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) | |
c8c06efa | 26 | * mapping->i_mmap_rwsem |
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) | |
c4843a75 GT |
33 | * mem_cgroup_{begin,end}_page_stat (memcg->move_lock) |
34 | * mapping->tree_lock (widely used) | |
250df6ed | 35 | * inode->i_lock (in set_page_dirty's __mark_inode_dirty) |
f758eeab | 36 | * bdi.wb->list_lock (in set_page_dirty's __mark_inode_dirty) |
82591e6e NP |
37 | * sb_lock (within inode_lock in fs/fs-writeback.c) |
38 | * mapping->tree_lock (widely used, in set_page_dirty, | |
39 | * in arch-dependent flush_dcache_mmap_lock, | |
f758eeab | 40 | * within bdi.wb->list_lock in __sync_single_inode) |
6a46079c | 41 | * |
5a505085 | 42 | * anon_vma->rwsem,mapping->i_mutex (memory_failure, collect_procs_anon) |
9b679320 | 43 | * ->tasklist_lock |
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> | |
b95f1b31 | 56 | #include <linux/export.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> |
ef5d437f | 61 | #include <linux/backing-dev.h> |
1da177e4 LT |
62 | |
63 | #include <asm/tlbflush.h> | |
64 | ||
72b252ae MG |
65 | #include <trace/events/tlb.h> |
66 | ||
b291f000 NP |
67 | #include "internal.h" |
68 | ||
fdd2e5f8 | 69 | static struct kmem_cache *anon_vma_cachep; |
5beb4930 | 70 | static struct kmem_cache *anon_vma_chain_cachep; |
fdd2e5f8 AB |
71 | |
72 | static inline struct anon_vma *anon_vma_alloc(void) | |
73 | { | |
01d8b20d PZ |
74 | struct anon_vma *anon_vma; |
75 | ||
76 | anon_vma = kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL); | |
77 | if (anon_vma) { | |
78 | atomic_set(&anon_vma->refcount, 1); | |
7a3ef208 KK |
79 | anon_vma->degree = 1; /* Reference for first vma */ |
80 | anon_vma->parent = anon_vma; | |
01d8b20d PZ |
81 | /* |
82 | * Initialise the anon_vma root to point to itself. If called | |
83 | * from fork, the root will be reset to the parents anon_vma. | |
84 | */ | |
85 | anon_vma->root = anon_vma; | |
86 | } | |
87 | ||
88 | return anon_vma; | |
fdd2e5f8 AB |
89 | } |
90 | ||
01d8b20d | 91 | static inline void anon_vma_free(struct anon_vma *anon_vma) |
fdd2e5f8 | 92 | { |
01d8b20d | 93 | VM_BUG_ON(atomic_read(&anon_vma->refcount)); |
88c22088 PZ |
94 | |
95 | /* | |
4fc3f1d6 | 96 | * Synchronize against page_lock_anon_vma_read() such that |
88c22088 PZ |
97 | * we can safely hold the lock without the anon_vma getting |
98 | * freed. | |
99 | * | |
100 | * Relies on the full mb implied by the atomic_dec_and_test() from | |
101 | * put_anon_vma() against the acquire barrier implied by | |
4fc3f1d6 | 102 | * down_read_trylock() from page_lock_anon_vma_read(). This orders: |
88c22088 | 103 | * |
4fc3f1d6 IM |
104 | * page_lock_anon_vma_read() VS put_anon_vma() |
105 | * down_read_trylock() atomic_dec_and_test() | |
88c22088 | 106 | * LOCK MB |
4fc3f1d6 | 107 | * atomic_read() rwsem_is_locked() |
88c22088 PZ |
108 | * |
109 | * LOCK should suffice since the actual taking of the lock must | |
110 | * happen _before_ what follows. | |
111 | */ | |
7f39dda9 | 112 | might_sleep(); |
5a505085 | 113 | if (rwsem_is_locked(&anon_vma->root->rwsem)) { |
4fc3f1d6 | 114 | anon_vma_lock_write(anon_vma); |
08b52706 | 115 | anon_vma_unlock_write(anon_vma); |
88c22088 PZ |
116 | } |
117 | ||
fdd2e5f8 AB |
118 | kmem_cache_free(anon_vma_cachep, anon_vma); |
119 | } | |
1da177e4 | 120 | |
dd34739c | 121 | static inline struct anon_vma_chain *anon_vma_chain_alloc(gfp_t gfp) |
5beb4930 | 122 | { |
dd34739c | 123 | return kmem_cache_alloc(anon_vma_chain_cachep, gfp); |
5beb4930 RR |
124 | } |
125 | ||
e574b5fd | 126 | static void anon_vma_chain_free(struct anon_vma_chain *anon_vma_chain) |
5beb4930 RR |
127 | { |
128 | kmem_cache_free(anon_vma_chain_cachep, anon_vma_chain); | |
129 | } | |
130 | ||
6583a843 KC |
131 | static void anon_vma_chain_link(struct vm_area_struct *vma, |
132 | struct anon_vma_chain *avc, | |
133 | struct anon_vma *anon_vma) | |
134 | { | |
135 | avc->vma = vma; | |
136 | avc->anon_vma = anon_vma; | |
137 | list_add(&avc->same_vma, &vma->anon_vma_chain); | |
bf181b9f | 138 | anon_vma_interval_tree_insert(avc, &anon_vma->rb_root); |
6583a843 KC |
139 | } |
140 | ||
d9d332e0 LT |
141 | /** |
142 | * anon_vma_prepare - attach an anon_vma to a memory region | |
143 | * @vma: the memory region in question | |
144 | * | |
145 | * This makes sure the memory mapping described by 'vma' has | |
146 | * an 'anon_vma' attached to it, so that we can associate the | |
147 | * anonymous pages mapped into it with that anon_vma. | |
148 | * | |
149 | * The common case will be that we already have one, but if | |
23a0790a | 150 | * not we either need to find an adjacent mapping that we |
d9d332e0 LT |
151 | * can re-use the anon_vma from (very common when the only |
152 | * reason for splitting a vma has been mprotect()), or we | |
153 | * allocate a new one. | |
154 | * | |
155 | * Anon-vma allocations are very subtle, because we may have | |
4fc3f1d6 | 156 | * optimistically looked up an anon_vma in page_lock_anon_vma_read() |
d9d332e0 LT |
157 | * and that may actually touch the spinlock even in the newly |
158 | * allocated vma (it depends on RCU to make sure that the | |
159 | * anon_vma isn't actually destroyed). | |
160 | * | |
161 | * As a result, we need to do proper anon_vma locking even | |
162 | * for the new allocation. At the same time, we do not want | |
163 | * to do any locking for the common case of already having | |
164 | * an anon_vma. | |
165 | * | |
166 | * This must be called with the mmap_sem held for reading. | |
167 | */ | |
1da177e4 LT |
168 | int anon_vma_prepare(struct vm_area_struct *vma) |
169 | { | |
170 | struct anon_vma *anon_vma = vma->anon_vma; | |
5beb4930 | 171 | struct anon_vma_chain *avc; |
1da177e4 LT |
172 | |
173 | might_sleep(); | |
174 | if (unlikely(!anon_vma)) { | |
175 | struct mm_struct *mm = vma->vm_mm; | |
d9d332e0 | 176 | struct anon_vma *allocated; |
1da177e4 | 177 | |
dd34739c | 178 | avc = anon_vma_chain_alloc(GFP_KERNEL); |
5beb4930 RR |
179 | if (!avc) |
180 | goto out_enomem; | |
181 | ||
1da177e4 | 182 | anon_vma = find_mergeable_anon_vma(vma); |
d9d332e0 LT |
183 | allocated = NULL; |
184 | if (!anon_vma) { | |
1da177e4 LT |
185 | anon_vma = anon_vma_alloc(); |
186 | if (unlikely(!anon_vma)) | |
5beb4930 | 187 | goto out_enomem_free_avc; |
1da177e4 | 188 | allocated = anon_vma; |
1da177e4 LT |
189 | } |
190 | ||
4fc3f1d6 | 191 | anon_vma_lock_write(anon_vma); |
1da177e4 LT |
192 | /* page_table_lock to protect against threads */ |
193 | spin_lock(&mm->page_table_lock); | |
194 | if (likely(!vma->anon_vma)) { | |
195 | vma->anon_vma = anon_vma; | |
6583a843 | 196 | anon_vma_chain_link(vma, avc, anon_vma); |
7a3ef208 KK |
197 | /* vma reference or self-parent link for new root */ |
198 | anon_vma->degree++; | |
1da177e4 | 199 | allocated = NULL; |
31f2b0eb | 200 | avc = NULL; |
1da177e4 LT |
201 | } |
202 | spin_unlock(&mm->page_table_lock); | |
08b52706 | 203 | anon_vma_unlock_write(anon_vma); |
31f2b0eb ON |
204 | |
205 | if (unlikely(allocated)) | |
01d8b20d | 206 | put_anon_vma(allocated); |
31f2b0eb | 207 | if (unlikely(avc)) |
5beb4930 | 208 | anon_vma_chain_free(avc); |
1da177e4 LT |
209 | } |
210 | return 0; | |
5beb4930 RR |
211 | |
212 | out_enomem_free_avc: | |
213 | anon_vma_chain_free(avc); | |
214 | out_enomem: | |
215 | return -ENOMEM; | |
1da177e4 LT |
216 | } |
217 | ||
bb4aa396 LT |
218 | /* |
219 | * This is a useful helper function for locking the anon_vma root as | |
220 | * we traverse the vma->anon_vma_chain, looping over anon_vma's that | |
221 | * have the same vma. | |
222 | * | |
223 | * Such anon_vma's should have the same root, so you'd expect to see | |
224 | * just a single mutex_lock for the whole traversal. | |
225 | */ | |
226 | static inline struct anon_vma *lock_anon_vma_root(struct anon_vma *root, struct anon_vma *anon_vma) | |
227 | { | |
228 | struct anon_vma *new_root = anon_vma->root; | |
229 | if (new_root != root) { | |
230 | if (WARN_ON_ONCE(root)) | |
5a505085 | 231 | up_write(&root->rwsem); |
bb4aa396 | 232 | root = new_root; |
5a505085 | 233 | down_write(&root->rwsem); |
bb4aa396 LT |
234 | } |
235 | return root; | |
236 | } | |
237 | ||
238 | static inline void unlock_anon_vma_root(struct anon_vma *root) | |
239 | { | |
240 | if (root) | |
5a505085 | 241 | up_write(&root->rwsem); |
bb4aa396 LT |
242 | } |
243 | ||
5beb4930 RR |
244 | /* |
245 | * Attach the anon_vmas from src to dst. | |
246 | * Returns 0 on success, -ENOMEM on failure. | |
7a3ef208 KK |
247 | * |
