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> |
33c3fc71 | 62 | #include <linux/page_idle.h> |
1da177e4 LT |
63 | |
64 | #include <asm/tlbflush.h> | |
65 | ||
72b252ae MG |
66 | #include <trace/events/tlb.h> |
67 | ||
b291f000 NP |
68 | #include "internal.h" |
69 | ||
fdd2e5f8 | 70 | static struct kmem_cache *anon_vma_cachep; |
5beb4930 | 71 | static struct kmem_cache *anon_vma_chain_cachep; |
fdd2e5f8 AB |
72 | |
73 | static inline struct anon_vma *anon_vma_alloc(void) | |
74 | { | |
01d8b20d PZ |
75 | struct anon_vma *anon_vma; |
76 | ||
77 | anon_vma = kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL); | |
78 | if (anon_vma) { | |
79 | atomic_set(&anon_vma->refcount, 1); | |
7a3ef208 KK |
80 | anon_vma->degree = 1; /* Reference for first vma */ |
81 | anon_vma->parent = anon_vma; | |
01d8b20d PZ |
82 | /* |
83 | * Initialise the anon_vma root to point to itself. If called | |
84 | * from fork, the root will be reset to the parents anon_vma. | |
85 | */ | |
86 | anon_vma->root = anon_vma; | |
87 | } | |
88 | ||
89 | return anon_vma; | |
fdd2e5f8 AB |
90 | } |
91 | ||
01d8b20d | 92 | static inline void anon_vma_free(struct anon_vma *anon_vma) |
fdd2e5f8 | 93 | { |
01d8b20d | 94 | VM_BUG_ON(atomic_read(&anon_vma->refcount)); |
88c22088 PZ |
95 | |
96 | /* | |
4fc3f1d6 | 97 | * Synchronize against page_lock_anon_vma_read() such that |
88c22088 PZ |
98 | * we can safely hold the lock without the anon_vma getting |
99 | * freed. | |
100 | * | |
101 | * Relies on the full mb implied by the atomic_dec_and_test() from | |
102 | * put_anon_vma() against the acquire barrier implied by | |
4fc3f1d6 | 103 | * down_read_trylock() from page_lock_anon_vma_read(). This orders: |
88c22088 | 104 | * |
4fc3f1d6 IM |
105 | * page_lock_anon_vma_read() VS put_anon_vma() |
106 | * down_read_trylock() atomic_dec_and_test() | |
88c22088 | 107 | * LOCK MB |
4fc3f1d6 | 108 | * atomic_read() rwsem_is_locked() |
88c22088 PZ |
109 | * |
110 | * LOCK should suffice since the actual taking of the lock must | |
111 | * happen _before_ what follows. | |
112 | */ | |
7f39dda9 | 113 | might_sleep(); |
5a505085 | 114 | if (rwsem_is_locked(&anon_vma->root->rwsem)) { |
4fc3f1d6 | 115 | anon_vma_lock_write(anon_vma); |
08b52706 | 116 | anon_vma_unlock_write(anon_vma); |
88c22088 PZ |
117 | } |
118 | ||
fdd2e5f8 AB |
119 | kmem_cache_free(anon_vma_cachep, anon_vma); |
120 | } | |
1da177e4 | 121 | |
dd34739c | 122 | static inline struct anon_vma_chain *anon_vma_chain_alloc(gfp_t gfp) |
5beb4930 | 123 | { |
dd34739c | 124 | return kmem_cache_alloc(anon_vma_chain_cachep, gfp); |
5beb4930 RR |
125 | } |
126 | ||
e574b5fd | 127 | static void anon_vma_chain_free(struct anon_vma_chain *anon_vma_chain) |
5beb4930 RR |
128 | { |
129 | kmem_cache_free(anon_vma_chain_cachep, anon_vma_chain); | |
130 | } | |
131 | ||
6583a843 KC |
132 | static void anon_vma_chain_link(struct vm_area_struct *vma, |
133 | struct anon_vma_chain *avc, | |
134 | struct anon_vma *anon_vma) | |
135 | { | |
136 | avc->vma = vma; | |
137 | avc->anon_vma = anon_vma; | |
138 | list_add(&avc->same_vma, &vma->anon_vma_chain); | |
bf181b9f | 139 | anon_vma_interval_tree_insert(avc, &anon_vma->rb_root); |
6583a843 KC |
140 | } |
141 | ||
d9d332e0 LT |
142 | /** |
143 | * anon_vma_prepare - attach an anon_vma to a memory region | |
144 | * @vma: the memory region in question | |
145 | * | |
146 | * This makes sure the memory mapping described by 'vma' has | |
147 | * an 'anon_vma' attached to it, so that we can associate the | |
148 | * anonymous pages mapped into it with that anon_vma. | |
149 | * | |
150 | * The common case will be that we already have one, but if | |
23a0790a | 151 | * not we either need to find an adjacent mapping that we |
d9d332e0 LT |
152 | * can re-use the anon_vma from (very common when the only |
153 | * reason for splitting a vma has been mprotect()), or we | |
154 | * allocate a new one. | |
155 | * | |
156 | * Anon-vma allocations are very subtle, because we may have | |
4fc3f1d6 | 157 | * optimistically looked up an anon_vma in page_lock_anon_vma_read() |
d9d332e0 LT |
158 | * and that may actually touch the spinlock even in the newly |
159 | * allocated vma (it depends on RCU to make sure that the | |
160 | * anon_vma isn't actually destroyed). | |
161 | * | |
162 | * As a result, we need to do proper anon_vma locking even | |
163 | * for the new allocation. At the same time, we do not want | |
164 | * to do any locking for the common case of already having | |
165 | * an anon_vma. | |
166 | * | |
167 | * This must be called with the mmap_sem held for reading. | |
168 | */ | |
1da177e4 LT |
169 | int anon_vma_prepare(struct vm_area_struct *vma) |
170 | { | |
171 | struct anon_vma *anon_vma = vma->anon_vma; | |
5beb4930 | 172 | struct anon_vma_chain *avc; |
1da177e4 LT |
173 | |
174 | might_sleep(); | |
175 | if (unlikely(!anon_vma)) { | |
176 | struct mm_struct *mm = vma->vm_mm; | |
d9d332e0 | 177 | struct anon_vma *allocated; |
1da177e4 | 178 | |
dd34739c | 179 | avc = anon_vma_chain_alloc(GFP_KERNEL); |
5beb4930 RR |
180 | if (!avc) |
181 | goto out_enomem; | |
182 | ||
1da177e4 | 183 | anon_vma = find_mergeable_anon_vma(vma); |
d9d332e0 LT |
184 | allocated = NULL; |
185 | if (!anon_vma) { | |
1da177e4 LT |
186 | anon_vma = anon_vma_alloc(); |
187 | if (unlikely(!anon_vma)) | |
5beb4930 | 188 | goto out_enomem_free_avc; |
1da177e4 | 189 | allocated = anon_vma; |
1da177e4 LT |
190 | } |
191 | ||
4fc3f1d6 | 192 | anon_vma_lock_write(anon_vma); |
1da177e4 LT |
193 | /* page_table_lock to protect against threads */ |
194 | spin_lock(&mm->page_table_lock); | |
195 | if (likely(!vma->anon_vma)) { | |
196 | vma->anon_vma = anon_vma; | |
6583a843 | 197 | anon_vma_chain_link(vma, avc, anon_vma); |
7a3ef208 KK |
198 | /* vma reference or self-parent link for new root */ |
199 | anon_vma->degree++; | |
1da177e4 | 200 | allocated = NULL; |
31f2b0eb | 201 | avc = NULL; |
1da177e4 LT |
202 | } |
203 | spin_unlock(&mm->page_table_lock); | |
08b52706 | 204 | anon_vma_unlock_write(anon_vma); |
31f2b0eb ON |
205 | |
206 | if (unlikely(allocated)) | |
01d8b20d | 207 | put_anon_vma(allocated); |
31f2b0eb | 208 | if (unlikely(avc)) |
5beb4930 | 209 | anon_vma_chain_free(avc); |
1da177e4 LT |
210 | } |
211 | return 0; | |
5beb4930 RR |
212 | |
213 | out_enomem_free_avc: | |
214 | anon_vma_chain_free(avc); | |
215 | out_enomem: | |
216 | return -ENOMEM; | |
1da177e4 LT |
217 | } |
218 | ||
bb4aa396 LT |
219 | /* |
220 | * This is a useful helper function for locking the anon_vma root as | |
221 | * we traverse the vma->anon_vma_chain, looping over anon_vma's that | |
222 | * have the same vma. | |
223 | * | |
224 | * Such anon_vma's should have the same root, so you'd expect to see | |
225 | * just a single mutex_lock for the whole traversal. | |
226 | */ | |
227 | static inline struct anon_vma *lock_anon_vma_root(struct anon_vma *root, struct anon_vma *anon_vma) | |
228 | { | |
229 | struct anon_vma *new_root = anon_vma->root; | |
230 | if (new_root != root) { | |
231 | if (WARN_ON_ONCE(root)) | |
5a505085 | 232 | up_write(&root->rwsem); |
bb4aa396 | 233 | root = new_root; |
5a505085 | 234 | down_write(&root->rwsem); |
bb4aa396 LT |
235 | } |
236 | return root; | |
237 | } | |
238 | ||
239 | static inline void unlock_anon_vma_root(struct anon_vma *root) | |
240 | { | |
241 | if (root) | |
5a505085 | 242 | up_write(&root->rwsem); |
bb4aa396 LT |
243 | } |
244 | ||
5beb4930 RR |
245 | /* |
246 | * Attach the anon_vmas from src to dst. | |
247 | * Returns 0 on success, -ENOMEM on failure. | |
7a3ef208 KK |
248 | * |
249 | * If dst->anon_vma is NULL this function tries to find and reuse existing | |
250 | * anon_vma which has no vmas and only one child anon_vma. This prevents | |
251 | * degradation of anon_vma hierarchy to endless linear chain in case of | |
252 | * constantly forking task. On the other hand, an anon_vma with more than one | |
253 | * child isn't reused even if there was no alive vma, thus rmap walker has a | |
254 | * good chance of avoiding scanning the whole hierarchy when it searches where | |
255 | * page is mapped. | |
5beb4930 RR |
256 | */ |
257 | int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src) | |
1da177e4 | 258 | { |
5beb4930 | 259 | struct anon_vma_chain *avc, *pavc; |
