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; | |
629 | put_cpu(); | |
630 | } | |
631 | ||
632 | static void set_tlb_ubc_flush_pending(struct mm_struct *mm, | |
633 | struct page *page) | |
634 | { | |
635 | struct tlbflush_unmap_batch *tlb_ubc = ¤t->tlb_ubc; | |
636 | ||
637 | cpumask_or(&tlb_ubc->cpumask, &tlb_ubc->cpumask, mm_cpumask(mm)); | |
638 | tlb_ubc->flush_required = true; | |
639 | } | |
640 | ||
641 | /* | |
642 | * Returns true if the TLB flush should be deferred to the end of a batch of | |
643 | * unmap operations to reduce IPIs. | |
644 | */ | |
645 | static bool should_defer_flush(struct mm_struct *mm, enum ttu_flags flags) | |
646 | { | |
647 | bool should_defer = false; | |
648 | ||
649 | if (!(flags & TTU_BATCH_FLUSH)) | |
650 | return false; | |
651 | ||
652 | /* If remote CPUs need to be flushed then defer batch the flush */ | |
653 | if (cpumask_any_but(mm_cpumask(mm), get_cpu()) < nr_cpu_ids) | |
654 | should_defer = true; | |
655 | put_cpu(); | |
656 | ||
657 | return should_defer; | |
658 | } | |
659 | #else | |
660 | static void set_tlb_ubc_flush_pending(struct mm_struct *mm, | |
661 | struct page *page) | |
662 | { | |
663 | } | |
664 | ||
665 | static bool should_defer_flush(struct mm_struct *mm, enum ttu_flags flags) | |
666 | { | |
667 | return false; | |
668 | } | |
669 | #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */ | |
670 | ||
1da177e4 | 671 | /* |
bf89c8c8 | 672 | * At what user virtual address is page expected in vma? |
ab941e0f | 673 | * Caller should check the page is actually part of the vma. |
1da177e4 LT |
674 | */ |
675 | unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma) | |
676 | { | |
86c2ad19 | 677 | unsigned long address; |
21d0d443 | 678 | if (PageAnon(page)) { |
4829b906 HD |
679 | struct anon_vma *page__anon_vma = page_anon_vma(page); |
680 | /* | |
681 | * Note: swapoff's unuse_vma() is more efficient with this | |
682 | * check, and needs it to match anon_vma when KSM is active. | |
683 | */ | |
684 | if (!vma->anon_vma || !page__anon_vma || | |
685 | vma->anon_vma->root != page__anon_vma->root) | |
21d0d443 | 686 | return -EFAULT; |
27ba0644 KS |
687 | } else if (page->mapping) { |
688 | if (!vma->vm_file || vma->vm_file->f_mapping != page->mapping) | |
1da177e4 LT |
689 | return -EFAULT; |
690 | } else | |
691 | return -EFAULT; | |
86c2ad19 ML |
692 | address = __vma_address(page, vma); |
693 | if (unlikely(address < vma->vm_start || address >= vma->vm_end)) | |
694 | return -EFAULT; | |
695 | return address; | |
1da177e4 LT |
696 | } |
697 | ||
6219049a BL |
698 | pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address) |
699 | { | |
700 | pgd_t *pgd; | |
701 | pud_t *pud; | |
702 | pmd_t *pmd = NULL; | |
f72e7dcd | 703 | pmd_t pmde; |
6219049a BL |
704 | |
705 | pgd = pgd_offset(mm, address); | |
706 | if (!pgd_present(*pgd)) | |
707 | goto out; | |
708 | ||
709 | pud = pud_offset(pgd, address); | |
710 | if (!pud_present(*pud)) | |
711 | goto out; | |
712 | ||
713 | pmd = pmd_offset(pud, address); | |
f72e7dcd | 714 | /* |
8809aa2d | 715 | * Some THP functions use the sequence pmdp_huge_clear_flush(), set_pmd_at() |
f72e7dcd HD |
716 | * without holding anon_vma lock for write. So when looking for a |
717 | * genuine pmde (in which to find pte), test present and !THP together. | |
718 | */ | |
e37c6982 CB |
719 | pmde = *pmd; |
720 | barrier(); | |
f72e7dcd | 721 | if (!pmd_present(pmde) || pmd_trans_huge(pmde)) |
6219049a BL |
722 | pmd = NULL; |
723 | out: | |
724 | return pmd; | |
725 | } | |
726 | ||
81b4082d ND |
727 | /* |
728 | * Check that @page is mapped at @address into @mm. | |
729 | * | |
479db0bf NP |
730 | * If @sync is false, page_check_address may perform a racy check to avoid |
731 | * the page table lock when the pte is not present (helpful when reclaiming | |
732 | * highly shared pages). | |
733 | * | |
b8072f09 | 734 | * On success returns with pte mapped and locked. |
81b4082d | 735 | */ |
e9a81a82 | 736 | pte_t *__page_check_address(struct page *page, struct mm_struct *mm, |
479db0bf | 737 | unsigned long address, spinlock_t **ptlp, int sync) |
81b4082d | 738 | { |
81b4082d ND |
739 | pmd_t *pmd; |
740 | pte_t *pte; | |
c0718806 | 741 | spinlock_t *ptl; |
81b4082d | 742 | |
0fe6e20b | 743 | if (unlikely(PageHuge(page))) { |
98398c32 | 744 | /* when pud is not present, pte will be NULL */ |
0fe6e20b | 745 | pte = huge_pte_offset(mm, address); |
98398c32 JW |
746 | if (!pte) |
747 | return NULL; | |
748 | ||
cb900f41 | 749 | ptl = huge_pte_lockptr(page_hstate(page), mm, pte); |
0fe6e20b NH |
750 | goto check; |
751 | } | |
752 | ||
6219049a BL |
753 | pmd = mm_find_pmd(mm, address); |
754 | if (!pmd) | |
c0718806 HD |
755 | return NULL; |
756 | ||
c0718806 HD |
757 | pte = pte_offset_map(pmd, address); |
758 | /* Make a quick check before getting the lock */ | |
479db0bf | 759 | if (!sync && !pte_present(*pte)) { |
c0718806 HD |
760 | pte_unmap(pte); |
761 | return NULL; | |
762 | } | |
763 | ||
4c21e2f2 | 764 | ptl = pte_lockptr(mm, pmd); |
0fe6e20b | 765 | check: |
c0718806 HD |
766 | spin_lock(ptl); |
767 | if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) { | |
768 | *ptlp = ptl; | |
769 | return pte; | |
81b4082d | 770 | } |
c0718806 HD |
771 | pte_unmap_unlock(pte, ptl); |
772 | return NULL; | |
81b4082d ND |
773 | } |
774 | ||
b291f000 NP |
775 | /** |
776 | * page_mapped_in_vma - check whether a page is really mapped in a VMA | |