248 | * If dst->anon_vma is NULL this function tries to find and reuse existing | |
249 | * anon_vma which has no vmas and only one child anon_vma. This prevents | |
250 | * degradation of anon_vma hierarchy to endless linear chain in case of | |
251 | * constantly forking task. On the other hand, an anon_vma with more than one | |
252 | * child isn't reused even if there was no alive vma, thus rmap walker has a | |
253 | * good chance of avoiding scanning the whole hierarchy when it searches where | |
254 | * page is mapped. | |
5beb4930 RR |
255 | */ |
256 | int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src) | |
1da177e4 | 257 | { |
5beb4930 | 258 | struct anon_vma_chain *avc, *pavc; |
bb4aa396 | 259 | struct anon_vma *root = NULL; |
5beb4930 | 260 | |
646d87b4 | 261 | list_for_each_entry_reverse(pavc, &src->anon_vma_chain, same_vma) { |
bb4aa396 LT |
262 | struct anon_vma *anon_vma; |
263 | ||
dd34739c LT |
264 | avc = anon_vma_chain_alloc(GFP_NOWAIT | __GFP_NOWARN); |
265 | if (unlikely(!avc)) { | |
266 | unlock_anon_vma_root(root); | |
267 | root = NULL; | |
268 | avc = anon_vma_chain_alloc(GFP_KERNEL); | |
269 | if (!avc) | |
270 | goto enomem_failure; | |
271 | } | |
bb4aa396 LT |
272 | anon_vma = pavc->anon_vma; |
273 | root = lock_anon_vma_root(root, anon_vma); | |
274 | anon_vma_chain_link(dst, avc, anon_vma); | |
7a3ef208 KK |
275 | |
276 | /* | |
277 | * Reuse existing anon_vma if its degree lower than two, | |
278 | * that means it has no vma and only one anon_vma child. | |
279 | * | |
280 | * Do not chose parent anon_vma, otherwise first child | |
281 | * will always reuse it. Root anon_vma is never reused: | |
282 | * it has self-parent reference and at least one child. | |
283 | */ | |
284 | if (!dst->anon_vma && anon_vma != src->anon_vma && | |
285 | anon_vma->degree < 2) | |
286 | dst->anon_vma = anon_vma; | |
5beb4930 | 287 | } |
7a3ef208 KK |
288 | if (dst->anon_vma) |
289 | dst->anon_vma->degree++; | |
bb4aa396 | 290 | unlock_anon_vma_root(root); |
5beb4930 | 291 | return 0; |
1da177e4 | 292 | |
5beb4930 | 293 | enomem_failure: |
3fe89b3e LY |
294 | /* |
295 | * dst->anon_vma is dropped here otherwise its degree can be incorrectly | |
296 | * decremented in unlink_anon_vmas(). | |
297 | * We can safely do this because callers of anon_vma_clone() don't care | |
298 | * about dst->anon_vma if anon_vma_clone() failed. | |
299 | */ | |
300 | dst->anon_vma = NULL; | |
5beb4930 RR |
301 | unlink_anon_vmas(dst); |
302 | return -ENOMEM; | |
1da177e4 LT |
303 | } |
304 | ||
5beb4930 RR |
305 | /* |
306 | * Attach vma to its own anon_vma, as well as to the anon_vmas that | |
307 | * the corresponding VMA in the parent process is attached to. | |
308 | * Returns 0 on success, non-zero on failure. | |
309 | */ | |
310 | int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma) | |
1da177e4 | 311 | { |
5beb4930 RR |
312 | struct anon_vma_chain *avc; |
313 | struct anon_vma *anon_vma; | |
c4ea95d7 | 314 | int error; |
1da177e4 | 315 | |
5beb4930 RR |
316 | /* Don't bother if the parent process has no anon_vma here. */ |
317 | if (!pvma->anon_vma) | |
318 | return 0; | |
319 | ||
7a3ef208 KK |
320 | /* Drop inherited anon_vma, we'll reuse existing or allocate new. */ |
321 | vma->anon_vma = NULL; | |
322 | ||
5beb4930 RR |
323 | /* |
324 | * First, attach the new VMA to the parent VMA's anon_vmas, | |
325 | * so rmap can find non-COWed pages in child processes. | |
326 | */ | |
c4ea95d7 DF |
327 | error = anon_vma_clone(vma, pvma); |
328 | if (error) | |
329 | return error; | |
5beb4930 | 330 | |
7a3ef208 KK |
331 | /* An existing anon_vma has been reused, all done then. */ |
332 | if (vma->anon_vma) | |
333 | return 0; | |
334 | ||
5beb4930 RR |
335 | /* Then add our own anon_vma. */ |
336 | anon_vma = anon_vma_alloc(); | |
337 | if (!anon_vma) | |
338 | goto out_error; | |
dd34739c | 339 | avc = anon_vma_chain_alloc(GFP_KERNEL); |
5beb4930 RR |
340 | if (!avc) |
341 | goto out_error_free_anon_vma; | |
5c341ee1 RR |
342 | |
343 | /* | |
344 | * The root anon_vma's spinlock is the lock actually used when we | |
345 | * lock any of the anon_vmas in this anon_vma tree. | |
346 | */ | |
347 | anon_vma->root = pvma->anon_vma->root; | |
7a3ef208 | 348 | anon_vma->parent = pvma->anon_vma; |
76545066 | 349 | /* |
01d8b20d PZ |
350 | * With refcounts, an anon_vma can stay around longer than the |
351 | * process it belongs to. The root anon_vma needs to be pinned until | |
352 | * this anon_vma is freed, because the lock lives in the root. | |
76545066 RR |
353 | */ |
354 | get_anon_vma(anon_vma->root); | |
5beb4930 RR |
355 | /* Mark this anon_vma as the one where our new (COWed) pages go. */ |
356 | vma->anon_vma = anon_vma; | |
4fc3f1d6 | 357 | anon_vma_lock_write(anon_vma); |
5c341ee1 | 358 | anon_vma_chain_link(vma, avc, anon_vma); |
7a3ef208 | 359 | anon_vma->parent->degree++; |
08b52706 | 360 | anon_vma_unlock_write(anon_vma); |
5beb4930 RR |
361 | |
362 | return 0; | |
363 | ||
364 | out_error_free_anon_vma: | |
01d8b20d | 365 | put_anon_vma(anon_vma); |
5beb4930 | 366 | out_error: |
4946d54c | 367 | unlink_anon_vmas(vma); |
5beb4930 | 368 | return -ENOMEM; |
1da177e4 LT |
369 | } |
370 | ||
5beb4930 RR |
371 | void unlink_anon_vmas(struct vm_area_struct *vma) |
372 | { | |
373 | struct anon_vma_chain *avc, *next; | |
eee2acba | 374 | struct anon_vma *root = NULL; |
5beb4930 | 375 | |
5c341ee1 RR |
376 | /* |
377 | * Unlink each anon_vma chained to the VMA. This list is ordered | |
378 | * from newest to oldest, ensuring the root anon_vma gets freed last. | |
379 | */ | |
5beb4930 | 380 | list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) { |
eee2acba PZ |
381 | struct anon_vma *anon_vma = avc->anon_vma; |
382 | ||
383 | root = lock_anon_vma_root(root, anon_vma); | |
bf181b9f | 384 | anon_vma_interval_tree_remove(avc, &anon_vma->rb_root); |
eee2acba PZ |
385 | |
386 | /* | |
387 | * Leave empty anon_vmas on the list - we'll need | |
388 | * to free them outside the lock. | |
389 | */ | |
7a3ef208 KK |
390 | if (RB_EMPTY_ROOT(&anon_vma->rb_root)) { |
391 | anon_vma->parent->degree--; | |
eee2acba | 392 | continue; |
7a3ef208 | 393 | } |
eee2acba PZ |
394 | |
395 | list_del(&avc->same_vma); | |
396 | anon_vma_chain_free(avc); | |
397 | } | |
7a3ef208 KK |
398 | if (vma->anon_vma) |
399 | vma->anon_vma->degree--; | |
eee2acba PZ |
400 | unlock_anon_vma_root(root); |
401 | ||
402 | /* | |
403 | * Iterate the list once more, it now only contains empty and unlinked | |
404 | * anon_vmas, destroy them. Could not do before due to __put_anon_vma() | |
5a505085 | 405 | * needing to write-acquire the anon_vma->root->rwsem. |
eee2acba PZ |
406 | */ |
407 | list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) { | |
408 | struct anon_vma *anon_vma = avc->anon_vma; | |
409 | ||
7a3ef208 | 410 | BUG_ON(anon_vma->degree); |
eee2acba PZ |
411 | put_anon_vma(anon_vma); |
412 | ||
5beb4930 RR |
413 | list_del(&avc->same_vma); |
414 | anon_vma_chain_free(avc); | |
415 | } | |
416 | } | |
417 | ||
51cc5068 | 418 | static void anon_vma_ctor(void *data) |
1da177e4 | 419 | { |
a35afb83 | 420 | struct anon_vma *anon_vma = data; |
1da177e4 | 421 | |
5a505085 | 422 | init_rwsem(&anon_vma->rwsem); |
83813267 | 423 | atomic_set(&anon_vma->refcount, 0); |
bf181b9f | 424 | anon_vma->rb_root = RB_ROOT; |
1da177e4 LT |
425 | } |
426 | ||
427 | void __init anon_vma_init(void) | |
428 | { | |
429 | anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma), | |
20c2df83 | 430 | 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor); |
5beb4930 | 431 | anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain, SLAB_PANIC); |
1da177e4 LT |
432 | } |
433 | ||
434 | /* | |
6111e4ca PZ |
435 | * Getting a lock on a stable anon_vma from a page off the LRU is tricky! |
436 | * | |
437 | * Since there is no serialization what so ever against page_remove_rmap() | |
438 | * the best this function can do is return a locked anon_vma that might | |
439 | * have been relevant to this page. | |
440 | * | |
441 | * The page might have been remapped to a different anon_vma or the anon_vma | |
442 | * returned may already be freed (and even reused). | |
443 | * | |
bc658c96 PZ |
444 | * In case it was remapped to a different anon_vma, the new anon_vma will be a |
445 | * child of the old anon_vma, and the anon_vma lifetime rules will therefore | |
446 | * ensure that any anon_vma obtained from the page will still be valid for as | |
447 | * long as we observe page_mapped() [ hence all those page_mapped() tests ]. | |
448 | * | |
6111e4ca PZ |
