]>
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> | |
6e84f315 | 49 | #include <linux/sched/mm.h> |
29930025 | 50 | #include <linux/sched/task.h> |
1da177e4 LT |
51 | #include <linux/pagemap.h> |
52 | #include <linux/swap.h> | |
53 | #include <linux/swapops.h> | |
54 | #include <linux/slab.h> | |
55 | #include <linux/init.h> | |
5ad64688 | 56 | #include <linux/ksm.h> |
1da177e4 LT |
57 | #include <linux/rmap.h> |
58 | #include <linux/rcupdate.h> | |
b95f1b31 | 59 | #include <linux/export.h> |
8a9f3ccd | 60 | #include <linux/memcontrol.h> |
cddb8a5c | 61 | #include <linux/mmu_notifier.h> |
64cdd548 | 62 | #include <linux/migrate.h> |
0fe6e20b | 63 | #include <linux/hugetlb.h> |
ef5d437f | 64 | #include <linux/backing-dev.h> |
33c3fc71 | 65 | #include <linux/page_idle.h> |
1da177e4 LT |
66 | |
67 | #include <asm/tlbflush.h> | |
68 | ||
72b252ae MG |
69 | #include <trace/events/tlb.h> |
70 | ||
b291f000 NP |
71 | #include "internal.h" |
72 | ||
fdd2e5f8 | 73 | static struct kmem_cache *anon_vma_cachep; |
5beb4930 | 74 | static struct kmem_cache *anon_vma_chain_cachep; |
fdd2e5f8 AB |
75 | |
76 | static inline struct anon_vma *anon_vma_alloc(void) | |
77 | { | |
01d8b20d PZ |
78 | struct anon_vma *anon_vma; |
79 | ||
80 | anon_vma = kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL); | |
81 | if (anon_vma) { | |
82 | atomic_set(&anon_vma->refcount, 1); | |
7a3ef208 KK |
83 | anon_vma->degree = 1; /* Reference for first vma */ |
84 | anon_vma->parent = anon_vma; | |
01d8b20d PZ |
85 | /* |
86 | * Initialise the anon_vma root to point to itself. If called | |
87 | * from fork, the root will be reset to the parents anon_vma. | |
88 | */ | |
89 | anon_vma->root = anon_vma; | |
90 | } | |
91 | ||
92 | return anon_vma; | |
fdd2e5f8 AB |
93 | } |
94 | ||
01d8b20d | 95 | static inline void anon_vma_free(struct anon_vma *anon_vma) |
fdd2e5f8 | 96 | { |
01d8b20d | 97 | VM_BUG_ON(atomic_read(&anon_vma->refcount)); |
88c22088 PZ |
98 | |
99 | /* | |
4fc3f1d6 | 100 | * Synchronize against page_lock_anon_vma_read() such that |
88c22088 PZ |
101 | * we can safely hold the lock without the anon_vma getting |
102 | * freed. | |
103 | * | |
104 | * Relies on the full mb implied by the atomic_dec_and_test() from | |
105 | * put_anon_vma() against the acquire barrier implied by | |
4fc3f1d6 | 106 | * down_read_trylock() from page_lock_anon_vma_read(). This orders: |
88c22088 | 107 | * |
4fc3f1d6 IM |
108 | * page_lock_anon_vma_read() VS put_anon_vma() |
109 | * down_read_trylock() atomic_dec_and_test() | |
88c22088 | 110 | * LOCK MB |
4fc3f1d6 | 111 | * atomic_read() rwsem_is_locked() |
88c22088 PZ |
112 | * |
113 | * LOCK should suffice since the actual taking of the lock must | |
114 | * happen _before_ what follows. | |
115 | */ | |
7f39dda9 | 116 | might_sleep(); |
5a505085 | 117 | if (rwsem_is_locked(&anon_vma->root->rwsem)) { |
4fc3f1d6 | 118 | anon_vma_lock_write(anon_vma); |
08b52706 | 119 | anon_vma_unlock_write(anon_vma); |
88c22088 PZ |
120 | } |
121 | ||
fdd2e5f8 AB |
122 | kmem_cache_free(anon_vma_cachep, anon_vma); |
123 | } | |
1da177e4 | 124 | |
dd34739c | 125 | static inline struct anon_vma_chain *anon_vma_chain_alloc(gfp_t gfp) |
5beb4930 | 126 | { |
dd34739c | 127 | return kmem_cache_alloc(anon_vma_chain_cachep, gfp); |
5beb4930 RR |
128 | } |
129 | ||
e574b5fd | 130 | static void anon_vma_chain_free(struct anon_vma_chain *anon_vma_chain) |
5beb4930 RR |
131 | { |
132 | kmem_cache_free(anon_vma_chain_cachep, anon_vma_chain); | |
133 | } | |
134 | ||
6583a843 KC |
135 | static void anon_vma_chain_link(struct vm_area_struct *vma, |
136 | struct anon_vma_chain *avc, | |
137 | struct anon_vma *anon_vma) | |
138 | { | |
139 | avc->vma = vma; | |
140 | avc->anon_vma = anon_vma; | |
141 | list_add(&avc->same_vma, &vma->anon_vma_chain); | |
bf181b9f | 142 | anon_vma_interval_tree_insert(avc, &anon_vma->rb_root); |
6583a843 KC |
143 | } |
144 | ||
d9d332e0 | 145 | /** |
d5a187da | 146 | * __anon_vma_prepare - attach an anon_vma to a memory region |
d9d332e0 LT |
147 | * @vma: the memory region in question |
148 | * | |
149 | * This makes sure the memory mapping described by 'vma' has | |
150 | * an 'anon_vma' attached to it, so that we can associate the | |
151 | * anonymous pages mapped into it with that anon_vma. | |
152 | * | |
d5a187da VB |
153 | * The common case will be that we already have one, which |
154 | * is handled inline by anon_vma_prepare(). But if | |
23a0790a | 155 | * not we either need to find an adjacent mapping that we |
d9d332e0 LT |
156 | * can re-use the anon_vma from (very common when the only |
157 | * reason for splitting a vma has been mprotect()), or we | |
158 | * allocate a new one. | |
159 | * | |
160 | * Anon-vma allocations are very subtle, because we may have | |
4fc3f1d6 | 161 | * optimistically looked up an anon_vma in page_lock_anon_vma_read() |
d9d332e0 LT |
162 | * and that may actually touch the spinlock even in the newly |
163 | * allocated vma (it depends on RCU to make sure that the | |
164 | * anon_vma isn't actually destroyed). | |
165 | * | |
166 | * As a result, we need to do proper anon_vma locking even | |
167 | * for the new allocation. At the same time, we do not want | |
168 | * to do any locking for the common case of already having | |
169 | * an anon_vma. | |
170 | * | |
171 | * This must be called with the mmap_sem held for reading. | |
172 | */ | |
d5a187da | 173 | int __anon_vma_prepare(struct vm_area_struct *vma) |
1da177e4 | 174 | { |
d5a187da VB |
175 | struct mm_struct *mm = vma->vm_mm; |
176 | struct anon_vma *anon_vma, *allocated; | |
5beb4930 | 177 | struct anon_vma_chain *avc; |
1da177e4 LT |
178 | |
179 | might_sleep(); | |
1da177e4 | 180 | |
d5a187da VB |
181 | avc = anon_vma_chain_alloc(GFP_KERNEL); |
182 | if (!avc) | |
183 | goto out_enomem; | |
184 | ||
185 | anon_vma = find_mergeable_anon_vma(vma); | |
186 | allocated = NULL; | |
187 | if (!anon_vma) { | |
188 | anon_vma = anon_vma_alloc(); | |
189 | if (unlikely(!anon_vma)) | |
190 | goto out_enomem_free_avc; | |
191 | allocated = anon_vma; | |
192 | } | |
5beb4930 | 193 | |
d5a187da VB |
194 | anon_vma_lock_write(anon_vma); |
195 | /* page_table_lock to protect against threads */ | |
196 | spin_lock(&mm->page_table_lock); | |
197 | if (likely(!vma->anon_vma)) { | |
198 | vma->anon_vma = anon_vma; | |
199 | anon_vma_chain_link(vma, avc, anon_vma); | |