bb4aa396 | 260 | struct anon_vma *root = NULL; |
5beb4930 | 261 | |
646d87b4 | 262 | list_for_each_entry_reverse(pavc, &src->anon_vma_chain, same_vma) { |
bb4aa396 LT |
263 | struct anon_vma *anon_vma; |
264 | ||
dd34739c LT |
265 | avc = anon_vma_chain_alloc(GFP_NOWAIT | __GFP_NOWARN); |
266 | if (unlikely(!avc)) { | |
267 | unlock_anon_vma_root(root); | |
268 | root = NULL; | |
269 | avc = anon_vma_chain_alloc(GFP_KERNEL); | |
270 | if (!avc) | |
271 | goto enomem_failure; | |
272 | } | |
bb4aa396 LT |
273 | anon_vma = pavc->anon_vma; |
274 | root = lock_anon_vma_root(root, anon_vma); | |
275 | anon_vma_chain_link(dst, avc, anon_vma); | |
7a3ef208 KK |
276 | |
277 | /* | |
278 | * Reuse existing anon_vma if its degree lower than two, | |
279 | * that means it has no vma and only one anon_vma child. | |
280 | * | |
281 | * Do not chose parent anon_vma, otherwise first child | |
282 | * will always reuse it. Root anon_vma is never reused: | |
283 | * it has self-parent reference and at least one child. | |
284 | */ | |
285 | if (!dst->anon_vma && anon_vma != src->anon_vma && | |
286 | anon_vma->degree < 2) | |
287 | dst->anon_vma = anon_vma; | |
5beb4930 | 288 | } |
7a3ef208 KK |
289 | if (dst->anon_vma) |
290 | dst->anon_vma->degree++; | |
bb4aa396 | 291 | unlock_anon_vma_root(root); |
5beb4930 | 292 | return 0; |
1da177e4 | 293 | |
5beb4930 | 294 | enomem_failure: |
3fe89b3e LY |
295 | /* |
296 | * dst->anon_vma is dropped here otherwise its degree can be incorrectly | |
297 | * decremented in unlink_anon_vmas(). | |
298 | * We can safely do this because callers of anon_vma_clone() don't care | |
299 | * about dst->anon_vma if anon_vma_clone() failed. | |
300 | */ | |
301 | dst->anon_vma = NULL; | |
5beb4930 RR |
302 | unlink_anon_vmas(dst); |
303 | return -ENOMEM; | |
1da177e4 LT |
304 | } |
305 | ||
5beb4930 RR |
306 | /* |
307 | * Attach vma to its own anon_vma, as well as to the anon_vmas that | |
308 | * the corresponding VMA in the parent process is attached to. | |
309 | * Returns 0 on success, non-zero on failure. | |
310 | */ | |
311 | int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma) | |
1da177e4 | 312 | { |
5beb4930 RR |
313 | struct anon_vma_chain *avc; |
314 | struct anon_vma *anon_vma; | |
c4ea95d7 | 315 | int error; |
1da177e4 | 316 | |
5beb4930 RR |
317 | /* Don't bother if the parent process has no anon_vma here. */ |
318 | if (!pvma->anon_vma) | |
319 | return 0; | |
320 | ||
7a3ef208 KK |
321 | /* Drop inherited anon_vma, we'll reuse existing or allocate new. */ |
322 | vma->anon_vma = NULL; | |
323 | ||
5beb4930 RR |
324 | /* |
325 | * First, attach the new VMA to the parent VMA's anon_vmas, | |
326 | * so rmap can find non-COWed pages in child processes. | |
327 | */ | |
c4ea95d7 DF |
328 | error = anon_vma_clone(vma, pvma); |
329 | if (error) | |
330 | return error; | |
5beb4930 | 331 | |
7a3ef208 KK |
332 | /* An existing anon_vma has been reused, all done then. */ |
333 | if (vma->anon_vma) | |
334 | return 0; | |
335 | ||
5beb4930 RR |
336 | /* Then add our own anon_vma. */ |
337 | anon_vma = anon_vma_alloc(); | |
338 | if (!anon_vma) | |
339 | goto out_error; | |
dd34739c | 340 | avc = anon_vma_chain_alloc(GFP_KERNEL); |
5beb4930 RR |
341 | if (!avc) |
342 | goto out_error_free_anon_vma; | |
5c341ee1 RR |
343 | |
344 | /* | |
345 | * The root anon_vma's spinlock is the lock actually used when we | |
346 | * lock any of the anon_vmas in this anon_vma tree. | |
347 | */ | |
348 | anon_vma->root = pvma->anon_vma->root; | |
7a3ef208 | 349 | anon_vma->parent = pvma->anon_vma; |
76545066 | 350 | /* |
01d8b20d PZ |
351 | * With refcounts, an anon_vma can stay around longer than the |
352 | * process it belongs to. The root anon_vma needs to be pinned until | |
353 | * this anon_vma is freed, because the lock lives in the root. | |
76545066 RR |
354 | */ |
355 | get_anon_vma(anon_vma->root); | |
5beb4930 RR |
356 | /* Mark this anon_vma as the one where our new (COWed) pages go. */ |
357 | vma->anon_vma = anon_vma; | |
4fc3f1d6 | 358 | anon_vma_lock_write(anon_vma); |
5c341ee1 | 359 | anon_vma_chain_link(vma, avc, anon_vma); |
7a3ef208 | 360 | anon_vma->parent->degree++; |
08b52706 | 361 | anon_vma_unlock_write(anon_vma); |
5beb4930 RR |
362 | |
363 | return 0; | |
364 | ||
365 | out_error_free_anon_vma: | |
01d8b20d | 366 | put_anon_vma(anon_vma); |
5beb4930 | 367 | out_error: |
4946d54c | 368 | unlink_anon_vmas(vma); |
5beb4930 | 369 | return -ENOMEM; |
1da177e4 LT |
370 | } |
371 | ||
5beb4930 RR |
372 | void unlink_anon_vmas(struct vm_area_struct *vma) |
373 | { | |
374 | struct anon_vma_chain *avc, *next; | |
eee2acba | 375 | struct anon_vma *root = NULL; |
5beb4930 | 376 | |
5c341ee1 RR |
377 | /* |
378 | * Unlink each anon_vma chained to the VMA. This list is ordered | |
379 | * from newest to oldest, ensuring the root anon_vma gets freed last. | |
380 | */ | |
5beb4930 | 381 | list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) { |
eee2acba PZ |
382 | struct anon_vma *anon_vma = avc->anon_vma; |
383 | ||
384 | root = lock_anon_vma_root(root, anon_vma); | |
bf181b9f | 385 | anon_vma_interval_tree_remove(avc, &anon_vma->rb_root); |
eee2acba PZ |
386 | |
387 | /* | |
388 | * Leave empty anon_vmas on the list - we'll need | |
389 | * to free them outside the lock. | |
390 | */ | |
7a3ef208 KK |
391 | if (RB_EMPTY_ROOT(&anon_vma->rb_root)) { |
392 | anon_vma->parent->degree--; | |
eee2acba | 393 | continue; |
7a3ef208 | 394 | } |
eee2acba PZ |
395 | |
396 | list_del(&avc->same_vma); | |
397 | anon_vma_chain_free(avc); | |
398 | } | |
7a3ef208 KK |
399 | if (vma->anon_vma) |
400 | vma->anon_vma->degree--; | |
eee2acba PZ |
401 | unlock_anon_vma_root(root); |
402 | ||
403 | /* | |
404 | * Iterate the list once more, it now only contains empty and unlinked | |
405 | * anon_vmas, destroy them. Could not do before due to __put_anon_vma() | |
5a505085 | 406 | * needing to write-acquire the anon_vma->root->rwsem. |
eee2acba PZ |
407 | */ |
408 | list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) { | |
409 | struct anon_vma *anon_vma = avc->anon_vma; | |
410 | ||
7a3ef208 | 411 | BUG_ON(anon_vma->degree); |
eee2acba PZ |
412 | put_anon_vma(anon_vma); |
413 | ||
5beb4930 RR |
414 | list_del(&avc->same_vma); |
415 | anon_vma_chain_free(avc); | |
416 | } | |
417 | } | |
418 | ||
51cc5068 | 419 | static void anon_vma_ctor(void *data) |
1da177e4 | 420 | { |
a35afb83 | 421 | struct anon_vma *anon_vma = data; |
1da177e4 | 422 | |
5a505085 | 423 | init_rwsem(&anon_vma->rwsem); |
83813267 | 424 | atomic_set(&anon_vma->refcount, 0); |
bf181b9f | 425 | anon_vma->rb_root = RB_ROOT; |
1da177e4 LT |
426 | } |
427 | ||
428 | void __init anon_vma_init(void) | |
429 | { | |
430 | anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma), | |
5d097056 VD |
431 | 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC|SLAB_ACCOUNT, |
432 | anon_vma_ctor); | |
433 | anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain, | |
434 | SLAB_PANIC|SLAB_ACCOUNT); | |
1da177e4 LT |
435 | } |
436 | ||
437 | /* | |
6111e4ca PZ |
438 | * Getting a lock on a stable anon_vma from a page off the LRU is tricky! |
439 | * | |
440 | * Since there is no serialization what so ever against page_remove_rmap() | |
441 | * the best this function can do is return a locked anon_vma that might | |
442 | * have been relevant to this page. | |
443 | * | |
444 | * The page might have been remapped to a different anon_vma or the anon_vma | |
445 | * returned may already be freed (and even reused). | |
446 | * | |
bc658c96 PZ |
447 | * In case it was remapped to a different anon_vma, the new anon_vma will be a |
448 | * child of the old anon_vma, and the anon_vma lifetime rules will therefore | |
449 | * ensure that any anon_vma obtained from the page will still be valid for as | |
450 | * long as we observe page_mapped() [ hence all those page_mapped() tests ]. | |
451 | * | |
6111e4ca PZ |
452 | * All users of this function must be very careful when walking the anon_vma |
453 | * chain and verify that the page in question is indeed mapped in it | |
454 | * [ something equivalent to page_mapped_in_vma() ]. | |
455 | * | |
456 | * Since anon_vma's slab is DESTROY_BY_RCU and we know from page_remove_rmap() | |
457 | * that the anon_vma pointer from page->mapping is valid if there is a | |
