]>
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) | |
88f306b6 KS |
26 | * hugetlbfs_i_mmap_rwsem_key (in huge_pmd_share) |
27 | * mapping->i_mmap_rwsem | |
28 | * anon_vma->rwsem | |
29 | * mm->page_table_lock or pte_lock | |
a52633d8 | 30 | * zone_lru_lock (in mark_page_accessed, isolate_lru_page) |
88f306b6 KS |
31 | * swap_lock (in swap_duplicate, swap_info_get) |
32 | * mmlist_lock (in mmput, drain_mmlist and others) | |
33 | * mapping->private_lock (in __set_page_dirty_buffers) | |
34 | * mem_cgroup_{begin,end}_page_stat (memcg->move_lock) | |
35 | * mapping->tree_lock (widely used) | |
36 | * inode->i_lock (in set_page_dirty's __mark_inode_dirty) | |
37 | * bdi.wb->list_lock (in set_page_dirty's __mark_inode_dirty) | |
38 | * sb_lock (within inode_lock in fs/fs-writeback.c) | |
39 | * mapping->tree_lock (widely used, in set_page_dirty, | |
40 | * in arch-dependent flush_dcache_mmap_lock, | |
41 | * within bdi.wb->list_lock in __sync_single_inode) | |
6a46079c | 42 | * |
5a505085 | 43 | * anon_vma->rwsem,mapping->i_mutex (memory_failure, collect_procs_anon) |
9b679320 | 44 | * ->tasklist_lock |
6a46079c | 45 | * pte map lock |
1da177e4 LT |
46 | */ |
47 | ||
48 | #include <linux/mm.h> | |
49 | #include <linux/pagemap.h> | |
50 | #include <linux/swap.h> | |
51 | #include <linux/swapops.h> | |
52 | #include <linux/slab.h> | |
53 | #include <linux/init.h> | |
5ad64688 | 54 | #include <linux/ksm.h> |
1da177e4 LT |
55 | #include <linux/rmap.h> |
56 | #include <linux/rcupdate.h> | |
b95f1b31 | 57 | #include <linux/export.h> |
8a9f3ccd | 58 | #include <linux/memcontrol.h> |
cddb8a5c | 59 | #include <linux/mmu_notifier.h> |
64cdd548 | 60 | #include <linux/migrate.h> |
0fe6e20b | 61 | #include <linux/hugetlb.h> |
ef5d437f | 62 | #include <linux/backing-dev.h> |
33c3fc71 | 63 | #include <linux/page_idle.h> |
1da177e4 LT |
64 | |
65 | #include <asm/tlbflush.h> | |
66 | ||
72b252ae MG |
67 | #include <trace/events/tlb.h> |
68 | ||
b291f000 NP |
69 | #include "internal.h" |
70 | ||
fdd2e5f8 | 71 | static struct kmem_cache *anon_vma_cachep; |
5beb4930 | 72 | static struct kmem_cache *anon_vma_chain_cachep; |
fdd2e5f8 AB |
73 | |
74 | static inline struct anon_vma *anon_vma_alloc(void) | |
75 | { | |
01d8b20d PZ |
76 | struct anon_vma *anon_vma; |
77 | ||
78 | anon_vma = kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL); | |
79 | if (anon_vma) { | |
80 | atomic_set(&anon_vma->refcount, 1); | |
7a3ef208 KK |
81 | anon_vma->degree = 1; /* Reference for first vma */ |
82 | anon_vma->parent = anon_vma; | |
01d8b20d PZ |
83 | /* |
84 | * Initialise the anon_vma root to point to itself. If called | |
85 | * from fork, the root will be reset to the parents anon_vma. | |
86 | */ | |
87 | anon_vma->root = anon_vma; | |
88 | } | |
89 | ||
90 | return anon_vma; | |
fdd2e5f8 AB |
91 | } |
92 | ||
01d8b20d | 93 | static inline void anon_vma_free(struct anon_vma *anon_vma) |
fdd2e5f8 | 94 | { |
01d8b20d | 95 | VM_BUG_ON(atomic_read(&anon_vma->refcount)); |
88c22088 PZ |
96 | |
97 | /* | |
4fc3f1d6 | 98 | * Synchronize against page_lock_anon_vma_read() such that |
88c22088 PZ |
99 | * we can safely hold the lock without the anon_vma getting |
100 | * freed. | |
101 | * | |
102 | * Relies on the full mb implied by the atomic_dec_and_test() from | |
103 | * put_anon_vma() against the acquire barrier implied by | |
4fc3f1d6 | 104 | * down_read_trylock() from page_lock_anon_vma_read(). This orders: |
88c22088 | 105 | * |
4fc3f1d6 IM |
106 | * page_lock_anon_vma_read() VS put_anon_vma() |
107 | * down_read_trylock() atomic_dec_and_test() | |
88c22088 | 108 | * LOCK MB |
4fc3f1d6 | 109 | * atomic_read() rwsem_is_locked() |
88c22088 PZ |
110 | * |
111 | * LOCK should suffice since the actual taking of the lock must | |
112 | * happen _before_ what follows. | |
113 | */ | |
7f39dda9 | 114 | might_sleep(); |
5a505085 | 115 | if (rwsem_is_locked(&anon_vma->root->rwsem)) { |
4fc3f1d6 | 116 | anon_vma_lock_write(anon_vma); |
08b52706 | 117 | anon_vma_unlock_write(anon_vma); |
88c22088 PZ |
118 | } |
119 | ||
fdd2e5f8 AB |
120 | kmem_cache_free(anon_vma_cachep, anon_vma); |
121 | } | |
1da177e4 | 122 | |
dd34739c | 123 | static inline struct anon_vma_chain *anon_vma_chain_alloc(gfp_t gfp) |
5beb4930 | 124 | { |
dd34739c | 125 | return kmem_cache_alloc(anon_vma_chain_cachep, gfp); |
5beb4930 RR |
126 | } |
127 | ||
e574b5fd | 128 | static void anon_vma_chain_free(struct anon_vma_chain *anon_vma_chain) |
5beb4930 RR |
129 | { |
130 | kmem_cache_free(anon_vma_chain_cachep, anon_vma_chain); | |
131 | } | |
132 | ||
6583a843 KC |
133 | static void anon_vma_chain_link(struct vm_area_struct *vma, |
134 | struct anon_vma_chain *avc, | |
135 | struct anon_vma *anon_vma) | |
136 | { | |
137 | avc->vma = vma; | |
138 | avc->anon_vma = anon_vma; | |
139 | list_add(&avc->same_vma, &vma->anon_vma_chain); | |
bf181b9f | 140 | anon_vma_interval_tree_insert(avc, &anon_vma->rb_root); |
6583a843 KC |
141 | } |
142 | ||
d9d332e0 | 143 | /** |
d5a187da | 144 | * __anon_vma_prepare - attach an anon_vma to a memory region |
d9d332e0 LT |
145 | * @vma: the memory region in question |
146 | * | |
147 | * This makes sure the memory mapping described by 'vma' has | |
148 | * an 'anon_vma' attached to it, so that we can associate the | |
149 | * anonymous pages mapped into it with that anon_vma. | |
150 | * | |
d5a187da VB |
151 | * The common case will be that we already have one, which |
152 | * is handled inline by anon_vma_prepare(). But if | |
23a0790a | 153 | * not we either need to find an adjacent mapping that we |
d9d332e0 LT |
154 | * can re-use the anon_vma from (very common when the only |
155 | * reason for splitting a vma has been mprotect()), or we | |
156 | * allocate a new one. | |
157 | * | |
158 | * Anon-vma allocations are very subtle, because we may have | |
4fc3f1d6 | 159 | * optimistically looked up an anon_vma in page_lock_anon_vma_read() |
d9d332e0 LT |
160 | * and that may actually touch the spinlock even in the newly |
161 | * allocated vma (it depends on RCU to make sure that the | |
162 | * anon_vma isn't actually destroyed). | |
163 | * | |
164 | * As a result, we need to do proper anon_vma locking even | |
165 | * for the new allocation. At the same time, we do not want | |
166 | * to do any locking for the common case of already having | |
167 | * an anon_vma. | |
168 | * | |
169 | * This must be called with the mmap_sem held for reading. | |
170 | */ | |
d5a187da | 171 | int __anon_vma_prepare(struct vm_area_struct *vma) |
1da177e4 | 172 | { |
d5a187da VB |
173 | struct mm_struct *mm = vma->vm_mm; |
174 | struct anon_vma *anon_vma, *allocated; | |
5beb4930 | 175 | struct anon_vma_chain *avc; |
1da177e4 LT |
176 | |
177 | might_sleep(); | |
1da177e4 | 178 | |
d5a187da VB |
179 | avc = anon_vma_chain_alloc(GFP_KERNEL); |
180 | if (!avc) | |
181 | goto out_enomem; | |
182 | ||
183 | anon_vma = find_mergeable_anon_vma(vma); | |
184 | allocated = NULL; | |
185 | if (!anon_vma) { | |
186 | anon_vma = anon_vma_alloc(); | |