777 | * @page: the page to test | |
778 | * @vma: the VMA to test | |
779 | * | |
780 | * Returns 1 if the page is mapped into the page tables of the VMA, 0 | |
781 | * if the page is not mapped into the page tables of this VMA. Only | |
782 | * valid for normal file or anonymous VMAs. | |
783 | */ | |
6a46079c | 784 | int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma) |
b291f000 NP |
785 | { |
786 | unsigned long address; | |
787 | pte_t *pte; | |
788 | spinlock_t *ptl; | |
789 | ||
86c2ad19 ML |
790 | address = __vma_address(page, vma); |
791 | if (unlikely(address < vma->vm_start || address >= vma->vm_end)) | |
b291f000 NP |
792 | return 0; |
793 | pte = page_check_address(page, vma->vm_mm, address, &ptl, 1); | |
794 | if (!pte) /* the page is not in this mm */ | |
795 | return 0; | |
796 | pte_unmap_unlock(pte, ptl); | |
797 | ||
798 | return 1; | |
799 | } | |
800 | ||
9f32624b JK |
801 | struct page_referenced_arg { |
802 | int mapcount; | |
803 | int referenced; | |
804 | unsigned long vm_flags; | |
805 | struct mem_cgroup *memcg; | |
806 | }; | |
1da177e4 | 807 | /* |
9f32624b | 808 | * arg: page_referenced_arg will be passed |
1da177e4 | 809 | */ |
ac769501 | 810 | static int page_referenced_one(struct page *page, struct vm_area_struct *vma, |
9f32624b | 811 | unsigned long address, void *arg) |
1da177e4 LT |
812 | { |
813 | struct mm_struct *mm = vma->vm_mm; | |
117b0791 | 814 | spinlock_t *ptl; |
1da177e4 | 815 | int referenced = 0; |
9f32624b | 816 | struct page_referenced_arg *pra = arg; |
1da177e4 | 817 | |
71e3aac0 AA |
818 | if (unlikely(PageTransHuge(page))) { |
819 | pmd_t *pmd; | |
820 | ||
2da28bfd AA |
821 | /* |
822 | * rmap might return false positives; we must filter | |
823 | * these out using page_check_address_pmd(). | |
824 | */ | |
71e3aac0 | 825 | pmd = page_check_address_pmd(page, mm, address, |
117b0791 KS |
826 | PAGE_CHECK_ADDRESS_PMD_FLAG, &ptl); |
827 | if (!pmd) | |
9f32624b | 828 | return SWAP_AGAIN; |
2da28bfd AA |
829 | |
830 | if (vma->vm_flags & VM_LOCKED) { | |
117b0791 | 831 | spin_unlock(ptl); |
9f32624b JK |
832 | pra->vm_flags |= VM_LOCKED; |
833 | return SWAP_FAIL; /* To break the loop */ | |
2da28bfd AA |
834 | } |
835 | ||
836 | /* go ahead even if the pmd is pmd_trans_splitting() */ | |
837 | if (pmdp_clear_flush_young_notify(vma, address, pmd)) | |
71e3aac0 | 838 | referenced++; |
117b0791 | 839 | spin_unlock(ptl); |
71e3aac0 AA |
840 | } else { |
841 | pte_t *pte; | |
71e3aac0 | 842 | |
2da28bfd AA |
843 | /* |
844 | * rmap might return false positives; we must filter | |
845 | * these out using page_check_address(). | |
846 | */ | |
71e3aac0 AA |
847 | pte = page_check_address(page, mm, address, &ptl, 0); |
848 | if (!pte) | |
9f32624b | 849 | return SWAP_AGAIN; |
71e3aac0 | 850 | |
2da28bfd AA |
851 | if (vma->vm_flags & VM_LOCKED) { |
852 | pte_unmap_unlock(pte, ptl); | |
9f32624b JK |
853 | pra->vm_flags |= VM_LOCKED; |
854 | return SWAP_FAIL; /* To break the loop */ | |
2da28bfd AA |
855 | } |
856 | ||
71e3aac0 AA |
857 | if (ptep_clear_flush_young_notify(vma, address, pte)) { |
858 | /* | |
859 | * Don't treat a reference through a sequentially read | |
860 | * mapping as such. If the page has been used in | |
861 | * another mapping, we will catch it; if this other | |
862 | * mapping is already gone, the unmap path will have | |
863 | * set PG_referenced or activated the page. | |
864 | */ | |
64363aad | 865 | if (likely(!(vma->vm_flags & VM_SEQ_READ))) |
71e3aac0 AA |
866 | referenced++; |
867 | } | |
868 | pte_unmap_unlock(pte, ptl); | |
869 | } | |
870 | ||
9f32624b JK |
871 | if (referenced) { |
872 | pra->referenced++; | |
873 | pra->vm_flags |= vma->vm_flags; | |
1da177e4 | 874 | } |
34bbd704 | 875 | |
9f32624b JK |
876 | pra->mapcount--; |
877 | if (!pra->mapcount) | |
878 | return SWAP_SUCCESS; /* To break the loop */ | |
879 | ||
880 | return SWAP_AGAIN; | |
1da177e4 LT |
881 | } |
882 | ||
9f32624b | 883 | static bool invalid_page_referenced_vma(struct vm_area_struct *vma, void *arg) |
1da177e4 | 884 | { |
9f32624b JK |
885 | struct page_referenced_arg *pra = arg; |
886 | struct mem_cgroup *memcg = pra->memcg; | |
1da177e4 | 887 | |
9f32624b JK |
888 | if (!mm_match_cgroup(vma->vm_mm, memcg)) |
889 | return true; | |
1da177e4 | 890 | |
9f32624b | 891 | return false; |
1da177e4 LT |
892 | } |
893 | ||
894 | /** | |
895 | * page_referenced - test if the page was referenced | |
896 | * @page: the page to test | |
897 | * @is_locked: caller holds lock on the page | |
72835c86 | 898 | * @memcg: target memory cgroup |
6fe6b7e3 | 899 | * @vm_flags: collect encountered vma->vm_flags who actually referenced the page |
1da177e4 LT |
900 | * |
901 | * Quick test_and_clear_referenced for all mappings to a page, | |
902 | * returns the number of ptes which referenced the page. | |
903 | */ | |
6fe6b7e3 WF |
904 | int page_referenced(struct page *page, |
905 | int is_locked, | |
72835c86 | 906 | struct mem_cgroup *memcg, |
6fe6b7e3 | 907 | unsigned long *vm_flags) |
1da177e4 | 908 | { |
9f32624b | 909 | int ret; |
5ad64688 | 910 | int we_locked = 0; |
9f32624b JK |
911 | struct page_referenced_arg pra = { |
912 | .mapcount = page_mapcount(page), | |
913 | .memcg = memcg, | |
914 | }; | |
915 | struct rmap_walk_control rwc = { | |
916 | .rmap_one = page_referenced_one, | |
917 | .arg = (void *)&pra, | |
918 | .anon_lock = page_lock_anon_vma_read, | |
919 | }; | |