458 | * mapcount, we can dereference the anon_vma after observing those. | |
1da177e4 | 459 | */ |
746b18d4 | 460 | struct anon_vma *page_get_anon_vma(struct page *page) |
1da177e4 | 461 | { |
746b18d4 | 462 | struct anon_vma *anon_vma = NULL; |
1da177e4 LT |
463 | unsigned long anon_mapping; |
464 | ||
465 | rcu_read_lock(); | |
4db0c3c2 | 466 | anon_mapping = (unsigned long)READ_ONCE(page->mapping); |
3ca7b3c5 | 467 | if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON) |
1da177e4 LT |
468 | goto out; |
469 | if (!page_mapped(page)) | |
470 | goto out; | |
471 | ||
472 | anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON); | |
746b18d4 PZ |
473 | if (!atomic_inc_not_zero(&anon_vma->refcount)) { |
474 | anon_vma = NULL; | |
475 | goto out; | |
476 | } | |
f1819427 HD |
477 | |
478 | /* | |
479 | * If this page is still mapped, then its anon_vma cannot have been | |
746b18d4 PZ |
480 | * freed. But if it has been unmapped, we have no security against the |
481 | * anon_vma structure being freed and reused (for another anon_vma: | |
482 | * SLAB_DESTROY_BY_RCU guarantees that - so the atomic_inc_not_zero() | |
483 | * above cannot corrupt). | |
f1819427 | 484 | */ |
746b18d4 | 485 | if (!page_mapped(page)) { |
7f39dda9 | 486 | rcu_read_unlock(); |
746b18d4 | 487 | put_anon_vma(anon_vma); |
7f39dda9 | 488 | return NULL; |
746b18d4 | 489 | } |
1da177e4 LT |
490 | out: |
491 | rcu_read_unlock(); | |
746b18d4 PZ |
492 | |
493 | return anon_vma; | |
494 | } | |
495 | ||
88c22088 PZ |
496 | /* |
497 | * Similar to page_get_anon_vma() except it locks the anon_vma. | |
498 | * | |
499 | * Its a little more complex as it tries to keep the fast path to a single | |
500 | * atomic op -- the trylock. If we fail the trylock, we fall back to getting a | |
501 | * reference like with page_get_anon_vma() and then block on the mutex. | |
502 | */ | |
4fc3f1d6 | 503 | struct anon_vma *page_lock_anon_vma_read(struct page *page) |
746b18d4 | 504 | { |
88c22088 | 505 | struct anon_vma *anon_vma = NULL; |
eee0f252 | 506 | struct anon_vma *root_anon_vma; |
88c22088 | 507 | unsigned long anon_mapping; |
746b18d4 | 508 | |
88c22088 | 509 | rcu_read_lock(); |
4db0c3c2 | 510 | anon_mapping = (unsigned long)READ_ONCE(page->mapping); |
88c22088 PZ |
511 | if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON) |
512 | goto out; | |
513 | if (!page_mapped(page)) | |
514 | goto out; | |
515 | ||
516 | anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON); | |
4db0c3c2 | 517 | root_anon_vma = READ_ONCE(anon_vma->root); |
4fc3f1d6 | 518 | if (down_read_trylock(&root_anon_vma->rwsem)) { |
88c22088 | 519 | /* |
eee0f252 HD |
520 | * If the page is still mapped, then this anon_vma is still |
521 | * its anon_vma, and holding the mutex ensures that it will | |
bc658c96 | 522 | * not go away, see anon_vma_free(). |
88c22088 | 523 | */ |
eee0f252 | 524 | if (!page_mapped(page)) { |
4fc3f1d6 | 525 | up_read(&root_anon_vma->rwsem); |
88c22088 PZ |
526 | anon_vma = NULL; |
527 | } | |
528 | goto out; | |
529 | } | |
746b18d4 | 530 | |
88c22088 PZ |
531 | /* trylock failed, we got to sleep */ |
532 | if (!atomic_inc_not_zero(&anon_vma->refcount)) { | |
533 | anon_vma = NULL; | |
534 | goto out; | |
535 | } | |
536 | ||
537 | if (!page_mapped(page)) { | |
7f39dda9 | 538 | rcu_read_unlock(); |
88c22088 | 539 | put_anon_vma(anon_vma); |
7f39dda9 | 540 | return NULL; |
88c22088 PZ |
541 | } |
542 | ||
543 | /* we pinned the anon_vma, its safe to sleep */ | |
544 | rcu_read_unlock(); | |
4fc3f1d6 | 545 | anon_vma_lock_read(anon_vma); |
88c22088 PZ |
546 | |
547 | if (atomic_dec_and_test(&anon_vma->refcount)) { | |
548 | /* | |
549 | * Oops, we held the last refcount, release the lock | |
550 | * and bail -- can't simply use put_anon_vma() because | |
4fc3f1d6 | 551 | * we'll deadlock on the anon_vma_lock_write() recursion. |
88c22088 | 552 | */ |
4fc3f1d6 | 553 | anon_vma_unlock_read(anon_vma); |
88c22088 PZ |
554 | __put_anon_vma(anon_vma); |
555 | anon_vma = NULL; | |
556 | } | |
557 | ||
558 | return anon_vma; | |
559 | ||
560 | out: | |
561 | rcu_read_unlock(); | |
746b18d4 | 562 | return anon_vma; |
34bbd704 ON |
563 | } |
564 | ||
4fc3f1d6 | 565 | void page_unlock_anon_vma_read(struct anon_vma *anon_vma) |
34bbd704 | 566 | { |
4fc3f1d6 | 567 | anon_vma_unlock_read(anon_vma); |
1da177e4 LT |
568 | } |
569 | ||
72b252ae MG |
570 | #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH |
571 | static void percpu_flush_tlb_batch_pages(void *data) | |
572 | { | |
573 | /* | |
574 | * All TLB entries are flushed on the assumption that it is | |
575 | * cheaper to flush all TLBs and let them be refilled than | |
576 | * flushing individual PFNs. Note that we do not track mm's | |
577 | * to flush as that might simply be multiple full TLB flushes | |
578 | * for no gain. | |
579 | */ | |
580 | count_vm_tlb_event(NR_TLB_REMOTE_FLUSH_RECEIVED); | |
581 | flush_tlb_local(); | |
582 | } | |
583 | ||
584 | /* | |
585 | * Flush TLB entries for recently unmapped pages from remote CPUs. It is | |
586 | * important if a PTE was dirty when it was unmapped that it's flushed | |
587 | * before any IO is initiated on the page to prevent lost writes. Similarly, | |
588 | * it must be flushed before freeing to prevent data leakage. | |
589 | */ | |
590 | void try_to_unmap_flush(void) | |
591 | { | |
592 | struct tlbflush_unmap_batch *tlb_ubc = ¤t->tlb_ubc; | |
593 | int cpu; | |
594 | ||
595 | if (!tlb_ubc->flush_required) | |
596 | return; | |
597 | ||
598 | cpu = get_cpu(); | |
599 | ||
600 | trace_tlb_flush(TLB_REMOTE_SHOOTDOWN, -1UL); | |
601 | ||
602 | if (cpumask_test_cpu(cpu, &tlb_ubc->cpumask)) | |
603 | percpu_flush_tlb_batch_pages(&tlb_ubc->cpumask); | |
604 | ||
605 | if (cpumask_any_but(&tlb_ubc->cpumask, cpu) < nr_cpu_ids) { | |
606 | smp_call_function_many(&tlb_ubc->cpumask, | |
607 | percpu_flush_tlb_batch_pages, (void *)tlb_ubc, true); | |
608 | } | |
609 | cpumask_clear(&tlb_ubc->cpumask); | |
610 | tlb_ubc->flush_required = false; | |
d950c947 | 611 | tlb_ubc->writable = false; |
72b252ae MG |
612 | put_cpu(); |
613 | } | |
614 | ||
d950c947 MG |
615 | /* Flush iff there are potentially writable TLB entries that can race with IO */ |
616 | void try_to_unmap_flush_dirty(void) | |
617 | { | |
618 | struct tlbflush_unmap_batch *tlb_ubc = ¤t->tlb_ubc; | |
619 | ||
620 | if (tlb_ubc->writable) | |
621 | try_to_unmap_flush(); | |
622 | } | |
623 | ||
72b252ae | 624 | static void set_tlb_ubc_flush_pending(struct mm_struct *mm, |
d950c947 | 625 | struct page *page, bool writable) |
72b252ae MG |
626 | { |
627 | struct tlbflush_unmap_batch *tlb_ubc = ¤t->tlb_ubc; | |
628 | ||
629 | cpumask_or(&tlb_ubc->cpumask, &tlb_ubc->cpumask, mm_cpumask(mm)); | |
630 | tlb_ubc->flush_required = true; | |
d950c947 MG |
631 | |
632 | /* | |
633 | * If the PTE was dirty then it's best to assume it's writable. The | |
634 | * caller must use try_to_unmap_flush_dirty() or try_to_unmap_flush() | |
635 | * before the page is queued for IO. | |
636 | */ | |
637 | if (writable) | |
638 | tlb_ubc->writable = true; | |
72b252ae MG |
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, | |
d950c947 | 661 | struct page *page, bool writable) |
72b252ae MG |
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 | ||
8749cfea | 801 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
1da177e4 | 802 | /* |
8749cfea VD |
803 | * Check that @page is mapped at @address into @mm. In contrast to |
804 | * page_check_address(), this function can handle transparent huge pages. | |
805 | * | |
806 | * On success returns true with pte mapped and locked. For PMD-mapped | |
807 | * transparent huge pages *@ptep is set to NULL. | |
1da177e4 | 808 | */ |
8749cfea VD |
809 | bool page_check_address_transhuge(struct page *page, struct mm_struct *mm, |
810 | unsigned long address, pmd_t **pmdp, | |
811 | pte_t **ptep, spinlock_t **ptlp) | |
1da177e4 | 812 | { |
b20ce5e0 KS |
813 | pgd_t *pgd; |
814 | pud_t *pud; | |
815 | pmd_t *pmd; | |
816 | pte_t *pte; | |
8749cfea | 817 | spinlock_t *ptl; |
1da177e4 | 818 | |
b20ce5e0 KS |
819 | if (unlikely(PageHuge(page))) { |
820 | /* when pud is not present, pte will be NULL */ | |
821 | pte = huge_pte_offset(mm, address); | |
822 | if (!pte) | |
8749cfea | 823 | return false; |