449 | * All users of this function must be very careful when walking the anon_vma |
450 | * chain and verify that the page in question is indeed mapped in it | |
451 | * [ something equivalent to page_mapped_in_vma() ]. | |
452 | * | |
453 | * Since anon_vma's slab is DESTROY_BY_RCU and we know from page_remove_rmap() | |
454 | * that the anon_vma pointer from page->mapping is valid if there is a | |
455 | * mapcount, we can dereference the anon_vma after observing those. | |
1da177e4 | 456 | */ |
746b18d4 | 457 | struct anon_vma *page_get_anon_vma(struct page *page) |
1da177e4 | 458 | { |
746b18d4 | 459 | struct anon_vma *anon_vma = NULL; |
1da177e4 LT |
460 | unsigned long anon_mapping; |
461 | ||
462 | rcu_read_lock(); | |
4db0c3c2 | 463 | anon_mapping = (unsigned long)READ_ONCE(page->mapping); |
3ca7b3c5 | 464 | if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON) |
1da177e4 LT |
465 | goto out; |
466 | if (!page_mapped(page)) | |
467 | goto out; | |
468 | ||
469 | anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON); | |
746b18d4 PZ |
470 | if (!atomic_inc_not_zero(&anon_vma->refcount)) { |
471 | anon_vma = NULL; | |
472 | goto out; | |
473 | } | |
f1819427 HD |
474 | |
475 | /* | |
476 | * If this page is still mapped, then its anon_vma cannot have been | |
746b18d4 PZ |
477 | * freed. But if it has been unmapped, we have no security against the |
478 | * anon_vma structure being freed and reused (for another anon_vma: | |
479 | * SLAB_DESTROY_BY_RCU guarantees that - so the atomic_inc_not_zero() | |
480 | * above cannot corrupt). | |
f1819427 | 481 | */ |
746b18d4 | 482 | if (!page_mapped(page)) { |
7f39dda9 | 483 | rcu_read_unlock(); |
746b18d4 | 484 | put_anon_vma(anon_vma); |
7f39dda9 | 485 | return NULL; |
746b18d4 | 486 | } |
1da177e4 LT |
487 | out: |
488 | rcu_read_unlock(); | |
746b18d4 PZ |
489 | |
490 | return anon_vma; | |
491 | } | |
492 | ||
88c22088 PZ |
493 | /* |
494 | * Similar to page_get_anon_vma() except it locks the anon_vma. | |
495 | * | |
496 | * Its a little more complex as it tries to keep the fast path to a single | |
497 | * atomic op -- the trylock. If we fail the trylock, we fall back to getting a | |
498 | * reference like with page_get_anon_vma() and then block on the mutex. | |
499 | */ | |
4fc3f1d6 | 500 | struct anon_vma *page_lock_anon_vma_read(struct page *page) |
746b18d4 | 501 | { |
88c22088 | 502 | struct anon_vma *anon_vma = NULL; |
eee0f252 | 503 | struct anon_vma *root_anon_vma; |
88c22088 | 504 | unsigned long anon_mapping; |
746b18d4 | 505 | |
88c22088 | 506 | rcu_read_lock(); |
4db0c3c2 | 507 | anon_mapping = (unsigned long)READ_ONCE(page->mapping); |
88c22088 PZ |
508 | if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON) |
509 | goto out; | |
510 | if (!page_mapped(page)) | |
511 | goto out; | |
512 | ||
513 | anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON); | |
4db0c3c2 | 514 | root_anon_vma = READ_ONCE(anon_vma->root); |
4fc3f1d6 | 515 | if (down_read_trylock(&root_anon_vma->rwsem)) { |
88c22088 | 516 | /* |
eee0f252 HD |
517 | * If the page is still mapped, then this anon_vma is still |
518 | * its anon_vma, and holding the mutex ensures that it will | |
bc658c96 | 519 | * not go away, see anon_vma_free(). |
88c22088 | 520 | */ |
eee0f252 | 521 | if (!page_mapped(page)) { |
4fc3f1d6 | 522 | up_read(&root_anon_vma->rwsem); |
88c22088 PZ |
523 | anon_vma = NULL; |
524 | } | |
525 | goto out; | |
526 | } | |
746b18d4 | 527 | |
88c22088 PZ |
528 | /* trylock failed, we got to sleep */ |
529 | if (!atomic_inc_not_zero(&anon_vma->refcount)) { | |
530 | anon_vma = NULL; | |
531 | goto out; | |
532 | } | |
533 | ||
534 | if (!page_mapped(page)) { | |
7f39dda9 | 535 | rcu_read_unlock(); |
88c22088 | 536 | put_anon_vma(anon_vma); |
7f39dda9 | 537 | return NULL; |
88c22088 PZ |
538 | } |
539 | ||
540 | /* we pinned the anon_vma, its safe to sleep */ | |
541 | rcu_read_unlock(); | |
4fc3f1d6 | 542 | anon_vma_lock_read(anon_vma); |
88c22088 PZ |
543 | |
544 | if (atomic_dec_and_test(&anon_vma->refcount)) { | |
545 | /* | |
546 | * Oops, we held the last refcount, release the lock | |
547 | * and bail -- can't simply use put_anon_vma() because | |
4fc3f1d6 | 548 | * we'll deadlock on the anon_vma_lock_write() recursion. |
88c22088 | 549 | */ |
4fc3f1d6 | 550 | anon_vma_unlock_read(anon_vma); |
88c22088 PZ |
551 | __put_anon_vma(anon_vma); |
552 | anon_vma = NULL; | |
553 | } | |
554 | ||
555 | return anon_vma; | |
556 | ||
557 | out: | |
558 | rcu_read_unlock(); | |
746b18d4 | 559 | return anon_vma; |
34bbd704 ON |
560 | } |
561 | ||
4fc3f1d6 | 562 | void page_unlock_anon_vma_read(struct anon_vma *anon_vma) |
34bbd704 | 563 | { |
4fc3f1d6 | 564 | anon_vma_unlock_read(anon_vma); |
1da177e4 LT |
565 | } |
566 | ||
567 | /* | |
3ad33b24 | 568 | * At what user virtual address is page expected in @vma? |
1da177e4 | 569 | */ |
86c2ad19 ML |
570 | static inline unsigned long |
571 | __vma_address(struct page *page, struct vm_area_struct *vma) | |
1da177e4 | 572 | { |
a0f7a756 | 573 | pgoff_t pgoff = page_to_pgoff(page); |
86c2ad19 ML |
574 | return vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); |
575 | } | |
576 | ||
577 | inline unsigned long | |
578 | vma_address(struct page *page, struct vm_area_struct *vma) | |
579 | { | |
580 | unsigned long address = __vma_address(page, vma); | |
581 | ||
582 | /* page should be within @vma mapping range */ | |
81d1b09c | 583 | VM_BUG_ON_VMA(address < vma->vm_start || address >= vma->vm_end, vma); |
86c2ad19 | 584 | |
1da177e4 LT |
585 | return address; |
586 | } | |
587 | ||
72b252ae MG |
588 | #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH |
589 | static void percpu_flush_tlb_batch_pages(void *data) | |
590 | { | |
591 | /* | |
592 | * All TLB entries are flushed on the assumption that it is | |
593 | * cheaper to flush all TLBs and let them be refilled than | |
594 | * flushing individual PFNs. Note that we do not track mm's | |
595 | * to flush as that might simply be multiple full TLB flushes | |
596 | * for no gain. | |
597 | */ | |
598 | count_vm_tlb_event(NR_TLB_REMOTE_FLUSH_RECEIVED); | |
599 | flush_tlb_local(); | |
600 | } | |
601 | ||
602 | /* | |
603 | * Flush TLB entries for recently unmapped pages from remote CPUs. It is | |
604 | * important if a PTE was dirty when it was unmapped that it's flushed | |
605 | * before any IO is initiated on the page to prevent lost writes. Similarly, | |
606 | * it must be flushed before freeing to prevent data leakage. | |
607 | */ | |
608 | void try_to_unmap_flush(void) | |
609 | { | |
610 | struct tlbflush_unmap_batch *tlb_ubc = ¤t->tlb_ubc; | |
611 | int cpu; | |
612 | ||
613 | if (!tlb_ubc->flush_required) | |
614 | return; | |
615 | ||
616 | cpu = get_cpu(); | |
617 | ||
618 | trace_tlb_flush(TLB_REMOTE_SHOOTDOWN, -1UL); | |
619 | ||
620 | if (cpumask_test_cpu(cpu, &tlb_ubc->cpumask)) | |
621 | percpu_flush_tlb_batch_pages(&tlb_ubc->cpumask); | |
622 | ||
623 | if (cpumask_any_but(&tlb_ubc->cpumask, cpu) < nr_cpu_ids) { | |
624 | smp_call_function_many(&tlb_ubc->cpumask, | |
625 | percpu_flush_tlb_batch_pages, (void *)tlb_ubc, true); | |
626 | } | |
627 | cpumask_clear(&tlb_ubc->cpumask); | |
628 | tlb_ubc->flush_required = false; | |
d950c947 | 629 | tlb_ubc->writable = false; |
72b252ae MG |
630 | put_cpu(); |
631 | } | |
632 | ||
d950c947 MG |
633 | /* Flush iff there are potentially writable TLB entries that can race with IO */ |
634 | void try_to_unmap_flush_dirty(void) | |
635 | { | |
636 | struct tlbflush_unmap_batch *tlb_ubc = ¤t->tlb_ubc; | |
637 | ||
638 | if (tlb_ubc->writable) | |
639 | try_to_unmap_flush(); | |
640 | } | |
641 | ||
72b252ae | 642 | static void set_tlb_ubc_flush_pending(struct mm_struct *mm, |
d950c947 | 643 | struct page *page, bool writable) |
72b252ae MG |
644 | { |
645 | struct tlbflush_unmap_batch *tlb_ubc = ¤t->tlb_ubc; | |
646 | ||
647 | cpumask_or(&tlb_ubc->cpumask, &tlb_ubc->cpumask, mm_cpumask(mm)); | |
648 | tlb_ubc->flush_required = true; | |
d950c947 MG |
649 | |
650 | /* | |
651 | * If the PTE was dirty then it's best to assume it's writable. The | |
652 | * caller must use try_to_unmap_flush_dirty() or try_to_unmap_flush() | |
653 | * before the page is queued for IO. | |
654 | */ | |
655 | if (writable) | |
656 | tlb_ubc->writable = true; | |
72b252ae MG |
657 | } |
658 | ||
659 | /* | |
660 | * Returns true if the TLB flush should be deferred to the end of a batch of | |
661 | * unmap operations to reduce IPIs. | |
662 | */ | |
663 | static bool should_defer_flush(struct mm_struct *mm, enum ttu_flags flags) | |