200 | /* vma reference or self-parent link for new root */ | |
201 | anon_vma->degree++; | |
d9d332e0 | 202 | allocated = NULL; |
d5a187da VB |
203 | avc = NULL; |
204 | } | |
205 | spin_unlock(&mm->page_table_lock); | |
206 | anon_vma_unlock_write(anon_vma); | |
1da177e4 | 207 | |
d5a187da VB |
208 | if (unlikely(allocated)) |
209 | put_anon_vma(allocated); | |
210 | if (unlikely(avc)) | |
211 | anon_vma_chain_free(avc); | |
31f2b0eb | 212 | |
1da177e4 | 213 | return 0; |
5beb4930 RR |
214 | |
215 | out_enomem_free_avc: | |
216 | anon_vma_chain_free(avc); | |
217 | out_enomem: | |
218 | return -ENOMEM; | |
1da177e4 LT |
219 | } |
220 | ||
bb4aa396 LT |
221 | /* |
222 | * This is a useful helper function for locking the anon_vma root as | |
223 | * we traverse the vma->anon_vma_chain, looping over anon_vma's that | |
224 | * have the same vma. | |
225 | * | |
226 | * Such anon_vma's should have the same root, so you'd expect to see | |
227 | * just a single mutex_lock for the whole traversal. | |
228 | */ | |
229 | static inline struct anon_vma *lock_anon_vma_root(struct anon_vma *root, struct anon_vma *anon_vma) | |
230 | { | |
231 | struct anon_vma *new_root = anon_vma->root; | |
232 | if (new_root != root) { | |
233 | if (WARN_ON_ONCE(root)) | |
5a505085 | 234 | up_write(&root->rwsem); |
bb4aa396 | 235 | root = new_root; |
5a505085 | 236 | down_write(&root->rwsem); |
bb4aa396 LT |
237 | } |
238 | return root; | |
239 | } | |
240 | ||
241 | static inline void unlock_anon_vma_root(struct anon_vma *root) | |
242 | { | |
243 | if (root) | |
5a505085 | 244 | up_write(&root->rwsem); |
bb4aa396 LT |
245 | } |
246 | ||
5beb4930 RR |
247 | /* |
248 | * Attach the anon_vmas from src to dst. | |
249 | * Returns 0 on success, -ENOMEM on failure. | |
7a3ef208 KK |
250 | * |
251 | * If dst->anon_vma is NULL this function tries to find and reuse existing | |
252 | * anon_vma which has no vmas and only one child anon_vma. This prevents | |
253 | * degradation of anon_vma hierarchy to endless linear chain in case of | |
254 | * constantly forking task. On the other hand, an anon_vma with more than one | |
255 | * child isn't reused even if there was no alive vma, thus rmap walker has a | |
256 | * good chance of avoiding scanning the whole hierarchy when it searches where | |
257 | * page is mapped. | |
5beb4930 RR |
258 | */ |
259 | int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src) | |
1da177e4 | 260 | { |
5beb4930 | 261 | struct anon_vma_chain *avc, *pavc; |
bb4aa396 | 262 | struct anon_vma *root = NULL; |
5beb4930 | 263 | |
646d87b4 | 264 | list_for_each_entry_reverse(pavc, &src->anon_vma_chain, same_vma) { |
bb4aa396 LT |
265 | struct anon_vma *anon_vma; |
266 | ||
dd34739c LT |
267 | avc = anon_vma_chain_alloc(GFP_NOWAIT | __GFP_NOWARN); |
268 | if (unlikely(!avc)) { | |
269 | unlock_anon_vma_root(root); | |
270 | root = NULL; | |
271 | avc = anon_vma_chain_alloc(GFP_KERNEL); | |
272 | if (!avc) | |
273 | goto enomem_failure; | |
274 | } | |
bb4aa396 LT |
275 | anon_vma = pavc->anon_vma; |
276 | root = lock_anon_vma_root(root, anon_vma); | |
277 | anon_vma_chain_link(dst, avc, anon_vma); | |
7a3ef208 KK |
278 | |
279 | /* | |
280 | * Reuse existing anon_vma if its degree lower than two, | |
281 | * that means it has no vma and only one anon_vma child. | |
282 | * | |
283 | * Do not chose parent anon_vma, otherwise first child | |
284 | * will always reuse it. Root anon_vma is never reused: | |
285 | * it has self-parent reference and at least one child. | |
286 | */ | |
287 | if (!dst->anon_vma && anon_vma != src->anon_vma && | |
288 | anon_vma->degree < 2) | |
289 | dst->anon_vma = anon_vma; | |
5beb4930 | 290 | } |
7a3ef208 KK |
291 | if (dst->anon_vma) |
292 | dst->anon_vma->degree++; | |
bb4aa396 | 293 | unlock_anon_vma_root(root); |
5beb4930 | 294 | return 0; |
1da177e4 | 295 | |
5beb4930 | 296 | enomem_failure: |
3fe89b3e LY |
297 | /* |
298 | * dst->anon_vma is dropped here otherwise its degree can be incorrectly | |
299 | * decremented in unlink_anon_vmas(). | |
300 | * We can safely do this because callers of anon_vma_clone() don't care | |
301 | * about dst->anon_vma if anon_vma_clone() failed. | |
302 | */ | |
303 | dst->anon_vma = NULL; | |
5beb4930 RR |
304 | unlink_anon_vmas(dst); |
305 | return -ENOMEM; | |
1da177e4 LT |
306 | } |
307 | ||
5beb4930 RR |
308 | /* |
309 | * Attach vma to its own anon_vma, as well as to the anon_vmas that | |
310 | * the corresponding VMA in the parent process is attached to. | |
311 | * Returns 0 on success, non-zero on failure. | |
312 | */ | |
313 | int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma) | |
1da177e4 | 314 | { |
5beb4930 RR |
315 | struct anon_vma_chain *avc; |
316 | struct anon_vma *anon_vma; | |
c4ea95d7 | 317 | int error; |
1da177e4 | 318 | |
5beb4930 RR |
319 | /* Don't bother if the parent process has no anon_vma here. */ |
320 | if (!pvma->anon_vma) | |
321 | return 0; | |
322 | ||
7a3ef208 KK |
323 | /* Drop inherited anon_vma, we'll reuse existing or allocate new. */ |
324 | vma->anon_vma = NULL; | |
325 | ||
5beb4930 RR |
326 | /* |
327 | * First, attach the new VMA to the parent VMA's anon_vmas, | |
328 | * so rmap can find non-COWed pages in child processes. | |
329 | */ | |
c4ea95d7 DF |
330 | error = anon_vma_clone(vma, pvma); |
331 | if (error) | |
332 | return error; | |
5beb4930 | 333 | |
7a3ef208 KK |
334 | /* An existing anon_vma has been reused, all done then. */ |
335 | if (vma->anon_vma) | |
336 | return 0; | |
337 | ||
5beb4930 RR |
338 | /* Then add our own anon_vma. */ |
339 | anon_vma = anon_vma_alloc(); | |
340 | if (!anon_vma) | |
341 | goto out_error; | |
dd34739c | 342 | avc = anon_vma_chain_alloc(GFP_KERNEL); |
5beb4930 RR |
343 | if (!avc) |
344 | goto out_error_free_anon_vma; | |
5c341ee1 RR |
345 | |
346 | /* | |
347 | * The root anon_vma's spinlock is the lock actually used when we | |
348 | * lock any of the anon_vmas in this anon_vma tree. | |
349 | */ | |
350 | anon_vma->root = pvma->anon_vma->root; | |
7a3ef208 | 351 | anon_vma->parent = pvma->anon_vma; |
76545066 | 352 | /* |
01d8b20d PZ |
353 | * With refcounts, an anon_vma can stay around longer than the |
354 | * process it belongs to. The root anon_vma needs to be pinned until | |
355 | * this anon_vma is freed, because the lock lives in the root. | |
76545066 RR |
356 | */ |
357 | get_anon_vma(anon_vma->root); | |
5beb4930 RR |
