663 | if (!pte) /* the page is not in this mm */ | |
664 | return 0; | |
665 | pte_unmap_unlock(pte, ptl); | |
666 | ||
667 | return 1; | |
668 | } | |
669 | ||
1da177e4 LT |
670 | /* |
671 | * Subfunctions of page_referenced: page_referenced_one called | |
672 | * repeatedly from either page_referenced_anon or page_referenced_file. | |
673 | */ | |
5ad64688 HD |
674 | int page_referenced_one(struct page *page, struct vm_area_struct *vma, |
675 | unsigned long address, unsigned int *mapcount, | |
676 | unsigned long *vm_flags) | |
1da177e4 LT |
677 | { |
678 | struct mm_struct *mm = vma->vm_mm; | |
1da177e4 LT |
679 | int referenced = 0; |
680 | ||
71e3aac0 AA |
681 | if (unlikely(PageTransHuge(page))) { |
682 | pmd_t *pmd; | |
683 | ||
684 | spin_lock(&mm->page_table_lock); | |
2da28bfd AA |
685 | /* |
686 | * rmap might return false positives; we must filter | |
687 | * these out using page_check_address_pmd(). | |
688 | */ | |
71e3aac0 AA |
689 | pmd = page_check_address_pmd(page, mm, address, |
690 | PAGE_CHECK_ADDRESS_PMD_FLAG); | |
2da28bfd AA |
691 | if (!pmd) { |
692 | spin_unlock(&mm->page_table_lock); | |
693 | goto out; | |
694 | } | |
695 | ||
696 | if (vma->vm_flags & VM_LOCKED) { | |
697 | spin_unlock(&mm->page_table_lock); | |
698 | *mapcount = 0; /* break early from loop */ | |
699 | *vm_flags |= VM_LOCKED; | |
700 | goto out; | |
701 | } | |
702 | ||
703 | /* go ahead even if the pmd is pmd_trans_splitting() */ | |
704 | if (pmdp_clear_flush_young_notify(vma, address, pmd)) | |
71e3aac0 AA |
705 | referenced++; |
706 | spin_unlock(&mm->page_table_lock); | |
707 | } else { | |
708 | pte_t *pte; | |
709 | spinlock_t *ptl; | |
710 | ||
2da28bfd AA |
711 | /* |
712 | * rmap might return false positives; we must filter | |
713 | * these out using page_check_address(). | |
714 | */ | |
71e3aac0 AA |
715 | pte = page_check_address(page, mm, address, &ptl, 0); |
716 | if (!pte) | |
717 | goto out; | |
718 | ||
2da28bfd AA |
719 | if (vma->vm_flags & VM_LOCKED) { |
720 | pte_unmap_unlock(pte, ptl); | |
721 | *mapcount = 0; /* break early from loop */ | |
722 | *vm_flags |= VM_LOCKED; | |
723 | goto out; | |
724 | } | |
725 | ||
71e3aac0 AA |
726 | if (ptep_clear_flush_young_notify(vma, address, pte)) { |
727 | /* | |
728 | * Don't treat a reference through a sequentially read | |
729 | * mapping as such. If the page has been used in | |
730 | * another mapping, we will catch it; if this other | |
731 | * mapping is already gone, the unmap path will have | |
732 | * set PG_referenced or activated the page. | |
733 | */ | |
734 | if (likely(!VM_SequentialReadHint(vma))) | |
735 | referenced++; | |
736 | } | |
737 | pte_unmap_unlock(pte, ptl); | |
738 | } | |
739 | ||
c0718806 | 740 | (*mapcount)--; |
273f047e | 741 | |
6fe6b7e3 WF |
742 | if (referenced) |
743 | *vm_flags |= vma->vm_flags; | |
273f047e | 744 | out: |
1da177e4 LT |
745 | return referenced; |
746 | } | |
747 | ||
bed7161a | 748 | static int page_referenced_anon(struct page *page, |
72835c86 | 749 | struct mem_cgroup *memcg, |
6fe6b7e3 | 750 | unsigned long *vm_flags) |
1da177e4 LT |
751 | { |
752 | unsigned int mapcount; | |
753 | struct anon_vma *anon_vma; | |
bf181b9f | 754 | pgoff_t pgoff; |
5beb4930 | 755 | struct anon_vma_chain *avc; |
1da177e4 LT |
756 | int referenced = 0; |
757 | ||
4fc3f1d6 | 758 | anon_vma = page_lock_anon_vma_read(page); |
1da177e4 LT |
759 | if (!anon_vma) |
760 | return referenced; | |
761 | ||
762 | mapcount = page_mapcount(page); | |
bf181b9f ML |
763 | pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); |
764 | anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) { | |
5beb4930 | 765 | struct vm_area_struct *vma = avc->vma; |
1cb1729b | 766 | unsigned long address = vma_address(page, vma); |
bed7161a BS |
767 | /* |
768 | * If we are reclaiming on behalf of a cgroup, skip | |
769 | * counting on behalf of references from different | |
770 | * cgroups | |
771 | */ | |
72835c86 | 772 | if (memcg && !mm_match_cgroup(vma->vm_mm, memcg)) |
bed7161a | 773 | continue; |
1cb1729b | 774 | referenced += page_referenced_one(page, vma, address, |
6fe6b7e3 | 775 | &mapcount, vm_flags); |
1da177e4 LT |
776 | if (!mapcount) |
777 | break; | |
778 | } | |
34bbd704 | 779 | |
4fc3f1d6 | 780 | page_unlock_anon_vma_read(anon_vma); |
1da177e4 LT |
781 | return referenced; |
782 | } | |
783 | ||
784 | /** | |
785 | * page_referenced_file - referenced check for object-based rmap | |
786 | * @page: the page we're checking references on. | |
72835c86 | 787 | * @memcg: target memory control group |
6fe6b7e3 | 788 | * @vm_flags: collect encountered vma->vm_flags who actually referenced the page |
1da177e4 LT |
789 | * |
790 | * For an object-based mapped page, find all the places it is mapped and | |
791 | * check/clear the referenced flag. This is done by following the page->mapping | |
792 | * pointer, then walking the chain of vmas it holds. It returns the number | |
793 | * of references it found. | |
794 | * | |
795 | * This function is only called from page_referenced for object-based pages. | |
796 | */ | |
bed7161a | 797 | static int page_referenced_file(struct page *page, |
72835c86 | 798 | struct mem_cgroup *memcg, |
6fe6b7e3 | 799 | unsigned long *vm_flags) |
1da177e4 LT |
800 | { |
801 | unsigned int mapcount; | |
802 | struct address_space *mapping = page->mapping; | |
803 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
804 | struct vm_area_struct *vma; | |
1da177e4 LT |
805 | int referenced = 0; |
806 | ||
807 | /* | |
808 | * The caller's checks on page->mapping and !PageAnon have made | |
809 | * sure that this is a file page: the check for page->mapping | |
810 | * excludes the case just before it gets set on an anon page. | |
811 | */ | |
812 | BUG_ON(PageAnon(page)); | |
813 | ||
814 | /* | |
815 | * The page lock not only makes sure that page->mapping cannot | |
816 | * suddenly be NULLified by truncation, it makes sure that the | |
817 | * structure at mapping cannot be freed and reused yet, | |
3d48ae45 | 818 | * so we can safely take mapping->i_mmap_mutex. |