1da177e4 | 920 | |
6fe6b7e3 | 921 | *vm_flags = 0; |
9f32624b JK |
922 | if (!page_mapped(page)) |
923 | return 0; | |
924 | ||
925 | if (!page_rmapping(page)) | |
926 | return 0; | |
927 | ||
928 | if (!is_locked && (!PageAnon(page) || PageKsm(page))) { | |
929 | we_locked = trylock_page(page); | |
930 | if (!we_locked) | |
931 | return 1; | |
1da177e4 | 932 | } |
9f32624b JK |
933 | |
934 | /* | |
935 | * If we are reclaiming on behalf of a cgroup, skip | |
936 | * counting on behalf of references from different | |
937 | * cgroups | |
938 | */ | |
939 | if (memcg) { | |
940 | rwc.invalid_vma = invalid_page_referenced_vma; | |
941 | } | |
942 | ||
943 | ret = rmap_walk(page, &rwc); | |
944 | *vm_flags = pra.vm_flags; | |
945 | ||
946 | if (we_locked) | |
947 | unlock_page(page); | |
948 | ||
949 | return pra.referenced; | |
1da177e4 LT |
950 | } |
951 | ||
1cb1729b | 952 | static int page_mkclean_one(struct page *page, struct vm_area_struct *vma, |
9853a407 | 953 | unsigned long address, void *arg) |
d08b3851 PZ |
954 | { |
955 | struct mm_struct *mm = vma->vm_mm; | |
c2fda5fe | 956 | pte_t *pte; |
d08b3851 PZ |
957 | spinlock_t *ptl; |
958 | int ret = 0; | |
9853a407 | 959 | int *cleaned = arg; |
d08b3851 | 960 | |
479db0bf | 961 | pte = page_check_address(page, mm, address, &ptl, 1); |
d08b3851 PZ |
962 | if (!pte) |
963 | goto out; | |
964 | ||
c2fda5fe PZ |
965 | if (pte_dirty(*pte) || pte_write(*pte)) { |
966 | pte_t entry; | |
d08b3851 | 967 | |
c2fda5fe | 968 | flush_cache_page(vma, address, pte_pfn(*pte)); |
2ec74c3e | 969 | entry = ptep_clear_flush(vma, address, pte); |
c2fda5fe PZ |
970 | entry = pte_wrprotect(entry); |
971 | entry = pte_mkclean(entry); | |
d6e88e67 | 972 | set_pte_at(mm, address, pte, entry); |
c2fda5fe PZ |
973 | ret = 1; |
974 | } | |
d08b3851 | 975 | |
d08b3851 | 976 | pte_unmap_unlock(pte, ptl); |
2ec74c3e | 977 | |
9853a407 | 978 | if (ret) { |
2ec74c3e | 979 | mmu_notifier_invalidate_page(mm, address); |
9853a407 JK |
980 | (*cleaned)++; |
981 | } | |
d08b3851 | 982 | out: |
9853a407 | 983 | return SWAP_AGAIN; |
d08b3851 PZ |
984 | } |
985 | ||
9853a407 | 986 | static bool invalid_mkclean_vma(struct vm_area_struct *vma, void *arg) |
d08b3851 | 987 | { |
9853a407 | 988 | if (vma->vm_flags & VM_SHARED) |
871beb8c | 989 | return false; |
d08b3851 | 990 | |
871beb8c | 991 | return true; |
d08b3851 PZ |
992 | } |
993 | ||
994 | int page_mkclean(struct page *page) | |
995 | { | |
9853a407 JK |
996 | int cleaned = 0; |
997 | struct address_space *mapping; | |
998 | struct rmap_walk_control rwc = { | |
999 | .arg = (void *)&cleaned, | |
1000 | .rmap_one = page_mkclean_one, | |
1001 | .invalid_vma = invalid_mkclean_vma, | |
1002 | }; | |
d08b3851 PZ |
1003 | |
1004 | BUG_ON(!PageLocked(page)); | |
1005 | ||
9853a407 JK |
1006 | if (!page_mapped(page)) |
1007 | return 0; | |
1008 | ||
1009 | mapping = page_mapping(page); | |
1010 | if (!mapping) | |
1011 | return 0; | |
1012 | ||
1013 | rmap_walk(page, &rwc); | |
d08b3851 | 1014 | |
9853a407 | 1015 | return cleaned; |
d08b3851 | 1016 | } |
60b59bea | 1017 | EXPORT_SYMBOL_GPL(page_mkclean); |
d08b3851 | 1018 | |
c44b6743 RR |
1019 | /** |
1020 | * page_move_anon_rmap - move a page to our anon_vma | |
1021 | * @page: the page to move to our anon_vma | |
1022 | * @vma: the vma the page belongs to | |
1023 | * @address: the user virtual address mapped | |
1024 | * | |
1025 | * When a page belongs exclusively to one process after a COW event, | |
1026 | * that page can be moved into the anon_vma that belongs to just that | |
1027 | * process, so the rmap code will not search the parent or sibling | |
1028 | * processes. | |
1029 | */ | |
1030 | void page_move_anon_rmap(struct page *page, | |
1031 | struct vm_area_struct *vma, unsigned long address) | |
1032 | { | |
1033 | struct anon_vma *anon_vma = vma->anon_vma; | |
1034 | ||
309381fe | 1035 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
81d1b09c | 1036 | VM_BUG_ON_VMA(!anon_vma, vma); |
309381fe | 1037 | VM_BUG_ON_PAGE(page->index != linear_page_index(vma, address), page); |
c44b6743 RR |
1038 | |
1039 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; | |
414e2fb8 VD |
1040 | /* |
1041 | * Ensure that anon_vma and the PAGE_MAPPING_ANON bit are written | |
1042 | * simultaneously, so a concurrent reader (eg page_referenced()'s | |
1043 | * PageAnon()) will not see one without the other. | |
1044 | */ | |
1045 | WRITE_ONCE(page->mapping, (struct address_space *) anon_vma); | |
c44b6743 RR |
1046 | } |
1047 | ||
9617d95e | 1048 | /** |
4e1c1975 AK |
1049 | * __page_set_anon_rmap - set up new anonymous rmap |
1050 | * @page: Page to add to rmap | |
1051 | * @vma: VM area to add page to. | |
1052 | * @address: User virtual address of the mapping | |
e8a03feb | 1053 | * @exclusive: the page is exclusively owned by the current process |
9617d95e NP |
1054 | */ |
1055 | static void __page_set_anon_rmap(struct page *page, | |
e8a03feb | 1056 | struct vm_area_struct *vma, unsigned long address, int exclusive) |
9617d95e | 1057 | { |
e8a03feb | 1058 | struct anon_vma *anon_vma = vma->anon_vma; |
ea90002b | 1059 | |
e8a03feb | 1060 | BUG_ON(!anon_vma); |
ea90002b | 1061 | |
4e1c1975 AK |
1062 | if (PageAnon(page)) |
1063 | return; | |
1064 | ||
ea90002b | 1065 | /* |
e8a03feb RR |
1066 | * If the page isn't exclusively mapped into this vma, |
1067 | * we must use the _oldest_ possible anon_vma for the | |