2da28bfd | 824 | |
b20ce5e0 | 825 | ptl = huge_pte_lockptr(page_hstate(page), mm, pte); |
8749cfea | 826 | pmd = NULL; |
b20ce5e0 KS |
827 | goto check_pte; |
828 | } | |
829 | ||
830 | pgd = pgd_offset(mm, address); | |
831 | if (!pgd_present(*pgd)) | |
8749cfea | 832 | return false; |
b20ce5e0 KS |
833 | pud = pud_offset(pgd, address); |
834 | if (!pud_present(*pud)) | |
8749cfea | 835 | return false; |
b20ce5e0 KS |
836 | pmd = pmd_offset(pud, address); |
837 | ||
838 | if (pmd_trans_huge(*pmd)) { | |
b20ce5e0 KS |
839 | ptl = pmd_lock(mm, pmd); |
840 | if (!pmd_present(*pmd)) | |
841 | goto unlock_pmd; | |
842 | if (unlikely(!pmd_trans_huge(*pmd))) { | |
117b0791 | 843 | spin_unlock(ptl); |
b20ce5e0 KS |
844 | goto map_pte; |
845 | } | |
846 | ||
847 | if (pmd_page(*pmd) != page) | |
848 | goto unlock_pmd; | |
849 | ||
8749cfea | 850 | pte = NULL; |
b20ce5e0 KS |
851 | goto found; |
852 | unlock_pmd: | |
853 | spin_unlock(ptl); | |
8749cfea | 854 | return false; |
71e3aac0 | 855 | } else { |
b20ce5e0 | 856 | pmd_t pmde = *pmd; |
71e3aac0 | 857 | |
b20ce5e0 KS |
858 | barrier(); |
859 | if (!pmd_present(pmde) || pmd_trans_huge(pmde)) | |
8749cfea | 860 | return false; |
b20ce5e0 KS |
861 | } |
862 | map_pte: | |
863 | pte = pte_offset_map(pmd, address); | |
864 | if (!pte_present(*pte)) { | |
865 | pte_unmap(pte); | |
8749cfea | 866 | return false; |
b20ce5e0 | 867 | } |
71e3aac0 | 868 | |
b20ce5e0 KS |
869 | ptl = pte_lockptr(mm, pmd); |
870 | check_pte: | |
871 | spin_lock(ptl); | |
2da28bfd | 872 | |
b20ce5e0 KS |
873 | if (!pte_present(*pte)) { |
874 | pte_unmap_unlock(pte, ptl); | |
8749cfea | 875 | return false; |
b20ce5e0 KS |
876 | } |
877 | ||
878 | /* THP can be referenced by any subpage */ | |
879 | if (pte_pfn(*pte) - page_to_pfn(page) >= hpage_nr_pages(page)) { | |
880 | pte_unmap_unlock(pte, ptl); | |
8749cfea | 881 | return false; |
b20ce5e0 | 882 | } |
8749cfea VD |
883 | found: |
884 | *ptep = pte; | |
885 | *pmdp = pmd; | |
886 | *ptlp = ptl; | |
887 | return true; | |
888 | } | |
889 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ | |
890 | ||
891 | struct page_referenced_arg { | |
892 | int mapcount; | |
893 | int referenced; | |
894 | unsigned long vm_flags; | |
895 | struct mem_cgroup *memcg; | |
896 | }; | |
897 | /* | |
898 | * arg: page_referenced_arg will be passed | |
899 | */ | |
900 | static int page_referenced_one(struct page *page, struct vm_area_struct *vma, | |
901 | unsigned long address, void *arg) | |
902 | { | |
903 | struct mm_struct *mm = vma->vm_mm; | |
904 | struct page_referenced_arg *pra = arg; | |
905 | pmd_t *pmd; | |
906 | pte_t *pte; | |
907 | spinlock_t *ptl; | |
908 | int referenced = 0; | |
909 | ||
910 | if (!page_check_address_transhuge(page, mm, address, &pmd, &pte, &ptl)) | |
911 | return SWAP_AGAIN; | |
b20ce5e0 KS |
912 | |
913 | if (vma->vm_flags & VM_LOCKED) { | |
8749cfea VD |
914 | if (pte) |
915 | pte_unmap(pte); | |
916 | spin_unlock(ptl); | |
b20ce5e0 KS |
917 | pra->vm_flags |= VM_LOCKED; |
918 | return SWAP_FAIL; /* To break the loop */ | |
71e3aac0 AA |
919 | } |
920 | ||
8749cfea VD |
921 | if (pte) { |
922 | if (ptep_clear_flush_young_notify(vma, address, pte)) { | |
923 | /* | |
924 | * Don't treat a reference through a sequentially read | |
925 | * mapping as such. If the page has been used in | |
926 | * another mapping, we will catch it; if this other | |
927 | * mapping is already gone, the unmap path will have | |
928 | * set PG_referenced or activated the page. | |
929 | */ | |
930 | if (likely(!(vma->vm_flags & VM_SEQ_READ))) | |
931 | referenced++; | |
932 | } | |
933 | pte_unmap(pte); | |
934 | } else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) { | |
935 | if (pmdp_clear_flush_young_notify(vma, address, pmd)) | |
b20ce5e0 | 936 | referenced++; |
8749cfea VD |
937 | } else { |
938 | /* unexpected pmd-mapped page? */ | |
939 | WARN_ON_ONCE(1); | |
b20ce5e0 | 940 | } |
8749cfea | 941 | spin_unlock(ptl); |
b20ce5e0 | 942 | |
33c3fc71 VD |
943 | if (referenced) |
944 | clear_page_idle(page); | |
945 | if (test_and_clear_page_young(page)) | |
946 | referenced++; | |
947 | ||
9f32624b JK |
948 | if (referenced) { |
949 | pra->referenced++; | |
950 | pra->vm_flags |= vma->vm_flags; | |
1da177e4 | 951 | } |
34bbd704 | 952 | |
9f32624b JK |
953 | pra->mapcount--; |
954 | if (!pra->mapcount) | |
955 | return SWAP_SUCCESS; /* To break the loop */ | |
956 | ||
957 | return SWAP_AGAIN; | |
1da177e4 LT |
958 | } |
959 | ||
9f32624b | 960 | static bool invalid_page_referenced_vma(struct vm_area_struct *vma, void *arg) |
1da177e4 | 961 | { |
9f32624b JK |
962 | struct page_referenced_arg *pra = arg; |
963 | struct mem_cgroup *memcg = pra->memcg; | |
1da177e4 | 964 | |
9f32624b JK |
965 | if (!mm_match_cgroup(vma->vm_mm, memcg)) |
966 | return true; | |
1da177e4 | 967 | |
9f32624b | 968 | return false; |
1da177e4 LT |
969 | } |
970 | ||
971 | /** | |
972 | * page_referenced - test if the page was referenced | |
973 | * @page: the page to test | |
974 | * @is_locked: caller holds lock on the page | |
72835c86 | 975 | * @memcg: target memory cgroup |
6fe6b7e3 | 976 | * @vm_flags: collect encountered vma->vm_flags who actually referenced the page |
1da177e4 LT |
977 | * |
978 | * Quick test_and_clear_referenced for all mappings to a page, | |
979 | * returns the number of ptes which referenced the page. | |
980 | */ | |
6fe6b7e3 WF |
981 | int page_referenced(struct page *page, |
982 | int is_locked, | |
72835c86 | 983 | struct mem_cgroup *memcg, |
6fe6b7e3 | 984 | unsigned long *vm_flags) |
1da177e4 | 985 | { |
9f32624b | 986 | int ret; |
5ad64688 | 987 | int we_locked = 0; |
9f32624b | 988 | struct page_referenced_arg pra = { |
b20ce5e0 | 989 | .mapcount = total_mapcount(page), |
9f32624b JK |
990 | .memcg = memcg, |
991 | }; | |
992 | struct rmap_walk_control rwc = { | |
993 | .rmap_one = page_referenced_one, | |
994 | .arg = (void *)&pra, | |
995 | .anon_lock = page_lock_anon_vma_read, | |
996 | }; | |
1da177e4 | 997 | |
6fe6b7e3 | 998 | *vm_flags = 0; |
9f32624b JK |
999 | if (!page_mapped(page)) |
1000 | return 0; | |
1001 | ||
1002 | if (!page_rmapping(page)) | |
1003 | return 0; | |
1004 | ||
1005 | if (!is_locked && (!PageAnon(page) || PageKsm(page))) { | |
1006 | we_locked = trylock_page(page); | |
1007 | if (!we_locked) | |
1008 | return 1; | |
1da177e4 | 1009 | } |
9f32624b JK |
1010 | |
1011 | /* | |
1012 | * If we are reclaiming on behalf of a cgroup, skip | |
1013 | * counting on behalf of references from different | |
1014 | * cgroups | |
1015 | */ | |
1016 | if (memcg) { | |
1017 | rwc.invalid_vma = invalid_page_referenced_vma; | |
1018 | } | |
1019 | ||
1020 | ret = rmap_walk(page, &rwc); | |
1021 | *vm_flags = pra.vm_flags; | |
1022 | ||
1023 | if (we_locked) | |
1024 | unlock_page(page); | |
1025 | ||
1026 | return pra.referenced; | |
1da177e4 LT |
1027 | } |
1028 | ||
1cb1729b | 1029 | static int page_mkclean_one(struct page *page, struct vm_area_struct *vma, |
9853a407 | 1030 | unsigned long address, void *arg) |
d08b3851 PZ |
1031 | { |
1032 | struct mm_struct *mm = vma->vm_mm; | |
c2fda5fe | 1033 | pte_t *pte; |
d08b3851 PZ |
1034 | spinlock_t *ptl; |
1035 | int ret = 0; | |
9853a407 | 1036 | int *cleaned = arg; |
d08b3851 | 1037 | |
479db0bf | 1038 | pte = page_check_address(page, mm, address, &ptl, 1); |
d08b3851 PZ |
1039 | if (!pte) |
1040 | goto out; | |
1041 | ||
c2fda5fe PZ |
1042 | if (pte_dirty(*pte) || pte_write(*pte)) { |
1043 | pte_t entry; | |
d08b3851 | 1044 | |
c2fda5fe | 1045 | flush_cache_page(vma, address, pte_pfn(*pte)); |
2ec74c3e | 1046 | entry = ptep_clear_flush(vma, address, pte); |
c2fda5fe PZ |
1047 | entry = pte_wrprotect(entry); |
1048 | entry = pte_mkclean(entry); | |
d6e88e67 | 1049 | set_pte_at(mm, address, pte, entry); |
c2fda5fe PZ |
1050 | ret = 1; |
1051 | } | |
d08b3851 | 1052 | |
d08b3851 | 1053 | pte_unmap_unlock(pte, ptl); |
2ec74c3e | 1054 | |
9853a407 | 1055 | if (ret) { |
2ec74c3e | 1056 | mmu_notifier_invalidate_page(mm, address); |
9853a407 JK |
1057 | (*cleaned)++; |
1058 | } | |
d08b3851 | 1059 | out: |
9853a407 | 1060 | return SWAP_AGAIN; |
d08b3851 PZ |
1061 | } |
1062 | ||
9853a407 | 1063 | static bool invalid_mkclean_vma(struct vm_area_struct *vma, void *arg) |
d08b3851 | 1064 | { |