1da177e4 LT |
819 | */ |
820 | BUG_ON(!PageLocked(page)); | |
821 | ||
6fa3eb70 S |
822 | //To avoid deadlock |
823 | if (!mutex_trylock(&mapping->i_mmap_mutex)) | |
824 | return 1; //put in active list | |
1da177e4 LT |
825 | |
826 | /* | |
3d48ae45 | 827 | * i_mmap_mutex does not stabilize mapcount at all, but mapcount |
1da177e4 LT |
828 | * is more likely to be accurate if we note it after spinning. |
829 | */ | |
830 | mapcount = page_mapcount(page); | |
831 | ||
6b2dbba8 | 832 | vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) { |
1cb1729b | 833 | unsigned long address = vma_address(page, vma); |
bed7161a BS |
834 | /* |
835 | * If we are reclaiming on behalf of a cgroup, skip | |
836 | * counting on behalf of references from different | |
837 | * cgroups | |
838 | */ | |
72835c86 | 839 | if (memcg && !mm_match_cgroup(vma->vm_mm, memcg)) |
bed7161a | 840 | continue; |
1cb1729b | 841 | referenced += page_referenced_one(page, vma, address, |
6fe6b7e3 | 842 | &mapcount, vm_flags); |
1da177e4 LT |
843 | if (!mapcount) |
844 | break; | |
845 | } | |
846 | ||
3d48ae45 | 847 | mutex_unlock(&mapping->i_mmap_mutex); |
1da177e4 LT |
848 | return referenced; |
849 | } | |
850 | ||
851 | /** | |
852 | * page_referenced - test if the page was referenced | |
853 | * @page: the page to test | |
854 | * @is_locked: caller holds lock on the page | |
72835c86 | 855 | * @memcg: target memory cgroup |
6fe6b7e3 | 856 | * @vm_flags: collect encountered vma->vm_flags who actually referenced the page |
1da177e4 LT |
857 | * |
858 | * Quick test_and_clear_referenced for all mappings to a page, | |
859 | * returns the number of ptes which referenced the page. | |
860 | */ | |
6fe6b7e3 WF |
861 | int page_referenced(struct page *page, |
862 | int is_locked, | |
72835c86 | 863 | struct mem_cgroup *memcg, |
6fe6b7e3 | 864 | unsigned long *vm_flags) |
1da177e4 LT |
865 | { |
866 | int referenced = 0; | |
5ad64688 | 867 | int we_locked = 0; |
1da177e4 | 868 | |
6fe6b7e3 | 869 | *vm_flags = 0; |
3ca7b3c5 | 870 | if (page_mapped(page) && page_rmapping(page)) { |
5ad64688 HD |
871 | if (!is_locked && (!PageAnon(page) || PageKsm(page))) { |
872 | we_locked = trylock_page(page); | |
873 | if (!we_locked) { | |
874 | referenced++; | |
875 | goto out; | |
876 | } | |
877 | } | |
878 | if (unlikely(PageKsm(page))) | |
72835c86 | 879 | referenced += page_referenced_ksm(page, memcg, |
5ad64688 HD |
880 | vm_flags); |
881 | else if (PageAnon(page)) | |
72835c86 | 882 | referenced += page_referenced_anon(page, memcg, |
6fe6b7e3 | 883 | vm_flags); |
5ad64688 | 884 | else if (page->mapping) |
72835c86 | 885 | referenced += page_referenced_file(page, memcg, |
6fe6b7e3 | 886 | vm_flags); |
5ad64688 | 887 | if (we_locked) |
1da177e4 | 888 | unlock_page(page); |
50a15981 MS |
889 | |
890 | if (page_test_and_clear_young(page_to_pfn(page))) | |
891 | referenced++; | |
1da177e4 | 892 | } |
5ad64688 | 893 | out: |
1da177e4 LT |
894 | return referenced; |
895 | } | |
896 | ||
1cb1729b HD |
897 | static int page_mkclean_one(struct page *page, struct vm_area_struct *vma, |
898 | unsigned long address) | |
d08b3851 PZ |
899 | { |
900 | struct mm_struct *mm = vma->vm_mm; | |
c2fda5fe | 901 | pte_t *pte; |
d08b3851 PZ |
902 | spinlock_t *ptl; |
903 | int ret = 0; | |
904 | ||
479db0bf | 905 | pte = page_check_address(page, mm, address, &ptl, 1); |
d08b3851 PZ |
906 | if (!pte) |
907 | goto out; | |
908 | ||
c2fda5fe PZ |
909 | if (pte_dirty(*pte) || pte_write(*pte)) { |
910 | pte_t entry; | |
d08b3851 | 911 | |
c2fda5fe | 912 | flush_cache_page(vma, address, pte_pfn(*pte)); |
2ec74c3e | 913 | entry = ptep_clear_flush(vma, address, pte); |
c2fda5fe PZ |
914 | entry = pte_wrprotect(entry); |
915 | entry = pte_mkclean(entry); | |
d6e88e67 | 916 | set_pte_at(mm, address, pte, entry); |
c2fda5fe PZ |
917 | ret = 1; |
918 | } | |
d08b3851 | 919 | |
d08b3851 | 920 | pte_unmap_unlock(pte, ptl); |
2ec74c3e SG |
921 | |
922 | if (ret) | |
923 | mmu_notifier_invalidate_page(mm, address); | |
d08b3851 PZ |
924 | out: |
925 | return ret; | |
926 | } | |
927 | ||
928 | static int page_mkclean_file(struct address_space *mapping, struct page *page) | |
929 | { | |
930 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
931 | struct vm_area_struct *vma; | |
d08b3851 PZ |
932 | int ret = 0; |
933 | ||
934 | BUG_ON(PageAnon(page)); | |
935 | ||
3d48ae45 | 936 | mutex_lock(&mapping->i_mmap_mutex); |
6b2dbba8 | 937 | vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) { |
1cb1729b HD |
938 | if (vma->vm_flags & VM_SHARED) { |
939 | unsigned long address = vma_address(page, vma); | |
1cb1729b HD |
940 | ret += page_mkclean_one(page, vma, address); |
941 | } | |
d08b3851 | 942 | } |
3d48ae45 | 943 | mutex_unlock(&mapping->i_mmap_mutex); |
d08b3851 PZ |
944 | return ret; |
945 | } | |
946 | ||
947 | int page_mkclean(struct page *page) | |
948 | { | |
949 | int ret = 0; | |
950 | ||
951 | BUG_ON(!PageLocked(page)); | |
952 | ||
953 | if (page_mapped(page)) { | |
954 | struct address_space *mapping = page_mapping(page); | |
ef5d437f | 955 | if (mapping) |
d08b3851 PZ |
956 | ret = page_mkclean_file(mapping, page); |
957 | } | |
958 | ||
959 | return ret; | |
960 | } | |
60b59bea | 961 | EXPORT_SYMBOL_GPL(page_mkclean); |
d08b3851 | 962 | |
c44b6743 RR |
963 | /** |
964 | * page_move_anon_rmap - move a page to our anon_vma | |
965 | * @page: the page to move to our anon_vma | |
966 | * @vma: the vma the page belongs to | |
967 | * @address: the user virtual address mapped | |
968 | * | |
969 | * When a page belongs exclusively to one process after a COW event, | |
970 | * that page can be moved into the anon_vma that belongs to just that | |
971 | * process, so the rmap code will not search the parent or sibling | |
972 | * processes. | |
973 | */ | |
974 | void page_move_anon_rmap(struct page *page, | |
975 | struct vm_area_struct *vma, unsigned long address) | |
976 | { | |
977 | struct anon_vma *anon_vma = vma->anon_vma; | |
978 | ||
979 | VM_BUG_ON(!PageLocked(page)); | |
980 | VM_BUG_ON(!anon_vma); | |
981 | VM_BUG_ON(page->index != linear_page_index(vma, address)); | |