1068 | * page mapping! | |
ea90002b | 1069 | */ |
4e1c1975 | 1070 | if (!exclusive) |
288468c3 | 1071 | anon_vma = anon_vma->root; |
9617d95e | 1072 | |
9617d95e NP |
1073 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; |
1074 | page->mapping = (struct address_space *) anon_vma; | |
9617d95e | 1075 | page->index = linear_page_index(vma, address); |
9617d95e NP |
1076 | } |
1077 | ||
c97a9e10 | 1078 | /** |
43d8eac4 | 1079 | * __page_check_anon_rmap - sanity check anonymous rmap addition |
c97a9e10 NP |
1080 | * @page: the page to add the mapping to |
1081 | * @vma: the vm area in which the mapping is added | |
1082 | * @address: the user virtual address mapped | |
1083 | */ | |
1084 | static void __page_check_anon_rmap(struct page *page, | |
1085 | struct vm_area_struct *vma, unsigned long address) | |
1086 | { | |
1087 | #ifdef CONFIG_DEBUG_VM | |
1088 | /* | |
1089 | * The page's anon-rmap details (mapping and index) are guaranteed to | |
1090 | * be set up correctly at this point. | |
1091 | * | |
1092 | * We have exclusion against page_add_anon_rmap because the caller | |
1093 | * always holds the page locked, except if called from page_dup_rmap, | |
1094 | * in which case the page is already known to be setup. | |
1095 | * | |
1096 | * We have exclusion against page_add_new_anon_rmap because those pages | |
1097 | * are initially only visible via the pagetables, and the pte is locked | |
1098 | * over the call to page_add_new_anon_rmap. | |
1099 | */ | |
44ab57a0 | 1100 | BUG_ON(page_anon_vma(page)->root != vma->anon_vma->root); |
c97a9e10 NP |
1101 | BUG_ON(page->index != linear_page_index(vma, address)); |
1102 | #endif | |
1103 | } | |
1104 | ||
1da177e4 LT |
1105 | /** |
1106 | * page_add_anon_rmap - add pte mapping to an anonymous page | |
1107 | * @page: the page to add the mapping to | |
1108 | * @vma: the vm area in which the mapping is added | |
1109 | * @address: the user virtual address mapped | |
1110 | * | |
5ad64688 | 1111 | * The caller needs to hold the pte lock, and the page must be locked in |
80e14822 HD |
1112 | * the anon_vma case: to serialize mapping,index checking after setting, |
1113 | * and to ensure that PageAnon is not being upgraded racily to PageKsm | |
1114 | * (but PageKsm is never downgraded to PageAnon). | |
1da177e4 LT |
1115 | */ |
1116 | void page_add_anon_rmap(struct page *page, | |
1117 | struct vm_area_struct *vma, unsigned long address) | |
ad8c2ee8 RR |
1118 | { |
1119 | do_page_add_anon_rmap(page, vma, address, 0); | |
1120 | } | |
1121 | ||
1122 | /* | |
1123 | * Special version of the above for do_swap_page, which often runs | |
1124 | * into pages that are exclusively owned by the current process. | |
1125 | * Everybody else should continue to use page_add_anon_rmap above. | |
1126 | */ | |
1127 | void do_page_add_anon_rmap(struct page *page, | |
1128 | struct vm_area_struct *vma, unsigned long address, int exclusive) | |
1da177e4 | 1129 | { |
5ad64688 | 1130 | int first = atomic_inc_and_test(&page->_mapcount); |
79134171 | 1131 | if (first) { |
bea04b07 JZ |
1132 | /* |
1133 | * We use the irq-unsafe __{inc|mod}_zone_page_stat because | |
1134 | * these counters are not modified in interrupt context, and | |
1135 | * pte lock(a spinlock) is held, which implies preemption | |
1136 | * disabled. | |
1137 | */ | |
3cd14fcd | 1138 | if (PageTransHuge(page)) |
79134171 AA |
1139 | __inc_zone_page_state(page, |
1140 | NR_ANON_TRANSPARENT_HUGEPAGES); | |
3cd14fcd KS |
1141 | __mod_zone_page_state(page_zone(page), NR_ANON_PAGES, |
1142 | hpage_nr_pages(page)); | |
79134171 | 1143 | } |
5ad64688 HD |
1144 | if (unlikely(PageKsm(page))) |
1145 | return; | |
1146 | ||
309381fe | 1147 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
5dbe0af4 | 1148 | /* address might be in next vma when migration races vma_adjust */ |
5ad64688 | 1149 | if (first) |
ad8c2ee8 | 1150 | __page_set_anon_rmap(page, vma, address, exclusive); |
69029cd5 | 1151 | else |
c97a9e10 | 1152 | __page_check_anon_rmap(page, vma, address); |
1da177e4 LT |
1153 | } |
1154 | ||
43d8eac4 | 1155 | /** |
9617d95e NP |
1156 | * page_add_new_anon_rmap - add pte mapping to a new anonymous page |
1157 | * @page: the page to add the mapping to | |
1158 | * @vma: the vm area in which the mapping is added | |
1159 | * @address: the user virtual address mapped | |
1160 | * | |
1161 | * Same as page_add_anon_rmap but must only be called on *new* pages. | |
1162 | * This means the inc-and-test can be bypassed. | |
c97a9e10 | 1163 | * Page does not have to be locked. |
9617d95e NP |
1164 | */ |
1165 | void page_add_new_anon_rmap(struct page *page, | |
1166 | struct vm_area_struct *vma, unsigned long address) | |
1167 | { | |
81d1b09c | 1168 | VM_BUG_ON_VMA(address < vma->vm_start || address >= vma->vm_end, vma); |
cbf84b7a HD |
1169 | SetPageSwapBacked(page); |
1170 | atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */ | |
3cd14fcd | 1171 | if (PageTransHuge(page)) |
79134171 | 1172 | __inc_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES); |
3cd14fcd KS |
1173 | __mod_zone_page_state(page_zone(page), NR_ANON_PAGES, |
1174 | hpage_nr_pages(page)); | |
e8a03feb | 1175 | __page_set_anon_rmap(page, vma, address, 1); |
9617d95e NP |
1176 | } |
1177 | ||
1da177e4 LT |
1178 | /** |
1179 | * page_add_file_rmap - add pte mapping to a file page | |
1180 | * @page: the page to add the mapping to | |
1181 | * | |
b8072f09 | 1182 | * The caller needs to hold the pte lock. |