187 | if (unlikely(!anon_vma)) | |
188 | goto out_enomem_free_avc; | |
189 | allocated = anon_vma; | |
190 | } | |
5beb4930 | 191 | |
d5a187da VB |
192 | anon_vma_lock_write(anon_vma); |
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; | |
197 | anon_vma_chain_link(vma, avc, anon_vma); | |
198 | /* vma reference or self-parent link for new root */ | |
199 | anon_vma->degree++; | |
d9d332e0 | 200 | allocated = NULL; |
d5a187da VB |
201 | avc = NULL; |
202 | } | |
203 | spin_unlock(&mm->page_table_lock); | |
204 | anon_vma_unlock_write(anon_vma); | |
1da177e4 | 205 | |
d5a187da VB |
206 | if (unlikely(allocated)) |
207 | put_anon_vma(allocated); | |
208 | if (unlikely(avc)) | |
209 | anon_vma_chain_free(avc); | |
31f2b0eb | 210 | |
1da177e4 | 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 | ||
e4c5800a | 411 | VM_WARN_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 | 570 | #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH |
72b252ae MG |
571 | /* |
572 | * Flush TLB entries for recently unmapped pages from remote CPUs. It is | |
573 | * important if a PTE was dirty when it was unmapped that it's flushed | |
574 | * before any IO is initiated on the page to prevent lost writes. Similarly, | |
575 | * it must be flushed before freeing to prevent data leakage. | |
576 | */ | |
577 | void try_to_unmap_flush(void) | |
578 | { | |
579 | struct tlbflush_unmap_batch *tlb_ubc = ¤t->tlb_ubc; | |
580 | int cpu; | |
581 | ||
582 | if (!tlb_ubc->flush_required) | |
583 | return; | |
584 | ||
585 | cpu = get_cpu(); | |
586 | ||
858eaaa7 NA |
587 | if (cpumask_test_cpu(cpu, &tlb_ubc->cpumask)) { |
588 | count_vm_tlb_event(NR_TLB_LOCAL_FLUSH_ALL); | |
589 | local_flush_tlb(); | |
590 | trace_tlb_flush(TLB_LOCAL_SHOOTDOWN, TLB_FLUSH_ALL); | |
72b252ae | 591 | } |
858eaaa7 NA |
592 | |
593 | if (cpumask_any_but(&tlb_ubc->cpumask, cpu) < nr_cpu_ids) | |
594 | flush_tlb_others(&tlb_ubc->cpumask, NULL, 0, TLB_FLUSH_ALL); | |
72b252ae MG |
595 | cpumask_clear(&tlb_ubc->cpumask); |
596 | tlb_ubc->flush_required = false; | |
d950c947 | 597 | tlb_ubc->writable = false; |
72b252ae MG |
598 | put_cpu(); |
599 | } | |
600 | ||
d950c947 MG |
601 | /* Flush iff there are potentially writable TLB entries that can race with IO */ |
602 | void try_to_unmap_flush_dirty(void) | |
603 | { | |
604 | struct tlbflush_unmap_batch *tlb_ubc = ¤t->tlb_ubc; | |
605 | ||
606 | if (tlb_ubc->writable) | |
607 | try_to_unmap_flush(); | |
608 | } | |
609 | ||
c7ab0d2f | 610 | static void set_tlb_ubc_flush_pending(struct mm_struct *mm, bool writable) |
72b252ae MG |
611 | { |
612 | struct tlbflush_unmap_batch *tlb_ubc = ¤t->tlb_ubc; | |
613 | ||
614 | cpumask_or(&tlb_ubc->cpumask, &tlb_ubc->cpumask, mm_cpumask(mm)); | |
615 | tlb_ubc->flush_required = true; | |
d950c947 MG |
616 | |
617 | /* | |
618 | * If the PTE was dirty then it's best to assume it's writable. The | |
619 | * caller must use try_to_unmap_flush_dirty() or try_to_unmap_flush() | |
620 | * before the page is queued for IO. | |
621 | */ | |
622 | if (writable) | |
623 | tlb_ubc->writable = true; | |
72b252ae MG |
624 | } |
625 | ||
626 | /* | |
627 | * Returns true if the TLB flush should be deferred to the end of a batch of | |
628 | * unmap operations to reduce IPIs. | |
629 | */ | |
630 | static bool should_defer_flush(struct mm_struct *mm, enum ttu_flags flags) | |
631 | { | |
632 | bool should_defer = false; | |
633 | ||
634 | if (!(flags & TTU_BATCH_FLUSH)) | |
635 | return false; | |
636 | ||
637 | /* If remote CPUs need to be flushed then defer batch the flush */ | |
638 | if (cpumask_any_but(mm_cpumask(mm), get_cpu()) < nr_cpu_ids) | |
639 | should_defer = true; | |
640 | put_cpu(); | |
641 | ||
642 | return should_defer; | |
643 | } | |
644 | #else | |
c7ab0d2f | 645 | static void set_tlb_ubc_flush_pending(struct mm_struct *mm, bool writable) |
72b252ae MG |
646 | { |
647 | } | |
648 | ||
649 | static bool should_defer_flush(struct mm_struct *mm, enum ttu_flags flags) | |
650 | { | |
651 | return false; | |
652 | } | |
653 | #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */ | |
654 | ||
1da177e4 | 655 | /* |
bf89c8c8 | 656 | * At what user virtual address is page expected in vma? |
ab941e0f | 657 | * Caller should check the page is actually part of the vma. |
1da177e4 LT |
658 | */ |
659 | unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma) | |
660 | { | |
86c2ad19 | 661 | unsigned long address; |
21d0d443 | 662 | if (PageAnon(page)) { |
4829b906 HD |
663 | struct anon_vma *page__anon_vma = page_anon_vma(page); |
664 | /* | |
665 | * Note: swapoff's unuse_vma() is more efficient with this | |
666 | * check, and needs it to match anon_vma when KSM is active. | |
667 | */ | |
668 | if (!vma->anon_vma || !page__anon_vma || | |
669 | vma->anon_vma->root != page__anon_vma->root) | |
21d0d443 | 670 | return -EFAULT; |
27ba0644 KS |
671 | } else if (page->mapping) { |
672 | if (!vma->vm_file || vma->vm_file->f_mapping != page->mapping) | |
1da177e4 LT |
673 | return -EFAULT; |
674 | } else | |
675 | return -EFAULT; | |
86c2ad19 ML |
676 | address = __vma_address(page, vma); |
677 | if (unlikely(address < vma->vm_start || address >= vma->vm_end)) | |
678 | return -EFAULT; | |
679 | return address; | |
1da177e4 LT |
680 | } |
681 | ||
6219049a BL |
682 | pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address) |
683 | { | |
684 | pgd_t *pgd; | |
685 | pud_t *pud; | |
686 | pmd_t *pmd = NULL; | |
f72e7dcd | 687 | pmd_t pmde; |
6219049a BL |
688 | |
689 | pgd = pgd_offset(mm, address); | |
690 | if (!pgd_present(*pgd)) | |
691 | goto out; | |
692 | ||
693 | pud = pud_offset(pgd, address); | |
694 | if (!pud_present(*pud)) | |
695 | goto out; | |
696 | ||
697 | pmd = pmd_offset(pud, address); | |
f72e7dcd | 698 | /* |
8809aa2d | 699 | * Some THP functions use the sequence pmdp_huge_clear_flush(), set_pmd_at() |
f72e7dcd HD |
700 | * without holding anon_vma lock for write. So when looking for a |
701 | * genuine pmde (in which to find pte), test present and !THP together. | |
702 | */ | |
e37c6982 CB |
703 | pmde = *pmd; |
704 | barrier(); | |
f72e7dcd | 705 | if (!pmd_present(pmde) || pmd_trans_huge(pmde)) |
6219049a BL |
706 | pmd = NULL; |
707 | out: | |
708 | return pmd; | |
709 | } | |
710 | ||
8749cfea VD |
711 | struct page_referenced_arg { |
712 | int mapcount; | |
713 | int referenced; | |
714 | unsigned long vm_flags; | |
715 | struct mem_cgroup *memcg; | |
716 | }; | |
717 | /* | |
718 | * arg: page_referenced_arg will be passed | |
719 | */ | |
720 | static int page_referenced_one(struct page *page, struct vm_area_struct *vma, | |
721 | unsigned long address, void *arg) | |
722 | { | |
8749cfea | 723 | struct page_referenced_arg *pra = arg; |
8eaedede KS |
724 | struct page_vma_mapped_walk pvmw = { |
725 | .page = page, | |
726 | .vma = vma, | |
727 | .address = address, | |
728 | }; | |
8749cfea VD |
729 | int referenced = 0; |
730 | ||
8eaedede KS |
731 | while (page_vma_mapped_walk(&pvmw)) { |
732 | address = pvmw.address; | |
b20ce5e0 | 733 | |