982 | ||
983 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; | |
984 | page->mapping = (struct address_space *) anon_vma; | |
985 | } | |
986 | ||
9617d95e | 987 | /** |
4e1c1975 AK |
988 | * __page_set_anon_rmap - set up new anonymous rmap |
989 | * @page: Page to add to rmap | |
990 | * @vma: VM area to add page to. | |
991 | * @address: User virtual address of the mapping | |
e8a03feb | 992 | * @exclusive: the page is exclusively owned by the current process |
9617d95e NP |
993 | */ |
994 | static void __page_set_anon_rmap(struct page *page, | |
e8a03feb | 995 | struct vm_area_struct *vma, unsigned long address, int exclusive) |
9617d95e | 996 | { |
e8a03feb | 997 | struct anon_vma *anon_vma = vma->anon_vma; |
ea90002b | 998 | |
e8a03feb | 999 | BUG_ON(!anon_vma); |
ea90002b | 1000 | |
4e1c1975 AK |
1001 | if (PageAnon(page)) |
1002 | return; | |
1003 | ||
ea90002b | 1004 | /* |
e8a03feb RR |
1005 | * If the page isn't exclusively mapped into this vma, |
1006 | * we must use the _oldest_ possible anon_vma for the | |
1007 | * page mapping! | |
ea90002b | 1008 | */ |
4e1c1975 | 1009 | if (!exclusive) |
288468c3 | 1010 | anon_vma = anon_vma->root; |
9617d95e | 1011 | |
9617d95e NP |
1012 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; |
1013 | page->mapping = (struct address_space *) anon_vma; | |
9617d95e | 1014 | page->index = linear_page_index(vma, address); |
9617d95e NP |
1015 | } |
1016 | ||
c97a9e10 | 1017 | /** |
43d8eac4 | 1018 | * __page_check_anon_rmap - sanity check anonymous rmap addition |
c97a9e10 NP |
1019 | * @page: the page to add the mapping to |
1020 | * @vma: the vm area in which the mapping is added | |
1021 | * @address: the user virtual address mapped | |
1022 | */ | |
1023 | static void __page_check_anon_rmap(struct page *page, | |
1024 | struct vm_area_struct *vma, unsigned long address) | |
1025 | { | |
1026 | #ifdef CONFIG_DEBUG_VM | |
1027 | /* | |
1028 | * The page's anon-rmap details (mapping and index) are guaranteed to | |
1029 | * be set up correctly at this point. | |
1030 | * | |
1031 | * We have exclusion against page_add_anon_rmap because the caller | |
1032 | * always holds the page locked, except if called from page_dup_rmap, | |
1033 | * in which case the page is already known to be setup. | |
1034 | * | |
1035 | * We have exclusion against page_add_new_anon_rmap because those pages | |
1036 | * are initially only visible via the pagetables, and the pte is locked | |
1037 | * over the call to page_add_new_anon_rmap. | |
1038 | */ | |
44ab57a0 | 1039 | BUG_ON(page_anon_vma(page)->root != vma->anon_vma->root); |
c97a9e10 NP |
1040 | BUG_ON(page->index != linear_page_index(vma, address)); |
1041 | #endif | |
1042 | } | |
1043 | ||
1da177e4 LT |
1044 | /** |
1045 | * page_add_anon_rmap - add pte mapping to an anonymous page | |
1046 | * @page: the page to add the mapping to | |
1047 | * @vma: the vm area in which the mapping is added | |
1048 | * @address: the user virtual address mapped | |
1049 | * | |
5ad64688 | 1050 | * The caller needs to hold the pte lock, and the page must be locked in |
80e14822 HD |
1051 | * the anon_vma case: to serialize mapping,index checking after setting, |
1052 | * and to ensure that PageAnon is not being upgraded racily to PageKsm | |
1053 | * (but PageKsm is never downgraded to PageAnon). | |
1da177e4 LT |
1054 | */ |
1055 | void page_add_anon_rmap(struct page *page, | |
1056 | struct vm_area_struct *vma, unsigned long address) | |
ad8c2ee8 RR |
1057 | { |
1058 | do_page_add_anon_rmap(page, vma, address, 0); | |
1059 | } | |
1060 | ||
1061 | /* | |
1062 | * Special version of the above for do_swap_page, which often runs | |
1063 | * into pages that are exclusively owned by the current process. | |
1064 | * Everybody else should continue to use page_add_anon_rmap above. | |
1065 | */ | |
1066 | void do_page_add_anon_rmap(struct page *page, | |
1067 | struct vm_area_struct *vma, unsigned long address, int exclusive) | |
1da177e4 | 1068 | { |
5ad64688 | 1069 | int first = atomic_inc_and_test(&page->_mapcount); |
79134171 AA |
1070 | if (first) { |
1071 | if (!PageTransHuge(page)) | |
1072 | __inc_zone_page_state(page, NR_ANON_PAGES); | |
1073 | else | |
1074 | __inc_zone_page_state(page, | |
1075 | NR_ANON_TRANSPARENT_HUGEPAGES); | |
1076 | } | |
5ad64688 HD |
1077 | if (unlikely(PageKsm(page))) |
1078 | return; | |
1079 | ||
c97a9e10 | 1080 | VM_BUG_ON(!PageLocked(page)); |
5dbe0af4 | 1081 | /* address might be in next vma when migration races vma_adjust */ |
5ad64688 | 1082 | if (first) |
ad8c2ee8 | 1083 | __page_set_anon_rmap(page, vma, address, exclusive); |
69029cd5 | 1084 | else |
c97a9e10 | 1085 | __page_check_anon_rmap(page, vma, address); |
1da177e4 LT |
1086 | } |
1087 | ||
43d8eac4 | 1088 | /** |
9617d95e NP |
1089 | * page_add_new_anon_rmap - add pte mapping to a new anonymous page |
1090 | * @page: the page to add the mapping to | |
1091 | * @vma: the vm area in which the mapping is added | |
1092 | * @address: the user virtual address mapped | |
1093 | * | |
1094 | * Same as page_add_anon_rmap but must only be called on *new* pages. | |
1095 | * This means the inc-and-test can be bypassed. | |
c97a9e10 | 1096 | * Page does not have to be locked. |
9617d95e NP |
1097 | */ |
1098 | void page_add_new_anon_rmap(struct page *page, | |
1099 | struct vm_area_struct *vma, unsigned long address) | |
1100 | { | |
b5934c53 | 1101 | VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); |
cbf84b7a HD |
1102 | SetPageSwapBacked(page); |
1103 | atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */ | |
79134171 AA |
1104 | if (!PageTransHuge(page)) |
1105 | __inc_zone_page_state(page, NR_ANON_PAGES); | |
1106 | else | |
1107 | __inc_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES); | |
e8a03feb | 1108 | __page_set_anon_rmap(page, vma, address, 1); |
39b5f29a | 1109 | if (!mlocked_vma_newpage(vma, page)) |
cbf84b7a | 1110 | lru_cache_add_lru(page, LRU_ACTIVE_ANON); |
b5934c53 HD |
1111 | else |
1112 | add_page_to_unevictable_list(page); | |
9617d95e NP |