358 | /* Mark this anon_vma as the one where our new (COWed) pages go. */ |
359 | vma->anon_vma = anon_vma; | |
4fc3f1d6 | 360 | anon_vma_lock_write(anon_vma); |
5c341ee1 | 361 | anon_vma_chain_link(vma, avc, anon_vma); |
7a3ef208 | 362 | anon_vma->parent->degree++; |
08b52706 | 363 | anon_vma_unlock_write(anon_vma); |
5beb4930 RR |
364 | |
365 | return 0; | |
366 | ||
367 | out_error_free_anon_vma: | |
01d8b20d | 368 | put_anon_vma(anon_vma); |
5beb4930 | 369 | out_error: |
4946d54c | 370 | unlink_anon_vmas(vma); |
5beb4930 | 371 | return -ENOMEM; |
1da177e4 LT |
372 | } |
373 | ||
5beb4930 RR |
374 | void unlink_anon_vmas(struct vm_area_struct *vma) |
375 | { | |
376 | struct anon_vma_chain *avc, *next; | |
eee2acba | 377 | struct anon_vma *root = NULL; |
5beb4930 | 378 | |
5c341ee1 RR |
379 | /* |
380 | * Unlink each anon_vma chained to the VMA. This list is ordered | |
381 | * from newest to oldest, ensuring the root anon_vma gets freed last. | |
382 | */ | |
5beb4930 | 383 | list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) { |
eee2acba PZ |
384 | struct anon_vma *anon_vma = avc->anon_vma; |
385 | ||
386 | root = lock_anon_vma_root(root, anon_vma); | |
bf181b9f | 387 | anon_vma_interval_tree_remove(avc, &anon_vma->rb_root); |
eee2acba PZ |
388 | |
389 | /* | |
390 | * Leave empty anon_vmas on the list - we'll need | |
391 | * to free them outside the lock. | |
392 | */ | |
7a3ef208 KK |
393 | if (RB_EMPTY_ROOT(&anon_vma->rb_root)) { |
394 | anon_vma->parent->degree--; | |
eee2acba | 395 | continue; |
7a3ef208 | 396 | } |
eee2acba PZ |
397 | |
398 | list_del(&avc->same_vma); | |
399 | anon_vma_chain_free(avc); | |
400 | } | |
7a3ef208 KK |
401 | if (vma->anon_vma) |
402 | vma->anon_vma->degree--; | |
eee2acba PZ |
403 | unlock_anon_vma_root(root); |
404 | ||
405 | /* | |
406 | * Iterate the list once more, it now only contains empty and unlinked | |
407 | * anon_vmas, destroy them. Could not do before due to __put_anon_vma() | |
5a505085 | 408 | * needing to write-acquire the anon_vma->root->rwsem. |
eee2acba PZ |
409 | */ |
410 | list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) { | |
411 | struct anon_vma *anon_vma = avc->anon_vma; | |
412 | ||
e4c5800a | 413 | VM_WARN_ON(anon_vma->degree); |
eee2acba PZ |
414 | put_anon_vma(anon_vma); |
415 | ||
5beb4930 RR |
416 | list_del(&avc->same_vma); |
417 | anon_vma_chain_free(avc); | |
418 | } | |
419 | } | |
420 | ||
51cc5068 | 421 | static void anon_vma_ctor(void *data) |
1da177e4 | 422 | { |
a35afb83 | 423 | struct anon_vma *anon_vma = data; |
1da177e4 | 424 | |
5a505085 | 425 | init_rwsem(&anon_vma->rwsem); |
83813267 | 426 | atomic_set(&anon_vma->refcount, 0); |
bf181b9f | 427 | anon_vma->rb_root = RB_ROOT; |
1da177e4 LT |
428 | } |
429 | ||
430 | void __init anon_vma_init(void) | |
431 | { | |
432 | anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma), | |
5d097056 VD |
433 | 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC|SLAB_ACCOUNT, |
434 | anon_vma_ctor); | |
435 | anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain, | |
436 | SLAB_PANIC|SLAB_ACCOUNT); | |
1da177e4 LT |
437 | } |
438 | ||
439 | /* | |
6111e4ca PZ |
440 | * Getting a lock on a stable anon_vma from a page off the LRU is tricky! |
441 | * | |
442 | * Since there is no serialization what so ever against page_remove_rmap() | |
443 | * the best this function can do is return a locked anon_vma that might | |
444 | * have been relevant to this page. | |
445 | * | |
446 | * The page might have been remapped to a different anon_vma or the anon_vma | |
447 | * returned may already be freed (and even reused). | |
448 | * | |
bc658c96 PZ |
449 | * In case it was remapped to a different anon_vma, the new anon_vma will be a |
450 | * child of the old anon_vma, and the anon_vma lifetime rules will therefore | |
451 | * ensure that any anon_vma obtained from the page will still be valid for as | |
452 | * long as we observe page_mapped() [ hence all those page_mapped() tests ]. | |
453 | * | |
6111e4ca PZ |
454 | * All users of this function must be very careful when walking the anon_vma |
455 | * chain and verify that the page in question is indeed mapped in it | |
456 | * [ something equivalent to page_mapped_in_vma() ]. | |
457 | * | |
458 | * Since anon_vma's slab is DESTROY_BY_RCU and we know from page_remove_rmap() | |
459 | * that the anon_vma pointer from page->mapping is valid if there is a | |
460 | * mapcount, we can dereference the anon_vma after observing those. | |
1da177e4 | 461 | */ |
746b18d4 | 462 | struct anon_vma *page_get_anon_vma(struct page *page) |
1da177e4 | 463 | { |
746b18d4 | 464 | struct anon_vma *anon_vma = NULL; |
1da177e4 LT |
465 | unsigned long anon_mapping; |
466 | ||
467 | rcu_read_lock(); | |
4db0c3c2 | 468 | anon_mapping = (unsigned long)READ_ONCE(page->mapping); |
3ca7b3c5 | 469 | if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON) |
1da177e4 LT |
470 | goto out; |
471 | if (!page_mapped(page)) | |
472 | goto out; | |
473 | ||
474 | anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON); | |
746b18d4 PZ |
475 | if (!atomic_inc_not_zero(&anon_vma->refcount)) { |
476 | anon_vma = NULL; | |
477 | goto out; | |
478 | } | |
f1819427 HD |
479 | |
480 | /* | |
481 | * If this page is still mapped, then its anon_vma cannot have been | |
746b18d4 PZ |
482 | * freed. But if it has been unmapped, we have no security against the |
483 | * anon_vma structure being freed and reused (for another anon_vma: | |
484 | * SLAB_DESTROY_BY_RCU guarantees that - so the atomic_inc_not_zero() | |
485 | * above cannot corrupt). | |
f1819427 | 486 | */ |
746b18d4 | 487 | if (!page_mapped(page)) { |
7f39dda9 | 488 | rcu_read_unlock(); |
746b18d4 | 489 | put_anon_vma(anon_vma); |
7f39dda9 | 490 | return NULL; |
746b18d4 | 491 | } |
1da177e4 LT |
492 | out: |
493 | rcu_read_unlock(); | |
746b18d4 PZ |
494 | |
495 | return anon_vma; | |
496 | } | |
497 | ||
88c22088 PZ |
498 | /* |
499 | * Similar to page_get_anon_vma() except it locks the anon_vma. | |
500 | * | |
501 | * Its a little more complex as it tries to keep the fast path to a single | |
502 | * atomic op -- the trylock. If we fail the trylock, we fall back to getting a | |
503 | * reference like with page_get_anon_vma() and then block on the mutex. | |
504 | */ | |
4fc3f1d6 | 505 | struct anon_vma *page_lock_anon_vma_read(struct page *page) |
746b18d4 | 506 | { |