1da177e4 LT |
1183 | */ |
1184 | void page_add_file_rmap(struct page *page) | |
1185 | { | |
d7365e78 | 1186 | struct mem_cgroup *memcg; |
89c06bd5 | 1187 | |
6de22619 | 1188 | memcg = mem_cgroup_begin_page_stat(page); |
d69b042f | 1189 | if (atomic_inc_and_test(&page->_mapcount)) { |
65ba55f5 | 1190 | __inc_zone_page_state(page, NR_FILE_MAPPED); |
d7365e78 | 1191 | mem_cgroup_inc_page_stat(memcg, MEM_CGROUP_STAT_FILE_MAPPED); |
d69b042f | 1192 | } |
6de22619 | 1193 | mem_cgroup_end_page_stat(memcg); |
1da177e4 LT |
1194 | } |
1195 | ||
8186eb6a JW |
1196 | static void page_remove_file_rmap(struct page *page) |
1197 | { | |
1198 | struct mem_cgroup *memcg; | |
8186eb6a | 1199 | |
6de22619 | 1200 | memcg = mem_cgroup_begin_page_stat(page); |
8186eb6a JW |
1201 | |
1202 | /* page still mapped by someone else? */ | |
1203 | if (!atomic_add_negative(-1, &page->_mapcount)) | |
1204 | goto out; | |
1205 | ||
1206 | /* Hugepages are not counted in NR_FILE_MAPPED for now. */ | |
1207 | if (unlikely(PageHuge(page))) | |
1208 | goto out; | |
1209 | ||
1210 | /* | |
1211 | * We use the irq-unsafe __{inc|mod}_zone_page_stat because | |
1212 | * these counters are not modified in interrupt context, and | |
1213 | * pte lock(a spinlock) is held, which implies preemption disabled. | |
1214 | */ | |
1215 | __dec_zone_page_state(page, NR_FILE_MAPPED); | |
1216 | mem_cgroup_dec_page_stat(memcg, MEM_CGROUP_STAT_FILE_MAPPED); | |
1217 | ||
1218 | if (unlikely(PageMlocked(page))) | |
1219 | clear_page_mlock(page); | |
1220 | out: | |
6de22619 | 1221 | mem_cgroup_end_page_stat(memcg); |
8186eb6a JW |
1222 | } |
1223 | ||
1da177e4 LT |
1224 | /** |
1225 | * page_remove_rmap - take down pte mapping from a page | |
1226 | * @page: page to remove mapping from | |
1227 | * | |
b8072f09 | 1228 | * The caller needs to hold the pte lock. |
1da177e4 | 1229 | */ |
edc315fd | 1230 | void page_remove_rmap(struct page *page) |
1da177e4 | 1231 | { |
8186eb6a JW |
1232 | if (!PageAnon(page)) { |
1233 | page_remove_file_rmap(page); | |
1234 | return; | |
1235 | } | |
89c06bd5 | 1236 | |
b904dcfe KM |
1237 | /* page still mapped by someone else? */ |
1238 | if (!atomic_add_negative(-1, &page->_mapcount)) | |
8186eb6a JW |
1239 | return; |
1240 | ||
1241 | /* Hugepages are not counted in NR_ANON_PAGES for now. */ | |
1242 | if (unlikely(PageHuge(page))) | |
1243 | return; | |
b904dcfe | 1244 | |
0fe6e20b | 1245 | /* |
bea04b07 JZ |
1246 | * We use the irq-unsafe __{inc|mod}_zone_page_stat because |
1247 | * these counters are not modified in interrupt context, and | |
bea04b07 | 1248 | * pte lock(a spinlock) is held, which implies preemption disabled. |
0fe6e20b | 1249 | */ |
8186eb6a JW |
1250 | if (PageTransHuge(page)) |
1251 | __dec_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES); | |
1252 | ||
1253 | __mod_zone_page_state(page_zone(page), NR_ANON_PAGES, | |
1254 | -hpage_nr_pages(page)); | |
1255 | ||
e6c509f8 HD |
1256 | if (unlikely(PageMlocked(page))) |
1257 | clear_page_mlock(page); | |
8186eb6a | 1258 | |
b904dcfe KM |
1259 | /* |
1260 | * It would be tidy to reset the PageAnon mapping here, | |
1261 | * but that might overwrite a racing page_add_anon_rmap | |
1262 | * which increments mapcount after us but sets mapping | |
1263 | * before us: so leave the reset to free_hot_cold_page, | |
1264 | * and remember that it's only reliable while mapped. | |
1265 | * Leaving it set also helps swapoff to reinstate ptes | |
1266 | * faster for those pages still in swapcache. | |
1267 | */ | |
1da177e4 LT |
1268 | } |
1269 | ||
1270 | /* | |
52629506 | 1271 | * @arg: enum ttu_flags will be passed to this argument |
1da177e4 | 1272 | */ |
ac769501 | 1273 | static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma, |
52629506 | 1274 | unsigned long address, void *arg) |
1da177e4 LT |
1275 | { |
1276 | struct mm_struct *mm = vma->vm_mm; | |
1da177e4 LT |
1277 | pte_t *pte; |
1278 | pte_t pteval; | |
c0718806 | 1279 | spinlock_t *ptl; |
1da177e4 | 1280 | int ret = SWAP_AGAIN; |
52629506 | 1281 | enum ttu_flags flags = (enum ttu_flags)arg; |
1da177e4 | 1282 | |
479db0bf | 1283 | pte = page_check_address(page, mm, address, &ptl, 0); |
c0718806 | 1284 | if (!pte) |
81b4082d | 1285 | goto out; |
1da177e4 LT |
1286 | |
1287 | /* | |
1288 | * If the page is mlock()d, we cannot swap it out. | |
1289 | * If it's recently referenced (perhaps page_referenced | |
1290 | * skipped over this mm) then we should reactivate it. | |
1291 | */ | |
14fa31b8 | 1292 | if (!(flags & TTU_IGNORE_MLOCK)) { |
caed0f48 KM |
1293 | if (vma->vm_flags & VM_LOCKED) |
1294 | goto out_mlock; | |
1295 | ||
daa5ba76 | 1296 | if (flags & TTU_MUNLOCK) |
53f79acb | 1297 | goto out_unmap; |
14fa31b8 AK |
1298 | } |
1299 | if (!(flags & TTU_IGNORE_ACCESS)) { | |
b291f000 NP |
1300 | if (ptep_clear_flush_young_notify(vma, address, pte)) { |
1301 | ret = SWAP_FAIL; | |
1302 | goto out_unmap; | |
1303 | } | |
1304 | } | |
1da177e4 | 1305 | |
1da177e4 LT |
1306 | /* Nuke the page table entry. */ |
1307 | flush_cache_page(vma, address, page_to_pfn(page)); | |
72b252ae MG |
1308 | if (should_defer_flush(mm, flags)) { |
1309 | /* | |
1310 | * We clear the PTE but do not flush so potentially a remote | |
1311 | * CPU could still be writing to the page. If the entry was | |
1312 | * previously clean then the architecture must guarantee that | |
1313 | * a clear->dirty transition on a cached TLB entry is written | |