1113 | } |
1114 | ||
1da177e4 LT |
1115 | /** |
1116 | * page_add_file_rmap - add pte mapping to a file page | |
1117 | * @page: the page to add the mapping to | |
1118 | * | |
b8072f09 | 1119 | * The caller needs to hold the pte lock. |
1da177e4 LT |
1120 | */ |
1121 | void page_add_file_rmap(struct page *page) | |
1122 | { | |
89c06bd5 KH |
1123 | bool locked; |
1124 | unsigned long flags; | |
1125 | ||
1126 | mem_cgroup_begin_update_page_stat(page, &locked, &flags); | |
d69b042f | 1127 | if (atomic_inc_and_test(&page->_mapcount)) { |
65ba55f5 | 1128 | __inc_zone_page_state(page, NR_FILE_MAPPED); |
2a7106f2 | 1129 | mem_cgroup_inc_page_stat(page, MEMCG_NR_FILE_MAPPED); |
d69b042f | 1130 | } |
89c06bd5 | 1131 | mem_cgroup_end_update_page_stat(page, &locked, &flags); |
1da177e4 LT |
1132 | } |
1133 | ||
1134 | /** | |
1135 | * page_remove_rmap - take down pte mapping from a page | |
1136 | * @page: page to remove mapping from | |
1137 | * | |
b8072f09 | 1138 | * The caller needs to hold the pte lock. |
1da177e4 | 1139 | */ |
edc315fd | 1140 | void page_remove_rmap(struct page *page) |
1da177e4 | 1141 | { |
89c06bd5 KH |
1142 | bool anon = PageAnon(page); |
1143 | bool locked; | |
1144 | unsigned long flags; | |
1145 | ||
1146 | /* | |
1147 | * The anon case has no mem_cgroup page_stat to update; but may | |
1148 | * uncharge_page() below, where the lock ordering can deadlock if | |
1149 | * we hold the lock against page_stat move: so avoid it on anon. | |
1150 | */ | |
1151 | if (!anon) | |
1152 | mem_cgroup_begin_update_page_stat(page, &locked, &flags); | |
1153 | ||
b904dcfe KM |
1154 | /* page still mapped by someone else? */ |
1155 | if (!atomic_add_negative(-1, &page->_mapcount)) | |
89c06bd5 | 1156 | goto out; |
b904dcfe | 1157 | |
0fe6e20b NH |
1158 | /* |
1159 | * Hugepages are not counted in NR_ANON_PAGES nor NR_FILE_MAPPED | |
1160 | * and not charged by memcg for now. | |
1161 | */ | |
1162 | if (unlikely(PageHuge(page))) | |
89c06bd5 KH |
1163 | goto out; |
1164 | if (anon) { | |
b904dcfe | 1165 | mem_cgroup_uncharge_page(page); |
79134171 AA |
1166 | if (!PageTransHuge(page)) |
1167 | __dec_zone_page_state(page, NR_ANON_PAGES); | |
1168 | else | |
1169 | __dec_zone_page_state(page, | |
1170 | NR_ANON_TRANSPARENT_HUGEPAGES); | |
b904dcfe KM |
1171 | } else { |
1172 | __dec_zone_page_state(page, NR_FILE_MAPPED); | |
2a7106f2 | 1173 | mem_cgroup_dec_page_stat(page, MEMCG_NR_FILE_MAPPED); |
e6c509f8 | 1174 | mem_cgroup_end_update_page_stat(page, &locked, &flags); |
b904dcfe | 1175 | } |
e6c509f8 HD |
1176 | if (unlikely(PageMlocked(page))) |
1177 | clear_page_mlock(page); | |
b904dcfe KM |
1178 | /* |
1179 | * It would be tidy to reset the PageAnon mapping here, | |
1180 | * but that might overwrite a racing page_add_anon_rmap | |
1181 | * which increments mapcount after us but sets mapping | |
1182 | * before us: so leave the reset to free_hot_cold_page, | |
1183 | * and remember that it's only reliable while mapped. | |
1184 | * Leaving it set also helps swapoff to reinstate ptes | |
1185 | * faster for those pages still in swapcache. | |
1186 | */ | |
e6c509f8 | 1187 | return; |
89c06bd5 KH |
1188 | out: |
1189 | if (!anon) | |
1190 | mem_cgroup_end_update_page_stat(page, &locked, &flags); | |
1da177e4 LT |
1191 | } |
1192 | ||
1193 | /* | |
1194 | * Subfunctions of try_to_unmap: try_to_unmap_one called | |
99ef0315 | 1195 | * repeatedly from try_to_unmap_ksm, try_to_unmap_anon or try_to_unmap_file. |
1da177e4 | 1196 | */ |
5ad64688 HD |
1197 | int try_to_unmap_one(struct page *page, struct vm_area_struct *vma, |
1198 | unsigned long address, enum ttu_flags flags) | |
1da177e4 LT |
1199 | { |
1200 | struct mm_struct *mm = vma->vm_mm; | |
1da177e4 LT |
1201 | pte_t *pte; |
1202 | pte_t pteval; | |
c0718806 | 1203 | spinlock_t *ptl; |
1da177e4 LT |
1204 | int ret = SWAP_AGAIN; |
1205 | ||
479db0bf | 1206 | pte = page_check_address(page, mm, address, &ptl, 0); |
c0718806 | 1207 | if (!pte) |
81b4082d | 1208 | goto out; |
1da177e4 LT |
1209 | |
1210 | /* | |
1211 | * If the page is mlock()d, we cannot swap it out. | |
1212 | * If it's recently referenced (perhaps page_referenced | |
1213 | * skipped over this mm) then we should reactivate it. | |
1214 | */ | |
14fa31b8 | 1215 | if (!(flags & TTU_IGNORE_MLOCK)) { |
caed0f48 KM |
1216 | if (vma->vm_flags & VM_LOCKED) |
1217 | goto out_mlock; | |
1218 | ||
af8e3354 | 1219 | if (TTU_ACTION(flags) == TTU_MUNLOCK) |
53f79acb | 1220 | goto out_unmap; |
14fa31b8 AK |
1221 | } |
1222 | if (!(flags & TTU_IGNORE_ACCESS)) { | |
b291f000 NP |
1223 | if (ptep_clear_flush_young_notify(vma, address, pte)) { |
1224 | ret = SWAP_FAIL; | |
1225 | goto out_unmap; | |
1226 | } | |
1227 | } | |
1da177e4 | 1228 | |
1da177e4 LT |
1229 | /* Nuke the page table entry. */ |
1230 | flush_cache_page(vma, address, page_to_pfn(page)); | |
2ec74c3e | 1231 | pteval = ptep_clear_flush(vma, address, pte); |
1da177e4 LT |
1232 | |
1233 | /* Move the dirty bit to the physical page now the pte is gone. */ | |
1234 | if (pte_dirty(pteval)) | |
1235 | set_page_dirty(page); | |
1236 | ||
365e9c87 HD |
1237 | /* Update high watermark before we lower rss */ |
1238 | update_hiwater_rss(mm); | |
1239 | ||
888b9f7c | 1240 | if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) { |
5f24ae58 NH |
1241 | if (!PageHuge(page)) { |
1242 | if (PageAnon(page)) | |
1243 | dec_mm_counter(mm, MM_ANONPAGES); | |
1244 | else | |
1245 | dec_mm_counter(mm, MM_FILEPAGES); | |
1246 | } | |
888b9f7c | 1247 | set_pte_at(mm, address, pte, |
5f24ae58 | 1248 | swp_entry_to_pte(make_hwpoison_entry(page))); |
888b9f7c | 1249 | } else if (PageAnon(page)) { |
4c21e2f2 | 1250 | swp_entry_t entry = { .val = page_private(page) }; |
0697212a CL |
1251 | |
1252 | if (PageSwapCache(page)) { | |
1253 | /* | |
1254 | * Store the swap location in the pte. | |
1255 | * See handle_pte_fault() ... | |
1256 | */ | |
570a335b HD |
1257 | if (swap_duplicate(entry) < 0) { |
1258 | set_pte_at(mm, address, pte, pteval); | |
1259 | ret = SWAP_FAIL; | |
1260 | goto out_unmap; | |
1261 | } | |
0697212a CL |