664 | { | |
665 | bool should_defer = false; | |
666 | ||
667 | if (!(flags & TTU_BATCH_FLUSH)) | |
668 | return false; | |
669 | ||
670 | /* If remote CPUs need to be flushed then defer batch the flush */ | |
671 | if (cpumask_any_but(mm_cpumask(mm), get_cpu()) < nr_cpu_ids) | |
672 | should_defer = true; | |
673 | put_cpu(); | |
674 | ||
675 | return should_defer; | |
676 | } | |
677 | #else | |
678 | static void set_tlb_ubc_flush_pending(struct mm_struct *mm, | |
d950c947 | 679 | struct page *page, bool writable) |
72b252ae MG |
680 | { |
681 | } | |
682 | ||
683 | static bool should_defer_flush(struct mm_struct *mm, enum ttu_flags flags) | |
684 | { | |
685 | return false; | |
686 | } | |
687 | #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */ | |
688 | ||
1da177e4 | 689 | /* |
bf89c8c8 | 690 | * At what user virtual address is page expected in vma? |
ab941e0f | 691 | * Caller should check the page is actually part of the vma. |
1da177e4 LT |
692 | */ |
693 | unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma) | |
694 | { | |
86c2ad19 | 695 | unsigned long address; |
21d0d443 | 696 | if (PageAnon(page)) { |
4829b906 HD |
697 | struct anon_vma *page__anon_vma = page_anon_vma(page); |
698 | /* | |
699 | * Note: swapoff's unuse_vma() is more efficient with this | |
700 | * check, and needs it to match anon_vma when KSM is active. | |
701 | */ | |
702 | if (!vma->anon_vma || !page__anon_vma || | |
703 | vma->anon_vma->root != page__anon_vma->root) | |
21d0d443 | 704 | return -EFAULT; |
27ba0644 KS |
705 | } else if (page->mapping) { |
706 | if (!vma->vm_file || vma->vm_file->f_mapping != page->mapping) | |
1da177e4 LT |
707 | return -EFAULT; |
708 | } else | |
709 | return -EFAULT; | |
86c2ad19 ML |
710 | address = __vma_address(page, vma); |
711 | if (unlikely(address < vma->vm_start || address >= vma->vm_end)) | |
712 | return -EFAULT; | |
713 | return address; | |
1da177e4 LT |
714 | } |
715 | ||
6219049a BL |
716 | pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address) |
717 | { | |
718 | pgd_t *pgd; | |
719 | pud_t *pud; | |
720 | pmd_t *pmd = NULL; | |
f72e7dcd | 721 | pmd_t pmde; |
6219049a BL |
722 | |
723 | pgd = pgd_offset(mm, address); | |
724 | if (!pgd_present(*pgd)) | |
725 | goto out; | |
726 | ||
727 | pud = pud_offset(pgd, address); | |
728 | if (!pud_present(*pud)) | |
729 | goto out; | |
730 | ||
731 | pmd = pmd_offset(pud, address); | |
f72e7dcd | 732 | /* |
8809aa2d | 733 | * Some THP functions use the sequence pmdp_huge_clear_flush(), set_pmd_at() |
f72e7dcd HD |
734 | * without holding anon_vma lock for write. So when looking for a |
735 | * genuine pmde (in which to find pte), test present and !THP together. | |
736 | */ | |
e37c6982 CB |
737 | pmde = *pmd; |
738 | barrier(); | |
f72e7dcd | 739 | if (!pmd_present(pmde) || pmd_trans_huge(pmde)) |
6219049a BL |
740 | pmd = NULL; |
741 | out: | |
742 | return pmd; | |
743 | } | |
744 | ||
81b4082d ND |
745 | /* |
746 | * Check that @page is mapped at @address into @mm. | |
747 | * | |
479db0bf NP |
748 | * If @sync is false, page_check_address may perform a racy check to avoid |
749 | * the page table lock when the pte is not present (helpful when reclaiming | |
750 | * highly shared pages). | |
751 | * | |
b8072f09 | 752 | * On success returns with pte mapped and locked. |
81b4082d | 753 | */ |
e9a81a82 | 754 | pte_t *__page_check_address(struct page *page, struct mm_struct *mm, |
479db0bf | 755 | unsigned long address, spinlock_t **ptlp, int sync) |
81b4082d | 756 | { |
81b4082d ND |
757 | pmd_t *pmd; |
758 | pte_t *pte; | |
c0718806 | 759 | spinlock_t *ptl; |
81b4082d | 760 | |
0fe6e20b | 761 | if (unlikely(PageHuge(page))) { |
98398c32 | 762 | /* when pud is not present, pte will be NULL */ |
0fe6e20b | 763 | pte = huge_pte_offset(mm, address); |
98398c32 JW |
764 | if (!pte) |
765 | return NULL; | |
766 | ||
cb900f41 | 767 | ptl = huge_pte_lockptr(page_hstate(page), mm, pte); |
0fe6e20b NH |
768 | goto check; |
769 | } | |
770 | ||
6219049a BL |
771 | pmd = mm_find_pmd(mm, address); |
772 | if (!pmd) | |
c0718806 HD |
773 | return NULL; |
774 | ||
c0718806 HD |
775 | pte = pte_offset_map(pmd, address); |
776 | /* Make a quick check before getting the lock */ | |
479db0bf | 777 | if (!sync && !pte_present(*pte)) { |
c0718806 HD |
778 | pte_unmap(pte); |
779 | return NULL; | |
780 | } | |
781 | ||
4c21e2f2 | 782 | ptl = pte_lockptr(mm, pmd); |
0fe6e20b | 783 | check: |
c0718806 HD |
784 | spin_lock(ptl); |
785 | if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) { | |
786 | *ptlp = ptl; | |
787 | return pte; | |
81b4082d | 788 | } |
c0718806 HD |
789 | pte_unmap_unlock(pte, ptl); |
790 | return NULL; | |
81b4082d ND |
791 | } |
792 | ||
b291f000 NP |
793 | /** |
794 | * page_mapped_in_vma - check whether a page is really mapped in a VMA | |
795 | * @page: the page to test | |
796 | * @vma: the VMA to test | |
797 | * | |
798 | * Returns 1 if the page is mapped into the page tables of the VMA, 0 | |
799 | * if the page is not mapped into the page tables of this VMA. Only | |
800 | * valid for normal file or anonymous VMAs. | |
801 | */ | |
6a46079c | 802 | int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma) |
b291f000 NP |
803 | { |
804 | unsigned long address; | |
805 | pte_t *pte; | |
806 | spinlock_t *ptl; | |
807 | ||
86c2ad19 ML |
808 | address = __vma_address(page, vma); |
809 | if (unlikely(address < vma->vm_start || address >= vma->vm_end)) | |
b291f000 NP |
810 | return 0; |
811 | pte = page_check_address(page, vma->vm_mm, address, &ptl, 1); | |
812 | if (!pte) /* the page is not in this mm */ | |
813 | return 0; | |
814 | pte_unmap_unlock(pte, ptl); | |
815 | ||
816 | return 1; | |
817 | } | |
818 | ||
9f32624b JK |
819 | struct page_referenced_arg { |
820 | int mapcount; | |
821 | int referenced; | |
822 | unsigned long vm_flags; | |
823 | struct mem_cgroup *memcg; | |
824 | }; | |
1da177e4 | 825 | /* |
9f32624b | 826 | * arg: page_referenced_arg will be passed |
1da177e4 | 827 | */ |
ac769501 | 828 | static int page_referenced_one(struct page *page, struct vm_area_struct *vma, |
9f32624b | 829 | unsigned long address, void *arg) |
1da177e4 LT |
830 | { |
831 | struct mm_struct *mm = vma->vm_mm; | |
117b0791 | 832 | spinlock_t *ptl; |
1da177e4 | 833 | int referenced = 0; |
9f32624b | 834 | struct page_referenced_arg *pra = arg; |
1da177e4 | 835 | |
71e3aac0 AA |
836 | if (unlikely(PageTransHuge(page))) { |
837 | pmd_t *pmd; | |
838 | ||
2da28bfd AA |
839 | /* |
840 | * rmap might return false positives; we must filter | |
841 | * these out using page_check_address_pmd(). | |
842 | */ | |
71e3aac0 | 843 | pmd = page_check_address_pmd(page, mm, address, |
117b0791 KS |
844 | PAGE_CHECK_ADDRESS_PMD_FLAG, &ptl); |
845 | if (!pmd) | |
9f32624b | 846 | return SWAP_AGAIN; |
2da28bfd AA |
847 | |
848 | if (vma->vm_flags & VM_LOCKED) { | |
117b0791 | 849 | spin_unlock(ptl); |
9f32624b JK |
850 | pra->vm_flags |= VM_LOCKED; |
851 | return SWAP_FAIL; /* To break the loop */ | |
2da28bfd AA |
852 | } |
853 | ||
854 | /* go ahead even if the pmd is pmd_trans_splitting() */ | |
855 | if (pmdp_clear_flush_young_notify(vma, address, pmd)) | |
71e3aac0 | 856 | referenced++; |
117b0791 | 857 | spin_unlock(ptl); |
71e3aac0 AA |
858 | } else { |
859 | pte_t *pte; | |
71e3aac0 | 860 | |
2da28bfd AA |
861 | /* |
862 | * rmap might return false positives; we must filter | |
863 | * these out using page_check_address(). | |
864 | */ | |
71e3aac0 AA |
865 | pte = page_check_address(page, mm, address, &ptl, 0); |
866 | if (!pte) | |
9f32624b | 867 | return SWAP_AGAIN; |
71e3aac0 | 868 | |
2da28bfd AA |
869 | if (vma->vm_flags & VM_LOCKED) { |
870 | pte_unmap_unlock(pte, ptl); | |
9f32624b JK |
871 | pra->vm_flags |= VM_LOCKED; |
872 | return SWAP_FAIL; /* To break the loop */ | |
2da28bfd AA |
873 | } |
874 | ||
71e3aac0 AA |
875 | if (ptep_clear_flush_young_notify(vma, address, pte)) { |
876 | /* | |
877 | * Don't treat a reference through a sequentially read | |
878 | * mapping as such. If the page has been used in | |
879 | * another mapping, we will catch it; if this other | |
880 | * mapping is already gone, the unmap path will have | |
881 | * set PG_referenced or activated the page. | |
882 | */ | |
64363aad | 883 | if (likely(!(vma->vm_flags & VM_SEQ_READ))) |
71e3aac0 AA |
884 | referenced++; |
885 | } | |
886 | pte_unmap_unlock(pte, ptl); | |
887 | } | |
888 | ||
9f32624b JK |
889 | if (referenced) { |
890 | pra->referenced++; | |
891 | pra->vm_flags |= vma->vm_flags; | |