8eaedede KS |
734 | if (vma->vm_flags & VM_LOCKED) { |
735 | page_vma_mapped_walk_done(&pvmw); | |
736 | pra->vm_flags |= VM_LOCKED; | |
737 | return SWAP_FAIL; /* To break the loop */ | |
738 | } | |
71e3aac0 | 739 | |
8eaedede KS |
740 | if (pvmw.pte) { |
741 | if (ptep_clear_flush_young_notify(vma, address, | |
742 | pvmw.pte)) { | |
743 | /* | |
744 | * Don't treat a reference through | |
745 | * a sequentially read mapping as such. | |
746 | * If the page has been used in another mapping, | |
747 | * we will catch it; if this other mapping is | |
748 | * already gone, the unmap path will have set | |
749 | * PG_referenced or activated the page. | |
750 | */ | |
751 | if (likely(!(vma->vm_flags & VM_SEQ_READ))) | |
752 | referenced++; | |
753 | } | |
754 | } else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) { | |
755 | if (pmdp_clear_flush_young_notify(vma, address, | |
756 | pvmw.pmd)) | |
8749cfea | 757 | referenced++; |
8eaedede KS |
758 | } else { |
759 | /* unexpected pmd-mapped page? */ | |
760 | WARN_ON_ONCE(1); | |
8749cfea | 761 | } |
8eaedede KS |
762 | |
763 | pra->mapcount--; | |
b20ce5e0 | 764 | } |
b20ce5e0 | 765 | |
33c3fc71 VD |
766 | if (referenced) |
767 | clear_page_idle(page); | |
768 | if (test_and_clear_page_young(page)) | |
769 | referenced++; | |
770 | ||
9f32624b JK |
771 | if (referenced) { |
772 | pra->referenced++; | |
773 | pra->vm_flags |= vma->vm_flags; | |
1da177e4 | 774 | } |
34bbd704 | 775 | |
9f32624b JK |
776 | if (!pra->mapcount) |
777 | return SWAP_SUCCESS; /* To break the loop */ | |
778 | ||
779 | return SWAP_AGAIN; | |
1da177e4 LT |
780 | } |
781 | ||
9f32624b | 782 | static bool invalid_page_referenced_vma(struct vm_area_struct *vma, void *arg) |
1da177e4 | 783 | { |
9f32624b JK |
784 | struct page_referenced_arg *pra = arg; |
785 | struct mem_cgroup *memcg = pra->memcg; | |
1da177e4 | 786 | |
9f32624b JK |
787 | if (!mm_match_cgroup(vma->vm_mm, memcg)) |
788 | return true; | |
1da177e4 | 789 | |
9f32624b | 790 | return false; |
1da177e4 LT |
791 | } |
792 | ||
793 | /** | |
794 | * page_referenced - test if the page was referenced | |
795 | * @page: the page to test | |
796 | * @is_locked: caller holds lock on the page | |
72835c86 | 797 | * @memcg: target memory cgroup |
6fe6b7e3 | 798 | * @vm_flags: collect encountered vma->vm_flags who actually referenced the page |
1da177e4 LT |
799 | * |
800 | * Quick test_and_clear_referenced for all mappings to a page, | |
801 | * returns the number of ptes which referenced the page. | |
802 | */ | |
6fe6b7e3 WF |
803 | int page_referenced(struct page *page, |
804 | int is_locked, | |
72835c86 | 805 | struct mem_cgroup *memcg, |
6fe6b7e3 | 806 | unsigned long *vm_flags) |
1da177e4 | 807 | { |
9f32624b | 808 | int ret; |
5ad64688 | 809 | int we_locked = 0; |
9f32624b | 810 | struct page_referenced_arg pra = { |
b20ce5e0 | 811 | .mapcount = total_mapcount(page), |
9f32624b JK |
812 | .memcg = memcg, |
813 | }; | |
814 | struct rmap_walk_control rwc = { | |
815 | .rmap_one = page_referenced_one, | |
816 | .arg = (void *)&pra, | |
817 | .anon_lock = page_lock_anon_vma_read, | |
818 | }; | |
1da177e4 | 819 | |
6fe6b7e3 | 820 | *vm_flags = 0; |
9f32624b JK |
821 | if (!page_mapped(page)) |
822 | return 0; | |
823 | ||
824 | if (!page_rmapping(page)) | |
825 | return 0; | |
826 | ||
827 | if (!is_locked && (!PageAnon(page) || PageKsm(page))) { | |
828 | we_locked = trylock_page(page); | |
829 | if (!we_locked) | |
830 | return 1; | |
1da177e4 | 831 | } |
9f32624b JK |
832 | |
833 | /* | |
834 | * If we are reclaiming on behalf of a cgroup, skip | |
835 | * counting on behalf of references from different | |
836 | * cgroups | |
837 | */ | |
838 | if (memcg) { | |
839 | rwc.invalid_vma = invalid_page_referenced_vma; | |
840 | } | |
841 | ||
842 | ret = rmap_walk(page, &rwc); | |
843 | *vm_flags = pra.vm_flags; | |
844 | ||
845 | if (we_locked) | |
846 | unlock_page(page); | |
847 | ||
848 | return pra.referenced; | |
1da177e4 LT |
849 | } |
850 | ||
1cb1729b | 851 | static int page_mkclean_one(struct page *page, struct vm_area_struct *vma, |
9853a407 | 852 | unsigned long address, void *arg) |
d08b3851 | 853 | { |
f27176cf KS |
854 | struct page_vma_mapped_walk pvmw = { |
855 | .page = page, | |
856 | .vma = vma, | |
857 | .address = address, | |
858 | .flags = PVMW_SYNC, | |
859 | }; | |
9853a407 | 860 | int *cleaned = arg; |
d08b3851 | 861 | |
f27176cf KS |
862 | while (page_vma_mapped_walk(&pvmw)) { |
863 | int ret = 0; | |
864 | address = pvmw.address; | |
865 | if (pvmw.pte) { | |
866 | pte_t entry; | |
867 | pte_t *pte = pvmw.pte; | |
868 | ||
869 | if (!pte_dirty(*pte) && !pte_write(*pte)) | |
870 | continue; | |
871 | ||
872 | flush_cache_page(vma, address, pte_pfn(*pte)); | |
873 | entry = ptep_clear_flush(vma, address, pte); | |
874 | entry = pte_wrprotect(entry); | |
875 | entry = pte_mkclean(entry); | |
876 | set_pte_at(vma->vm_mm, address, pte, entry); | |
877 | ret = 1; | |
878 | } else { | |
879 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE | |
880 | pmd_t *pmd = pvmw.pmd; | |
881 | pmd_t entry; | |
882 | ||
883 | if (!pmd_dirty(*pmd) && !pmd_write(*pmd)) | |
884 | continue; | |
885 | ||
886 | flush_cache_page(vma, address, page_to_pfn(page)); | |
887 | entry = pmdp_huge_clear_flush(vma, address, pmd); | |
888 | entry = pmd_wrprotect(entry); | |
889 | entry = pmd_mkclean(entry); | |
890 | set_pmd_at(vma->vm_mm, address, pmd, entry); | |
891 | ret = 1; | |
892 | #else | |
893 | /* unexpected pmd-mapped page? */ | |
894 | WARN_ON_ONCE(1); | |
895 | #endif | |
896 | } | |
d08b3851 | 897 | |
f27176cf KS |
898 | if (ret) { |
899 | mmu_notifier_invalidate_page(vma->vm_mm, address); | |
900 | (*cleaned)++; | |
901 | } | |
c2fda5fe | 902 | } |
d08b3851 | 903 | |
9853a407 | 904 | return SWAP_AGAIN; |
d08b3851 PZ |
905 | } |
906 | ||
9853a407 | 907 | static bool invalid_mkclean_vma(struct vm_area_struct *vma, void *arg) |
d08b3851 | 908 | { |
9853a407 | 909 | if (vma->vm_flags & VM_SHARED) |
871beb8c | 910 | return false; |
d08b3851 | 911 | |
871beb8c | 912 | return true; |
d08b3851 PZ |
913 | } |
914 | ||
915 | int page_mkclean(struct page *page) | |
916 | { | |
9853a407 JK |
917 | int cleaned = 0; |
918 | struct address_space *mapping; | |
919 | struct rmap_walk_control rwc = { | |
920 | .arg = (void *)&cleaned, | |
921 | .rmap_one = page_mkclean_one, | |
922 | .invalid_vma = invalid_mkclean_vma, | |
923 | }; | |
d08b3851 PZ |
924 | |
925 | BUG_ON(!PageLocked(page)); | |
926 | ||
9853a407 JK |
927 | if (!page_mapped(page)) |
928 | return 0; | |
929 | ||
930 | mapping = page_mapping(page); | |
931 | if (!mapping) | |
932 | return 0; | |
933 | ||
934 | rmap_walk(page, &rwc); | |
d08b3851 | 935 | |
9853a407 | 936 | return cleaned; |
d08b3851 | 937 | } |
60b59bea | 938 | EXPORT_SYMBOL_GPL(page_mkclean); |
d08b3851 | 939 | |
c44b6743 RR |
940 | /** |
941 | * page_move_anon_rmap - move a page to our anon_vma | |
942 | * @page: the page to move to our anon_vma | |
943 | * @vma: the vma the page belongs to | |
c44b6743 RR |