1262 | if (list_empty(&mm->mmlist)) { |
1263 | spin_lock(&mmlist_lock); | |
1264 | if (list_empty(&mm->mmlist)) | |
1265 | list_add(&mm->mmlist, &init_mm.mmlist); | |
1266 | spin_unlock(&mmlist_lock); | |
1267 | } | |
d559db08 | 1268 | dec_mm_counter(mm, MM_ANONPAGES); |
b084d435 | 1269 | inc_mm_counter(mm, MM_SWAPENTS); |
ce1744f4 | 1270 | } else if (IS_ENABLED(CONFIG_MIGRATION)) { |
0697212a CL |
1271 | /* |
1272 | * Store the pfn of the page in a special migration | |
1273 | * pte. do_swap_page() will wait until the migration | |
1274 | * pte is removed and then restart fault handling. | |
1275 | */ | |
14fa31b8 | 1276 | BUG_ON(TTU_ACTION(flags) != TTU_MIGRATION); |
0697212a | 1277 | entry = make_migration_entry(page, pte_write(pteval)); |
1da177e4 LT |
1278 | } |
1279 | set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); | |
1280 | BUG_ON(pte_file(*pte)); | |
ce1744f4 KK |
1281 | } else if (IS_ENABLED(CONFIG_MIGRATION) && |
1282 | (TTU_ACTION(flags) == TTU_MIGRATION)) { | |
04e62a29 CL |
1283 | /* Establish migration entry for a file page */ |
1284 | swp_entry_t entry; | |
1285 | entry = make_migration_entry(page, pte_write(pteval)); | |
1286 | set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); | |
1287 | } else | |
d559db08 | 1288 | dec_mm_counter(mm, MM_FILEPAGES); |
1da177e4 | 1289 | |
edc315fd | 1290 | page_remove_rmap(page); |
1da177e4 LT |
1291 | page_cache_release(page); |
1292 | ||
1293 | out_unmap: | |
c0718806 | 1294 | pte_unmap_unlock(pte, ptl); |
2ec74c3e SG |
1295 | if (ret != SWAP_FAIL) |
1296 | mmu_notifier_invalidate_page(mm, address); | |
caed0f48 KM |
1297 | out: |
1298 | return ret; | |
53f79acb | 1299 | |
caed0f48 KM |
1300 | out_mlock: |
1301 | pte_unmap_unlock(pte, ptl); | |
1302 | ||
1303 | ||
1304 | /* | |
1305 | * We need mmap_sem locking, Otherwise VM_LOCKED check makes | |
1306 | * unstable result and race. Plus, We can't wait here because | |
5a505085 | 1307 | * we now hold anon_vma->rwsem or mapping->i_mmap_mutex. |
caed0f48 KM |
1308 | * if trylock failed, the page remain in evictable lru and later |
1309 | * vmscan could retry to move the page to unevictable lru if the | |
1310 | * page is actually mlocked. | |
1311 | */ | |
1312 | if (down_read_trylock(&vma->vm_mm->mmap_sem)) { | |
1313 | if (vma->vm_flags & VM_LOCKED) { | |
1314 | mlock_vma_page(page); | |
1315 | ret = SWAP_MLOCK; | |
53f79acb | 1316 | } |
caed0f48 | 1317 | up_read(&vma->vm_mm->mmap_sem); |
53f79acb | 1318 | } |
1da177e4 LT |
1319 | return ret; |
1320 | } | |
1321 | ||
1322 | /* | |
1323 | * objrmap doesn't work for nonlinear VMAs because the assumption that | |
1324 | * offset-into-file correlates with offset-into-virtual-addresses does not hold. | |
1325 | * Consequently, given a particular page and its ->index, we cannot locate the | |
1326 | * ptes which are mapping that page without an exhaustive linear search. | |
1327 | * | |
1328 | * So what this code does is a mini "virtual scan" of each nonlinear VMA which | |
1329 | * maps the file to which the target page belongs. The ->vm_private_data field | |
1330 | * holds the current cursor into that scan. Successive searches will circulate | |
1331 | * around the vma's virtual address space. | |
1332 | * | |
1333 | * So as more replacement pressure is applied to the pages in a nonlinear VMA, | |
1334 | * more scanning pressure is placed against them as well. Eventually pages | |
1335 | * will become fully unmapped and are eligible for eviction. | |
1336 | * | |
1337 | * For very sparsely populated VMAs this is a little inefficient - chances are | |
1338 | * there there won't be many ptes located within the scan cluster. In this case | |
1339 | * maybe we could scan further - to the end of the pte page, perhaps. | |
b291f000 NP |
1340 | * |
1341 | * Mlocked pages: check VM_LOCKED under mmap_sem held for read, if we can | |
1342 | * acquire it without blocking. If vma locked, mlock the pages in the cluster, | |
1343 | * rather than unmapping them. If we encounter the "check_page" that vmscan is | |
1344 | * trying to unmap, return SWAP_MLOCK, else default SWAP_AGAIN. | |
1da177e4 LT |
1345 | */ |
1346 | #define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE) | |
1347 | #define CLUSTER_MASK (~(CLUSTER_SIZE - 1)) | |
1348 | ||
b291f000 NP |
1349 | static int try_to_unmap_cluster(unsigned long cursor, unsigned int *mapcount, |
1350 | struct vm_area_struct *vma, struct page *check_page) | |
1da177e4 LT |
1351 | { |
1352 | struct mm_struct *mm = vma->vm_mm; | |
1da177e4 | 1353 | pmd_t *pmd; |
c0718806 | 1354 | pte_t *pte; |
1da177e4 | 1355 | pte_t pteval; |
c0718806 | 1356 | spinlock_t *ptl; |
1da177e4 LT |
1357 | struct page *page; |
1358 | unsigned long address; | |
2ec74c3e SG |
1359 | unsigned long mmun_start; /* For mmu_notifiers */ |
1360 | unsigned long mmun_end; /* For mmu_notifiers */ | |
1da177e4 | 1361 | unsigned long end; |
b291f000 NP |
1362 | int ret = SWAP_AGAIN; |
1363 | int locked_vma = 0; | |
1da177e4 | 1364 | |
1da177e4 LT |
1365 | address = (vma->vm_start + cursor) & CLUSTER_MASK; |
1366 | end = address + CLUSTER_SIZE; | |
1367 | if (address < vma->vm_start) | |
1368 | address = vma->vm_start; | |
1369 | if (end > vma->vm_end) | |
1370 | end = vma->vm_end; | |
1371 | ||
6219049a BL |
1372 | pmd = mm_find_pmd(mm, address); |
1373 | if (!pmd) | |
b291f000 NP |
1374 | return ret; |
1375 | ||
2ec74c3e SG |
1376 | mmun_start = address; |
1377 | mmun_end = end; | |
1378 | mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); | |
1379 | ||
b291f000 | 1380 | /* |
af8e3354 | 1381 | * If we can acquire the mmap_sem for read, and vma is VM_LOCKED, |
b291f000 NP |
1382 | * keep the sem while scanning the cluster for mlocking pages. |
1383 | */ | |
af8e3354 | 1384 | if (down_read_trylock(&vma->vm_mm->mmap_sem)) { |
b291f000 NP |
1385 | locked_vma = (vma->vm_flags & VM_LOCKED); |
1386 | if (!locked_vma) | |
1387 | up_read(&vma->vm_mm->mmap_sem); /* don't need it */ | |
1388 | } | |
c0718806 HD |
1389 | |
1390 | pte = pte_offset_map_lock(mm, pmd, address, &ptl); | |
1da177e4 | 1391 | |
365e9c87 HD |