88c22088 | 507 | struct anon_vma *anon_vma = NULL; |
eee0f252 | 508 | struct anon_vma *root_anon_vma; |
88c22088 | 509 | unsigned long anon_mapping; |
746b18d4 | 510 | |
88c22088 | 511 | rcu_read_lock(); |
4db0c3c2 | 512 | anon_mapping = (unsigned long)READ_ONCE(page->mapping); |
88c22088 PZ |
513 | if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON) |
514 | goto out; | |
515 | if (!page_mapped(page)) | |
516 | goto out; | |
517 | ||
518 | anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON); | |
4db0c3c2 | 519 | root_anon_vma = READ_ONCE(anon_vma->root); |
4fc3f1d6 | 520 | if (down_read_trylock(&root_anon_vma->rwsem)) { |
88c22088 | 521 | /* |
eee0f252 HD |
522 | * If the page is still mapped, then this anon_vma is still |
523 | * its anon_vma, and holding the mutex ensures that it will | |
bc658c96 | 524 | * not go away, see anon_vma_free(). |
88c22088 | 525 | */ |
eee0f252 | 526 | if (!page_mapped(page)) { |
4fc3f1d6 | 527 | up_read(&root_anon_vma->rwsem); |
88c22088 PZ |
528 | anon_vma = NULL; |
529 | } | |
530 | goto out; | |
531 | } | |
746b18d4 | 532 | |
88c22088 PZ |
533 | /* trylock failed, we got to sleep */ |
534 | if (!atomic_inc_not_zero(&anon_vma->refcount)) { | |
535 | anon_vma = NULL; | |
536 | goto out; | |
537 | } | |
538 | ||
539 | if (!page_mapped(page)) { | |
7f39dda9 | 540 | rcu_read_unlock(); |
88c22088 | 541 | put_anon_vma(anon_vma); |
7f39dda9 | 542 | return NULL; |
88c22088 PZ |
543 | } |
544 | ||
545 | /* we pinned the anon_vma, its safe to sleep */ | |
546 | rcu_read_unlock(); | |
4fc3f1d6 | 547 | anon_vma_lock_read(anon_vma); |
88c22088 PZ |
548 | |
549 | if (atomic_dec_and_test(&anon_vma->refcount)) { | |
550 | /* | |
551 | * Oops, we held the last refcount, release the lock | |
552 | * and bail -- can't simply use put_anon_vma() because | |
4fc3f1d6 | 553 | * we'll deadlock on the anon_vma_lock_write() recursion. |
88c22088 | 554 | */ |
4fc3f1d6 | 555 | anon_vma_unlock_read(anon_vma); |
88c22088 PZ |
556 | __put_anon_vma(anon_vma); |
557 | anon_vma = NULL; | |
558 | } | |
559 | ||
560 | return anon_vma; | |
561 | ||
562 | out: | |
563 | rcu_read_unlock(); | |
746b18d4 | 564 | return anon_vma; |
34bbd704 ON |
565 | } |
566 | ||
4fc3f1d6 | 567 | void page_unlock_anon_vma_read(struct anon_vma *anon_vma) |
34bbd704 | 568 | { |
4fc3f1d6 | 569 | anon_vma_unlock_read(anon_vma); |
1da177e4 LT |
570 | } |
571 | ||
72b252ae | 572 | #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH |
72b252ae MG |
573 | /* |
574 | * Flush TLB entries for recently unmapped pages from remote CPUs. It is | |
575 | * important if a PTE was dirty when it was unmapped that it's flushed | |
576 | * before any IO is initiated on the page to prevent lost writes. Similarly, | |
577 | * it must be flushed before freeing to prevent data leakage. | |
578 | */ | |
579 | void try_to_unmap_flush(void) | |
580 | { | |
581 | struct tlbflush_unmap_batch *tlb_ubc = ¤t->tlb_ubc; | |
582 | int cpu; | |
583 | ||
584 | if (!tlb_ubc->flush_required) | |
585 | return; | |
586 | ||
587 | cpu = get_cpu(); | |
588 | ||
858eaaa7 NA |
589 | if (cpumask_test_cpu(cpu, &tlb_ubc->cpumask)) { |
590 | count_vm_tlb_event(NR_TLB_LOCAL_FLUSH_ALL); | |
591 | local_flush_tlb(); | |
592 | trace_tlb_flush(TLB_LOCAL_SHOOTDOWN, TLB_FLUSH_ALL); | |
72b252ae | 593 | } |
858eaaa7 NA |
594 | |
595 | if (cpumask_any_but(&tlb_ubc->cpumask, cpu) < nr_cpu_ids) | |
596 | flush_tlb_others(&tlb_ubc->cpumask, NULL, 0, TLB_FLUSH_ALL); | |
72b252ae MG |
597 | cpumask_clear(&tlb_ubc->cpumask); |
598 | tlb_ubc->flush_required = false; | |
d950c947 | 599 | tlb_ubc->writable = false; |
72b252ae MG |
600 | put_cpu(); |
601 | } | |
602 | ||
d950c947 MG |
603 | /* Flush iff there are potentially writable TLB entries that can race with IO */ |
604 | void try_to_unmap_flush_dirty(void) | |
605 | { | |
606 | struct tlbflush_unmap_batch *tlb_ubc = ¤t->tlb_ubc; | |
607 | ||
608 | if (tlb_ubc->writable) | |
609 | try_to_unmap_flush(); | |
610 | } | |
611 | ||
c7ab0d2f | 612 | static void set_tlb_ubc_flush_pending(struct mm_struct *mm, bool writable) |
72b252ae MG |
613 | { |
614 | struct tlbflush_unmap_batch *tlb_ubc = ¤t->tlb_ubc; | |
615 | ||
616 | cpumask_or(&tlb_ubc->cpumask, &tlb_ubc->cpumask, mm_cpumask(mm)); | |
617 | tlb_ubc->flush_required = true; | |
d950c947 MG |
618 | |
619 | /* | |
620 | * If the PTE was dirty then it's best to assume it's writable. The | |
621 | * caller must use try_to_unmap_flush_dirty() or try_to_unmap_flush() | |
622 | * before the page is queued for IO. | |
623 | */ | |
624 | if (writable) | |
625 | tlb_ubc->writable = true; | |
72b252ae MG |
626 | } |
627 | ||
628 | /* | |
629 | * Returns true if the TLB flush should be deferred to the end of a batch of | |
630 | * unmap operations to reduce IPIs. | |
631 | */ | |
632 | static bool should_defer_flush(struct mm_struct *mm, enum ttu_flags flags) | |
633 | { | |
634 | bool should_defer = false; | |
635 | ||
636 | if (!(flags & TTU_BATCH_FLUSH)) | |
637 | return false; | |
638 | ||
639 | /* If remote CPUs need to be flushed then defer batch the flush */ | |
640 | if (cpumask_any_but(mm_cpumask(mm), get_cpu()) < nr_cpu_ids) | |
641 | should_defer = true; | |
642 | put_cpu(); | |
643 | ||
644 | return should_defer; | |
645 | } | |
646 | #else | |
c7ab0d2f | 647 | static void set_tlb_ubc_flush_pending(struct mm_struct *mm, bool writable) |
72b252ae MG |
648 | { |
649 | } | |
650 | ||
651 | static bool should_defer_flush(struct mm_struct *mm, enum ttu_flags flags) | |
652 | { | |
653 | return false; | |
654 | } | |
655 | #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */ | |
656 | ||
1da177e4 | 657 | /* |
bf89c8c8 | 658 | * At what user virtual address is page expected in vma? |
ab941e0f | 659 | * Caller should check the page is actually part of the vma. |
1da177e4 LT |
660 | */ |
661 | unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma) | |
662 | { | |
86c2ad19 | 663 | unsigned long address; |
21d0d443 | 664 | if (PageAnon(page)) { |
4829b906 HD |
665 | struct anon_vma *page__anon_vma = page_anon_vma(page); |
666 | /* | |
667 | * Note: swapoff's unuse_vma() is more efficient with this | |
668 | * check, and needs it to match anon_vma when KSM is active. | |
669 | */ | |
670 | if (!vma->anon_vma || !page__anon_vma || | |
671 | vma->anon_vma->root != page__anon_vma->root) | |
21d0d443 | 672 | return -EFAULT; |
27ba0644 KS |
673 | } else if (page->mapping) { |