944 | * |
945 | * When a page belongs exclusively to one process after a COW event, | |
946 | * that page can be moved into the anon_vma that belongs to just that | |
947 | * process, so the rmap code will not search the parent or sibling | |
948 | * processes. | |
949 | */ | |
5a49973d | 950 | void page_move_anon_rmap(struct page *page, struct vm_area_struct *vma) |
c44b6743 RR |
951 | { |
952 | struct anon_vma *anon_vma = vma->anon_vma; | |
953 | ||
5a49973d HD |
954 | page = compound_head(page); |
955 | ||
309381fe | 956 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
81d1b09c | 957 | VM_BUG_ON_VMA(!anon_vma, vma); |
c44b6743 RR |
958 | |
959 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; | |
414e2fb8 VD |
960 | /* |
961 | * Ensure that anon_vma and the PAGE_MAPPING_ANON bit are written | |
962 | * simultaneously, so a concurrent reader (eg page_referenced()'s | |
963 | * PageAnon()) will not see one without the other. | |
964 | */ | |
965 | WRITE_ONCE(page->mapping, (struct address_space *) anon_vma); | |
c44b6743 RR |
966 | } |
967 | ||
9617d95e | 968 | /** |
4e1c1975 AK |
969 | * __page_set_anon_rmap - set up new anonymous rmap |
970 | * @page: Page to add to rmap | |
971 | * @vma: VM area to add page to. | |
972 | * @address: User virtual address of the mapping | |
e8a03feb | 973 | * @exclusive: the page is exclusively owned by the current process |
9617d95e NP |
974 | */ |
975 | static void __page_set_anon_rmap(struct page *page, | |
e8a03feb | 976 | struct vm_area_struct *vma, unsigned long address, int exclusive) |
9617d95e | 977 | { |
e8a03feb | 978 | struct anon_vma *anon_vma = vma->anon_vma; |
ea90002b | 979 | |
e8a03feb | 980 | BUG_ON(!anon_vma); |
ea90002b | 981 | |
4e1c1975 AK |
982 | if (PageAnon(page)) |
983 | return; | |
984 | ||
ea90002b | 985 | /* |
e8a03feb RR |
986 | * If the page isn't exclusively mapped into this vma, |
987 | * we must use the _oldest_ possible anon_vma for the | |
988 | * page mapping! | |
ea90002b | 989 | */ |
4e1c1975 | 990 | if (!exclusive) |
288468c3 | 991 | anon_vma = anon_vma->root; |
9617d95e | 992 | |
9617d95e NP |
993 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; |
994 | page->mapping = (struct address_space *) anon_vma; | |
9617d95e | 995 | page->index = linear_page_index(vma, address); |
9617d95e NP |
996 | } |
997 | ||
c97a9e10 | 998 | /** |
43d8eac4 | 999 | * __page_check_anon_rmap - sanity check anonymous rmap addition |
c97a9e10 NP |
1000 | * @page: the page to add the mapping to |
1001 | * @vma: the vm area in which the mapping is added | |
1002 | * @address: the user virtual address mapped | |
1003 | */ | |
1004 | static void __page_check_anon_rmap(struct page *page, | |
1005 | struct vm_area_struct *vma, unsigned long address) | |
1006 | { | |
1007 | #ifdef CONFIG_DEBUG_VM | |
1008 | /* | |
1009 | * The page's anon-rmap details (mapping and index) are guaranteed to | |
1010 | * be set up correctly at this point. | |
1011 | * | |
1012 | * We have exclusion against page_add_anon_rmap because the caller | |
1013 | * always holds the page locked, except if called from page_dup_rmap, | |
1014 | * in which case the page is already known to be setup. | |
1015 | * | |
1016 | * We have exclusion against page_add_new_anon_rmap because those pages | |
1017 | * are initially only visible via the pagetables, and the pte is locked | |
1018 | * over the call to page_add_new_anon_rmap. | |
1019 | */ | |
44ab57a0 | 1020 | BUG_ON(page_anon_vma(page)->root != vma->anon_vma->root); |
53f9263b | 1021 | BUG_ON(page_to_pgoff(page) != linear_page_index(vma, address)); |
c97a9e10 NP |
1022 | #endif |
1023 | } | |
1024 | ||
1da177e4 LT |
1025 | /** |
1026 | * page_add_anon_rmap - add pte mapping to an anonymous page | |
1027 | * @page: the page to add the mapping to | |
1028 | * @vma: the vm area in which the mapping is added | |
1029 | * @address: the user virtual address mapped | |
d281ee61 | 1030 | * @compound: charge the page as compound or small page |
1da177e4 | 1031 | * |
5ad64688 | 1032 | * The caller needs to hold the pte lock, and the page must be locked in |
80e14822 HD |
1033 | * the anon_vma case: to serialize mapping,index checking after setting, |
1034 | * and to ensure that PageAnon is not being upgraded racily to PageKsm | |
1035 | * (but PageKsm is never downgraded to PageAnon). | |
1da177e4 LT |
1036 | */ |
1037 | void page_add_anon_rmap(struct page *page, | |
d281ee61 | 1038 | struct vm_area_struct *vma, unsigned long address, bool compound) |
ad8c2ee8 | 1039 | { |
d281ee61 | 1040 | do_page_add_anon_rmap(page, vma, address, compound ? RMAP_COMPOUND : 0); |
ad8c2ee8 RR |
1041 | } |
1042 | ||
1043 | /* | |
1044 | * Special version of the above for do_swap_page, which often runs | |
1045 | * into pages that are exclusively owned by the current process. | |
1046 | * Everybody else should continue to use page_add_anon_rmap above. | |
1047 | */ | |
1048 | void do_page_add_anon_rmap(struct page *page, | |
d281ee61 | 1049 | struct vm_area_struct *vma, unsigned long address, int flags) |
1da177e4 | 1050 | { |
53f9263b KS |
1051 | bool compound = flags & RMAP_COMPOUND; |
1052 | bool first; | |
1053 | ||
e9b61f19 KS |
1054 | if (compound) { |
1055 | atomic_t *mapcount; | |
53f9263b | 1056 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
e9b61f19 KS |
1057 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
1058 | mapcount = compound_mapcount_ptr(page); | |
1059 | first = atomic_inc_and_test(mapcount); | |
53f9263b KS |
1060 | } else { |
1061 | first = atomic_inc_and_test(&page->_mapcount); | |
1062 | } | |
1063 | ||
79134171 | 1064 | if (first) { |
d281ee61 | 1065 | int nr = compound ? hpage_nr_pages(page) : 1; |
bea04b07 JZ |
1066 | /* |
1067 | * We use the irq-unsafe __{inc|mod}_zone_page_stat because | |
1068 | * these counters are not modified in interrupt context, and | |
1069 | * pte lock(a spinlock) is held, which implies preemption | |
1070 | * disabled. | |
1071 | */ | |
65c45377 | 1072 | if (compound) |
11fb9989 | 1073 | __inc_node_page_state(page, NR_ANON_THPS); |
4b9d0fab | 1074 | __mod_node_page_state(page_pgdat(page), NR_ANON_MAPPED, nr); |
79134171 | 1075 | } |
5ad64688 HD |
1076 | if (unlikely(PageKsm(page))) |
1077 | return; | |
1078 | ||
309381fe | 1079 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
53f9263b | 1080 | |
5dbe0af4 | 1081 | /* address might be in next vma when migration races vma_adjust */ |
5ad64688 | 1082 | if (first) |
d281ee61 KS |
1083 | __page_set_anon_rmap(page, vma, address, |
1084 | flags & RMAP_EXCLUSIVE); | |
69029cd5 | 1085 | else |
c97a9e10 | 1086 | __page_check_anon_rmap(page, vma, address); |
1da177e4 LT |
1087 | } |
1088 | ||
43d8eac4 | 1089 | /** |
9617d95e NP |
1090 | * page_add_new_anon_rmap - add pte mapping to a new anonymous page |
1091 | * @page: the page to add the mapping to | |
1092 | * @vma: the vm area in which the mapping is added | |
1093 | * @address: the user virtual address mapped | |
d281ee61 | 1094 | * @compound: charge the page as compound or small page |