1314 | * through and traps if the PTE is unmapped. | |
1315 | */ | |
1316 | pteval = ptep_get_and_clear(mm, address, pte); | |
1317 | ||
1318 | /* Potentially writable TLBs must be flushed before IO */ | |
1319 | if (pte_dirty(pteval)) | |
1320 | flush_tlb_page(vma, address); | |
1321 | else | |
1322 | set_tlb_ubc_flush_pending(mm, page); | |
1323 | } else { | |
1324 | pteval = ptep_clear_flush(vma, address, pte); | |
1325 | } | |
1da177e4 LT |
1326 | |
1327 | /* Move the dirty bit to the physical page now the pte is gone. */ | |
1328 | if (pte_dirty(pteval)) | |
1329 | set_page_dirty(page); | |
1330 | ||
365e9c87 HD |
1331 | /* Update high watermark before we lower rss */ |
1332 | update_hiwater_rss(mm); | |
1333 | ||
888b9f7c | 1334 | if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) { |
5f24ae58 NH |
1335 | if (!PageHuge(page)) { |
1336 | if (PageAnon(page)) | |
1337 | dec_mm_counter(mm, MM_ANONPAGES); | |
1338 | else | |
1339 | dec_mm_counter(mm, MM_FILEPAGES); | |
1340 | } | |
888b9f7c | 1341 | set_pte_at(mm, address, pte, |
5f24ae58 | 1342 | swp_entry_to_pte(make_hwpoison_entry(page))); |
45961722 KW |
1343 | } else if (pte_unused(pteval)) { |
1344 | /* | |
1345 | * The guest indicated that the page content is of no | |
1346 | * interest anymore. Simply discard the pte, vmscan | |
1347 | * will take care of the rest. | |
1348 | */ | |
1349 | if (PageAnon(page)) | |
1350 | dec_mm_counter(mm, MM_ANONPAGES); | |
1351 | else | |
1352 | dec_mm_counter(mm, MM_FILEPAGES); | |
888b9f7c | 1353 | } else if (PageAnon(page)) { |
4c21e2f2 | 1354 | swp_entry_t entry = { .val = page_private(page) }; |
179ef71c | 1355 | pte_t swp_pte; |
0697212a CL |
1356 | |
1357 | if (PageSwapCache(page)) { | |
1358 | /* | |
1359 | * Store the swap location in the pte. | |
1360 | * See handle_pte_fault() ... | |
1361 | */ | |
570a335b HD |
1362 | if (swap_duplicate(entry) < 0) { |
1363 | set_pte_at(mm, address, pte, pteval); | |
1364 | ret = SWAP_FAIL; | |
1365 | goto out_unmap; | |
1366 | } | |
0697212a CL |
1367 | if (list_empty(&mm->mmlist)) { |
1368 | spin_lock(&mmlist_lock); | |
1369 | if (list_empty(&mm->mmlist)) | |
1370 | list_add(&mm->mmlist, &init_mm.mmlist); | |
1371 | spin_unlock(&mmlist_lock); | |
1372 | } | |
d559db08 | 1373 | dec_mm_counter(mm, MM_ANONPAGES); |
b084d435 | 1374 | inc_mm_counter(mm, MM_SWAPENTS); |
ce1744f4 | 1375 | } else if (IS_ENABLED(CONFIG_MIGRATION)) { |
0697212a CL |
1376 | /* |
1377 | * Store the pfn of the page in a special migration | |
1378 | * pte. do_swap_page() will wait until the migration | |
1379 | * pte is removed and then restart fault handling. | |
1380 | */ | |
daa5ba76 | 1381 | BUG_ON(!(flags & TTU_MIGRATION)); |
0697212a | 1382 | entry = make_migration_entry(page, pte_write(pteval)); |
1da177e4 | 1383 | } |
179ef71c CG |
1384 | swp_pte = swp_entry_to_pte(entry); |
1385 | if (pte_soft_dirty(pteval)) | |
1386 | swp_pte = pte_swp_mksoft_dirty(swp_pte); | |
1387 | set_pte_at(mm, address, pte, swp_pte); | |
ce1744f4 | 1388 | } else if (IS_ENABLED(CONFIG_MIGRATION) && |
daa5ba76 | 1389 | (flags & TTU_MIGRATION)) { |
04e62a29 CL |
1390 | /* Establish migration entry for a file page */ |
1391 | swp_entry_t entry; | |
1392 | entry = make_migration_entry(page, pte_write(pteval)); | |
1393 | set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); | |
1394 | } else | |
d559db08 | 1395 | dec_mm_counter(mm, MM_FILEPAGES); |
1da177e4 | 1396 | |
edc315fd | 1397 | page_remove_rmap(page); |
1da177e4 LT |
1398 | page_cache_release(page); |
1399 | ||
1400 | out_unmap: | |
c0718806 | 1401 | pte_unmap_unlock(pte, ptl); |
daa5ba76 | 1402 | if (ret != SWAP_FAIL && !(flags & TTU_MUNLOCK)) |
2ec74c3e | 1403 | mmu_notifier_invalidate_page(mm, address); |
caed0f48 KM |
1404 | out: |
1405 | return ret; | |
53f79acb | 1406 | |
caed0f48 KM |
1407 | out_mlock: |
1408 | pte_unmap_unlock(pte, ptl); | |
1409 | ||
1410 | ||
1411 | /* | |
1412 | * We need mmap_sem locking, Otherwise VM_LOCKED check makes | |
1413 | * unstable result and race. Plus, We can't wait here because | |
c8c06efa | 1414 | * we now hold anon_vma->rwsem or mapping->i_mmap_rwsem. |
caed0f48 KM |
1415 | * if trylock failed, the page remain in evictable lru and later |
1416 | * vmscan could retry to move the page to unevictable lru if the | |
1417 | * page is actually mlocked. | |
1418 | */ | |
1419 | if (down_read_trylock(&vma->vm_mm->mmap_sem)) { | |
1420 | if (vma->vm_flags & VM_LOCKED) { | |
1421 | mlock_vma_page(page); | |
1422 | ret = SWAP_MLOCK; | |
53f79acb | 1423 | } |
caed0f48 | 1424 | up_read(&vma->vm_mm->mmap_sem); |
53f79acb | 1425 | } |
1da177e4 LT |
1426 | return ret; |
1427 | } | |
1428 | ||
71e3aac0 | 1429 | bool is_vma_temporary_stack(struct vm_area_struct *vma) |
a8bef8ff MG |
1430 | { |
1431 | int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP); | |
1432 | ||
1433 | if (!maybe_stack) | |
1434 | return false; | |
1435 | ||
1436 | if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) == | |
1437 | VM_STACK_INCOMPLETE_SETUP) | |
1438 | return true; | |
1439 | ||
1440 | return false; | |
1441 | } | |
1442 | ||
52629506 JK |
1443 | static bool invalid_migration_vma(struct vm_area_struct *vma, void *arg) |
1444 | { | |
1445 | return is_vma_temporary_stack(vma); | |
1446 | } | |
1447 | ||
52629506 JK |
1448 | static int page_not_mapped(struct page *page) |
1449 | { | |
1450 | return !page_mapped(page); | |
1451 | }; | |