9853a407 | 1065 | if (vma->vm_flags & VM_SHARED) |
871beb8c | 1066 | return false; |
d08b3851 | 1067 | |
871beb8c | 1068 | return true; |
d08b3851 PZ |
1069 | } |
1070 | ||
1071 | int page_mkclean(struct page *page) | |
1072 | { | |
9853a407 JK |
1073 | int cleaned = 0; |
1074 | struct address_space *mapping; | |
1075 | struct rmap_walk_control rwc = { | |
1076 | .arg = (void *)&cleaned, | |
1077 | .rmap_one = page_mkclean_one, | |
1078 | .invalid_vma = invalid_mkclean_vma, | |
1079 | }; | |
d08b3851 PZ |
1080 | |
1081 | BUG_ON(!PageLocked(page)); | |
1082 | ||
9853a407 JK |
1083 | if (!page_mapped(page)) |
1084 | return 0; | |
1085 | ||
1086 | mapping = page_mapping(page); | |
1087 | if (!mapping) | |
1088 | return 0; | |
1089 | ||
1090 | rmap_walk(page, &rwc); | |
d08b3851 | 1091 | |
9853a407 | 1092 | return cleaned; |
d08b3851 | 1093 | } |
60b59bea | 1094 | EXPORT_SYMBOL_GPL(page_mkclean); |
d08b3851 | 1095 | |
c44b6743 RR |
1096 | /** |
1097 | * page_move_anon_rmap - move a page to our anon_vma | |
1098 | * @page: the page to move to our anon_vma | |
1099 | * @vma: the vma the page belongs to | |
1100 | * @address: the user virtual address mapped | |
1101 | * | |
1102 | * When a page belongs exclusively to one process after a COW event, | |
1103 | * that page can be moved into the anon_vma that belongs to just that | |
1104 | * process, so the rmap code will not search the parent or sibling | |
1105 | * processes. | |
1106 | */ | |
1107 | void page_move_anon_rmap(struct page *page, | |
1108 | struct vm_area_struct *vma, unsigned long address) | |
1109 | { | |
1110 | struct anon_vma *anon_vma = vma->anon_vma; | |
1111 | ||
309381fe | 1112 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
81d1b09c | 1113 | VM_BUG_ON_VMA(!anon_vma, vma); |
309381fe | 1114 | VM_BUG_ON_PAGE(page->index != linear_page_index(vma, address), page); |
c44b6743 RR |
1115 | |
1116 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; | |
414e2fb8 VD |
1117 | /* |
1118 | * Ensure that anon_vma and the PAGE_MAPPING_ANON bit are written | |
1119 | * simultaneously, so a concurrent reader (eg page_referenced()'s | |
1120 | * PageAnon()) will not see one without the other. | |
1121 | */ | |
1122 | WRITE_ONCE(page->mapping, (struct address_space *) anon_vma); | |
c44b6743 RR |
1123 | } |
1124 | ||
9617d95e | 1125 | /** |
4e1c1975 AK |
1126 | * __page_set_anon_rmap - set up new anonymous rmap |
1127 | * @page: Page to add to rmap | |
1128 | * @vma: VM area to add page to. | |
1129 | * @address: User virtual address of the mapping | |
e8a03feb | 1130 | * @exclusive: the page is exclusively owned by the current process |
9617d95e NP |
1131 | */ |
1132 | static void __page_set_anon_rmap(struct page *page, | |
e8a03feb | 1133 | struct vm_area_struct *vma, unsigned long address, int exclusive) |
9617d95e | 1134 | { |
e8a03feb | 1135 | struct anon_vma *anon_vma = vma->anon_vma; |
ea90002b | 1136 | |
e8a03feb | 1137 | BUG_ON(!anon_vma); |
ea90002b | 1138 | |
4e1c1975 AK |
1139 | if (PageAnon(page)) |
1140 | return; | |
1141 | ||
ea90002b | 1142 | /* |
e8a03feb RR |
1143 | * If the page isn't exclusively mapped into this vma, |
1144 | * we must use the _oldest_ possible anon_vma for the | |
1145 | * page mapping! | |
ea90002b | 1146 | */ |
4e1c1975 | 1147 | if (!exclusive) |
288468c3 | 1148 | anon_vma = anon_vma->root; |
9617d95e | 1149 | |
9617d95e NP |
1150 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; |
1151 | page->mapping = (struct address_space *) anon_vma; | |
9617d95e | 1152 | page->index = linear_page_index(vma, address); |
9617d95e NP |
1153 | } |
1154 | ||
c97a9e10 | 1155 | /** |
43d8eac4 | 1156 | * __page_check_anon_rmap - sanity check anonymous rmap addition |
c97a9e10 NP |
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 | static void __page_check_anon_rmap(struct page *page, | |
1162 | struct vm_area_struct *vma, unsigned long address) | |
1163 | { | |
1164 | #ifdef CONFIG_DEBUG_VM | |
1165 | /* | |
1166 | * The page's anon-rmap details (mapping and index) are guaranteed to | |
1167 | * be set up correctly at this point. | |
1168 | * | |
1169 | * We have exclusion against page_add_anon_rmap because the caller | |
1170 | * always holds the page locked, except if called from page_dup_rmap, | |
1171 | * in which case the page is already known to be setup. | |
1172 | * | |
1173 | * We have exclusion against page_add_new_anon_rmap because those pages | |
1174 | * are initially only visible via the pagetables, and the pte is locked | |
1175 | * over the call to page_add_new_anon_rmap. | |
1176 | */ | |
44ab57a0 | 1177 | BUG_ON(page_anon_vma(page)->root != vma->anon_vma->root); |
53f9263b | 1178 | BUG_ON(page_to_pgoff(page) != linear_page_index(vma, address)); |
c97a9e10 NP |
1179 | #endif |
1180 | } | |
1181 | ||
1da177e4 LT |
1182 | /** |
1183 | * page_add_anon_rmap - add pte mapping to an anonymous page | |
1184 | * @page: the page to add the mapping to | |
1185 | * @vma: the vm area in which the mapping is added | |
1186 | * @address: the user virtual address mapped | |
d281ee61 | 1187 | * @compound: charge the page as compound or small page |
1da177e4 | 1188 | * |
5ad64688 | 1189 | * The caller needs to hold the pte lock, and the page must be locked in |
80e14822 HD |
1190 | * the anon_vma case: to serialize mapping,index checking after setting, |
1191 | * and to ensure that PageAnon is not being upgraded racily to PageKsm | |
1192 | * (but PageKsm is never downgraded to PageAnon). | |
1da177e4 LT |
1193 | */ |
1194 | void page_add_anon_rmap(struct page *page, | |
d281ee61 | 1195 | struct vm_area_struct *vma, unsigned long address, bool compound) |
ad8c2ee8 | 1196 | { |
d281ee61 | 1197 | do_page_add_anon_rmap(page, vma, address, compound ? RMAP_COMPOUND : 0); |
ad8c2ee8 RR |
1198 | } |
1199 | ||
1200 | /* | |
1201 | * Special version of the above for do_swap_page, which often runs | |
1202 | * into pages that are exclusively owned by the current process. | |
1203 | * Everybody else should continue to use page_add_anon_rmap above. | |
1204 | */ | |
1205 | void do_page_add_anon_rmap(struct page *page, | |
d281ee61 | 1206 | struct vm_area_struct *vma, unsigned long address, int flags) |
1da177e4 | 1207 | { |
53f9263b KS |
1208 | bool compound = flags & RMAP_COMPOUND; |
1209 | bool first; | |
1210 | ||
e9b61f19 KS |
1211 | if (compound) { |
1212 | atomic_t *mapcount; | |
53f9263b | 1213 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
e9b61f19 KS |
1214 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
1215 | mapcount = compound_mapcount_ptr(page); | |
1216 | first = atomic_inc_and_test(mapcount); | |
53f9263b KS |
1217 | } else { |
1218 | first = atomic_inc_and_test(&page->_mapcount); | |
1219 | } | |
1220 | ||
79134171 | 1221 | if (first) { |
d281ee61 | 1222 | int nr = compound ? hpage_nr_pages(page) : 1; |
bea04b07 JZ |
1223 | /* |
1224 | * We use the irq-unsafe __{inc|mod}_zone_page_stat because | |
1225 | * these counters are not modified in interrupt context, and | |
1226 | * pte lock(a spinlock) is held, which implies preemption | |
1227 | * disabled. | |
1228 | */ | |
d281ee61 | 1229 | if (compound) { |
79134171 AA |
1230 | __inc_zone_page_state(page, |
1231 | NR_ANON_TRANSPARENT_HUGEPAGES); | |
d281ee61 KS |
1232 | } |
1233 | __mod_zone_page_state(page_zone(page), NR_ANON_PAGES, nr); | |
79134171 | 1234 | } |
5ad64688 HD |
1235 | if (unlikely(PageKsm(page))) |
1236 | return; | |
1237 | ||
309381fe | 1238 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
53f9263b | 1239 | |
5dbe0af4 | 1240 | /* address might be in next vma when migration races vma_adjust */ |
5ad64688 | 1241 | if (first) |
d281ee61 KS |
1242 | __page_set_anon_rmap(page, vma, address, |
1243 | flags & RMAP_EXCLUSIVE); | |
69029cd5 | 1244 | else |
c97a9e10 | 1245 | __page_check_anon_rmap(page, vma, address); |
1da177e4 LT |
1246 | } |
1247 | ||
43d8eac4 | 1248 | /** |
9617d95e NP |
1249 | * page_add_new_anon_rmap - add pte mapping to a new anonymous page |
1250 | * @page: the page to add the mapping to | |
1251 | * @vma: the vm area in which the mapping is added | |
1252 | * @address: the user virtual address mapped | |
d281ee61 | 1253 | * @compound: charge the page as compound or small page |
9617d95e NP |
1254 | * |
1255 | * Same as page_add_anon_rmap but must only be called on *new* pages. | |