9617d95e NP |
1095 | * |
1096 | * Same as page_add_anon_rmap but must only be called on *new* pages. | |
1097 | * This means the inc-and-test can be bypassed. | |
c97a9e10 | 1098 | * Page does not have to be locked. |
9617d95e NP |
1099 | */ |
1100 | void page_add_new_anon_rmap(struct page *page, | |
d281ee61 | 1101 | struct vm_area_struct *vma, unsigned long address, bool compound) |
9617d95e | 1102 | { |
d281ee61 KS |
1103 | int nr = compound ? hpage_nr_pages(page) : 1; |
1104 | ||
81d1b09c | 1105 | VM_BUG_ON_VMA(address < vma->vm_start || address >= vma->vm_end, vma); |
fa9949da | 1106 | __SetPageSwapBacked(page); |
d281ee61 KS |
1107 | if (compound) { |
1108 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
53f9263b KS |
1109 | /* increment count (starts at -1) */ |
1110 | atomic_set(compound_mapcount_ptr(page), 0); | |
11fb9989 | 1111 | __inc_node_page_state(page, NR_ANON_THPS); |
53f9263b KS |
1112 | } else { |
1113 | /* Anon THP always mapped first with PMD */ | |
1114 | VM_BUG_ON_PAGE(PageTransCompound(page), page); | |
1115 | /* increment count (starts at -1) */ | |
1116 | atomic_set(&page->_mapcount, 0); | |
d281ee61 | 1117 | } |
4b9d0fab | 1118 | __mod_node_page_state(page_pgdat(page), NR_ANON_MAPPED, nr); |
e8a03feb | 1119 | __page_set_anon_rmap(page, vma, address, 1); |
9617d95e NP |
1120 | } |
1121 | ||
1da177e4 LT |
1122 | /** |
1123 | * page_add_file_rmap - add pte mapping to a file page | |
1124 | * @page: the page to add the mapping to | |
1125 | * | |
b8072f09 | 1126 | * The caller needs to hold the pte lock. |
1da177e4 | 1127 | */ |
dd78fedd | 1128 | void page_add_file_rmap(struct page *page, bool compound) |
1da177e4 | 1129 | { |
dd78fedd KS |
1130 | int i, nr = 1; |
1131 | ||
1132 | VM_BUG_ON_PAGE(compound && !PageTransHuge(page), page); | |
62cccb8c | 1133 | lock_page_memcg(page); |
dd78fedd KS |
1134 | if (compound && PageTransHuge(page)) { |
1135 | for (i = 0, nr = 0; i < HPAGE_PMD_NR; i++) { | |
1136 | if (atomic_inc_and_test(&page[i]._mapcount)) | |
1137 | nr++; | |
1138 | } | |
1139 | if (!atomic_inc_and_test(compound_mapcount_ptr(page))) | |
1140 | goto out; | |
65c45377 | 1141 | VM_BUG_ON_PAGE(!PageSwapBacked(page), page); |
11fb9989 | 1142 | __inc_node_page_state(page, NR_SHMEM_PMDMAPPED); |
dd78fedd | 1143 | } else { |
c8efc390 KS |
1144 | if (PageTransCompound(page) && page_mapping(page)) { |
1145 | VM_WARN_ON_ONCE(!PageLocked(page)); | |
1146 | ||
9a73f61b KS |
1147 | SetPageDoubleMap(compound_head(page)); |
1148 | if (PageMlocked(page)) | |
1149 | clear_page_mlock(compound_head(page)); | |
1150 | } | |
dd78fedd KS |
1151 | if (!atomic_inc_and_test(&page->_mapcount)) |
1152 | goto out; | |
d69b042f | 1153 | } |
50658e2e | 1154 | __mod_node_page_state(page_pgdat(page), NR_FILE_MAPPED, nr); |
dd78fedd KS |
1155 | mem_cgroup_inc_page_stat(page, MEM_CGROUP_STAT_FILE_MAPPED); |
1156 | out: | |
62cccb8c | 1157 | unlock_page_memcg(page); |
1da177e4 LT |
1158 | } |
1159 | ||
dd78fedd | 1160 | static void page_remove_file_rmap(struct page *page, bool compound) |
8186eb6a | 1161 | { |
dd78fedd KS |
1162 | int i, nr = 1; |
1163 | ||
57dea93a | 1164 | VM_BUG_ON_PAGE(compound && !PageHead(page), page); |
62cccb8c | 1165 | lock_page_memcg(page); |
8186eb6a | 1166 | |
53f9263b KS |
1167 | /* Hugepages are not counted in NR_FILE_MAPPED for now. */ |
1168 | if (unlikely(PageHuge(page))) { | |
1169 | /* hugetlb pages are always mapped with pmds */ | |
1170 | atomic_dec(compound_mapcount_ptr(page)); | |
8186eb6a | 1171 | goto out; |
53f9263b | 1172 | } |
8186eb6a | 1173 | |
53f9263b | 1174 | /* page still mapped by someone else? */ |
dd78fedd KS |
1175 | if (compound && PageTransHuge(page)) { |
1176 | for (i = 0, nr = 0; i < HPAGE_PMD_NR; i++) { | |
1177 | if (atomic_add_negative(-1, &page[i]._mapcount)) | |
1178 | nr++; | |
1179 | } | |
1180 | if (!atomic_add_negative(-1, compound_mapcount_ptr(page))) | |
1181 | goto out; | |
65c45377 | 1182 | VM_BUG_ON_PAGE(!PageSwapBacked(page), page); |
11fb9989 | 1183 | __dec_node_page_state(page, NR_SHMEM_PMDMAPPED); |
dd78fedd KS |
1184 | } else { |
1185 | if (!atomic_add_negative(-1, &page->_mapcount)) | |
1186 | goto out; | |
1187 | } | |
8186eb6a JW |
1188 | |
1189 | /* | |
50658e2e | 1190 | * We use the irq-unsafe __{inc|mod}_zone_page_state because |
8186eb6a JW |
1191 | * these counters are not modified in interrupt context, and |
1192 | * pte lock(a spinlock) is held, which implies preemption disabled. | |
1193 | */ | |
50658e2e | 1194 | __mod_node_page_state(page_pgdat(page), NR_FILE_MAPPED, -nr); |
62cccb8c | 1195 | mem_cgroup_dec_page_stat(page, MEM_CGROUP_STAT_FILE_MAPPED); |
8186eb6a JW |
1196 | |
1197 | if (unlikely(PageMlocked(page))) | |
1198 | clear_page_mlock(page); | |
1199 | out: | |
62cccb8c | 1200 | unlock_page_memcg(page); |
8186eb6a JW |
1201 | } |
1202 | ||
53f9263b KS |
1203 | static void page_remove_anon_compound_rmap(struct page *page) |
1204 | { | |
1205 | int i, nr; | |
1206 | ||
1207 | if (!atomic_add_negative(-1, compound_mapcount_ptr(page))) | |
1208 | return; | |
1209 | ||
1210 | /* Hugepages are not counted in NR_ANON_PAGES for now. */ | |
1211 | if (unlikely(PageHuge(page))) | |
1212 | return; | |
1213 | ||
1214 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) | |
1215 | return; | |
1216 | ||
11fb9989 | 1217 | __dec_node_page_state(page, NR_ANON_THPS); |
53f9263b KS |
1218 | |
1219 | if (TestClearPageDoubleMap(page)) { | |
1220 | /* | |
1221 | * Subpages can be mapped with PTEs too. Check how many of | |
1222 | * themi are still mapped. | |
1223 | */ | |
1224 | for (i = 0, nr = 0; i < HPAGE_PMD_NR; i++) { | |
1225 | if (atomic_add_negative(-1, &page[i]._mapcount)) | |
1226 | nr++; | |
1227 | } | |
1228 | } else { | |
1229 | nr = HPAGE_PMD_NR; | |
1230 | } | |
1231 | ||
e90309c9 KS |
1232 | if (unlikely(PageMlocked(page))) |
1233 | clear_page_mlock(page); | |
1234 | ||
9a982250 | 1235 | if (nr) { |
4b9d0fab | 1236 | __mod_node_page_state(page_pgdat(page), NR_ANON_MAPPED, -nr); |
9a982250 KS |
1237 | deferred_split_huge_page(page); |
1238 | } | |
53f9263b KS |
1239 | } |
1240 | ||
1da177e4 LT |
1241 | /** |
1242 | * page_remove_rmap - take down pte mapping from a page | |
d281ee61 KS |
1243 | * @page: page to remove mapping from |
1244 | * @compound: uncharge the page as compound or small page | |
1da177e4 | 1245 | * |
b8072f09 | 1246 | * The caller needs to hold the pte lock. |
1da177e4 | 1247 | */ |
d281ee61 | 1248 | void page_remove_rmap(struct page *page, bool compound) |
1da177e4 | 1249 | { |
dd78fedd KS |
1250 | if (!PageAnon(page)) |
1251 | return page_remove_file_rmap(page, compound); | |
89c06bd5 | 1252 | |
53f9263b KS |
1253 | if (compound) |
1254 | return page_remove_anon_compound_rmap(page); | |
1255 | ||
b904dcfe KM |
1256 | /* page still mapped by someone else? */ |
1257 | if (!atomic_add_negative(-1, &page->_mapcount)) | |
8186eb6a JW |
1258 | return; |
1259 | ||
0fe6e20b | 1260 | /* |
bea04b07 JZ |
1261 | * We use the irq-unsafe __{inc|mod}_zone_page_stat because |
1262 | * these counters are not modified in interrupt context, and | |