674 | if (!vma->vm_file || vma->vm_file->f_mapping != page->mapping) | |
1da177e4 LT |
675 | return -EFAULT; |
676 | } else | |
677 | return -EFAULT; | |
86c2ad19 ML |
678 | address = __vma_address(page, vma); |
679 | if (unlikely(address < vma->vm_start || address >= vma->vm_end)) | |
680 | return -EFAULT; | |
681 | return address; | |
1da177e4 LT |
682 | } |
683 | ||
6219049a BL |
684 | pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address) |
685 | { | |
686 | pgd_t *pgd; | |
687 | pud_t *pud; | |
688 | pmd_t *pmd = NULL; | |
f72e7dcd | 689 | pmd_t pmde; |
6219049a BL |
690 | |
691 | pgd = pgd_offset(mm, address); | |
692 | if (!pgd_present(*pgd)) | |
693 | goto out; | |
694 | ||
695 | pud = pud_offset(pgd, address); | |
696 | if (!pud_present(*pud)) | |
697 | goto out; | |
698 | ||
699 | pmd = pmd_offset(pud, address); | |
f72e7dcd | 700 | /* |
8809aa2d | 701 | * Some THP functions use the sequence pmdp_huge_clear_flush(), set_pmd_at() |
f72e7dcd HD |
702 | * without holding anon_vma lock for write. So when looking for a |
703 | * genuine pmde (in which to find pte), test present and !THP together. | |
704 | */ | |
e37c6982 CB |
705 | pmde = *pmd; |
706 | barrier(); | |
f72e7dcd | 707 | if (!pmd_present(pmde) || pmd_trans_huge(pmde)) |
6219049a BL |
708 | pmd = NULL; |
709 | out: | |
710 | return pmd; | |
711 | } | |
712 | ||
8749cfea VD |
713 | struct page_referenced_arg { |
714 | int mapcount; | |
715 | int referenced; | |
716 | unsigned long vm_flags; | |
717 | struct mem_cgroup *memcg; | |
718 | }; | |
719 | /* | |
720 | * arg: page_referenced_arg will be passed | |
721 | */ | |
722 | static int page_referenced_one(struct page *page, struct vm_area_struct *vma, | |
723 | unsigned long address, void *arg) | |
724 | { | |
8749cfea | 725 | struct page_referenced_arg *pra = arg; |
8eaedede KS |
726 | struct page_vma_mapped_walk pvmw = { |
727 | .page = page, | |
728 | .vma = vma, | |
729 | .address = address, | |
730 | }; | |
8749cfea VD |
731 | int referenced = 0; |
732 | ||
8eaedede KS |
733 | while (page_vma_mapped_walk(&pvmw)) { |
734 | address = pvmw.address; | |
b20ce5e0 | 735 | |
8eaedede KS |
736 | if (vma->vm_flags & VM_LOCKED) { |
737 | page_vma_mapped_walk_done(&pvmw); | |
738 | pra->vm_flags |= VM_LOCKED; | |
739 | return SWAP_FAIL; /* To break the loop */ | |
740 | } | |
71e3aac0 | 741 | |
8eaedede KS |
742 | if (pvmw.pte) { |
743 | if (ptep_clear_flush_young_notify(vma, address, | |
744 | pvmw.pte)) { | |
745 | /* | |
746 | * Don't treat a reference through | |
747 | * a sequentially read mapping as such. | |
748 | * If the page has been used in another mapping, | |
749 | * we will catch it; if this other mapping is | |
750 | * already gone, the unmap path will have set | |
751 | * PG_referenced or activated the page. | |
752 | */ | |
753 | if (likely(!(vma->vm_flags & VM_SEQ_READ))) | |
754 | referenced++; | |
755 | } | |
756 | } else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) { | |
757 | if (pmdp_clear_flush_young_notify(vma, address, | |
758 | pvmw.pmd)) | |
8749cfea | 759 | referenced++; |
8eaedede KS |
760 | } else { |
761 | /* unexpected pmd-mapped page? */ | |
762 | WARN_ON_ONCE(1); | |
8749cfea | 763 | } |
8eaedede KS |
764 | |
765 | pra->mapcount--; | |
b20ce5e0 | 766 | } |
b20ce5e0 | 767 | |
33c3fc71 VD |
768 | if (referenced) |
769 | clear_page_idle(page); | |
770 | if (test_and_clear_page_young(page)) | |
771 | referenced++; | |
772 | ||
9f32624b JK |
773 | if (referenced) { |
774 | pra->referenced++; | |
775 | pra->vm_flags |= vma->vm_flags; | |
1da177e4 | 776 | } |
34bbd704 | 777 | |
9f32624b JK |
778 | if (!pra->mapcount) |
779 | return SWAP_SUCCESS; /* To break the loop */ | |
780 | ||
781 | return SWAP_AGAIN; | |
1da177e4 LT |
782 | } |
783 | ||
9f32624b | 784 | static bool invalid_page_referenced_vma(struct vm_area_struct *vma, void *arg) |
1da177e4 | 785 | { |
9f32624b JK |
786 | struct page_referenced_arg *pra = arg; |
787 | struct mem_cgroup *memcg = pra->memcg; | |
1da177e4 | 788 | |
9f32624b JK |
789 | if (!mm_match_cgroup(vma->vm_mm, memcg)) |
790 | return true; | |
1da177e4 | 791 | |
9f32624b | 792 | return false; |
1da177e4 LT |
793 | } |
794 | ||
795 | /** | |
796 | * page_referenced - test if the page was referenced | |
797 | * @page: the page to test | |
798 | * @is_locked: caller holds lock on the page | |
72835c86 | 799 | * @memcg: target memory cgroup |
6fe6b7e3 | 800 | * @vm_flags: collect encountered vma->vm_flags who actually referenced the page |
1da177e4 LT |
801 | * |
802 | * Quick test_and_clear_referenced for all mappings to a page, | |
803 | * returns the number of ptes which referenced the page. | |
804 | */ | |
6fe6b7e3 WF |
805 | int page_referenced(struct page *page, |
806 | int is_locked, | |
72835c86 | 807 | struct mem_cgroup *memcg, |
6fe6b7e3 | 808 | unsigned long *vm_flags) |
1da177e4 | 809 | { |
9f32624b | 810 | int ret; |
5ad64688 | 811 | int we_locked = 0; |
9f32624b | 812 | struct page_referenced_arg pra = { |
b20ce5e0 | 813 | .mapcount = total_mapcount(page), |
9f32624b JK |
814 | .memcg = memcg, |
815 | }; | |
816 | struct rmap_walk_control rwc = { | |
817 | .rmap_one = page_referenced_one, | |
818 | .arg = (void *)&pra, | |
819 | .anon_lock = page_lock_anon_vma_read, | |
820 | }; | |
1da177e4 | 821 | |
6fe6b7e3 | 822 | *vm_flags = 0; |
9f32624b JK |
823 | if (!page_mapped(page)) |
824 | return 0; | |
825 | ||
826 | if (!page_rmapping(page)) | |
827 | return 0; | |
828 | ||
829 | if (!is_locked && (!PageAnon(page) || PageKsm(page))) { | |
830 | we_locked = trylock_page(page); | |
831 | if (!we_locked) | |
832 | return 1; | |
1da177e4 | 833 | } |
9f32624b JK |
834 | |
835 | /* | |
836 | * If we are reclaiming on behalf of a cgroup, skip | |
837 | * counting on behalf of references from different | |
838 | * cgroups | |
839 | */ | |
840 | if (memcg) { | |
841 | rwc.invalid_vma = invalid_page_referenced_vma; | |
842 | } | |
843 | ||
844 | ret = rmap_walk(page, &rwc); | |
845 | *vm_flags = pra.vm_flags; | |
846 | ||
847 | if (we_locked) | |
848 | unlock_page(page); | |
849 | ||
850 | return pra.referenced; | |
1da177e4 LT |
851 | } |
852 | ||
1cb1729b | 853 | static int page_mkclean_one(struct page *page, struct vm_area_struct *vma, |
9853a407 | 854 | unsigned long address, void *arg) |
d08b3851 | 855 | { |
f27176cf KS |
856 | struct page_vma_mapped_walk pvmw = { |
857 | .page = page, | |