1256 | * This means the inc-and-test can be bypassed. | |
c97a9e10 | 1257 | * Page does not have to be locked. |
9617d95e NP |
1258 | */ |
1259 | void page_add_new_anon_rmap(struct page *page, | |
d281ee61 | 1260 | struct vm_area_struct *vma, unsigned long address, bool compound) |
9617d95e | 1261 | { |
d281ee61 KS |
1262 | int nr = compound ? hpage_nr_pages(page) : 1; |
1263 | ||
81d1b09c | 1264 | VM_BUG_ON_VMA(address < vma->vm_start || address >= vma->vm_end, vma); |
cbf84b7a | 1265 | SetPageSwapBacked(page); |
d281ee61 KS |
1266 | if (compound) { |
1267 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
53f9263b KS |
1268 | /* increment count (starts at -1) */ |
1269 | atomic_set(compound_mapcount_ptr(page), 0); | |
79134171 | 1270 | __inc_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES); |
53f9263b KS |
1271 | } else { |
1272 | /* Anon THP always mapped first with PMD */ | |
1273 | VM_BUG_ON_PAGE(PageTransCompound(page), page); | |
1274 | /* increment count (starts at -1) */ | |
1275 | atomic_set(&page->_mapcount, 0); | |
d281ee61 KS |
1276 | } |
1277 | __mod_zone_page_state(page_zone(page), NR_ANON_PAGES, nr); | |
e8a03feb | 1278 | __page_set_anon_rmap(page, vma, address, 1); |
9617d95e NP |
1279 | } |
1280 | ||
1da177e4 LT |
1281 | /** |
1282 | * page_add_file_rmap - add pte mapping to a file page | |
1283 | * @page: the page to add the mapping to | |
1284 | * | |
b8072f09 | 1285 | * The caller needs to hold the pte lock. |
1da177e4 LT |
1286 | */ |
1287 | void page_add_file_rmap(struct page *page) | |
1288 | { | |
d7365e78 | 1289 | struct mem_cgroup *memcg; |
89c06bd5 | 1290 | |
6de22619 | 1291 | memcg = mem_cgroup_begin_page_stat(page); |
d69b042f | 1292 | if (atomic_inc_and_test(&page->_mapcount)) { |
65ba55f5 | 1293 | __inc_zone_page_state(page, NR_FILE_MAPPED); |
d7365e78 | 1294 | mem_cgroup_inc_page_stat(memcg, MEM_CGROUP_STAT_FILE_MAPPED); |
d69b042f | 1295 | } |
6de22619 | 1296 | mem_cgroup_end_page_stat(memcg); |
1da177e4 LT |
1297 | } |
1298 | ||
8186eb6a JW |
1299 | static void page_remove_file_rmap(struct page *page) |
1300 | { | |
1301 | struct mem_cgroup *memcg; | |
8186eb6a | 1302 | |
6de22619 | 1303 | memcg = mem_cgroup_begin_page_stat(page); |
8186eb6a | 1304 | |
53f9263b KS |
1305 | /* Hugepages are not counted in NR_FILE_MAPPED for now. */ |
1306 | if (unlikely(PageHuge(page))) { | |
1307 | /* hugetlb pages are always mapped with pmds */ | |
1308 | atomic_dec(compound_mapcount_ptr(page)); | |
8186eb6a | 1309 | goto out; |
53f9263b | 1310 | } |
8186eb6a | 1311 | |
53f9263b KS |
1312 | /* page still mapped by someone else? */ |
1313 | if (!atomic_add_negative(-1, &page->_mapcount)) | |
8186eb6a JW |
1314 | goto out; |
1315 | ||
1316 | /* | |
1317 | * We use the irq-unsafe __{inc|mod}_zone_page_stat because | |
1318 | * these counters are not modified in interrupt context, and | |
1319 | * pte lock(a spinlock) is held, which implies preemption disabled. | |
1320 | */ | |
1321 | __dec_zone_page_state(page, NR_FILE_MAPPED); | |
1322 | mem_cgroup_dec_page_stat(memcg, MEM_CGROUP_STAT_FILE_MAPPED); | |
1323 | ||
1324 | if (unlikely(PageMlocked(page))) | |
1325 | clear_page_mlock(page); | |
1326 | out: | |
6de22619 | 1327 | mem_cgroup_end_page_stat(memcg); |
8186eb6a JW |
1328 | } |
1329 | ||
53f9263b KS |
1330 | static void page_remove_anon_compound_rmap(struct page *page) |
1331 | { | |
1332 | int i, nr; | |
1333 | ||
1334 | if (!atomic_add_negative(-1, compound_mapcount_ptr(page))) | |
1335 | return; | |
1336 | ||
1337 | /* Hugepages are not counted in NR_ANON_PAGES for now. */ | |
1338 | if (unlikely(PageHuge(page))) | |
1339 | return; | |
1340 | ||
1341 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) | |
1342 | return; | |
1343 | ||
1344 | __dec_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES); | |
1345 | ||
1346 | if (TestClearPageDoubleMap(page)) { | |
1347 | /* | |
1348 | * Subpages can be mapped with PTEs too. Check how many of | |
1349 | * themi are still mapped. | |
1350 | */ | |
1351 | for (i = 0, nr = 0; i < HPAGE_PMD_NR; i++) { | |
1352 | if (atomic_add_negative(-1, &page[i]._mapcount)) | |
1353 | nr++; | |
1354 | } | |
1355 | } else { | |
1356 | nr = HPAGE_PMD_NR; | |
1357 | } | |
1358 | ||
e90309c9 KS |
1359 | if (unlikely(PageMlocked(page))) |
1360 | clear_page_mlock(page); | |
1361 | ||
9a982250 | 1362 | if (nr) { |
53f9263b | 1363 | __mod_zone_page_state(page_zone(page), NR_ANON_PAGES, -nr); |
9a982250 KS |
1364 | deferred_split_huge_page(page); |
1365 | } | |
53f9263b KS |
1366 | } |
1367 | ||
1da177e4 LT |
1368 | /** |
1369 | * page_remove_rmap - take down pte mapping from a page | |
d281ee61 KS |
1370 | * @page: page to remove mapping from |
1371 | * @compound: uncharge the page as compound or small page | |
1da177e4 | 1372 | * |
b8072f09 | 1373 | * The caller needs to hold the pte lock. |
1da177e4 | 1374 | */ |
d281ee61 | 1375 | void page_remove_rmap(struct page *page, bool compound) |
1da177e4 | 1376 | { |
8186eb6a | 1377 | if (!PageAnon(page)) { |
d281ee61 | 1378 | VM_BUG_ON_PAGE(compound && !PageHuge(page), page); |
8186eb6a JW |
1379 | page_remove_file_rmap(page); |
1380 | return; | |
1381 | } | |
89c06bd5 | 1382 | |
53f9263b KS |
1383 | if (compound) |
1384 | return page_remove_anon_compound_rmap(page); | |
1385 | ||
b904dcfe KM |
1386 | /* page still mapped by someone else? */ |
1387 | if (!atomic_add_negative(-1, &page->_mapcount)) | |
8186eb6a JW |
1388 | return; |
1389 | ||
0fe6e20b | 1390 | /* |
bea04b07 JZ |
1391 | * We use the irq-unsafe __{inc|mod}_zone_page_stat because |
1392 | * these counters are not modified in interrupt context, and | |
bea04b07 | 1393 | * pte lock(a spinlock) is held, which implies preemption disabled. |
0fe6e20b | 1394 | */ |
53f9263b | 1395 | __dec_zone_page_state(page, NR_ANON_PAGES); |
8186eb6a | 1396 | |
e6c509f8 HD |
1397 | if (unlikely(PageMlocked(page))) |
1398 | clear_page_mlock(page); | |
8186eb6a | 1399 | |
9a982250 KS |
1400 | if (PageTransCompound(page)) |
1401 | deferred_split_huge_page(compound_head(page)); | |
1402 | ||
b904dcfe KM |
1403 | /* |
1404 | * It would be tidy to reset the PageAnon mapping here, | |
1405 | * but that might overwrite a racing page_add_anon_rmap | |
1406 | * which increments mapcount after us but sets mapping | |
1407 | * before us: so leave the reset to free_hot_cold_page, | |
1408 | * and remember that it's only reliable while mapped. | |
1409 | * Leaving it set also helps swapoff to reinstate ptes | |
1410 | * faster for those pages still in swapcache. | |
1411 | */ | |
1da177e4 LT |
1412 | } |
1413 | ||
854e9ed0 MK |
1414 | struct rmap_private { |
1415 | enum ttu_flags flags; | |
1416 | int lazyfreed; | |
1417 | }; | |
1418 | ||
1da177e4 | 1419 | /* |
52629506 | 1420 | * @arg: enum ttu_flags will be passed to this argument |
1da177e4 | 1421 | */ |
ac769501 | 1422 | static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma, |
52629506 | 1423 | unsigned long address, void *arg) |
1da177e4 LT |
1424 | { |
1425 | struct mm_struct *mm = vma->vm_mm; | |
1da177e4 LT |
1426 | pte_t *pte; |
1427 | pte_t pteval; | |
c0718806 | 1428 | spinlock_t *ptl; |
1da177e4 | 1429 | int ret = SWAP_AGAIN; |
854e9ed0 MK |
1430 | struct rmap_private *rp = arg; |
1431 | enum ttu_flags flags = rp->flags; | |
1da177e4 | 1432 | |
b87537d9 HD |
1433 | /* munlock has nothing to gain from examining un-locked vmas */ |
1434 | if ((flags & TTU_MUNLOCK) && !(vma->vm_flags & VM_LOCKED)) | |
1435 | goto out; | |
1436 | ||
479db0bf | 1437 | pte = page_check_address(page, mm, address, &ptl, 0); |
c0718806 | 1438 | if (!pte) |
81b4082d | 1439 | goto out; |
1da177e4 LT |
1440 | |
1441 | /* | |
1442 | * If the page is mlock()d, we cannot swap it out. | |
1443 | * If it's recently referenced (perhaps page_referenced | |
1444 | * skipped over this mm) then we should reactivate it. | |
1445 | */ | |
14fa31b8 | 1446 | if (!(flags & TTU_IGNORE_MLOCK)) { |
b87537d9 HD |
1447 | if (vma->vm_flags & VM_LOCKED) { |
1448 | /* Holding pte lock, we do *not* need mmap_sem here */ | |
1449 | mlock_vma_page(page); | |
1450 | ret = SWAP_MLOCK; | |
1451 | goto out_unmap; | |
1452 | } | |
daa5ba76 | 1453 | if (flags & TTU_MUNLOCK) |
53f79acb | 1454 | goto out_unmap; |
14fa31b8 AK |
1455 | } |
1456 | if (!(flags & TTU_IGNORE_ACCESS)) { | |
b291f000 NP |