bea04b07 | 1263 | * pte lock(a spinlock) is held, which implies preemption disabled. |
0fe6e20b | 1264 | */ |
4b9d0fab | 1265 | __dec_node_page_state(page, NR_ANON_MAPPED); |
8186eb6a | 1266 | |
e6c509f8 HD |
1267 | if (unlikely(PageMlocked(page))) |
1268 | clear_page_mlock(page); | |
8186eb6a | 1269 | |
9a982250 KS |
1270 | if (PageTransCompound(page)) |
1271 | deferred_split_huge_page(compound_head(page)); | |
1272 | ||
b904dcfe KM |
1273 | /* |
1274 | * It would be tidy to reset the PageAnon mapping here, | |
1275 | * but that might overwrite a racing page_add_anon_rmap | |
1276 | * which increments mapcount after us but sets mapping | |
1277 | * before us: so leave the reset to free_hot_cold_page, | |
1278 | * and remember that it's only reliable while mapped. | |
1279 | * Leaving it set also helps swapoff to reinstate ptes | |
1280 | * faster for those pages still in swapcache. | |
1281 | */ | |
1da177e4 LT |
1282 | } |
1283 | ||
854e9ed0 MK |
1284 | struct rmap_private { |
1285 | enum ttu_flags flags; | |
1286 | int lazyfreed; | |
1287 | }; | |
1288 | ||
1da177e4 | 1289 | /* |
52629506 | 1290 | * @arg: enum ttu_flags will be passed to this argument |
1da177e4 | 1291 | */ |
ac769501 | 1292 | static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma, |
52629506 | 1293 | unsigned long address, void *arg) |
1da177e4 LT |
1294 | { |
1295 | struct mm_struct *mm = vma->vm_mm; | |
c7ab0d2f KS |
1296 | struct page_vma_mapped_walk pvmw = { |
1297 | .page = page, | |
1298 | .vma = vma, | |
1299 | .address = address, | |
1300 | }; | |
1da177e4 | 1301 | pte_t pteval; |
c7ab0d2f | 1302 | struct page *subpage; |
1da177e4 | 1303 | int ret = SWAP_AGAIN; |
854e9ed0 MK |
1304 | struct rmap_private *rp = arg; |
1305 | enum ttu_flags flags = rp->flags; | |
1da177e4 | 1306 | |
b87537d9 HD |
1307 | /* munlock has nothing to gain from examining un-locked vmas */ |
1308 | if ((flags & TTU_MUNLOCK) && !(vma->vm_flags & VM_LOCKED)) | |
c7ab0d2f | 1309 | return SWAP_AGAIN; |
b87537d9 | 1310 | |
fec89c10 KS |
1311 | if (flags & TTU_SPLIT_HUGE_PMD) { |
1312 | split_huge_pmd_address(vma, address, | |
1313 | flags & TTU_MIGRATION, page); | |
fec89c10 KS |
1314 | } |
1315 | ||
c7ab0d2f KS |
1316 | while (page_vma_mapped_walk(&pvmw)) { |
1317 | subpage = page - page_to_pfn(page) + pte_pfn(*pvmw.pte); | |
1318 | address = pvmw.address; | |
1da177e4 | 1319 | |
c7ab0d2f KS |
1320 | /* Unexpected PMD-mapped THP? */ |
1321 | VM_BUG_ON_PAGE(!pvmw.pte, page); | |
1322 | ||
1323 | /* | |
1324 | * If the page is mlock()d, we cannot swap it out. | |
1325 | * If it's recently referenced (perhaps page_referenced | |
1326 | * skipped over this mm) then we should reactivate it. | |
1327 | */ | |
1328 | if (!(flags & TTU_IGNORE_MLOCK)) { | |
1329 | if (vma->vm_flags & VM_LOCKED) { | |
1330 | /* PTE-mapped THP are never mlocked */ | |
1331 | if (!PageTransCompound(page)) { | |
1332 | /* | |
1333 | * Holding pte lock, we do *not* need | |
1334 | * mmap_sem here | |
1335 | */ | |
1336 | mlock_vma_page(page); | |
1337 | } | |
1338 | ret = SWAP_MLOCK; | |
1339 | page_vma_mapped_walk_done(&pvmw); | |
1340 | break; | |
9a73f61b | 1341 | } |
c7ab0d2f KS |
1342 | if (flags & TTU_MUNLOCK) |
1343 | continue; | |
b87537d9 | 1344 | } |
c7ab0d2f KS |
1345 | |
1346 | if (!(flags & TTU_IGNORE_ACCESS)) { | |
1347 | if (ptep_clear_flush_young_notify(vma, address, | |
1348 | pvmw.pte)) { | |
1349 | ret = SWAP_FAIL; | |
1350 | page_vma_mapped_walk_done(&pvmw); | |
1351 | break; | |
1352 | } | |
b291f000 | 1353 | } |
1da177e4 | 1354 | |
c7ab0d2f KS |
1355 | /* Nuke the page table entry. */ |
1356 | flush_cache_page(vma, address, pte_pfn(*pvmw.pte)); | |
1357 | if (should_defer_flush(mm, flags)) { | |
1358 | /* | |
1359 | * We clear the PTE but do not flush so potentially | |
1360 | * a remote CPU could still be writing to the page. | |
1361 | * If the entry was previously clean then the | |
1362 | * architecture must guarantee that a clear->dirty | |
1363 | * transition on a cached TLB entry is written through | |
1364 | * and traps if the PTE is unmapped. | |
1365 | */ | |
1366 | pteval = ptep_get_and_clear(mm, address, pvmw.pte); | |
1367 | ||
1368 | set_tlb_ubc_flush_pending(mm, pte_dirty(pteval)); | |
1369 | } else { | |
1370 | pteval = ptep_clear_flush(vma, address, pvmw.pte); | |
1371 | } | |
72b252ae | 1372 | |
c7ab0d2f KS |
1373 | /* Move the dirty bit to the page. Now the pte is gone. */ |
1374 | if (pte_dirty(pteval)) | |
1375 | set_page_dirty(page); | |
1da177e4 | 1376 | |
c7ab0d2f KS |
1377 | /* Update high watermark before we lower rss */ |
1378 | update_hiwater_rss(mm); | |
1da177e4 | 1379 | |
c7ab0d2f KS |
1380 | if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) { |
1381 | if (PageHuge(page)) { | |
1382 | int nr = 1 << compound_order(page); | |
1383 | hugetlb_count_sub(nr, mm); | |
1384 | } else { | |
1385 | dec_mm_counter(mm, mm_counter(page)); | |
1386 | } | |
365e9c87 | 1387 | |
c7ab0d2f KS |
1388 | pteval = swp_entry_to_pte(make_hwpoison_entry(subpage)); |
1389 | set_pte_at(mm, address, pvmw.pte, pteval); | |
1390 | } else if (pte_unused(pteval)) { | |
1391 | /* | |
1392 | * The guest indicated that the page content is of no | |
1393 | * interest anymore. Simply discard the pte, vmscan | |
1394 | * will take care of the rest. | |
1395 | */ | |
eca56ff9 | 1396 | dec_mm_counter(mm, mm_counter(page)); |
c7ab0d2f KS |
1397 | } else if (IS_ENABLED(CONFIG_MIGRATION) && |
1398 | (flags & TTU_MIGRATION)) { | |
1399 | swp_entry_t entry; | |
1400 | pte_t swp_pte; | |
1401 | /* | |
1402 | * Store the pfn of the page in a special migration | |
1403 | * pte. do_swap_page() will wait until the migration | |
1404 | * pte is removed and then restart fault handling. | |
1405 | */ | |
1406 | entry = make_migration_entry(subpage, | |
1407 | pte_write(pteval)); | |
1408 | swp_pte = swp_entry_to_pte(entry); | |
1409 | if (pte_soft_dirty(pteval)) | |
1410 | swp_pte = pte_swp_mksoft_dirty(swp_pte); | |
1411 | set_pte_at(mm, address, pvmw.pte, swp_pte); | |
1412 | } else if (PageAnon(page)) { | |
1413 | swp_entry_t entry = { .val = page_private(subpage) }; | |
1414 | pte_t swp_pte; | |
1415 | /* | |
1416 | * Store the swap location in the pte. | |
1417 | * See handle_pte_fault() ... | |
1418 | */ | |
1419 | VM_BUG_ON_PAGE(!PageSwapCache(page), page); | |
1420 | ||
1421 | if (!PageDirty(page) && (flags & TTU_LZFREE)) { | |
1422 | /* It's a freeable page by MADV_FREE */ | |
1423 | dec_mm_counter(mm, MM_ANONPAGES); | |
1424 | rp->lazyfreed++; | |
1425 | goto discard; | |
1426 | } | |
854e9ed0 | 1427 | |
c7ab0d2f KS |
1428 | if (swap_duplicate(entry) < 0) { |
1429 | set_pte_at(mm, address, pvmw.pte, pteval); | |
1430 | ret = SWAP_FAIL; | |
1431 | page_vma_mapped_walk_done(&pvmw); | |
1432 | break; | |
1433 | } | |
1434 | if (list_empty(&mm->mmlist)) { | |
1435 | spin_lock(&mmlist_lock); | |
1436 | if (list_empty(&mm->mmlist)) | |
1437 | list_add(&mm->mmlist, &init_mm.mmlist); | |
1438 | spin_unlock(&mmlist_lock); | |
1439 | } | |