858 | .vma = vma, | |
859 | .address = address, | |
860 | .flags = PVMW_SYNC, | |
861 | }; | |
9853a407 | 862 | int *cleaned = arg; |
d08b3851 | 863 | |
f27176cf KS |
864 | while (page_vma_mapped_walk(&pvmw)) { |
865 | int ret = 0; | |
866 | address = pvmw.address; | |
867 | if (pvmw.pte) { | |
868 | pte_t entry; | |
869 | pte_t *pte = pvmw.pte; | |
870 | ||
871 | if (!pte_dirty(*pte) && !pte_write(*pte)) | |
872 | continue; | |
873 | ||
874 | flush_cache_page(vma, address, pte_pfn(*pte)); | |
875 | entry = ptep_clear_flush(vma, address, pte); | |
876 | entry = pte_wrprotect(entry); | |
877 | entry = pte_mkclean(entry); | |
878 | set_pte_at(vma->vm_mm, address, pte, entry); | |
879 | ret = 1; | |
880 | } else { | |
881 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE | |
882 | pmd_t *pmd = pvmw.pmd; | |
883 | pmd_t entry; | |
884 | ||
885 | if (!pmd_dirty(*pmd) && !pmd_write(*pmd)) | |
886 | continue; | |
887 | ||
888 | flush_cache_page(vma, address, page_to_pfn(page)); | |
889 | entry = pmdp_huge_clear_flush(vma, address, pmd); | |
890 | entry = pmd_wrprotect(entry); | |
891 | entry = pmd_mkclean(entry); | |
892 | set_pmd_at(vma->vm_mm, address, pmd, entry); | |
893 | ret = 1; | |
894 | #else | |
895 | /* unexpected pmd-mapped page? */ | |
896 | WARN_ON_ONCE(1); | |
897 | #endif | |
898 | } | |
d08b3851 | 899 | |
f27176cf KS |
900 | if (ret) { |
901 | mmu_notifier_invalidate_page(vma->vm_mm, address); | |
902 | (*cleaned)++; | |
903 | } | |
c2fda5fe | 904 | } |
d08b3851 | 905 | |
9853a407 | 906 | return SWAP_AGAIN; |
d08b3851 PZ |
907 | } |
908 | ||
9853a407 | 909 | static bool invalid_mkclean_vma(struct vm_area_struct *vma, void *arg) |
d08b3851 | 910 | { |
9853a407 | 911 | if (vma->vm_flags & VM_SHARED) |
871beb8c | 912 | return false; |
d08b3851 | 913 | |
871beb8c | 914 | return true; |
d08b3851 PZ |
915 | } |
916 | ||
917 | int page_mkclean(struct page *page) | |
918 | { | |
9853a407 JK |
919 | int cleaned = 0; |
920 | struct address_space *mapping; | |
921 | struct rmap_walk_control rwc = { | |
922 | .arg = (void *)&cleaned, | |
923 | .rmap_one = page_mkclean_one, | |
924 | .invalid_vma = invalid_mkclean_vma, | |
925 | }; | |
d08b3851 PZ |
926 | |
927 | BUG_ON(!PageLocked(page)); | |
928 | ||
9853a407 JK |
929 | if (!page_mapped(page)) |
930 | return 0; | |
931 | ||
932 | mapping = page_mapping(page); | |
933 | if (!mapping) | |
934 | return 0; | |
935 | ||
936 | rmap_walk(page, &rwc); | |
d08b3851 | 937 | |
9853a407 | 938 | return cleaned; |
d08b3851 | 939 | } |
60b59bea | 940 | EXPORT_SYMBOL_GPL(page_mkclean); |
d08b3851 | 941 | |
c44b6743 RR |
942 | /** |
943 | * page_move_anon_rmap - move a page to our anon_vma | |
944 | * @page: the page to move to our anon_vma | |
945 | * @vma: the vma the page belongs to | |
c44b6743 RR |
946 | * |
947 | * When a page belongs exclusively to one process after a COW event, | |
948 | * that page can be moved into the anon_vma that belongs to just that | |
949 | * process, so the rmap code will not search the parent or sibling | |
950 | * processes. | |
951 | */ | |
5a49973d | 952 | void page_move_anon_rmap(struct page *page, struct vm_area_struct *vma) |
c44b6743 RR |
953 | { |
954 | struct anon_vma *anon_vma = vma->anon_vma; | |
955 | ||
5a49973d HD |
956 | page = compound_head(page); |
957 | ||
309381fe | 958 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
81d1b09c | 959 | VM_BUG_ON_VMA(!anon_vma, vma); |
c44b6743 RR |
960 | |
961 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; | |
414e2fb8 VD |
962 | /* |
963 | * Ensure that anon_vma and the PAGE_MAPPING_ANON bit are written | |
964 | * simultaneously, so a concurrent reader (eg page_referenced()'s | |
965 | * PageAnon()) will not see one without the other. | |
966 | */ | |
967 | WRITE_ONCE(page->mapping, (struct address_space *) anon_vma); | |
c44b6743 RR |
968 | } |
969 | ||
9617d95e | 970 | /** |
4e1c1975 AK |
971 | * __page_set_anon_rmap - set up new anonymous rmap |
972 | * @page: Page to add to rmap | |
973 | * @vma: VM area to add page to. | |
974 | * @address: User virtual address of the mapping | |
e8a03feb | 975 | * @exclusive: the page is exclusively owned by the current process |
9617d95e NP |
976 | */ |
977 | static void __page_set_anon_rmap(struct page *page, | |
e8a03feb | 978 | struct vm_area_struct *vma, unsigned long address, int exclusive) |
9617d95e | 979 | { |
e8a03feb | 980 | struct anon_vma *anon_vma = vma->anon_vma; |
ea90002b | 981 | |
e8a03feb | 982 | BUG_ON(!anon_vma); |
ea90002b | 983 | |
4e1c1975 AK |
984 | if (PageAnon(page)) |
985 | return; | |
986 | ||
ea90002b | 987 | /* |
e8a03feb RR |
988 | * If the page isn't exclusively mapped into this vma, |
989 | * we must use the _oldest_ possible anon_vma for the | |
990 | * page mapping! | |
ea90002b | 991 | */ |
4e1c1975 | 992 | if (!exclusive) |
288468c3 | 993 | anon_vma = anon_vma->root; |
9617d95e | 994 | |
9617d95e NP |
995 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; |
996 | page->mapping = (struct address_space *) anon_vma; | |
9617d95e | 997 | page->index = linear_page_index(vma, address); |
9617d95e NP |
998 | } |
999 | ||
c97a9e10 | 1000 | /** |
43d8eac4 | 1001 | * __page_check_anon_rmap - sanity check anonymous rmap addition |
c97a9e10 NP |
1002 | * @page: the page to add the mapping to |
1003 | * @vma: the vm area in which the mapping is added | |
1004 | * @address: the user virtual address mapped | |
1005 | */ | |
1006 | static void __page_check_anon_rmap(struct page *page, | |
1007 | struct vm_area_struct *vma, unsigned long address) | |
1008 | { | |
1009 | #ifdef CONFIG_DEBUG_VM | |
1010 | /* | |
1011 | * The page's anon-rmap details (mapping and index) are guaranteed to | |
1012 | * be set up correctly at this point. | |
1013 | * | |
1014 | * We have exclusion against page_add_anon_rmap because the caller | |
1015 | * always holds the page locked, except if called from page_dup_rmap, | |
1016 | * in which case the page is already known to be setup. | |
1017 | * | |
1018 | * We have exclusion against page_add_new_anon_rmap because those pages | |
1019 | * are initially only visible via the pagetables, and the pte is locked | |
1020 | * over the call to page_add_new_anon_rmap. | |
1021 | */ | |
44ab57a0 | 1022 | BUG_ON(page_anon_vma(page)->root != vma->anon_vma->root); |
53f9263b | 1023 | BUG_ON(page_to_pgoff(page) != linear_page_index(vma, address)); |
c97a9e10 NP |