1457 | if (ptep_clear_flush_young_notify(vma, address, pte)) { |
1458 | ret = SWAP_FAIL; | |
1459 | goto out_unmap; | |
1460 | } | |
1461 | } | |
1da177e4 | 1462 | |
1da177e4 LT |
1463 | /* Nuke the page table entry. */ |
1464 | flush_cache_page(vma, address, page_to_pfn(page)); | |
72b252ae MG |
1465 | if (should_defer_flush(mm, flags)) { |
1466 | /* | |
1467 | * We clear the PTE but do not flush so potentially a remote | |
1468 | * CPU could still be writing to the page. If the entry was | |
1469 | * previously clean then the architecture must guarantee that | |
1470 | * a clear->dirty transition on a cached TLB entry is written | |
1471 | * through and traps if the PTE is unmapped. | |
1472 | */ | |
1473 | pteval = ptep_get_and_clear(mm, address, pte); | |
1474 | ||
d950c947 | 1475 | set_tlb_ubc_flush_pending(mm, page, pte_dirty(pteval)); |
72b252ae MG |
1476 | } else { |
1477 | pteval = ptep_clear_flush(vma, address, pte); | |
1478 | } | |
1da177e4 LT |
1479 | |
1480 | /* Move the dirty bit to the physical page now the pte is gone. */ | |
1481 | if (pte_dirty(pteval)) | |
1482 | set_page_dirty(page); | |
1483 | ||
365e9c87 HD |
1484 | /* Update high watermark before we lower rss */ |
1485 | update_hiwater_rss(mm); | |
1486 | ||
888b9f7c | 1487 | if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) { |
5d317b2b NH |
1488 | if (PageHuge(page)) { |
1489 | hugetlb_count_sub(1 << compound_order(page), mm); | |
1490 | } else { | |
eca56ff9 | 1491 | dec_mm_counter(mm, mm_counter(page)); |
5f24ae58 | 1492 | } |
888b9f7c | 1493 | set_pte_at(mm, address, pte, |
5f24ae58 | 1494 | swp_entry_to_pte(make_hwpoison_entry(page))); |
45961722 KW |
1495 | } else if (pte_unused(pteval)) { |
1496 | /* | |
1497 | * The guest indicated that the page content is of no | |
1498 | * interest anymore. Simply discard the pte, vmscan | |
1499 | * will take care of the rest. | |
1500 | */ | |
eca56ff9 | 1501 | dec_mm_counter(mm, mm_counter(page)); |
470f119f HD |
1502 | } else if (IS_ENABLED(CONFIG_MIGRATION) && (flags & TTU_MIGRATION)) { |
1503 | swp_entry_t entry; | |
1504 | pte_t swp_pte; | |
1505 | /* | |
1506 | * Store the pfn of the page in a special migration | |
1507 | * pte. do_swap_page() will wait until the migration | |
1508 | * pte is removed and then restart fault handling. | |
1509 | */ | |
1510 | entry = make_migration_entry(page, pte_write(pteval)); | |
1511 | swp_pte = swp_entry_to_pte(entry); | |
1512 | if (pte_soft_dirty(pteval)) | |
1513 | swp_pte = pte_swp_mksoft_dirty(swp_pte); | |
1514 | set_pte_at(mm, address, pte, swp_pte); | |
888b9f7c | 1515 | } else if (PageAnon(page)) { |
4c21e2f2 | 1516 | swp_entry_t entry = { .val = page_private(page) }; |
179ef71c | 1517 | pte_t swp_pte; |
470f119f HD |
1518 | /* |
1519 | * Store the swap location in the pte. | |
1520 | * See handle_pte_fault() ... | |
1521 | */ | |
1522 | VM_BUG_ON_PAGE(!PageSwapCache(page), page); | |
854e9ed0 MK |
1523 | |
1524 | if (!PageDirty(page) && (flags & TTU_LZFREE)) { | |
1525 | /* It's a freeable page by MADV_FREE */ | |
1526 | dec_mm_counter(mm, MM_ANONPAGES); | |
1527 | rp->lazyfreed++; | |
1528 | goto discard; | |
1529 | } | |
1530 | ||
470f119f HD |
1531 | if (swap_duplicate(entry) < 0) { |
1532 | set_pte_at(mm, address, pte, pteval); | |
1533 | ret = SWAP_FAIL; | |
1534 | goto out_unmap; | |
1535 | } | |
1536 | if (list_empty(&mm->mmlist)) { | |
1537 | spin_lock(&mmlist_lock); | |
1538 | if (list_empty(&mm->mmlist)) | |
1539 | list_add(&mm->mmlist, &init_mm.mmlist); | |
1540 | spin_unlock(&mmlist_lock); | |
1da177e4 | 1541 | } |
470f119f HD |
1542 | dec_mm_counter(mm, MM_ANONPAGES); |
1543 | inc_mm_counter(mm, MM_SWAPENTS); | |
179ef71c CG |
1544 | swp_pte = swp_entry_to_pte(entry); |
1545 | if (pte_soft_dirty(pteval)) | |
1546 | swp_pte = pte_swp_mksoft_dirty(swp_pte); | |
1547 | set_pte_at(mm, address, pte, swp_pte); | |
04e62a29 | 1548 | } else |
eca56ff9 | 1549 | dec_mm_counter(mm, mm_counter_file(page)); |
1da177e4 | 1550 | |
854e9ed0 | 1551 | discard: |
d281ee61 | 1552 | page_remove_rmap(page, PageHuge(page)); |
1da177e4 LT |
1553 | page_cache_release(page); |
1554 | ||
1555 | out_unmap: | |
c0718806 | 1556 | pte_unmap_unlock(pte, ptl); |
b87537d9 | 1557 | if (ret != SWAP_FAIL && ret != SWAP_MLOCK && !(flags & TTU_MUNLOCK)) |
2ec74c3e | 1558 | mmu_notifier_invalidate_page(mm, address); |
caed0f48 KM |
1559 | out: |
1560 | return ret; | |
1da177e4 LT |
1561 | } |
1562 | ||
71e3aac0 | 1563 | bool is_vma_temporary_stack(struct vm_area_struct *vma) |
a8bef8ff MG |
1564 | { |
1565 | int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP); | |
1566 | ||
1567 | if (!maybe_stack) | |
1568 | return false; | |
1569 | ||
1570 | if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) == | |
1571 | VM_STACK_INCOMPLETE_SETUP) | |
1572 | return true; | |
1573 | ||
1574 | return false; | |
1575 | } | |
1576 | ||
52629506 JK |
1577 | static bool invalid_migration_vma(struct vm_area_struct *vma, void *arg) |
1578 | { | |
1579 | return is_vma_temporary_stack(vma); | |
1580 | } | |
1581 | ||
52629506 JK |
1582 | static int page_not_mapped(struct page *page) |
1583 | { | |
1584 | return !page_mapped(page); | |
1585 | }; | |
1586 | ||
1da177e4 LT |
1587 | /** |
1588 | * try_to_unmap - try to remove all page table mappings to a page | |
1589 | * @page: the page to get unmapped | |
14fa31b8 | 1590 | * @flags: action and flags |
1da177e4 LT |
1591 | * |
1592 | * Tries to remove all the page table entries which are mapping this | |
1593 | * page, used in the pageout path. Caller must hold the page lock. | |
1594 | * Return values are: | |
1595 | * | |
1596 | * SWAP_SUCCESS - we succeeded in removing all mappings | |
1597 | * SWAP_AGAIN - we missed a mapping, try again later | |
1598 | * SWAP_FAIL - the page is unswappable | |
b291f000 | 1599 | * SWAP_MLOCK - page is mlocked. |
1da177e4 | 1600 | */ |
14fa31b8 | 1601 | int try_to_unmap(struct page *page, enum ttu_flags flags) |
1da177e4 LT |
1602 | { |
1603 | int ret; | |
854e9ed0 MK |
1604 | struct rmap_private rp = { |
1605 | .flags = flags, | |
1606 | .lazyfreed = 0, | |
1607 | }; | |
1608 | ||
52629506 JK |
1609 | struct rmap_walk_control rwc = { |
1610 | .rmap_one = try_to_unmap_one, | |
854e9ed0 | 1611 | .arg = &rp, |
52629506 | 1612 | .done = page_not_mapped, |
52629506 JK |
1613 | .anon_lock = page_lock_anon_vma_read, |
1614 | }; | |
1da177e4 | 1615 | |
309381fe | 1616 | VM_BUG_ON_PAGE(!PageHuge(page) && PageTransHuge(page), page); |
1da177e4 | 1617 | |
52629506 JK |
1618 | /* |
1619 | * During exec, a temporary VMA is setup and later moved. | |
1620 | * The VMA is moved under the anon_vma lock but not the | |
1621 | * page tables leading to a race where migration cannot | |
1622 | * find the migration ptes. Rather than increasing the | |
1623 | * locking requirements of exec(), migration skips | |
1624 | * temporary VMAs until after exec() completes. | |
1625 | */ | |
daa5ba76 | 1626 | if ((flags & TTU_MIGRATION) && !PageKsm(page) && PageAnon(page)) |
52629506 JK |
1627 | rwc.invalid_vma = invalid_migration_vma; |
1628 | ||
1629 | ret = rmap_walk(page, &rwc); | |
1630 | ||
854e9ed0 | 1631 | if (ret != SWAP_MLOCK && !page_mapped(page)) { |
1da177e4 | 1632 | ret = SWAP_SUCCESS; |
854e9ed0 MK |
1633 | if (rp.lazyfreed && !PageDirty(page)) |
1634 | ret = SWAP_LZFREE; | |
1635 | } | |
1da177e4 LT |
1636 | return ret; |
1637 | } | |
81b4082d | 1638 | |
b291f000 NP |
1639 | /** |
1640 | * try_to_munlock - try to munlock a page | |
1641 | * @page: the page to be munlocked | |
1642 | * | |
1643 | * Called from munlock code. Checks all of the VMAs mapping the page | |
1644 | * to make sure nobody else has this page mlocked. The page will be | |
1645 | * returned with PG_mlocked cleared if no other vmas have it mlocked. | |
1646 | * | |
1647 | * Return values are: | |
1648 | * | |
53f79acb | 1649 | * SWAP_AGAIN - no vma is holding page mlocked, or, |
b291f000 | 1650 | * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem |
5ad64688 | 1651 | * SWAP_FAIL - page cannot be located at present |
b291f000 NP |
1652 | * SWAP_MLOCK - page is now mlocked. |
1653 | */ | |
1654 | int try_to_munlock(struct page *page) | |
1655 | { | |
e8351ac9 | 1656 | int ret; |
854e9ed0 MK |
1657 | struct rmap_private rp = { |
1658 | .flags = TTU_MUNLOCK, | |
1659 | .lazyfreed = 0, | |