854e9ed0 | 1440 | dec_mm_counter(mm, MM_ANONPAGES); |
c7ab0d2f KS |
1441 | inc_mm_counter(mm, MM_SWAPENTS); |
1442 | swp_pte = swp_entry_to_pte(entry); | |
1443 | if (pte_soft_dirty(pteval)) | |
1444 | swp_pte = pte_swp_mksoft_dirty(swp_pte); | |
1445 | set_pte_at(mm, address, pvmw.pte, swp_pte); | |
1446 | } else | |
1447 | dec_mm_counter(mm, mm_counter_file(page)); | |
854e9ed0 | 1448 | discard: |
c7ab0d2f KS |
1449 | page_remove_rmap(subpage, PageHuge(page)); |
1450 | put_page(page); | |
2ec74c3e | 1451 | mmu_notifier_invalidate_page(mm, address); |
c7ab0d2f | 1452 | } |
caed0f48 | 1453 | return ret; |
1da177e4 LT |
1454 | } |
1455 | ||
71e3aac0 | 1456 | bool is_vma_temporary_stack(struct vm_area_struct *vma) |
a8bef8ff MG |
1457 | { |
1458 | int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP); | |
1459 | ||
1460 | if (!maybe_stack) | |
1461 | return false; | |
1462 | ||
1463 | if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) == | |
1464 | VM_STACK_INCOMPLETE_SETUP) | |
1465 | return true; | |
1466 | ||
1467 | return false; | |
1468 | } | |
1469 | ||
52629506 JK |
1470 | static bool invalid_migration_vma(struct vm_area_struct *vma, void *arg) |
1471 | { | |
1472 | return is_vma_temporary_stack(vma); | |
1473 | } | |
1474 | ||
2a52bcbc | 1475 | static int page_mapcount_is_zero(struct page *page) |
52629506 | 1476 | { |
c7ab0d2f | 1477 | return !total_mapcount(page); |
2a52bcbc | 1478 | } |
52629506 | 1479 | |
1da177e4 LT |
1480 | /** |
1481 | * try_to_unmap - try to remove all page table mappings to a page | |
1482 | * @page: the page to get unmapped | |
14fa31b8 | 1483 | * @flags: action and flags |
1da177e4 LT |
1484 | * |
1485 | * Tries to remove all the page table entries which are mapping this | |
1486 | * page, used in the pageout path. Caller must hold the page lock. | |
1487 | * Return values are: | |
1488 | * | |
1489 | * SWAP_SUCCESS - we succeeded in removing all mappings | |
1490 | * SWAP_AGAIN - we missed a mapping, try again later | |
1491 | * SWAP_FAIL - the page is unswappable | |
b291f000 | 1492 | * SWAP_MLOCK - page is mlocked. |
1da177e4 | 1493 | */ |
14fa31b8 | 1494 | int try_to_unmap(struct page *page, enum ttu_flags flags) |
1da177e4 LT |
1495 | { |
1496 | int ret; | |
854e9ed0 MK |
1497 | struct rmap_private rp = { |
1498 | .flags = flags, | |
1499 | .lazyfreed = 0, | |
1500 | }; | |
1501 | ||
52629506 JK |
1502 | struct rmap_walk_control rwc = { |
1503 | .rmap_one = try_to_unmap_one, | |
854e9ed0 | 1504 | .arg = &rp, |
2a52bcbc | 1505 | .done = page_mapcount_is_zero, |
52629506 JK |
1506 | .anon_lock = page_lock_anon_vma_read, |
1507 | }; | |
1da177e4 | 1508 | |
52629506 JK |
1509 | /* |
1510 | * During exec, a temporary VMA is setup and later moved. | |
1511 | * The VMA is moved under the anon_vma lock but not the | |
1512 | * page tables leading to a race where migration cannot | |
1513 | * find the migration ptes. Rather than increasing the | |
1514 | * locking requirements of exec(), migration skips | |
1515 | * temporary VMAs until after exec() completes. | |
1516 | */ | |
daa5ba76 | 1517 | if ((flags & TTU_MIGRATION) && !PageKsm(page) && PageAnon(page)) |
52629506 JK |
1518 | rwc.invalid_vma = invalid_migration_vma; |
1519 | ||
2a52bcbc KS |
1520 | if (flags & TTU_RMAP_LOCKED) |
1521 | ret = rmap_walk_locked(page, &rwc); | |
1522 | else | |
1523 | ret = rmap_walk(page, &rwc); | |
52629506 | 1524 | |
2a52bcbc | 1525 | if (ret != SWAP_MLOCK && !page_mapcount(page)) { |
1da177e4 | 1526 | ret = SWAP_SUCCESS; |
854e9ed0 MK |
1527 | if (rp.lazyfreed && !PageDirty(page)) |
1528 | ret = SWAP_LZFREE; | |
1529 | } | |
1da177e4 LT |
1530 | return ret; |
1531 | } | |
81b4082d | 1532 | |
2a52bcbc KS |
1533 | static int page_not_mapped(struct page *page) |
1534 | { | |
1535 | return !page_mapped(page); | |
1536 | }; | |
1537 | ||
b291f000 NP |
1538 | /** |
1539 | * try_to_munlock - try to munlock a page | |
1540 | * @page: the page to be munlocked | |
1541 | * | |
1542 | * Called from munlock code. Checks all of the VMAs mapping the page | |
1543 | * to make sure nobody else has this page mlocked. The page will be | |
1544 | * returned with PG_mlocked cleared if no other vmas have it mlocked. | |
1545 | * | |
1546 | * Return values are: | |
1547 | * | |
53f79acb | 1548 | * SWAP_AGAIN - no vma is holding page mlocked, or, |
b291f000 | 1549 | * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem |
5ad64688 | 1550 | * SWAP_FAIL - page cannot be located at present |
b291f000 NP |
1551 | * SWAP_MLOCK - page is now mlocked. |
1552 | */ | |
1553 | int try_to_munlock(struct page *page) | |
1554 | { | |
e8351ac9 | 1555 | int ret; |
854e9ed0 MK |
1556 | struct rmap_private rp = { |
1557 | .flags = TTU_MUNLOCK, | |
1558 | .lazyfreed = 0, | |
1559 | }; | |
1560 | ||
e8351ac9 JK |
1561 | struct rmap_walk_control rwc = { |
1562 | .rmap_one = try_to_unmap_one, | |
854e9ed0 | 1563 | .arg = &rp, |
e8351ac9 | 1564 | .done = page_not_mapped, |
e8351ac9 JK |
1565 | .anon_lock = page_lock_anon_vma_read, |
1566 | ||
1567 | }; | |
1568 | ||
309381fe | 1569 | VM_BUG_ON_PAGE(!PageLocked(page) || PageLRU(page), page); |
b291f000 | 1570 | |
e8351ac9 JK |
1571 | ret = rmap_walk(page, &rwc); |
1572 | return ret; | |
b291f000 | 1573 | } |
e9995ef9 | 1574 | |
01d8b20d | 1575 | void __put_anon_vma(struct anon_vma *anon_vma) |
76545066 | 1576 | { |
01d8b20d | 1577 | struct anon_vma *root = anon_vma->root; |
76545066 | 1578 | |
624483f3 | 1579 | anon_vma_free(anon_vma); |
01d8b20d PZ |
1580 | if (root != anon_vma && atomic_dec_and_test(&root->refcount)) |
1581 | anon_vma_free(root); | |
76545066 | 1582 | } |
76545066 | 1583 | |
0dd1c7bb JK |
1584 | static struct anon_vma *rmap_walk_anon_lock(struct page *page, |
1585 | struct rmap_walk_control *rwc) | |
faecd8dd JK |
1586 | { |
1587 | struct anon_vma *anon_vma; | |
1588 | ||
0dd1c7bb JK |
1589 | if (rwc->anon_lock) |
1590 | return rwc->anon_lock(page); | |
1591 | ||
faecd8dd JK |
1592 | /* |
1593 | * Note: remove_migration_ptes() cannot use page_lock_anon_vma_read() | |
1594 | * because that depends on page_mapped(); but not all its usages | |
1595 | * are holding mmap_sem. Users without mmap_sem are required to | |
1596 | * take a reference count to prevent the anon_vma disappearing | |
1597 | */ | |
1598 | anon_vma = page_anon_vma(page); | |
1599 | if (!anon_vma) | |
1600 | return NULL; | |
1601 | ||
1602 | anon_vma_lock_read(anon_vma); | |
1603 | return anon_vma; | |
1604 | } | |
1605 | ||
e9995ef9 | 1606 | /* |
e8351ac9 JK |
1607 | * rmap_walk_anon - do something to anonymous page using the object-based |
1608 | * rmap method | |
1609 | * @page: the page to be handled | |
1610 | * @rwc: control variable according to each walk type | |
1611 | * | |
1612 | * Find all the mappings of a page using the mapping pointer and the vma chains | |
1613 | * contained in the anon_vma struct it points to. | |
1614 | * | |
1615 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma | |
1616 | * where the page was found will be held for write. So, we won't recheck | |