1024 | #endif |
1025 | } | |
1026 | ||
1da177e4 LT |
1027 | /** |
1028 | * page_add_anon_rmap - add pte mapping to an anonymous page | |
1029 | * @page: the page to add the mapping to | |
1030 | * @vma: the vm area in which the mapping is added | |
1031 | * @address: the user virtual address mapped | |
d281ee61 | 1032 | * @compound: charge the page as compound or small page |
1da177e4 | 1033 | * |
5ad64688 | 1034 | * The caller needs to hold the pte lock, and the page must be locked in |
80e14822 HD |
1035 | * the anon_vma case: to serialize mapping,index checking after setting, |
1036 | * and to ensure that PageAnon is not being upgraded racily to PageKsm | |
1037 | * (but PageKsm is never downgraded to PageAnon). | |
1da177e4 LT |
1038 | */ |
1039 | void page_add_anon_rmap(struct page *page, | |
d281ee61 | 1040 | struct vm_area_struct *vma, unsigned long address, bool compound) |
ad8c2ee8 | 1041 | { |
d281ee61 | 1042 | do_page_add_anon_rmap(page, vma, address, compound ? RMAP_COMPOUND : 0); |
ad8c2ee8 RR |
1043 | } |
1044 | ||
1045 | /* | |
1046 | * Special version of the above for do_swap_page, which often runs | |
1047 | * into pages that are exclusively owned by the current process. | |
1048 | * Everybody else should continue to use page_add_anon_rmap above. | |
1049 | */ | |
1050 | void do_page_add_anon_rmap(struct page *page, | |
d281ee61 | 1051 | struct vm_area_struct *vma, unsigned long address, int flags) |
1da177e4 | 1052 | { |
53f9263b KS |
1053 | bool compound = flags & RMAP_COMPOUND; |
1054 | bool first; | |
1055 | ||
e9b61f19 KS |
1056 | if (compound) { |
1057 | atomic_t *mapcount; | |
53f9263b | 1058 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
e9b61f19 KS |
1059 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
1060 | mapcount = compound_mapcount_ptr(page); | |
1061 | first = atomic_inc_and_test(mapcount); | |
53f9263b KS |
1062 | } else { |
1063 | first = atomic_inc_and_test(&page->_mapcount); | |
1064 | } | |
1065 | ||
79134171 | 1066 | if (first) { |
d281ee61 | 1067 | int nr = compound ? hpage_nr_pages(page) : 1; |
bea04b07 JZ |
1068 | /* |
1069 | * We use the irq-unsafe __{inc|mod}_zone_page_stat because | |
1070 | * these counters are not modified in interrupt context, and | |
1071 | * pte lock(a spinlock) is held, which implies preemption | |
1072 | * disabled. | |
1073 | */ | |
65c45377 | 1074 | if (compound) |
11fb9989 | 1075 | __inc_node_page_state(page, NR_ANON_THPS); |
4b9d0fab | 1076 | __mod_node_page_state(page_pgdat(page), NR_ANON_MAPPED, nr); |
79134171 | 1077 | } |
5ad64688 HD |
1078 | if (unlikely(PageKsm(page))) |
1079 | return; | |
1080 | ||
309381fe | 1081 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
53f9263b | 1082 | |
5dbe0af4 | 1083 | /* address might be in next vma when migration races vma_adjust */ |
5ad64688 | 1084 | if (first) |
d281ee61 KS |
1085 | __page_set_anon_rmap(page, vma, address, |
1086 | flags & RMAP_EXCLUSIVE); | |
69029cd5 | 1087 | else |
c97a9e10 | 1088 | __page_check_anon_rmap(page, vma, address); |
1da177e4 LT |
1089 | } |
1090 | ||
43d8eac4 | 1091 | /** |
9617d95e NP |
1092 | * page_add_new_anon_rmap - add pte mapping to a new anonymous page |
1093 | * @page: the page to add the mapping to | |
1094 | * @vma: the vm area in which the mapping is added | |
1095 | * @address: the user virtual address mapped | |
d281ee61 | 1096 | * @compound: charge the page as compound or small page |
9617d95e NP |
1097 | * |
1098 | * Same as page_add_anon_rmap but must only be called on *new* pages. | |
1099 | * This means the inc-and-test can be bypassed. | |
c97a9e10 | 1100 | * Page does not have to be locked. |
9617d95e NP |
1101 | */ |
1102 | void page_add_new_anon_rmap(struct page *page, | |
d281ee61 | 1103 | struct vm_area_struct *vma, unsigned long address, bool compound) |
9617d95e | 1104 | { |
d281ee61 KS |
1105 | int nr = compound ? hpage_nr_pages(page) : 1; |
1106 | ||
81d1b09c | 1107 | VM_BUG_ON_VMA(address < vma->vm_start || address >= vma->vm_end, vma); |
fa9949da | 1108 | __SetPageSwapBacked(page); |
d281ee61 KS |
1109 | if (compound) { |
1110 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
53f9263b KS |
1111 | /* increment count (starts at -1) */ |
1112 | atomic_set(compound_mapcount_ptr(page), 0); | |
11fb9989 | 1113 | __inc_node_page_state(page, NR_ANON_THPS); |
53f9263b KS |
1114 | } else { |
1115 | /* Anon THP always mapped first with PMD */ | |
1116 | VM_BUG_ON_PAGE(PageTransCompound(page), page); | |
1117 | /* increment count (starts at -1) */ | |
1118 | atomic_set(&page->_mapcount, 0); | |
d281ee61 | 1119 | } |
4b9d0fab | 1120 | __mod_node_page_state(page_pgdat(page), NR_ANON_MAPPED, nr); |
e8a03feb | 1121 | __page_set_anon_rmap(page, vma, address, 1); |
9617d95e NP |
1122 | } |
1123 | ||
1da177e4 LT |
1124 | /** |
1125 | * page_add_file_rmap - add pte mapping to a file page | |
1126 | * @page: the page to add the mapping to | |
1127 | * | |
b8072f09 | 1128 | * The caller needs to hold the pte lock. |
1da177e4 | 1129 | */ |
dd78fedd | 1130 | void page_add_file_rmap(struct page *page, bool compound) |
1da177e4 | 1131 | { |
dd78fedd KS |
1132 | int i, nr = 1; |
1133 | ||
1134 | VM_BUG_ON_PAGE(compound && !PageTransHuge(page), page); | |
62cccb8c | 1135 | lock_page_memcg(page); |
dd78fedd KS |
1136 | if (compound && PageTransHuge(page)) { |
1137 | for (i = 0, nr = 0; i < HPAGE_PMD_NR; i++) { | |
1138 | if (atomic_inc_and_test(&page[i]._mapcount)) | |
1139 | nr++; | |
1140 | } | |
1141 | if (!atomic_inc_and_test(compound_mapcount_ptr(page))) | |
1142 | goto out; | |
65c45377 | 1143 | VM_BUG_ON_PAGE(!PageSwapBacked(page), page); |
11fb9989 | 1144 | __inc_node_page_state(page, NR_SHMEM_PMDMAPPED); |
dd78fedd | 1145 | } else { |
c8efc390 KS |
1146 | if (PageTransCompound(page) && page_mapping(page)) { |
1147 | VM_WARN_ON_ONCE(!PageLocked(page)); | |
1148 | ||
9a73f61b KS |
1149 | SetPageDoubleMap(compound_head(page)); |
1150 | if (PageMlocked(page)) | |
1151 | clear_page_mlock(compound_head(page)); | |
1152 | } | |
dd78fedd KS |
1153 | if (!atomic_inc_and_test(&page->_mapcount)) |
1154 | goto out; | |
d69b042f | 1155 | } |
50658e2e | 1156 | __mod_node_page_state(page_pgdat(page), NR_FILE_MAPPED, nr); |
dd78fedd KS |
1157 | mem_cgroup_inc_page_stat(page, MEM_CGROUP_STAT_FILE_MAPPED); |
1158 | out: | |
62cccb8c | 1159 | unlock_page_memcg(page); |
1da177e4 LT |
1160 | } |
1161 | ||