1660 | }; | |
1661 | ||
e8351ac9 JK |
1662 | struct rmap_walk_control rwc = { |
1663 | .rmap_one = try_to_unmap_one, | |
854e9ed0 | 1664 | .arg = &rp, |
e8351ac9 | 1665 | .done = page_not_mapped, |
e8351ac9 JK |
1666 | .anon_lock = page_lock_anon_vma_read, |
1667 | ||
1668 | }; | |
1669 | ||
309381fe | 1670 | VM_BUG_ON_PAGE(!PageLocked(page) || PageLRU(page), page); |
b291f000 | 1671 | |
e8351ac9 JK |
1672 | ret = rmap_walk(page, &rwc); |
1673 | return ret; | |
b291f000 | 1674 | } |
e9995ef9 | 1675 | |
01d8b20d | 1676 | void __put_anon_vma(struct anon_vma *anon_vma) |
76545066 | 1677 | { |
01d8b20d | 1678 | struct anon_vma *root = anon_vma->root; |
76545066 | 1679 | |
624483f3 | 1680 | anon_vma_free(anon_vma); |
01d8b20d PZ |
1681 | if (root != anon_vma && atomic_dec_and_test(&root->refcount)) |
1682 | anon_vma_free(root); | |
76545066 | 1683 | } |
76545066 | 1684 | |
0dd1c7bb JK |
1685 | static struct anon_vma *rmap_walk_anon_lock(struct page *page, |
1686 | struct rmap_walk_control *rwc) | |
faecd8dd JK |
1687 | { |
1688 | struct anon_vma *anon_vma; | |
1689 | ||
0dd1c7bb JK |
1690 | if (rwc->anon_lock) |
1691 | return rwc->anon_lock(page); | |
1692 | ||
faecd8dd JK |
1693 | /* |
1694 | * Note: remove_migration_ptes() cannot use page_lock_anon_vma_read() | |
1695 | * because that depends on page_mapped(); but not all its usages | |
1696 | * are holding mmap_sem. Users without mmap_sem are required to | |
1697 | * take a reference count to prevent the anon_vma disappearing | |
1698 | */ | |
1699 | anon_vma = page_anon_vma(page); | |
1700 | if (!anon_vma) | |
1701 | return NULL; | |
1702 | ||
1703 | anon_vma_lock_read(anon_vma); | |
1704 | return anon_vma; | |
1705 | } | |
1706 | ||
e9995ef9 | 1707 | /* |
e8351ac9 JK |
1708 | * rmap_walk_anon - do something to anonymous page using the object-based |
1709 | * rmap method | |
1710 | * @page: the page to be handled | |
1711 | * @rwc: control variable according to each walk type | |
1712 | * | |
1713 | * Find all the mappings of a page using the mapping pointer and the vma chains | |
1714 | * contained in the anon_vma struct it points to. | |
1715 | * | |
1716 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma | |
1717 | * where the page was found will be held for write. So, we won't recheck | |
1718 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be | |
1719 | * LOCKED. | |
e9995ef9 | 1720 | */ |
051ac83a | 1721 | static int rmap_walk_anon(struct page *page, struct rmap_walk_control *rwc) |
e9995ef9 HD |
1722 | { |
1723 | struct anon_vma *anon_vma; | |
b258d860 | 1724 | pgoff_t pgoff; |
5beb4930 | 1725 | struct anon_vma_chain *avc; |
e9995ef9 HD |
1726 | int ret = SWAP_AGAIN; |
1727 | ||
0dd1c7bb | 1728 | anon_vma = rmap_walk_anon_lock(page, rwc); |
e9995ef9 HD |
1729 | if (!anon_vma) |
1730 | return ret; | |
faecd8dd | 1731 | |
b258d860 | 1732 | pgoff = page_to_pgoff(page); |
bf181b9f | 1733 | anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) { |
5beb4930 | 1734 | struct vm_area_struct *vma = avc->vma; |
e9995ef9 | 1735 | unsigned long address = vma_address(page, vma); |
0dd1c7bb | 1736 | |
ad12695f AA |
1737 | cond_resched(); |
1738 | ||
0dd1c7bb JK |
1739 | if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg)) |
1740 | continue; | |
1741 | ||
051ac83a | 1742 | ret = rwc->rmap_one(page, vma, address, rwc->arg); |
e9995ef9 HD |
1743 | if (ret != SWAP_AGAIN) |
1744 | break; | |
0dd1c7bb JK |
1745 | if (rwc->done && rwc->done(page)) |
1746 | break; | |
e9995ef9 | 1747 | } |
4fc3f1d6 | 1748 | anon_vma_unlock_read(anon_vma); |
e9995ef9 HD |
1749 | return ret; |
1750 | } | |
1751 | ||
e8351ac9 JK |
1752 | /* |
1753 | * rmap_walk_file - do something to file page using the object-based rmap method | |
1754 | * @page: the page to be handled | |
1755 | * @rwc: control variable according to each walk type | |
1756 | * | |
1757 | * Find all the mappings of a page using the mapping pointer and the vma chains | |
1758 | * contained in the address_space struct it points to. | |
1759 | * | |
1760 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma | |
1761 | * where the page was found will be held for write. So, we won't recheck | |
1762 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be | |
1763 | * LOCKED. | |
1764 | */ | |
051ac83a | 1765 | static int rmap_walk_file(struct page *page, struct rmap_walk_control *rwc) |
e9995ef9 HD |
1766 | { |
1767 | struct address_space *mapping = page->mapping; | |
b258d860 | 1768 | pgoff_t pgoff; |
e9995ef9 | 1769 | struct vm_area_struct *vma; |
e9995ef9 HD |
1770 | int ret = SWAP_AGAIN; |
1771 | ||
9f32624b JK |
1772 | /* |
1773 | * The page lock not only makes sure that page->mapping cannot | |
1774 | * suddenly be NULLified by truncation, it makes sure that the | |
1775 | * structure at mapping cannot be freed and reused yet, | |
c8c06efa | 1776 | * so we can safely take mapping->i_mmap_rwsem. |
9f32624b | 1777 | */ |
81d1b09c | 1778 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
9f32624b | 1779 | |
e9995ef9 HD |
1780 | if (!mapping) |
1781 | return ret; | |
3dec0ba0 | 1782 | |
b258d860 | 1783 | pgoff = page_to_pgoff(page); |
3dec0ba0 | 1784 | i_mmap_lock_read(mapping); |
6b2dbba8 | 1785 | vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) { |
e9995ef9 | 1786 | unsigned long address = vma_address(page, vma); |
0dd1c7bb | 1787 | |
ad12695f AA |
1788 | cond_resched(); |
1789 | ||
0dd1c7bb JK |
1790 | if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg)) |
1791 | continue; | |
1792 | ||
051ac83a | 1793 | ret = rwc->rmap_one(page, vma, address, rwc->arg); |
e9995ef9 | 1794 | if (ret != SWAP_AGAIN) |
0dd1c7bb JK |
1795 | goto done; |
1796 | if (rwc->done && rwc->done(page)) | |
1797 | goto done; | |
e9995ef9 | 1798 | } |
0dd1c7bb | 1799 | |
0dd1c7bb | 1800 | done: |
3dec0ba0 | 1801 | i_mmap_unlock_read(mapping); |
e9995ef9 HD |
1802 | return ret; |
1803 | } | |
1804 | ||
051ac83a | 1805 | int rmap_walk(struct page *page, struct rmap_walk_control *rwc) |
e9995ef9 | 1806 | { |
e9995ef9 | 1807 | if (unlikely(PageKsm(page))) |
051ac83a | 1808 | return rmap_walk_ksm(page, rwc); |
e9995ef9 | 1809 | else if (PageAnon(page)) |
051ac83a | 1810 | return rmap_walk_anon(page, rwc); |
e9995ef9 | 1811 | else |
051ac83a | 1812 | return rmap_walk_file(page, rwc); |
e9995ef9 | 1813 | } |
0fe6e20b | 1814 | |
e3390f67 | 1815 | #ifdef CONFIG_HUGETLB_PAGE |
0fe6e20b NH |
1816 | /* |
1817 | * The following three functions are for anonymous (private mapped) hugepages. | |
1818 | * Unlike common anonymous pages, anonymous hugepages have no accounting code | |
1819 | * and no lru code, because we handle hugepages differently from common pages. | |
1820 | */ | |
1821 | static void __hugepage_set_anon_rmap(struct page *page, | |
1822 | struct vm_area_struct *vma, unsigned long address, int exclusive) | |
1823 | { | |
1824 | struct anon_vma *anon_vma = vma->anon_vma; | |
433abed6 | 1825 | |
0fe6e20b | 1826 | BUG_ON(!anon_vma); |
433abed6 NH |
1827 | |
1828 | if (PageAnon(page)) | |
1829 | return; | |
1830 | if (!exclusive) | |
1831 | anon_vma = anon_vma->root; | |
1832 | ||
0fe6e20b NH |
1833 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; |
1834 | page->mapping = (struct address_space *) anon_vma; | |
1835 | page->index = linear_page_index(vma, address); | |
1836 | } | |
1837 | ||
1838 | void hugepage_add_anon_rmap(struct page *page, | |
1839 | struct vm_area_struct *vma, unsigned long address) | |
1840 | { | |
1841 | struct anon_vma *anon_vma = vma->anon_vma; | |
1842 | int first; | |
a850ea30 NH |
1843 | |
1844 | BUG_ON(!PageLocked(page)); | |
0fe6e20b | 1845 | BUG_ON(!anon_vma); |
5dbe0af4 | 1846 | /* address might be in next vma when migration races vma_adjust */ |
53f9263b | 1847 | first = atomic_inc_and_test(compound_mapcount_ptr(page)); |
0fe6e20b NH |
1848 | if (first) |
1849 | __hugepage_set_anon_rmap(page, vma, address, 0); | |
1850 | } | |
1851 | ||
1852 | void hugepage_add_new_anon_rmap(struct page *page, | |
1853 | struct vm_area_struct *vma, unsigned long address) | |
1854 | { | |
1855 | BUG_ON(address < vma->vm_start || address >= vma->vm_end); | |
53f9263b | 1856 | atomic_set(compound_mapcount_ptr(page), 0); |
0fe6e20b NH |
1857 | __hugepage_set_anon_rmap(page, vma, address, 1); |
1858 | } | |
e3390f67 | 1859 | #endif /* CONFIG_HUGETLB_PAGE */ |