1392 | /* Update high watermark before we lower rss */ |
1393 | update_hiwater_rss(mm); | |
1394 | ||
c0718806 | 1395 | for (; address < end; pte++, address += PAGE_SIZE) { |
1da177e4 LT |
1396 | if (!pte_present(*pte)) |
1397 | continue; | |
6aab341e LT |
1398 | page = vm_normal_page(vma, address, *pte); |
1399 | BUG_ON(!page || PageAnon(page)); | |
1da177e4 | 1400 | |
b291f000 | 1401 | if (locked_vma) { |
77552735 VB |
1402 | if (page == check_page) { |
1403 | /* we know we have check_page locked */ | |
1404 | mlock_vma_page(page); | |
b291f000 | 1405 | ret = SWAP_MLOCK; |
77552735 VB |
1406 | } else if (trylock_page(page)) { |
1407 | /* | |
1408 | * If we can lock the page, perform mlock. | |
1409 | * Otherwise leave the page alone, it will be | |
1410 | * eventually encountered again later. | |
1411 | */ | |
1412 | mlock_vma_page(page); | |
1413 | unlock_page(page); | |
1414 | } | |
b291f000 NP |
1415 | continue; /* don't unmap */ |
1416 | } | |
1417 | ||
cddb8a5c | 1418 | if (ptep_clear_flush_young_notify(vma, address, pte)) |
1da177e4 LT |
1419 | continue; |
1420 | ||
1421 | /* Nuke the page table entry. */ | |
eca35133 | 1422 | flush_cache_page(vma, address, pte_pfn(*pte)); |
2ec74c3e | 1423 | pteval = ptep_clear_flush(vma, address, pte); |
1da177e4 LT |
1424 | |
1425 | /* If nonlinear, store the file page offset in the pte. */ | |
1426 | if (page->index != linear_page_index(vma, address)) | |
1427 | set_pte_at(mm, address, pte, pgoff_to_pte(page->index)); | |
1428 | ||
1429 | /* Move the dirty bit to the physical page now the pte is gone. */ | |
1430 | if (pte_dirty(pteval)) | |
1431 | set_page_dirty(page); | |
1432 | ||
edc315fd | 1433 | page_remove_rmap(page); |
1da177e4 | 1434 | page_cache_release(page); |
d559db08 | 1435 | dec_mm_counter(mm, MM_FILEPAGES); |
1da177e4 LT |
1436 | (*mapcount)--; |
1437 | } | |
c0718806 | 1438 | pte_unmap_unlock(pte - 1, ptl); |
2ec74c3e | 1439 | mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); |
b291f000 NP |
1440 | if (locked_vma) |
1441 | up_read(&vma->vm_mm->mmap_sem); | |
1442 | return ret; | |
1da177e4 LT |
1443 | } |
1444 | ||
71e3aac0 | 1445 | bool is_vma_temporary_stack(struct vm_area_struct *vma) |
a8bef8ff MG |
1446 | { |
1447 | int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP); | |
1448 | ||
1449 | if (!maybe_stack) | |
1450 | return false; | |
1451 | ||
1452 | if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) == | |
1453 | VM_STACK_INCOMPLETE_SETUP) | |
1454 | return true; | |
1455 | ||
1456 | return false; | |
1457 | } | |
1458 | ||
b291f000 NP |
1459 | /** |
1460 | * try_to_unmap_anon - unmap or unlock anonymous page using the object-based | |
1461 | * rmap method | |
1462 | * @page: the page to unmap/unlock | |
8051be5e | 1463 | * @flags: action and flags |
b291f000 NP |
1464 | * |
1465 | * Find all the mappings of a page using the mapping pointer and the vma chains | |
1466 | * contained in the anon_vma struct it points to. | |
1467 | * | |
1468 | * This function is only called from try_to_unmap/try_to_munlock for | |
1469 | * anonymous pages. | |
1470 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma | |
1471 | * where the page was found will be held for write. So, we won't recheck | |
1472 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be | |
1473 | * 'LOCKED. | |
1474 | */ | |
14fa31b8 | 1475 | static int try_to_unmap_anon(struct page *page, enum ttu_flags flags) |
1da177e4 LT |
1476 | { |
1477 | struct anon_vma *anon_vma; | |
bf181b9f | 1478 | pgoff_t pgoff; |
5beb4930 | 1479 | struct anon_vma_chain *avc; |
1da177e4 | 1480 | int ret = SWAP_AGAIN; |
b291f000 | 1481 | |
4fc3f1d6 | 1482 | anon_vma = page_lock_anon_vma_read(page); |
1da177e4 LT |
1483 | if (!anon_vma) |
1484 | return ret; | |
1485 | ||
bf181b9f ML |
1486 | pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); |
1487 | anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) { | |
5beb4930 | 1488 | struct vm_area_struct *vma = avc->vma; |
a8bef8ff MG |
1489 | unsigned long address; |
1490 | ||
1491 | /* | |
1492 | * During exec, a temporary VMA is setup and later moved. | |
1493 | * The VMA is moved under the anon_vma lock but not the | |
1494 | * page tables leading to a race where migration cannot | |
1495 | * find the migration ptes. Rather than increasing the | |
1496 | * locking requirements of exec(), migration skips | |
1497 | * temporary VMAs until after exec() completes. | |
1498 | */ | |
ce1744f4 | 1499 | if (IS_ENABLED(CONFIG_MIGRATION) && (flags & TTU_MIGRATION) && |
a8bef8ff MG |
1500 | is_vma_temporary_stack(vma)) |
1501 | continue; | |
1502 | ||
1503 | address = vma_address(page, vma); | |
1cb1729b | 1504 | ret = try_to_unmap_one(page, vma, address, flags); |
53f79acb HD |
1505 | if (ret != SWAP_AGAIN || !page_mapped(page)) |
1506 | break; | |
1da177e4 | 1507 | } |
34bbd704 | 1508 | |
4fc3f1d6 | 1509 | page_unlock_anon_vma_read(anon_vma); |
1da177e4 LT |
1510 | return ret; |
1511 | } | |
1512 | ||
1513 | /** | |
b291f000 NP |
1514 | * try_to_unmap_file - unmap/unlock file page using the object-based rmap method |
1515 | * @page: the page to unmap/unlock | |
14fa31b8 | 1516 | * @flags: action and flags |
1da177e4 LT |
1517 | * |
1518 | * Find all the mappings of a page using the mapping pointer and the vma chains | |
1519 | * contained in the address_space struct it points to. | |
1520 | * | |
b291f000 NP |
1521 | * This function is only called from try_to_unmap/try_to_munlock for |
1522 | * object-based pages. | |
1523 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma | |
1524 | * where the page was found will be held for write. So, we won't recheck | |
1525 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be | |
1526 | * 'LOCKED. | |
1da177e4 | 1527 | */ |
14fa31b8 | 1528 | static int try_to_unmap_file(struct page *page, enum ttu_flags flags) |
1da177e4 LT |
1529 | { |
1530 | struct address_space *mapping = page->mapping; | |
1531 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
1532 | struct vm_area_struct *vma; | |
1da177e4 LT |
1533 | int ret = SWAP_AGAIN; |
1534 | unsigned long cursor; | |