1452 | ||
1da177e4 LT |
1453 | /** |
1454 | * try_to_unmap - try to remove all page table mappings to a page | |
1455 | * @page: the page to get unmapped | |
14fa31b8 | 1456 | * @flags: action and flags |
1da177e4 LT |
1457 | * |
1458 | * Tries to remove all the page table entries which are mapping this | |
1459 | * page, used in the pageout path. Caller must hold the page lock. | |
1460 | * Return values are: | |
1461 | * | |
1462 | * SWAP_SUCCESS - we succeeded in removing all mappings | |
1463 | * SWAP_AGAIN - we missed a mapping, try again later | |
1464 | * SWAP_FAIL - the page is unswappable | |
b291f000 | 1465 | * SWAP_MLOCK - page is mlocked. |
1da177e4 | 1466 | */ |
14fa31b8 | 1467 | int try_to_unmap(struct page *page, enum ttu_flags flags) |
1da177e4 LT |
1468 | { |
1469 | int ret; | |
52629506 JK |
1470 | struct rmap_walk_control rwc = { |
1471 | .rmap_one = try_to_unmap_one, | |
1472 | .arg = (void *)flags, | |
1473 | .done = page_not_mapped, | |
52629506 JK |
1474 | .anon_lock = page_lock_anon_vma_read, |
1475 | }; | |
1da177e4 | 1476 | |
309381fe | 1477 | VM_BUG_ON_PAGE(!PageHuge(page) && PageTransHuge(page), page); |
1da177e4 | 1478 | |
52629506 JK |
1479 | /* |
1480 | * During exec, a temporary VMA is setup and later moved. | |
1481 | * The VMA is moved under the anon_vma lock but not the | |
1482 | * page tables leading to a race where migration cannot | |
1483 | * find the migration ptes. Rather than increasing the | |
1484 | * locking requirements of exec(), migration skips | |
1485 | * temporary VMAs until after exec() completes. | |
1486 | */ | |
daa5ba76 | 1487 | if ((flags & TTU_MIGRATION) && !PageKsm(page) && PageAnon(page)) |
52629506 JK |
1488 | rwc.invalid_vma = invalid_migration_vma; |
1489 | ||
1490 | ret = rmap_walk(page, &rwc); | |
1491 | ||
b291f000 | 1492 | if (ret != SWAP_MLOCK && !page_mapped(page)) |
1da177e4 LT |
1493 | ret = SWAP_SUCCESS; |
1494 | return ret; | |
1495 | } | |
81b4082d | 1496 | |
b291f000 NP |
1497 | /** |
1498 | * try_to_munlock - try to munlock a page | |
1499 | * @page: the page to be munlocked | |
1500 | * | |
1501 | * Called from munlock code. Checks all of the VMAs mapping the page | |
1502 | * to make sure nobody else has this page mlocked. The page will be | |
1503 | * returned with PG_mlocked cleared if no other vmas have it mlocked. | |
1504 | * | |
1505 | * Return values are: | |
1506 | * | |
53f79acb | 1507 | * SWAP_AGAIN - no vma is holding page mlocked, or, |
b291f000 | 1508 | * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem |
5ad64688 | 1509 | * SWAP_FAIL - page cannot be located at present |
b291f000 NP |
1510 | * SWAP_MLOCK - page is now mlocked. |
1511 | */ | |
1512 | int try_to_munlock(struct page *page) | |
1513 | { | |
e8351ac9 JK |
1514 | int ret; |
1515 | struct rmap_walk_control rwc = { | |
1516 | .rmap_one = try_to_unmap_one, | |
1517 | .arg = (void *)TTU_MUNLOCK, | |
1518 | .done = page_not_mapped, | |
e8351ac9 JK |
1519 | .anon_lock = page_lock_anon_vma_read, |
1520 | ||
1521 | }; | |
1522 | ||
309381fe | 1523 | VM_BUG_ON_PAGE(!PageLocked(page) || PageLRU(page), page); |
b291f000 | 1524 | |
e8351ac9 JK |
1525 | ret = rmap_walk(page, &rwc); |
1526 | return ret; | |
b291f000 | 1527 | } |
e9995ef9 | 1528 | |
01d8b20d | 1529 | void __put_anon_vma(struct anon_vma *anon_vma) |
76545066 | 1530 | { |
01d8b20d | 1531 | struct anon_vma *root = anon_vma->root; |
76545066 | 1532 | |
624483f3 | 1533 | anon_vma_free(anon_vma); |
01d8b20d PZ |
1534 | if (root != anon_vma && atomic_dec_and_test(&root->refcount)) |
1535 | anon_vma_free(root); | |
76545066 | 1536 | } |
76545066 | 1537 | |
0dd1c7bb JK |
1538 | static struct anon_vma *rmap_walk_anon_lock(struct page *page, |
1539 | struct rmap_walk_control *rwc) | |
faecd8dd JK |
1540 | { |
1541 | struct anon_vma *anon_vma; | |
1542 | ||
0dd1c7bb JK |
1543 | if (rwc->anon_lock) |
1544 | return rwc->anon_lock(page); | |
1545 | ||
faecd8dd JK |
1546 | /* |
1547 | * Note: remove_migration_ptes() cannot use page_lock_anon_vma_read() | |
1548 | * because that depends on page_mapped(); but not all its usages | |
1549 | * are holding mmap_sem. Users without mmap_sem are required to | |
1550 | * take a reference count to prevent the anon_vma disappearing | |
1551 | */ | |
1552 | anon_vma = page_anon_vma(page); | |
1553 | if (!anon_vma) | |
1554 | return NULL; | |
1555 | ||
1556 | anon_vma_lock_read(anon_vma); | |
1557 | return anon_vma; | |
1558 | } | |
1559 | ||
e9995ef9 | 1560 | /* |
e8351ac9 JK |
1561 | * rmap_walk_anon - do something to anonymous page using the object-based |
1562 | * rmap method | |
1563 | * @page: the page to be handled | |
1564 | * @rwc: control variable according to each walk type | |
1565 | * | |
1566 | * Find all the mappings of a page using the mapping pointer and the vma chains | |
1567 | * contained in the anon_vma struct it points to. | |
1568 | * | |
1569 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma | |
1570 | * where the page was found will be held for write. So, we won't recheck | |
1571 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be | |
1572 | * LOCKED. | |
e9995ef9 | 1573 | */ |
051ac83a | 1574 | static int rmap_walk_anon(struct page *page, struct rmap_walk_control *rwc) |
e9995ef9 HD |
1575 | { |
1576 | struct anon_vma *anon_vma; | |
b258d860 | 1577 | pgoff_t pgoff; |
5beb4930 | 1578 | struct anon_vma_chain *avc; |
e9995ef9 HD |
1579 | int ret = SWAP_AGAIN; |
1580 | ||