1da177e4 | 892 | } |
34bbd704 | 893 | |
9f32624b JK |
894 | pra->mapcount--; |
895 | if (!pra->mapcount) | |
896 | return SWAP_SUCCESS; /* To break the loop */ | |
897 | ||
898 | return SWAP_AGAIN; | |
1da177e4 LT |
899 | } |
900 | ||
9f32624b | 901 | static bool invalid_page_referenced_vma(struct vm_area_struct *vma, void *arg) |
1da177e4 | 902 | { |
9f32624b JK |
903 | struct page_referenced_arg *pra = arg; |
904 | struct mem_cgroup *memcg = pra->memcg; | |
1da177e4 | 905 | |
9f32624b JK |
906 | if (!mm_match_cgroup(vma->vm_mm, memcg)) |
907 | return true; | |
1da177e4 | 908 | |
9f32624b | 909 | return false; |
1da177e4 LT |
910 | } |
911 | ||
912 | /** | |
913 | * page_referenced - test if the page was referenced | |
914 | * @page: the page to test | |
915 | * @is_locked: caller holds lock on the page | |
72835c86 | 916 | * @memcg: target memory cgroup |
6fe6b7e3 | 917 | * @vm_flags: collect encountered vma->vm_flags who actually referenced the page |
1da177e4 LT |
918 | * |
919 | * Quick test_and_clear_referenced for all mappings to a page, | |
920 | * returns the number of ptes which referenced the page. | |
921 | */ | |
6fe6b7e3 WF |
922 | int page_referenced(struct page *page, |
923 | int is_locked, | |
72835c86 | 924 | struct mem_cgroup *memcg, |
6fe6b7e3 | 925 | unsigned long *vm_flags) |
1da177e4 | 926 | { |
9f32624b | 927 | int ret; |
5ad64688 | 928 | int we_locked = 0; |
9f32624b JK |
929 | struct page_referenced_arg pra = { |
930 | .mapcount = page_mapcount(page), | |
931 | .memcg = memcg, | |
932 | }; | |
933 | struct rmap_walk_control rwc = { | |
934 | .rmap_one = page_referenced_one, | |
935 | .arg = (void *)&pra, | |
936 | .anon_lock = page_lock_anon_vma_read, | |
937 | }; | |
1da177e4 | 938 | |
6fe6b7e3 | 939 | *vm_flags = 0; |
9f32624b JK |
940 | if (!page_mapped(page)) |
941 | return 0; | |
942 | ||
943 | if (!page_rmapping(page)) | |
944 | return 0; | |
945 | ||
946 | if (!is_locked && (!PageAnon(page) || PageKsm(page))) { | |
947 | we_locked = trylock_page(page); | |
948 | if (!we_locked) | |
949 | return 1; | |
1da177e4 | 950 | } |
9f32624b JK |
951 | |
952 | /* | |
953 | * If we are reclaiming on behalf of a cgroup, skip | |
954 | * counting on behalf of references from different | |
955 | * cgroups | |
956 | */ | |
957 | if (memcg) { | |
958 | rwc.invalid_vma = invalid_page_referenced_vma; | |
959 | } | |
960 | ||
961 | ret = rmap_walk(page, &rwc); | |
962 | *vm_flags = pra.vm_flags; | |
963 | ||
964 | if (we_locked) | |
965 | unlock_page(page); | |
966 | ||
967 | return pra.referenced; | |
1da177e4 LT |
968 | } |
969 | ||
1cb1729b | 970 | static int page_mkclean_one(struct page *page, struct vm_area_struct *vma, |
9853a407 | 971 | unsigned long address, void *arg) |
d08b3851 PZ |
972 | { |
973 | struct mm_struct *mm = vma->vm_mm; | |
c2fda5fe | 974 | pte_t *pte; |
d08b3851 PZ |
975 | spinlock_t *ptl; |
976 | int ret = 0; | |
9853a407 | 977 | int *cleaned = arg; |
d08b3851 | 978 | |
479db0bf | 979 | pte = page_check_address(page, mm, address, &ptl, 1); |
d08b3851 PZ |
980 | if (!pte) |
981 | goto out; | |
982 | ||
c2fda5fe PZ |
983 | if (pte_dirty(*pte) || pte_write(*pte)) { |
984 | pte_t entry; | |
d08b3851 | 985 | |
c2fda5fe | 986 | flush_cache_page(vma, address, pte_pfn(*pte)); |
2ec74c3e | 987 | entry = ptep_clear_flush(vma, address, pte); |
c2fda5fe PZ |
988 | entry = pte_wrprotect(entry); |
989 | entry = pte_mkclean(entry); | |
d6e88e67 | 990 | set_pte_at(mm, address, pte, entry); |
c2fda5fe PZ |
991 | ret = 1; |
992 | } | |
d08b3851 | 993 | |
d08b3851 | 994 | pte_unmap_unlock(pte, ptl); |
2ec74c3e | 995 | |
9853a407 | 996 | if (ret) { |
2ec74c3e | 997 | mmu_notifier_invalidate_page(mm, address); |
9853a407 JK |
998 | (*cleaned)++; |
999 | } | |
d08b3851 | 1000 | out: |
9853a407 | 1001 | return SWAP_AGAIN; |
d08b3851 PZ |
1002 | } |
1003 | ||
9853a407 | 1004 | static bool invalid_mkclean_vma(struct vm_area_struct *vma, void *arg) |
d08b3851 | 1005 | { |
9853a407 | 1006 | if (vma->vm_flags & VM_SHARED) |
871beb8c | 1007 | return false; |
d08b3851 | 1008 | |
871beb8c | 1009 | return true; |
d08b3851 PZ |
1010 | } |
1011 | ||
1012 | int page_mkclean(struct page *page) | |
1013 | { | |
9853a407 JK |
1014 | int cleaned = 0; |
1015 | struct address_space *mapping; | |
1016 | struct rmap_walk_control rwc = { | |
1017 | .arg = (void *)&cleaned, | |
1018 | .rmap_one = page_mkclean_one, | |
1019 | .invalid_vma = invalid_mkclean_vma, | |
1020 | }; | |
d08b3851 PZ |
1021 | |
1022 | BUG_ON(!PageLocked(page)); | |
1023 | ||
9853a407 JK |
1024 | if (!page_mapped(page)) |
1025 | return 0; | |
1026 | ||
1027 | mapping = page_mapping(page); | |
1028 | if (!mapping) | |
1029 | return 0; | |
1030 | ||
1031 | rmap_walk(page, &rwc); | |
d08b3851 | 1032 | |
9853a407 | 1033 | return cleaned; |
d08b3851 | 1034 | } |
60b59bea | 1035 | EXPORT_SYMBOL_GPL(page_mkclean); |
d08b3851 | 1036 | |
c44b6743 RR |
1037 | /** |
1038 | * page_move_anon_rmap - move a page to our anon_vma | |
1039 | * @page: the page to move to our anon_vma | |
1040 | * @vma: the vma the page belongs to | |
1041 | * @address: the user virtual address mapped | |
1042 | * | |
1043 | * When a page belongs exclusively to one process after a COW event, | |
1044 | * that page can be moved into the anon_vma that belongs to just that | |
1045 | * process, so the rmap code will not search the parent or sibling | |
1046 | * processes. | |
1047 | */ | |
1048 | void page_move_anon_rmap(struct page *page, | |
1049 | struct vm_area_struct *vma, unsigned long address) | |
1050 | { | |
1051 | struct anon_vma *anon_vma = vma->anon_vma; | |
1052 | ||
309381fe | 1053 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
81d1b09c | 1054 | VM_BUG_ON_VMA(!anon_vma, vma); |
309381fe | 1055 | VM_BUG_ON_PAGE(page->index != linear_page_index(vma, address), page); |
c44b6743 RR |
1056 | |
1057 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; | |
414e2fb8 VD |
1058 | /* |
1059 | * Ensure that anon_vma and the PAGE_MAPPING_ANON bit are written | |
1060 | * simultaneously, so a concurrent reader (eg page_referenced()'s | |
1061 | * PageAnon()) will not see one without the other. | |
1062 | */ | |
1063 | WRITE_ONCE(page->mapping, (struct address_space *) anon_vma); | |
c44b6743 RR |
1064 | } |
1065 | ||
9617d95e | 1066 | /** |
4e1c1975 AK |
1067 | * __page_set_anon_rmap - set up new anonymous rmap |
1068 | * @page: Page to add to rmap | |
1069 | * @vma: VM area to add page to. | |
1070 | * @address: User virtual address of the mapping | |
e8a03feb | 1071 | * @exclusive: the page is exclusively owned by the current process |
9617d95e NP |
1072 | */ |
1073 | static void __page_set_anon_rmap(struct page *page, | |
e8a03feb | 1074 | struct vm_area_struct *vma, unsigned long address, int exclusive) |
9617d95e | 1075 | { |
e8a03feb | 1076 | struct anon_vma *anon_vma = vma->anon_vma; |
ea90002b | 1077 | |
e8a03feb | 1078 | BUG_ON(!anon_vma); |
ea90002b | 1079 | |
4e1c1975 AK |
1080 | if (PageAnon(page)) |
1081 | return; | |
1082 | ||
ea90002b | 1083 | /* |
e8a03feb RR |
1084 | * If the page isn't exclusively mapped into this vma, |
1085 | * we must use the _oldest_ possible anon_vma for the | |
1086 | * page mapping! | |
ea90002b | 1087 | */ |
4e1c1975 | 1088 | if (!exclusive) |
288468c3 | 1089 | anon_vma = anon_vma->root; |
9617d95e | 1090 | |
9617d95e NP |
1091 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; |
1092 | page->mapping = (struct address_space *) anon_vma; | |
9617d95e | 1093 | page->index = linear_page_index(vma, address); |
9617d95e NP |
1094 | } |
1095 | ||
c97a9e10 | 1096 | /** |
43d8eac4 | 1097 | * __page_check_anon_rmap - sanity check anonymous rmap addition |
c97a9e10 NP |
1098 | * @page: the page to add the mapping to |
1099 | * @vma: the vm area in which the mapping is added | |
1100 | * @address: the user virtual address mapped | |
1101 | */ | |
1102 | static void __page_check_anon_rmap(struct page *page, | |
1103 | struct vm_area_struct *vma, unsigned long address) | |
1104 | { | |
1105 | #ifdef CONFIG_DEBUG_VM | |
1106 | /* | |
1107 | * The page's anon-rmap details (mapping and index) are guaranteed to | |
1108 | * be set up correctly at this point. | |
1109 | * | |
1110 | * We have exclusion against page_add_anon_rmap because the caller | |
1111 | * always holds the page locked, except if called from page_dup_rmap, | |
1112 | * in which case the page is already known to be setup. | |
1113 | * | |
1114 | * We have exclusion against page_add_new_anon_rmap because those pages | |
1115 | * are initially only visible via the pagetables, and the pte is locked | |