1535 | unsigned long max_nl_cursor = 0; | |
1536 | unsigned long max_nl_size = 0; | |
1537 | unsigned int mapcount; | |
6fa3eb70 S |
1538 | int retry = 0; |
1539 | ||
1da177e4 | 1540 | |
369a713e HD |
1541 | if (PageHuge(page)) |
1542 | pgoff = page->index << compound_order(page); | |
1543 | ||
6fa3eb70 S |
1544 | while (!mutex_trylock(&mapping->i_mmap_mutex)) { |
1545 | retry++; | |
1546 | if (!(retry % MUTEX_RETRY_RESCHED)) | |
1547 | cond_resched(); | |
1548 | if (retry > MUTEX_RETRY_COUNT) { | |
1549 | printk(KERN_ERR ">> failed to lock i_mmap_mutex in try_to_unmap_file <<\n"); | |
1550 | return SWAP_FAIL; | |
1551 | } | |
1552 | } | |
1553 | ||
1554 | ||
6b2dbba8 | 1555 | vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) { |
1cb1729b | 1556 | unsigned long address = vma_address(page, vma); |
1cb1729b | 1557 | ret = try_to_unmap_one(page, vma, address, flags); |
53f79acb HD |
1558 | if (ret != SWAP_AGAIN || !page_mapped(page)) |
1559 | goto out; | |
1da177e4 LT |
1560 | } |
1561 | ||
1562 | if (list_empty(&mapping->i_mmap_nonlinear)) | |
1563 | goto out; | |
1564 | ||
53f79acb HD |
1565 | /* |
1566 | * We don't bother to try to find the munlocked page in nonlinears. | |
1567 | * It's costly. Instead, later, page reclaim logic may call | |
1568 | * try_to_unmap(TTU_MUNLOCK) and recover PG_mlocked lazily. | |
1569 | */ | |
1570 | if (TTU_ACTION(flags) == TTU_MUNLOCK) | |
1571 | goto out; | |
1572 | ||
1da177e4 | 1573 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, |
6b2dbba8 | 1574 | shared.nonlinear) { |
1da177e4 LT |
1575 | cursor = (unsigned long) vma->vm_private_data; |
1576 | if (cursor > max_nl_cursor) | |
1577 | max_nl_cursor = cursor; | |
1578 | cursor = vma->vm_end - vma->vm_start; | |
1579 | if (cursor > max_nl_size) | |
1580 | max_nl_size = cursor; | |
1581 | } | |
1582 | ||
b291f000 | 1583 | if (max_nl_size == 0) { /* all nonlinears locked or reserved ? */ |
1da177e4 LT |
1584 | ret = SWAP_FAIL; |
1585 | goto out; | |
1586 | } | |
1587 | ||
1588 | /* | |
1589 | * We don't try to search for this page in the nonlinear vmas, | |
1590 | * and page_referenced wouldn't have found it anyway. Instead | |
1591 | * just walk the nonlinear vmas trying to age and unmap some. | |
1592 | * The mapcount of the page we came in with is irrelevant, | |
1593 | * but even so use it as a guide to how hard we should try? | |
1594 | */ | |
1595 | mapcount = page_mapcount(page); | |
1596 | if (!mapcount) | |
1597 | goto out; | |
3d48ae45 | 1598 | cond_resched(); |
1da177e4 LT |
1599 | |
1600 | max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK; | |
1601 | if (max_nl_cursor == 0) | |
1602 | max_nl_cursor = CLUSTER_SIZE; | |
1603 | ||
1604 | do { | |
1605 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, | |
6b2dbba8 | 1606 | shared.nonlinear) { |
1da177e4 | 1607 | cursor = (unsigned long) vma->vm_private_data; |
839b9685 | 1608 | while ( cursor < max_nl_cursor && |
1da177e4 | 1609 | cursor < vma->vm_end - vma->vm_start) { |
53f79acb HD |
1610 | if (try_to_unmap_cluster(cursor, &mapcount, |
1611 | vma, page) == SWAP_MLOCK) | |
1612 | ret = SWAP_MLOCK; | |
1da177e4 LT |
1613 | cursor += CLUSTER_SIZE; |
1614 | vma->vm_private_data = (void *) cursor; | |
1615 | if ((int)mapcount <= 0) | |
1616 | goto out; | |
1617 | } | |
1618 | vma->vm_private_data = (void *) max_nl_cursor; | |
1619 | } | |
3d48ae45 | 1620 | cond_resched(); |
1da177e4 LT |
1621 | max_nl_cursor += CLUSTER_SIZE; |
1622 | } while (max_nl_cursor <= max_nl_size); | |
1623 | ||
1624 | /* | |
1625 | * Don't loop forever (perhaps all the remaining pages are | |
1626 | * in locked vmas). Reset cursor on all unreserved nonlinear | |
1627 | * vmas, now forgetting on which ones it had fallen behind. | |
1628 | */ | |
6b2dbba8 | 1629 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.nonlinear) |
101d2be7 | 1630 | vma->vm_private_data = NULL; |
1da177e4 | 1631 | out: |
3d48ae45 | 1632 | mutex_unlock(&mapping->i_mmap_mutex); |
1da177e4 LT |
1633 | return ret; |
1634 | } | |
1635 | ||
1636 | /** | |
1637 | * try_to_unmap - try to remove all page table mappings to a page | |
1638 | * @page: the page to get unmapped | |
14fa31b8 | 1639 | * @flags: action and flags |
1da177e4 LT |
1640 | * |
1641 | * Tries to remove all the page table entries which are mapping this | |
1642 | * page, used in the pageout path. Caller must hold the page lock. | |
1643 | * Return values are: | |
1644 | * | |
1645 | * SWAP_SUCCESS - we succeeded in removing all mappings | |
1646 | * SWAP_AGAIN - we missed a mapping, try again later | |
1647 | * SWAP_FAIL - the page is unswappable | |
b291f000 | 1648 | * SWAP_MLOCK - page is mlocked. |
1da177e4 | 1649 | */ |
14fa31b8 | 1650 | int try_to_unmap(struct page *page, enum ttu_flags flags) |
1da177e4 LT |
1651 | { |
1652 | int ret; | |
1653 | ||
1da177e4 | 1654 | BUG_ON(!PageLocked(page)); |
91600e9e | 1655 | VM_BUG_ON(!PageHuge(page) && PageTransHuge(page)); |
1da177e4 | 1656 | |
5ad64688 HD |
1657 | if (unlikely(PageKsm(page))) |
1658 | ret = try_to_unmap_ksm(page, flags); | |
1659 | else if (PageAnon(page)) | |
14fa31b8 | 1660 | ret = try_to_unmap_anon(page, flags); |
1da177e4 | 1661 | else |
14fa31b8 | 1662 | ret = try_to_unmap_file(page, flags); |
b291f000 | 1663 | if (ret != SWAP_MLOCK && !page_mapped(page)) |
1da177e4 LT |
1664 | ret = SWAP_SUCCESS; |
1665 | return ret; | |
1666 | } | |
81b4082d | 1667 | |
b291f000 NP |
1668 | /** |
1669 | * try_to_munlock - try to munlock a page | |
1670 | * @page: the page to be munlocked | |
1671 | * | |
1672 | * Called from munlock code. Checks all of the VMAs mapping the page | |
1673 | * to make sure nobody else has this page mlocked. The page will be | |
1674 | * returned with PG_mlocked cleared if no other vmas have it mlocked. | |
1675 | * | |
1676 | * Return values are: | |
1677 | * | |
53f79acb | 1678 | * SWAP_AGAIN - no vma is holding page mlocked, or, |
b291f000 | 1679 | * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem |
5ad64688 | 1680 | * SWAP_FAIL - page cannot be located at present |
b291f000 NP |