0dd1c7bb | 1581 | anon_vma = rmap_walk_anon_lock(page, rwc); |
e9995ef9 HD |
1582 | if (!anon_vma) |
1583 | return ret; | |
faecd8dd | 1584 | |
b258d860 | 1585 | pgoff = page_to_pgoff(page); |
bf181b9f | 1586 | anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) { |
5beb4930 | 1587 | struct vm_area_struct *vma = avc->vma; |
e9995ef9 | 1588 | unsigned long address = vma_address(page, vma); |
0dd1c7bb JK |
1589 | |
1590 | if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg)) | |
1591 | continue; | |
1592 | ||
051ac83a | 1593 | ret = rwc->rmap_one(page, vma, address, rwc->arg); |
e9995ef9 HD |
1594 | if (ret != SWAP_AGAIN) |
1595 | break; | |
0dd1c7bb JK |
1596 | if (rwc->done && rwc->done(page)) |
1597 | break; | |
e9995ef9 | 1598 | } |
4fc3f1d6 | 1599 | anon_vma_unlock_read(anon_vma); |
e9995ef9 HD |
1600 | return ret; |
1601 | } | |
1602 | ||
e8351ac9 JK |
1603 | /* |
1604 | * rmap_walk_file - do something to file page using the object-based rmap method | |
1605 | * @page: the page to be handled | |
1606 | * @rwc: control variable according to each walk type | |
1607 | * | |
1608 | * Find all the mappings of a page using the mapping pointer and the vma chains | |
1609 | * contained in the address_space struct it points to. | |
1610 | * | |
1611 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma | |
1612 | * where the page was found will be held for write. So, we won't recheck | |
1613 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be | |
1614 | * LOCKED. | |
1615 | */ | |
051ac83a | 1616 | static int rmap_walk_file(struct page *page, struct rmap_walk_control *rwc) |
e9995ef9 HD |
1617 | { |
1618 | struct address_space *mapping = page->mapping; | |
b258d860 | 1619 | pgoff_t pgoff; |
e9995ef9 | 1620 | struct vm_area_struct *vma; |
e9995ef9 HD |
1621 | int ret = SWAP_AGAIN; |
1622 | ||
9f32624b JK |
1623 | /* |
1624 | * The page lock not only makes sure that page->mapping cannot | |
1625 | * suddenly be NULLified by truncation, it makes sure that the | |
1626 | * structure at mapping cannot be freed and reused yet, | |
c8c06efa | 1627 | * so we can safely take mapping->i_mmap_rwsem. |
9f32624b | 1628 | */ |
81d1b09c | 1629 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
9f32624b | 1630 | |
e9995ef9 HD |
1631 | if (!mapping) |
1632 | return ret; | |
3dec0ba0 | 1633 | |
b258d860 | 1634 | pgoff = page_to_pgoff(page); |
3dec0ba0 | 1635 | i_mmap_lock_read(mapping); |
6b2dbba8 | 1636 | vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) { |
e9995ef9 | 1637 | unsigned long address = vma_address(page, vma); |
0dd1c7bb JK |
1638 | |
1639 | if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg)) | |
1640 | continue; | |
1641 | ||
051ac83a | 1642 | ret = rwc->rmap_one(page, vma, address, rwc->arg); |
e9995ef9 | 1643 | if (ret != SWAP_AGAIN) |
0dd1c7bb JK |
1644 | goto done; |
1645 | if (rwc->done && rwc->done(page)) | |
1646 | goto done; | |
e9995ef9 | 1647 | } |
0dd1c7bb | 1648 | |
0dd1c7bb | 1649 | done: |
3dec0ba0 | 1650 | i_mmap_unlock_read(mapping); |
e9995ef9 HD |
1651 | return ret; |
1652 | } | |
1653 | ||
051ac83a | 1654 | int rmap_walk(struct page *page, struct rmap_walk_control *rwc) |
e9995ef9 | 1655 | { |
e9995ef9 | 1656 | if (unlikely(PageKsm(page))) |
051ac83a | 1657 | return rmap_walk_ksm(page, rwc); |
e9995ef9 | 1658 | else if (PageAnon(page)) |
051ac83a | 1659 | return rmap_walk_anon(page, rwc); |
e9995ef9 | 1660 | else |
051ac83a | 1661 | return rmap_walk_file(page, rwc); |
e9995ef9 | 1662 | } |
0fe6e20b | 1663 | |
e3390f67 | 1664 | #ifdef CONFIG_HUGETLB_PAGE |
0fe6e20b NH |
1665 | /* |
1666 | * The following three functions are for anonymous (private mapped) hugepages. | |
1667 | * Unlike common anonymous pages, anonymous hugepages have no accounting code | |
1668 | * and no lru code, because we handle hugepages differently from common pages. | |
1669 | */ | |
1670 | static void __hugepage_set_anon_rmap(struct page *page, | |
1671 | struct vm_area_struct *vma, unsigned long address, int exclusive) | |
1672 | { | |
1673 | struct anon_vma *anon_vma = vma->anon_vma; | |
433abed6 | 1674 | |
0fe6e20b | 1675 | BUG_ON(!anon_vma); |
433abed6 NH |
1676 | |
1677 | if (PageAnon(page)) | |
1678 | return; | |
1679 | if (!exclusive) | |
1680 | anon_vma = anon_vma->root; | |
1681 | ||
0fe6e20b NH |
1682 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; |
1683 | page->mapping = (struct address_space *) anon_vma; | |
1684 | page->index = linear_page_index(vma, address); | |
1685 | } | |
1686 | ||
1687 | void hugepage_add_anon_rmap(struct page *page, | |
1688 | struct vm_area_struct *vma, unsigned long address) | |
1689 | { | |
1690 | struct anon_vma *anon_vma = vma->anon_vma; | |
1691 | int first; | |
a850ea30 NH |
1692 | |
1693 | BUG_ON(!PageLocked(page)); | |
0fe6e20b | 1694 | BUG_ON(!anon_vma); |
5dbe0af4 | 1695 | /* address might be in next vma when migration races vma_adjust */ |
0fe6e20b NH |
1696 | first = atomic_inc_and_test(&page->_mapcount); |
1697 | if (first) | |
1698 | __hugepage_set_anon_rmap(page, vma, address, 0); | |
1699 | } | |
1700 | ||
1701 | void hugepage_add_new_anon_rmap(struct page *page, | |
1702 | struct vm_area_struct *vma, unsigned long address) | |
1703 | { | |
1704 | BUG_ON(address < vma->vm_start || address >= vma->vm_end); | |
1705 | atomic_set(&page->_mapcount, 0); | |
1706 | __hugepage_set_anon_rmap(page, vma, address, 1); | |
1707 | } | |
e3390f67 | 1708 | #endif /* CONFIG_HUGETLB_PAGE */ |