1116 | * over the call to page_add_new_anon_rmap. | |
1117 | */ | |
44ab57a0 | 1118 | BUG_ON(page_anon_vma(page)->root != vma->anon_vma->root); |
c97a9e10 NP |
1119 | BUG_ON(page->index != linear_page_index(vma, address)); |
1120 | #endif | |
1121 | } | |
1122 | ||
1da177e4 LT |
1123 | /** |
1124 | * page_add_anon_rmap - add pte mapping to an anonymous page | |
1125 | * @page: the page to add the mapping to | |
1126 | * @vma: the vm area in which the mapping is added | |
1127 | * @address: the user virtual address mapped | |
1128 | * | |
5ad64688 | 1129 | * The caller needs to hold the pte lock, and the page must be locked in |
80e14822 HD |
1130 | * the anon_vma case: to serialize mapping,index checking after setting, |
1131 | * and to ensure that PageAnon is not being upgraded racily to PageKsm | |
1132 | * (but PageKsm is never downgraded to PageAnon). | |
1da177e4 LT |
1133 | */ |
1134 | void page_add_anon_rmap(struct page *page, | |
1135 | struct vm_area_struct *vma, unsigned long address) | |
ad8c2ee8 RR |
1136 | { |
1137 | do_page_add_anon_rmap(page, vma, address, 0); | |
1138 | } | |
1139 | ||
1140 | /* | |
1141 | * Special version of the above for do_swap_page, which often runs | |
1142 | * into pages that are exclusively owned by the current process. | |
1143 | * Everybody else should continue to use page_add_anon_rmap above. | |
1144 | */ | |
1145 | void do_page_add_anon_rmap(struct page *page, | |
1146 | struct vm_area_struct *vma, unsigned long address, int exclusive) | |
1da177e4 | 1147 | { |
5ad64688 | 1148 | int first = atomic_inc_and_test(&page->_mapcount); |
79134171 | 1149 | if (first) { |
bea04b07 JZ |
1150 | /* |
1151 | * We use the irq-unsafe __{inc|mod}_zone_page_stat because | |
1152 | * these counters are not modified in interrupt context, and | |
1153 | * pte lock(a spinlock) is held, which implies preemption | |
1154 | * disabled. | |
1155 | */ | |
3cd14fcd | 1156 | if (PageTransHuge(page)) |
79134171 AA |
1157 | __inc_zone_page_state(page, |
1158 | NR_ANON_TRANSPARENT_HUGEPAGES); | |
3cd14fcd KS |
1159 | __mod_zone_page_state(page_zone(page), NR_ANON_PAGES, |
1160 | hpage_nr_pages(page)); | |
79134171 | 1161 | } |
5ad64688 HD |
1162 | if (unlikely(PageKsm(page))) |
1163 | return; | |
1164 | ||
309381fe | 1165 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
5dbe0af4 | 1166 | /* address might be in next vma when migration races vma_adjust */ |
5ad64688 | 1167 | if (first) |
ad8c2ee8 | 1168 | __page_set_anon_rmap(page, vma, address, exclusive); |
69029cd5 | 1169 | else |
c97a9e10 | 1170 | __page_check_anon_rmap(page, vma, address); |
1da177e4 LT |
1171 | } |
1172 | ||
43d8eac4 | 1173 | /** |
9617d95e NP |
1174 | * page_add_new_anon_rmap - add pte mapping to a new anonymous page |
1175 | * @page: the page to add the mapping to | |
1176 | * @vma: the vm area in which the mapping is added | |
1177 | * @address: the user virtual address mapped | |
1178 | * | |
1179 | * Same as page_add_anon_rmap but must only be called on *new* pages. | |
1180 | * This means the inc-and-test can be bypassed. | |
c97a9e10 | 1181 | * Page does not have to be locked. |
9617d95e NP |
1182 | */ |
1183 | void page_add_new_anon_rmap(struct page *page, | |
1184 | struct vm_area_struct *vma, unsigned long address) | |
1185 | { | |
81d1b09c | 1186 | VM_BUG_ON_VMA(address < vma->vm_start || address >= vma->vm_end, vma); |
cbf84b7a HD |
1187 | SetPageSwapBacked(page); |
1188 | atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */ | |
3cd14fcd | 1189 | if (PageTransHuge(page)) |
79134171 | 1190 | __inc_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES); |
3cd14fcd KS |
1191 | __mod_zone_page_state(page_zone(page), NR_ANON_PAGES, |
1192 | hpage_nr_pages(page)); | |
e8a03feb | 1193 | __page_set_anon_rmap(page, vma, address, 1); |
9617d95e NP |
1194 | } |
1195 | ||
1da177e4 LT |
1196 | /** |
1197 | * page_add_file_rmap - add pte mapping to a file page | |
1198 | * @page: the page to add the mapping to | |
1199 | * | |
b8072f09 | 1200 | * The caller needs to hold the pte lock. |
1da177e4 LT |
1201 | */ |
1202 | void page_add_file_rmap(struct page *page) | |
1203 | { | |
d7365e78 | 1204 | struct mem_cgroup *memcg; |
89c06bd5 | 1205 | |
6de22619 | 1206 | memcg = mem_cgroup_begin_page_stat(page); |
d69b042f | 1207 | if (atomic_inc_and_test(&page->_mapcount)) { |
65ba55f5 | 1208 | __inc_zone_page_state(page, NR_FILE_MAPPED); |
d7365e78 | 1209 | mem_cgroup_inc_page_stat(memcg, MEM_CGROUP_STAT_FILE_MAPPED); |
d69b042f | 1210 | } |
6de22619 | 1211 | mem_cgroup_end_page_stat(memcg); |
1da177e4 LT |
1212 | } |
1213 | ||
8186eb6a JW |
1214 | static void page_remove_file_rmap(struct page *page) |
1215 | { | |
1216 | struct mem_cgroup *memcg; | |
8186eb6a | 1217 | |
6de22619 | 1218 | memcg = mem_cgroup_begin_page_stat(page); |
8186eb6a JW |
1219 | |
1220 | /* page still mapped by someone else? */ | |
1221 | if (!atomic_add_negative(-1, &page->_mapcount)) | |
1222 | goto out; | |
1223 | ||
1224 | /* Hugepages are not counted in NR_FILE_MAPPED for now. */ | |
1225 | if (unlikely(PageHuge(page))) | |
1226 | goto out; | |
1227 | ||
1228 | /* | |
1229 | * We use the irq-unsafe __{inc|mod}_zone_page_stat because | |
1230 | * these counters are not modified in interrupt context, and | |
1231 | * pte lock(a spinlock) is held, which implies preemption disabled. | |
1232 | */ | |
1233 | __dec_zone_page_state(page, NR_FILE_MAPPED); | |
1234 | mem_cgroup_dec_page_stat(memcg, MEM_CGROUP_STAT_FILE_MAPPED); | |
1235 | ||
1236 | if (unlikely(PageMlocked(page))) | |
1237 | clear_page_mlock(page); | |
1238 | out: | |
6de22619 | 1239 | mem_cgroup_end_page_stat(memcg); |
8186eb6a JW |
1240 | } |
1241 | ||
1da177e4 LT |
1242 | /** |
1243 | * page_remove_rmap - take down pte mapping from a page | |
1244 | * @page: page to remove mapping from | |
1245 | * | |
b8072f09 | 1246 | * The caller needs to hold the pte lock. |
1da177e4 | 1247 | */ |
edc315fd | 1248 | void page_remove_rmap(struct page *page) |
1da177e4 | 1249 | { |
8186eb6a JW |
1250 | if (!PageAnon(page)) { |
1251 | page_remove_file_rmap(page); | |
1252 | return; | |
1253 | } | |
89c06bd5 | 1254 | |
b904dcfe KM |
1255 | /* page still mapped by someone else? */ |
1256 | if (!atomic_add_negative(-1, &page->_mapcount)) | |
8186eb6a JW |
1257 | return; |
1258 | ||
1259 | /* Hugepages are not counted in NR_ANON_PAGES for now. */ | |
1260 | if (unlikely(PageHuge(page))) | |
1261 | return; | |
b904dcfe | 1262 | |
0fe6e20b | 1263 | /* |
bea04b07 JZ |
1264 | * We use the irq-unsafe __{inc|mod}_zone_page_stat because |
1265 | * these counters are not modified in interrupt context, and | |
bea04b07 | 1266 | * pte lock(a spinlock) is held, which implies preemption disabled. |
0fe6e20b | 1267 | */ |
8186eb6a JW |
1268 | if (PageTransHuge(page)) |
1269 | __dec_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES); | |
1270 | ||
1271 | __mod_zone_page_state(page_zone(page), NR_ANON_PAGES, | |
1272 | -hpage_nr_pages(page)); | |
1273 | ||
e6c509f8 HD |
1274 | if (unlikely(PageMlocked(page))) |
1275 | clear_page_mlock(page); | |
8186eb6a | 1276 | |
b904dcfe KM |
1277 | /* |
1278 | * It would be tidy to reset the PageAnon mapping here, | |
1279 | * but that might overwrite a racing page_add_anon_rmap | |
1280 | * which increments mapcount after us but sets mapping | |
1281 | * before us: so leave the reset to free_hot_cold_page, | |
1282 | * and remember that it's only reliable while mapped. | |
1283 | * Leaving it set also helps swapoff to reinstate ptes | |
1284 | * faster for those pages still in swapcache. | |
1285 | */ | |
1da177e4 LT |
1286 | } |
1287 | ||
1288 | /* | |
52629506 | 1289 | * @arg: enum ttu_flags will be passed to this argument |
1da177e4 | 1290 | */ |
ac769501 | 1291 | static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma, |
52629506 | 1292 | unsigned long address, void *arg) |
1da177e4 LT |
1293 | { |
1294 | struct mm_struct *mm = vma->vm_mm; | |
1da177e4 LT |
1295 | pte_t *pte; |
1296 | pte_t pteval; | |
c0718806 | 1297 | spinlock_t *ptl; |
1da177e4 | 1298 | int ret = SWAP_AGAIN; |
52629506 | 1299 | enum ttu_flags flags = (enum ttu_flags)arg; |
1da177e4 | 1300 | |
479db0bf | 1301 | pte = page_check_address(page, mm, address, &ptl, 0); |
c0718806 | 1302 | if (!pte) |
81b4082d | 1303 | goto out; |
1da177e4 LT |
1304 | |
1305 | /* | |
1306 | * If the page is mlock()d, we cannot swap it out. | |
1307 | * If it's recently referenced (perhaps page_referenced | |
1308 | * skipped over this mm) then we should reactivate it. | |
1309 | */ | |