1617 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be | |
1618 | * LOCKED. | |
e9995ef9 | 1619 | */ |
b9773199 KS |
1620 | static int rmap_walk_anon(struct page *page, struct rmap_walk_control *rwc, |
1621 | bool locked) | |
e9995ef9 HD |
1622 | { |
1623 | struct anon_vma *anon_vma; | |
a8fa41ad | 1624 | pgoff_t pgoff_start, pgoff_end; |
5beb4930 | 1625 | struct anon_vma_chain *avc; |
e9995ef9 HD |
1626 | int ret = SWAP_AGAIN; |
1627 | ||
b9773199 KS |
1628 | if (locked) { |
1629 | anon_vma = page_anon_vma(page); | |
1630 | /* anon_vma disappear under us? */ | |
1631 | VM_BUG_ON_PAGE(!anon_vma, page); | |
1632 | } else { | |
1633 | anon_vma = rmap_walk_anon_lock(page, rwc); | |
1634 | } | |
e9995ef9 HD |
1635 | if (!anon_vma) |
1636 | return ret; | |
faecd8dd | 1637 | |
a8fa41ad KS |
1638 | pgoff_start = page_to_pgoff(page); |
1639 | pgoff_end = pgoff_start + hpage_nr_pages(page) - 1; | |
1640 | anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, | |
1641 | pgoff_start, pgoff_end) { | |
5beb4930 | 1642 | struct vm_area_struct *vma = avc->vma; |
e9995ef9 | 1643 | unsigned long address = vma_address(page, vma); |
0dd1c7bb | 1644 | |
ad12695f AA |
1645 | cond_resched(); |
1646 | ||
0dd1c7bb JK |
1647 | if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg)) |
1648 | continue; | |
1649 | ||
051ac83a | 1650 | ret = rwc->rmap_one(page, vma, address, rwc->arg); |
e9995ef9 HD |
1651 | if (ret != SWAP_AGAIN) |
1652 | break; | |
0dd1c7bb JK |
1653 | if (rwc->done && rwc->done(page)) |
1654 | break; | |
e9995ef9 | 1655 | } |
b9773199 KS |
1656 | |
1657 | if (!locked) | |
1658 | anon_vma_unlock_read(anon_vma); | |
e9995ef9 HD |
1659 | return ret; |
1660 | } | |
1661 | ||
e8351ac9 JK |
1662 | /* |
1663 | * rmap_walk_file - do something to file page using the object-based rmap method | |
1664 | * @page: the page to be handled | |
1665 | * @rwc: control variable according to each walk type | |
1666 | * | |
1667 | * Find all the mappings of a page using the mapping pointer and the vma chains | |
1668 | * contained in the address_space struct it points to. | |
1669 | * | |
1670 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma | |
1671 | * where the page was found will be held for write. So, we won't recheck | |
1672 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be | |
1673 | * LOCKED. | |
1674 | */ | |
b9773199 KS |
1675 | static int rmap_walk_file(struct page *page, struct rmap_walk_control *rwc, |
1676 | bool locked) | |
e9995ef9 | 1677 | { |
b9773199 | 1678 | struct address_space *mapping = page_mapping(page); |
a8fa41ad | 1679 | pgoff_t pgoff_start, pgoff_end; |
e9995ef9 | 1680 | struct vm_area_struct *vma; |
e9995ef9 HD |
1681 | int ret = SWAP_AGAIN; |
1682 | ||
9f32624b JK |
1683 | /* |
1684 | * The page lock not only makes sure that page->mapping cannot | |
1685 | * suddenly be NULLified by truncation, it makes sure that the | |
1686 | * structure at mapping cannot be freed and reused yet, | |
c8c06efa | 1687 | * so we can safely take mapping->i_mmap_rwsem. |
9f32624b | 1688 | */ |
81d1b09c | 1689 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
9f32624b | 1690 | |
e9995ef9 HD |
1691 | if (!mapping) |
1692 | return ret; | |
3dec0ba0 | 1693 | |
a8fa41ad KS |
1694 | pgoff_start = page_to_pgoff(page); |
1695 | pgoff_end = pgoff_start + hpage_nr_pages(page) - 1; | |
b9773199 KS |
1696 | if (!locked) |
1697 | i_mmap_lock_read(mapping); | |
a8fa41ad KS |
1698 | vma_interval_tree_foreach(vma, &mapping->i_mmap, |
1699 | pgoff_start, pgoff_end) { | |
e9995ef9 | 1700 | unsigned long address = vma_address(page, vma); |
0dd1c7bb | 1701 | |
ad12695f AA |
1702 | cond_resched(); |
1703 | ||
0dd1c7bb JK |
1704 | if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg)) |
1705 | continue; | |
1706 | ||
051ac83a | 1707 | ret = rwc->rmap_one(page, vma, address, rwc->arg); |
e9995ef9 | 1708 | if (ret != SWAP_AGAIN) |
0dd1c7bb JK |
1709 | goto done; |
1710 | if (rwc->done && rwc->done(page)) | |
1711 | goto done; | |
e9995ef9 | 1712 | } |
0dd1c7bb | 1713 | |
0dd1c7bb | 1714 | done: |
b9773199 KS |
1715 | if (!locked) |
1716 | i_mmap_unlock_read(mapping); | |
e9995ef9 HD |
1717 | return ret; |
1718 | } | |
1719 | ||
051ac83a | 1720 | int rmap_walk(struct page *page, struct rmap_walk_control *rwc) |
e9995ef9 | 1721 | { |
e9995ef9 | 1722 | if (unlikely(PageKsm(page))) |
051ac83a | 1723 | return rmap_walk_ksm(page, rwc); |
e9995ef9 | 1724 | else if (PageAnon(page)) |
b9773199 KS |
1725 | return rmap_walk_anon(page, rwc, false); |
1726 | else | |
1727 | return rmap_walk_file(page, rwc, false); | |
1728 | } | |
1729 | ||
1730 | /* Like rmap_walk, but caller holds relevant rmap lock */ | |
1731 | int rmap_walk_locked(struct page *page, struct rmap_walk_control *rwc) | |
1732 | { | |
1733 | /* no ksm support for now */ | |
1734 | VM_BUG_ON_PAGE(PageKsm(page), page); | |
1735 | if (PageAnon(page)) | |
1736 | return rmap_walk_anon(page, rwc, true); | |
e9995ef9 | 1737 | else |
b9773199 | 1738 | return rmap_walk_file(page, rwc, true); |
e9995ef9 | 1739 | } |
0fe6e20b | 1740 | |
e3390f67 | 1741 | #ifdef CONFIG_HUGETLB_PAGE |
0fe6e20b NH |
1742 | /* |
1743 | * The following three functions are for anonymous (private mapped) hugepages. | |
1744 | * Unlike common anonymous pages, anonymous hugepages have no accounting code | |
1745 | * and no lru code, because we handle hugepages differently from common pages. | |
1746 | */ | |
1747 | static void __hugepage_set_anon_rmap(struct page *page, | |
1748 | struct vm_area_struct *vma, unsigned long address, int exclusive) | |
1749 | { | |
1750 | struct anon_vma *anon_vma = vma->anon_vma; | |
433abed6 | 1751 | |
0fe6e20b | 1752 | BUG_ON(!anon_vma); |
433abed6 NH |
1753 | |
1754 | if (PageAnon(page)) | |
1755 | return; | |
1756 | if (!exclusive) | |
1757 | anon_vma = anon_vma->root; | |
1758 | ||
0fe6e20b NH |
1759 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; |
1760 | page->mapping = (struct address_space *) anon_vma; | |
1761 | page->index = linear_page_index(vma, address); | |
1762 | } | |
1763 | ||
1764 | void hugepage_add_anon_rmap(struct page *page, | |
1765 | struct vm_area_struct *vma, unsigned long address) | |
1766 | { | |
1767 | struct anon_vma *anon_vma = vma->anon_vma; | |
1768 | int first; | |
a850ea30 NH |
1769 | |
1770 | BUG_ON(!PageLocked(page)); | |
0fe6e20b | 1771 | BUG_ON(!anon_vma); |
5dbe0af4 | 1772 | /* address might be in next vma when migration races vma_adjust */ |
53f9263b | 1773 | first = atomic_inc_and_test(compound_mapcount_ptr(page)); |
0fe6e20b NH |
1774 | if (first) |
1775 | __hugepage_set_anon_rmap(page, vma, address, 0); | |
1776 | } | |
1777 | ||
1778 | void hugepage_add_new_anon_rmap(struct page *page, | |
1779 | struct vm_area_struct *vma, unsigned long address) | |
1780 | { | |
1781 | BUG_ON(address < vma->vm_start || address >= vma->vm_end); | |
53f9263b | 1782 | atomic_set(compound_mapcount_ptr(page), 0); |
0fe6e20b NH |
1783 | __hugepage_set_anon_rmap(page, vma, address, 1); |
1784 | } | |
e3390f67 | 1785 | #endif /* CONFIG_HUGETLB_PAGE */ |