1681 | * SWAP_MLOCK - page is now mlocked. |
1682 | */ | |
1683 | int try_to_munlock(struct page *page) | |
1684 | { | |
1685 | VM_BUG_ON(!PageLocked(page) || PageLRU(page)); | |
1686 | ||
5ad64688 HD |
1687 | if (unlikely(PageKsm(page))) |
1688 | return try_to_unmap_ksm(page, TTU_MUNLOCK); | |
1689 | else if (PageAnon(page)) | |
14fa31b8 | 1690 | return try_to_unmap_anon(page, TTU_MUNLOCK); |
b291f000 | 1691 | else |
14fa31b8 | 1692 | return try_to_unmap_file(page, TTU_MUNLOCK); |
b291f000 | 1693 | } |
e9995ef9 | 1694 | |
01d8b20d | 1695 | void __put_anon_vma(struct anon_vma *anon_vma) |
76545066 | 1696 | { |
01d8b20d | 1697 | struct anon_vma *root = anon_vma->root; |
76545066 | 1698 | |
65375ce7 | 1699 | anon_vma_free(anon_vma); |
6fa3eb70 | 1700 | if (root && root != anon_vma && atomic_dec_and_test(&root->refcount)) |
01d8b20d | 1701 | anon_vma_free(root); |
76545066 | 1702 | } |
76545066 | 1703 | |
e9995ef9 HD |
1704 | #ifdef CONFIG_MIGRATION |
1705 | /* | |
1706 | * rmap_walk() and its helpers rmap_walk_anon() and rmap_walk_file(): | |
1707 | * Called by migrate.c to remove migration ptes, but might be used more later. | |
1708 | */ | |
1709 | static int rmap_walk_anon(struct page *page, int (*rmap_one)(struct page *, | |
1710 | struct vm_area_struct *, unsigned long, void *), void *arg) | |
1711 | { | |
1712 | struct anon_vma *anon_vma; | |
bf181b9f | 1713 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); |
5beb4930 | 1714 | struct anon_vma_chain *avc; |
e9995ef9 HD |
1715 | int ret = SWAP_AGAIN; |
1716 | ||
1717 | /* | |
4fc3f1d6 | 1718 | * Note: remove_migration_ptes() cannot use page_lock_anon_vma_read() |
e9995ef9 | 1719 | * because that depends on page_mapped(); but not all its usages |
3f6c8272 MG |
1720 | * are holding mmap_sem. Users without mmap_sem are required to |
1721 | * take a reference count to prevent the anon_vma disappearing | |
e9995ef9 HD |
1722 | */ |
1723 | anon_vma = page_anon_vma(page); | |
1724 | if (!anon_vma) | |
1725 | return ret; | |
4fc3f1d6 | 1726 | anon_vma_lock_read(anon_vma); |
bf181b9f | 1727 | anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) { |
5beb4930 | 1728 | struct vm_area_struct *vma = avc->vma; |
e9995ef9 | 1729 | unsigned long address = vma_address(page, vma); |
e9995ef9 HD |
1730 | ret = rmap_one(page, vma, address, arg); |
1731 | if (ret != SWAP_AGAIN) | |
1732 | break; | |
1733 | } | |
4fc3f1d6 | 1734 | anon_vma_unlock_read(anon_vma); |
e9995ef9 HD |
1735 | return ret; |
1736 | } | |
1737 | ||
1738 | static int rmap_walk_file(struct page *page, int (*rmap_one)(struct page *, | |
1739 | struct vm_area_struct *, unsigned long, void *), void *arg) | |
1740 | { | |
1741 | struct address_space *mapping = page->mapping; | |
1742 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
1743 | struct vm_area_struct *vma; | |
e9995ef9 HD |
1744 | int ret = SWAP_AGAIN; |
1745 | ||
1746 | if (!mapping) | |
1747 | return ret; | |
3d48ae45 | 1748 | mutex_lock(&mapping->i_mmap_mutex); |
6b2dbba8 | 1749 | vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) { |
e9995ef9 | 1750 | unsigned long address = vma_address(page, vma); |
e9995ef9 HD |
1751 | ret = rmap_one(page, vma, address, arg); |
1752 | if (ret != SWAP_AGAIN) | |
1753 | break; | |
1754 | } | |
1755 | /* | |
1756 | * No nonlinear handling: being always shared, nonlinear vmas | |
1757 | * never contain migration ptes. Decide what to do about this | |
1758 | * limitation to linear when we need rmap_walk() on nonlinear. | |
1759 | */ | |
3d48ae45 | 1760 | mutex_unlock(&mapping->i_mmap_mutex); |
e9995ef9 HD |
1761 | return ret; |
1762 | } | |
1763 | ||
1764 | int rmap_walk(struct page *page, int (*rmap_one)(struct page *, | |
1765 | struct vm_area_struct *, unsigned long, void *), void *arg) | |
1766 | { | |
1767 | VM_BUG_ON(!PageLocked(page)); | |
1768 | ||
1769 | if (unlikely(PageKsm(page))) | |
1770 | return rmap_walk_ksm(page, rmap_one, arg); | |
1771 | else if (PageAnon(page)) | |
1772 | return rmap_walk_anon(page, rmap_one, arg); | |
1773 | else | |
1774 | return rmap_walk_file(page, rmap_one, arg); | |
1775 | } | |
1776 | #endif /* CONFIG_MIGRATION */ | |
0fe6e20b | 1777 | |
e3390f67 | 1778 | #ifdef CONFIG_HUGETLB_PAGE |
0fe6e20b NH |
1779 | /* |
1780 | * The following three functions are for anonymous (private mapped) hugepages. | |
1781 | * Unlike common anonymous pages, anonymous hugepages have no accounting code | |
1782 | * and no lru code, because we handle hugepages differently from common pages. | |
1783 | */ | |
1784 | static void __hugepage_set_anon_rmap(struct page *page, | |
1785 | struct vm_area_struct *vma, unsigned long address, int exclusive) | |
1786 | { | |
1787 | struct anon_vma *anon_vma = vma->anon_vma; | |
433abed6 | 1788 | |
0fe6e20b | 1789 | BUG_ON(!anon_vma); |
433abed6 NH |
1790 | |
1791 | if (PageAnon(page)) | |
1792 | return; | |
1793 | if (!exclusive) | |
1794 | anon_vma = anon_vma->root; | |
1795 | ||
0fe6e20b NH |
1796 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; |
1797 | page->mapping = (struct address_space *) anon_vma; | |
1798 | page->index = linear_page_index(vma, address); | |
1799 | } | |
1800 | ||
1801 | void hugepage_add_anon_rmap(struct page *page, | |
1802 | struct vm_area_struct *vma, unsigned long address) | |
1803 | { | |
1804 | struct anon_vma *anon_vma = vma->anon_vma; | |
1805 | int first; | |
a850ea30 NH |
1806 | |
1807 | BUG_ON(!PageLocked(page)); | |
0fe6e20b | 1808 | BUG_ON(!anon_vma); |
5dbe0af4 | 1809 | /* address might be in next vma when migration races vma_adjust */ |
0fe6e20b NH |
1810 | first = atomic_inc_and_test(&page->_mapcount); |
1811 | if (first) | |
1812 | __hugepage_set_anon_rmap(page, vma, address, 0); | |
1813 | } | |
1814 | ||
1815 | void hugepage_add_new_anon_rmap(struct page *page, | |
1816 | struct vm_area_struct *vma, unsigned long address) | |
1817 | { | |
1818 | BUG_ON(address < vma->vm_start || address >= vma->vm_end); | |
1819 | atomic_set(&page->_mapcount, 0); | |
1820 | __hugepage_set_anon_rmap(page, vma, address, 1); | |
1821 | } | |
e3390f67 | 1822 | #endif /* CONFIG_HUGETLB_PAGE */ |