14fa31b8 | 1310 | if (!(flags & TTU_IGNORE_MLOCK)) { |
caed0f48 KM |
1311 | if (vma->vm_flags & VM_LOCKED) |
1312 | goto out_mlock; | |
1313 | ||
daa5ba76 | 1314 | if (flags & TTU_MUNLOCK) |
53f79acb | 1315 | goto out_unmap; |
14fa31b8 AK |
1316 | } |
1317 | if (!(flags & TTU_IGNORE_ACCESS)) { | |
b291f000 NP |
1318 | if (ptep_clear_flush_young_notify(vma, address, pte)) { |
1319 | ret = SWAP_FAIL; | |
1320 | goto out_unmap; | |
1321 | } | |
1322 | } | |
1da177e4 | 1323 | |
1da177e4 LT |
1324 | /* Nuke the page table entry. */ |
1325 | flush_cache_page(vma, address, page_to_pfn(page)); | |
72b252ae MG |
1326 | if (should_defer_flush(mm, flags)) { |
1327 | /* | |
1328 | * We clear the PTE but do not flush so potentially a remote | |
1329 | * CPU could still be writing to the page. If the entry was | |
1330 | * previously clean then the architecture must guarantee that | |
1331 | * a clear->dirty transition on a cached TLB entry is written | |
1332 | * through and traps if the PTE is unmapped. | |
1333 | */ | |
1334 | pteval = ptep_get_and_clear(mm, address, pte); | |
1335 | ||
d950c947 | 1336 | set_tlb_ubc_flush_pending(mm, page, pte_dirty(pteval)); |
72b252ae MG |
1337 | } else { |
1338 | pteval = ptep_clear_flush(vma, address, pte); | |
1339 | } | |
1da177e4 LT |
1340 | |
1341 | /* Move the dirty bit to the physical page now the pte is gone. */ | |
1342 | if (pte_dirty(pteval)) | |
1343 | set_page_dirty(page); | |
1344 | ||
365e9c87 HD |
1345 | /* Update high watermark before we lower rss */ |
1346 | update_hiwater_rss(mm); | |
1347 | ||
888b9f7c | 1348 | if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) { |
5f24ae58 NH |
1349 | if (!PageHuge(page)) { |
1350 | if (PageAnon(page)) | |
1351 | dec_mm_counter(mm, MM_ANONPAGES); | |
1352 | else | |
1353 | dec_mm_counter(mm, MM_FILEPAGES); | |
1354 | } | |
888b9f7c | 1355 | set_pte_at(mm, address, pte, |
5f24ae58 | 1356 | swp_entry_to_pte(make_hwpoison_entry(page))); |
45961722 KW |
1357 | } else if (pte_unused(pteval)) { |
1358 | /* | |
1359 | * The guest indicated that the page content is of no | |
1360 | * interest anymore. Simply discard the pte, vmscan | |
1361 | * will take care of the rest. | |
1362 | */ | |
1363 | if (PageAnon(page)) | |
1364 | dec_mm_counter(mm, MM_ANONPAGES); | |
1365 | else | |
1366 | dec_mm_counter(mm, MM_FILEPAGES); | |
888b9f7c | 1367 | } else if (PageAnon(page)) { |
4c21e2f2 | 1368 | swp_entry_t entry = { .val = page_private(page) }; |
179ef71c | 1369 | pte_t swp_pte; |
0697212a CL |
1370 | |
1371 | if (PageSwapCache(page)) { | |
1372 | /* | |
1373 | * Store the swap location in the pte. | |
1374 | * See handle_pte_fault() ... | |
1375 | */ | |
570a335b HD |
1376 | if (swap_duplicate(entry) < 0) { |
1377 | set_pte_at(mm, address, pte, pteval); | |
1378 | ret = SWAP_FAIL; | |
1379 | goto out_unmap; | |
1380 | } | |
0697212a CL |
1381 | if (list_empty(&mm->mmlist)) { |
1382 | spin_lock(&mmlist_lock); | |
1383 | if (list_empty(&mm->mmlist)) | |
1384 | list_add(&mm->mmlist, &init_mm.mmlist); | |
1385 | spin_unlock(&mmlist_lock); | |
1386 | } | |
d559db08 | 1387 | dec_mm_counter(mm, MM_ANONPAGES); |
b084d435 | 1388 | inc_mm_counter(mm, MM_SWAPENTS); |
ce1744f4 | 1389 | } else if (IS_ENABLED(CONFIG_MIGRATION)) { |
0697212a CL |
1390 | /* |
1391 | * Store the pfn of the page in a special migration | |
1392 | * pte. do_swap_page() will wait until the migration | |
1393 | * pte is removed and then restart fault handling. | |
1394 | */ | |
daa5ba76 | 1395 | BUG_ON(!(flags & TTU_MIGRATION)); |
0697212a | 1396 | entry = make_migration_entry(page, pte_write(pteval)); |
1da177e4 | 1397 | } |
179ef71c CG |
1398 | swp_pte = swp_entry_to_pte(entry); |
1399 | if (pte_soft_dirty(pteval)) | |
1400 | swp_pte = pte_swp_mksoft_dirty(swp_pte); | |
1401 | set_pte_at(mm, address, pte, swp_pte); | |
ce1744f4 | 1402 | } else if (IS_ENABLED(CONFIG_MIGRATION) && |
daa5ba76 | 1403 | (flags & TTU_MIGRATION)) { |
04e62a29 CL |
1404 | /* Establish migration entry for a file page */ |
1405 | swp_entry_t entry; | |
1406 | entry = make_migration_entry(page, pte_write(pteval)); | |
1407 | set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); | |
1408 | } else | |
d559db08 | 1409 | dec_mm_counter(mm, MM_FILEPAGES); |
1da177e4 | 1410 | |
edc315fd | 1411 | page_remove_rmap(page); |
1da177e4 LT |
1412 | page_cache_release(page); |
1413 | ||
1414 | out_unmap: | |
c0718806 | 1415 | pte_unmap_unlock(pte, ptl); |
daa5ba76 | 1416 | if (ret != SWAP_FAIL && !(flags & TTU_MUNLOCK)) |
2ec74c3e | 1417 | mmu_notifier_invalidate_page(mm, address); |
caed0f48 KM |
1418 | out: |
1419 | return ret; | |
53f79acb | 1420 | |
caed0f48 KM |
1421 | out_mlock: |
1422 | pte_unmap_unlock(pte, ptl); | |
1423 | ||
1424 | ||
1425 | /* | |
1426 | * We need mmap_sem locking, Otherwise VM_LOCKED check makes | |
1427 | * unstable result and race. Plus, We can't wait here because | |
c8c06efa | 1428 | * we now hold anon_vma->rwsem or mapping->i_mmap_rwsem. |
caed0f48 KM |
1429 | * if trylock failed, the page remain in evictable lru and later |
1430 | * vmscan could retry to move the page to unevictable lru if the | |
1431 | * page is actually mlocked. | |
1432 | */ | |
1433 | if (down_read_trylock(&vma->vm_mm->mmap_sem)) { | |
1434 | if (vma->vm_flags & VM_LOCKED) { | |
1435 | mlock_vma_page(page); | |
1436 | ret = SWAP_MLOCK; | |
53f79acb | 1437 | } |
caed0f48 | 1438 | up_read(&vma->vm_mm->mmap_sem); |
53f79acb | 1439 | } |
1da177e4 LT |
1440 | return ret; |
1441 | } | |
1442 | ||
71e3aac0 | 1443 | bool is_vma_temporary_stack(struct vm_area_struct *vma) |
a8bef8ff MG |
1444 | { |
1445 | int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP); | |
1446 | ||
1447 | if (!maybe_stack) | |
1448 | return false; | |
1449 | ||
1450 | if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) == | |
1451 | VM_STACK_INCOMPLETE_SETUP) | |
1452 | return true; | |
1453 | ||
1454 | return false; | |
1455 | } | |
1456 | ||
52629506 JK |
1457 | static bool invalid_migration_vma(struct vm_area_struct *vma, void *arg) |
1458 | { | |
1459 | return is_vma_temporary_stack(vma); | |
1460 | } | |
1461 | ||
52629506 JK |
1462 | static int page_not_mapped(struct page *page) |
1463 | { | |
1464 | return !page_mapped(page); | |
1465 | }; | |
1466 | ||
1da177e4 LT |
1467 | /** |
1468 | * try_to_unmap - try to remove all page table mappings to a page | |
1469 | * @page: the page to get unmapped | |
14fa31b8 | 1470 | * @flags: action and flags |
1da177e4 LT |
1471 | * |
1472 | * Tries to remove all the page table entries which are mapping this | |
1473 | * page, used in the pageout path. Caller must hold the page lock. | |
1474 | * Return values are: | |
1475 | * | |
1476 | * SWAP_SUCCESS - we succeeded in removing all mappings | |
1477 | * SWAP_AGAIN - we missed a mapping, try again later | |
1478 | * SWAP_FAIL - the page is unswappable | |
b291f000 | 1479 | * SWAP_MLOCK - page is mlocked. |
1da177e4 | 1480 | */ |
14fa31b8 | 1481 | int try_to_unmap(struct page *page, enum ttu_flags flags) |
1da177e4 LT |
1482 | { |
1483 | int ret; | |
52629506 JK |
1484 | struct rmap_walk_control rwc = { |
1485 | .rmap_one = try_to_unmap_one, | |
1486 | .arg = (void *)flags, | |
1487 | .done = page_not_mapped, | |
52629506 JK |
1488 | .anon_lock = page_lock_anon_vma_read, |
1489 | }; | |
1da177e4 | 1490 | |
309381fe | 1491 | VM_BUG_ON_PAGE(!PageHuge(page) && PageTransHuge(page), page); |
1da177e4 | 1492 | |
52629506 JK |
1493 | /* |
1494 | * During exec, a temporary VMA is setup and later moved. | |
1495 | * The VMA is moved under the anon_vma lock but not the | |
1496 | * page tables leading to a race where migration cannot | |
1497 | * find the migration ptes. Rather than increasing the | |
1498 | * locking requirements of exec(), migration skips | |
1499 | * temporary VMAs until after exec() completes. | |
1500 | */ | |
daa5ba76 | 1501 | if ((flags & TTU_MIGRATION) && !PageKsm(page) && PageAnon(page)) |
52629506 JK |
1502 | rwc.invalid_vma = invalid_migration_vma; |
1503 | ||
1504 | ret = rmap_walk(page, &rwc); | |
1505 | ||
b291f000 | 1506 | if (ret != SWAP_MLOCK && !page_mapped(page)) |
1da177e4 LT |
1507 | ret = SWAP_SUCCESS; |
1508 | return ret; | |
1509 | } | |
81b4082d | 1510 | |
b291f000 NP |
1511 | /** |
1512 | * try_to_munlock - try to munlock a page | |
1513 | * @page: the page to be munlocked | |
1514 | * | |
1515 | * Called from munlock code. Checks all of the VMAs mapping the page | |
1516 | * to make sure nobody else has this page mlocked. The page will be | |
1517 | * returned with PG_mlocked cleared if no other vmas have it mlocked. | |
1518 | * | |
1519 | * Return values are: | |
1520 | * | |
53f79acb | 1521 | * SWAP_AGAIN - no vma is holding page mlocked, or, |
b291f000 | 1522 | * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem |
5ad64688 | 1523 | * SWAP_FAIL - page cannot be located at present |
b291f000 NP |
1524 | * SWAP_MLOCK - page is now mlocked. |
1525 | */ | |
1526 | int try_to_munlock(struct page *page) | |
1527 | { | |
e8351ac9 JK |
1528 | int ret; |
1529 | struct rmap_walk_control rwc = { | |
1530 | .rmap_one = try_to_unmap_one, | |
1531 | .arg = (void *)TTU_MUNLOCK, | |
1532 | .done = page_not_mapped, | |
e8351ac9 JK |
1533 | .anon_lock = page_lock_anon_vma_read, |
1534 | ||
1535 | }; | |
1536 | ||
309381fe | 1537 | VM_BUG_ON_PAGE(!PageLocked(page) || PageLRU(page), page); |
b291f000 | 1538 | |
e8351ac9 JK |
1539 | ret = rmap_walk(page, &rwc); |
1540 | return ret; | |
b291f000 | 1541 | } |
e9995ef9 | 1542 | |
01d8b20d | 1543 | void __put_anon_vma(struct anon_vma *anon_vma) |
76545066 | 1544 | { |
01d8b20d | 1545 | struct anon_vma *root = anon_vma->root; |
76545066 | 1546 | |
624483f3 | 1547 | anon_vma_free(anon_vma); |
01d8b20d PZ |
1548 | if (root != anon_vma && atomic_dec_and_test(&root->refcount)) |
1549 | anon_vma_free(root); | |
76545066 | 1550 | } |
76545066 | 1551 | |
0dd1c7bb JK |
1552 | static struct anon_vma *rmap_walk_anon_lock(struct page *page, |
1553 | struct rmap_walk_control *rwc) | |
faecd8dd JK |
1554 | { |
1555 | struct anon_vma *anon_vma; | |
1556 | ||
0dd1c7bb JK |
1557 | if (rwc->anon_lock) |
1558 | return rwc->anon_lock(page); | |
1559 | ||
faecd8dd JK |
1560 | /* |
1561 | * Note: remove_migration_ptes() cannot use page_lock_anon_vma_read() | |
1562 | * because that depends on page_mapped(); but not all its usages | |
1563 | * are holding mmap_sem. Users without mmap_sem are required to | |
1564 | * take a reference count to prevent the anon_vma disappearing | |
1565 | */ | |
1566 | anon_vma = page_anon_vma(page); | |
1567 | if (!anon_vma) | |
1568 | return NULL; | |
1569 | ||
1570 | anon_vma_lock_read(anon_vma); | |
1571 | return anon_vma; | |
1572 | } | |
1573 | ||
e9995ef9 | 1574 | /* |
e8351ac9 JK |
1575 | * rmap_walk_anon - do something to anonymous page using the object-based |
1576 | * rmap method | |
1577 | * @page: the page to be handled | |
1578 | * @rwc: control variable according to each walk type | |
1579 | * | |
1580 | * Find all the mappings of a page using the mapping pointer and the vma chains | |
1581 | * contained in the anon_vma struct it points to. | |
1582 | * | |
1583 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma | |
1584 | * where the page was found will be held for write. So, we won't recheck | |
1585 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be | |
1586 | * LOCKED. | |
e9995ef9 | 1587 | */ |
051ac83a | 1588 | static int rmap_walk_anon(struct page *page, struct rmap_walk_control *rwc) |
e9995ef9 HD |
1589 | { |
1590 | struct anon_vma *anon_vma; | |
b258d860 | 1591 | pgoff_t pgoff; |
5beb4930 | 1592 | struct anon_vma_chain *avc; |
e9995ef9 HD |
1593 | int ret = SWAP_AGAIN; |
1594 | ||
0dd1c7bb | 1595 | anon_vma = rmap_walk_anon_lock(page, rwc); |
e9995ef9 HD |
1596 | if (!anon_vma) |
1597 | return ret; | |
faecd8dd | 1598 | |
b258d860 | 1599 | pgoff = page_to_pgoff(page); |
bf181b9f | 1600 | anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) { |
5beb4930 | 1601 | struct vm_area_struct *vma = avc->vma; |
e9995ef9 | 1602 | unsigned long address = vma_address(page, vma); |
0dd1c7bb JK |
1603 | |
1604 | if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg)) | |
1605 | continue; | |
1606 | ||
051ac83a | 1607 | ret = rwc->rmap_one(page, vma, address, rwc->arg); |
e9995ef9 HD |
1608 | if (ret != SWAP_AGAIN) |
1609 | break; | |
0dd1c7bb JK |
1610 | if (rwc->done && rwc->done(page)) |
1611 | break; | |
e9995ef9 | 1612 | } |
4fc3f1d6 | 1613 | anon_vma_unlock_read(anon_vma); |
e9995ef9 HD |
1614 | return ret; |
1615 | } | |
1616 | ||
e8351ac9 JK |
1617 | /* |
1618 | * rmap_walk_file - do something to file page using the object-based rmap method | |
1619 | * @page: the page to be handled | |
1620 | * @rwc: control variable according to each walk type | |
1621 | * | |
1622 | * Find all the mappings of a page using the mapping pointer and the vma chains | |
1623 | * contained in the address_space struct it points to. | |
1624 | * | |
1625 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma | |
1626 | * where the page was found will be held for write. So, we won't recheck | |
1627 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be | |
1628 | * LOCKED. | |
1629 | */ | |
051ac83a | 1630 | static int rmap_walk_file(struct page *page, struct rmap_walk_control *rwc) |
e9995ef9 HD |
1631 | { |
1632 | struct address_space *mapping = page->mapping; | |
b258d860 | 1633 | pgoff_t pgoff; |
e9995ef9 | 1634 | struct vm_area_struct *vma; |
e9995ef9 HD |
1635 | int ret = SWAP_AGAIN; |
1636 | ||
9f32624b JK |
1637 | /* |
1638 | * The page lock not only makes sure that page->mapping cannot | |
1639 | * suddenly be NULLified by truncation, it makes sure that the | |
1640 | * structure at mapping cannot be freed and reused yet, | |
c8c06efa | 1641 | * so we can safely take mapping->i_mmap_rwsem. |
9f32624b | 1642 | */ |
81d1b09c | 1643 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
9f32624b | 1644 | |
e9995ef9 HD |
1645 | if (!mapping) |
1646 | return ret; | |
3dec0ba0 | 1647 | |
b258d860 | 1648 | pgoff = page_to_pgoff(page); |
3dec0ba0 | 1649 | i_mmap_lock_read(mapping); |
6b2dbba8 | 1650 | vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) { |
e9995ef9 | 1651 | unsigned long address = vma_address(page, vma); |
0dd1c7bb JK |
1652 | |
1653 | if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg)) | |
1654 | continue; | |
1655 | ||
051ac83a | 1656 | ret = rwc->rmap_one(page, vma, address, rwc->arg); |
e9995ef9 | 1657 | if (ret != SWAP_AGAIN) |
0dd1c7bb JK |
1658 | goto done; |
1659 | if (rwc->done && rwc->done(page)) | |
1660 | goto done; | |
e9995ef9 | 1661 | } |
0dd1c7bb | 1662 | |
0dd1c7bb | 1663 | done: |
3dec0ba0 | 1664 | i_mmap_unlock_read(mapping); |
e9995ef9 HD |
1665 | return ret; |
1666 | } | |
1667 | ||
051ac83a | 1668 | int rmap_walk(struct page *page, struct rmap_walk_control *rwc) |
e9995ef9 | 1669 | { |
e9995ef9 | 1670 | if (unlikely(PageKsm(page))) |
051ac83a | 1671 | return rmap_walk_ksm(page, rwc); |
e9995ef9 | 1672 | else if (PageAnon(page)) |
051ac83a | 1673 | return rmap_walk_anon(page, rwc); |
e9995ef9 | 1674 | else |
051ac83a | 1675 | return rmap_walk_file(page, rwc); |
e9995ef9 | 1676 | } |
0fe6e20b | 1677 | |
e3390f67 | 1678 | #ifdef CONFIG_HUGETLB_PAGE |
0fe6e20b NH |
1679 | /* |
1680 | * The following three functions are for anonymous (private mapped) hugepages. | |
1681 | * Unlike common anonymous pages, anonymous hugepages have no accounting code | |
1682 | * and no lru code, because we handle hugepages differently from common pages. | |
1683 | */ | |
1684 | static void __hugepage_set_anon_rmap(struct page *page, | |
1685 | struct vm_area_struct *vma, unsigned long address, int exclusive) | |
1686 | { | |
1687 | struct anon_vma *anon_vma = vma->anon_vma; | |
433abed6 | 1688 | |
0fe6e20b | 1689 | BUG_ON(!anon_vma); |
433abed6 NH |
1690 | |
1691 | if (PageAnon(page)) | |
1692 | return; | |
1693 | if (!exclusive) | |
1694 | anon_vma = anon_vma->root; | |
1695 | ||
0fe6e20b NH |
1696 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; |
1697 | page->mapping = (struct address_space *) anon_vma; | |
1698 | page->index = linear_page_index(vma, address); | |
1699 | } | |
1700 | ||
1701 | void hugepage_add_anon_rmap(struct page *page, | |
1702 | struct vm_area_struct *vma, unsigned long address) | |
1703 | { | |
1704 | struct anon_vma *anon_vma = vma->anon_vma; | |
1705 | int first; | |
a850ea30 NH |
1706 | |
1707 | BUG_ON(!PageLocked(page)); | |
0fe6e20b | 1708 | BUG_ON(!anon_vma); |
5dbe0af4 | 1709 | /* address might be in next vma when migration races vma_adjust */ |
0fe6e20b NH |
1710 | first = atomic_inc_and_test(&page->_mapcount); |
1711 | if (first) | |
1712 | __hugepage_set_anon_rmap(page, vma, address, 0); | |
1713 | } | |
1714 | ||
1715 | void hugepage_add_new_anon_rmap(struct page *page, | |
1716 | struct vm_area_struct *vma, unsigned long address) | |
1717 | { | |
1718 | BUG_ON(address < vma->vm_start || address >= vma->vm_end); | |
1719 | atomic_set(&page->_mapcount, 0); | |
1720 | __hugepage_set_anon_rmap(page, vma, address, 1); | |
1721 | } | |
e3390f67 | 1722 | #endif /* CONFIG_HUGETLB_PAGE */ |