dd78fedd | 1162 | static void page_remove_file_rmap(struct page *page, bool compound) |
8186eb6a | 1163 | { |
dd78fedd KS |
1164 | int i, nr = 1; |
1165 | ||
57dea93a | 1166 | VM_BUG_ON_PAGE(compound && !PageHead(page), page); |
62cccb8c | 1167 | lock_page_memcg(page); |
8186eb6a | 1168 | |
53f9263b KS |
1169 | /* Hugepages are not counted in NR_FILE_MAPPED for now. */ |
1170 | if (unlikely(PageHuge(page))) { | |
1171 | /* hugetlb pages are always mapped with pmds */ | |
1172 | atomic_dec(compound_mapcount_ptr(page)); | |
8186eb6a | 1173 | goto out; |
53f9263b | 1174 | } |
8186eb6a | 1175 | |
53f9263b | 1176 | /* page still mapped by someone else? */ |
dd78fedd KS |
1177 | if (compound && PageTransHuge(page)) { |
1178 | for (i = 0, nr = 0; i < HPAGE_PMD_NR; i++) { | |
1179 | if (atomic_add_negative(-1, &page[i]._mapcount)) | |
1180 | nr++; | |
1181 | } | |
1182 | if (!atomic_add_negative(-1, compound_mapcount_ptr(page))) | |
1183 | goto out; | |
65c45377 | 1184 | VM_BUG_ON_PAGE(!PageSwapBacked(page), page); |
11fb9989 | 1185 | __dec_node_page_state(page, NR_SHMEM_PMDMAPPED); |
dd78fedd KS |
1186 | } else { |
1187 | if (!atomic_add_negative(-1, &page->_mapcount)) | |
1188 | goto out; | |
1189 | } | |
8186eb6a JW |
1190 | |
1191 | /* | |
50658e2e | 1192 | * We use the irq-unsafe __{inc|mod}_zone_page_state because |
8186eb6a JW |
1193 | * these counters are not modified in interrupt context, and |
1194 | * pte lock(a spinlock) is held, which implies preemption disabled. | |
1195 | */ | |
50658e2e | 1196 | __mod_node_page_state(page_pgdat(page), NR_FILE_MAPPED, -nr); |
62cccb8c | 1197 | mem_cgroup_dec_page_stat(page, MEM_CGROUP_STAT_FILE_MAPPED); |
8186eb6a JW |
1198 | |
1199 | if (unlikely(PageMlocked(page))) | |
1200 | clear_page_mlock(page); | |
1201 | out: | |
62cccb8c | 1202 | unlock_page_memcg(page); |
8186eb6a JW |
1203 | } |
1204 | ||
53f9263b KS |
1205 | static void page_remove_anon_compound_rmap(struct page *page) |
1206 | { | |
1207 | int i, nr; | |
1208 | ||
1209 | if (!atomic_add_negative(-1, compound_mapcount_ptr(page))) | |
1210 | return; | |
1211 | ||
1212 | /* Hugepages are not counted in NR_ANON_PAGES for now. */ | |
1213 | if (unlikely(PageHuge(page))) | |
1214 | return; | |
1215 | ||
1216 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) | |
1217 | return; | |
1218 | ||
11fb9989 | 1219 | __dec_node_page_state(page, NR_ANON_THPS); |
53f9263b KS |
1220 | |
1221 | if (TestClearPageDoubleMap(page)) { | |
1222 | /* | |
1223 | * Subpages can be mapped with PTEs too. Check how many of | |
1224 | * themi are still mapped. | |
1225 | */ | |
1226 | for (i = 0, nr = 0; i < HPAGE_PMD_NR; i++) { | |
1227 | if (atomic_add_negative(-1, &page[i]._mapcount)) | |
1228 | nr++; | |
1229 | } | |
1230 | } else { | |
1231 | nr = HPAGE_PMD_NR; | |
1232 | } | |
1233 | ||
e90309c9 KS |
1234 | if (unlikely(PageMlocked(page))) |
1235 | clear_page_mlock(page); | |
1236 | ||
9a982250 | 1237 | if (nr) { |
4b9d0fab | 1238 | __mod_node_page_state(page_pgdat(page), NR_ANON_MAPPED, -nr); |
9a982250 KS |
1239 | deferred_split_huge_page(page); |
1240 | } | |
53f9263b KS |
1241 | } |
1242 | ||
1da177e4 LT |
1243 | /** |
1244 | * page_remove_rmap - take down pte mapping from a page | |
d281ee61 KS |
1245 | * @page: page to remove mapping from |
1246 | * @compound: uncharge the page as compound or small page | |
1da177e4 | 1247 | * |
b8072f09 | 1248 | * The caller needs to hold the pte lock. |
1da177e4 | 1249 | */ |
d281ee61 | 1250 | void page_remove_rmap(struct page *page, bool compound) |
1da177e4 | 1251 | { |
dd78fedd KS |
1252 | if (!PageAnon(page)) |
1253 | return page_remove_file_rmap(page, compound); | |
89c06bd5 | 1254 | |
53f9263b KS |
1255 | if (compound) |
1256 | return page_remove_anon_compound_rmap(page); | |
1257 | ||
b904dcfe KM |
1258 | /* page still mapped by someone else? */ |
1259 | if (!atomic_add_negative(-1, &page->_mapcount)) | |
8186eb6a JW |
1260 | return; |
1261 | ||
0fe6e20b | 1262 | /* |
bea04b07 JZ |
1263 | * We use the irq-unsafe __{inc|mod}_zone_page_stat because |
1264 | * these counters are not modified in interrupt context, and | |
bea04b07 | 1265 | * pte lock(a spinlock) is held, which implies preemption disabled. |
0fe6e20b | 1266 | */ |
4b9d0fab | 1267 | __dec_node_page_state(page, NR_ANON_MAPPED); |
8186eb6a | 1268 | |
e6c509f8 HD |
1269 | if (unlikely(PageMlocked(page))) |
1270 | clear_page_mlock(page); | |
8186eb6a | 1271 | |
9a982250 KS |
1272 | if (PageTransCompound(page)) |
1273 | deferred_split_huge_page(compound_head(page)); | |
1274 | ||
b904dcfe KM |
1275 | /* |
1276 | * It would be tidy to reset the PageAnon mapping here, | |
1277 | * but that might overwrite a racing page_add_anon_rmap | |
1278 | * which increments mapcount after us but sets mapping | |
1279 | * before us: so leave the reset to free_hot_cold_page, | |
1280 | * and remember that it's only reliable while mapped. | |
1281 | * Leaving it set also helps swapoff to reinstate ptes | |
1282 | * faster for those pages still in swapcache. | |
1283 | */ | |
1da177e4 LT |
1284 | } |
1285 | ||
854e9ed0 MK |
1286 | struct rmap_private { |
1287 | enum ttu_flags flags; | |
1288 | int lazyfreed; | |
1289 | }; | |
1290 | ||
1da177e4 | 1291 | /* |
52629506 | 1292 | * @arg: enum ttu_flags will be passed to this argument |
1da177e4 | 1293 | */ |
ac769501 | 1294 | static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma, |
52629506 | 1295 | unsigned long address, void *arg) |
1da177e4 LT |
1296 | { |
1297 | struct mm_struct *mm = vma->vm_mm; | |
c7ab0d2f KS |
1298 | struct page_vma_mapped_walk pvmw = { |
1299 | .page = page, | |
1300 | .vma = vma, | |
1301 | .address = address, | |
1302 | }; | |
1da177e4 | 1303 | pte_t pteval; |
c7ab0d2f | 1304 | struct page *subpage; |
1da177e4 | 1305 | int ret = SWAP_AGAIN; |
854e9ed0 MK |
1306 | struct rmap_private *rp = arg; |
1307 | enum ttu_flags flags = rp->flags; | |
1da177e4 | 1308 | |
b87537d9 HD |
1309 | /* munlock has nothing to gain from examining un-locked vmas */ |
1310 | if ((flags & TTU_MUNLOCK) && !(vma->vm_flags & VM_LOCKED)) | |
c7ab0d2f | 1311 | return SWAP_AGAIN; |
b87537d9 | 1312 | |
fec89c10 KS |
1313 | if (flags & TTU_SPLIT_HUGE_PMD) { |
1314 | split_huge_pmd_address(vma, address, | |
1315 | flags & TTU_MIGRATION, page); | |
fec89c10 KS |
1316 | } |
1317 | ||
c7ab0d2f KS |
1318 | while (page_vma_mapped_walk(&pvmw)) { |
1319 | subpage = page - page_to_pfn(page) + pte_pfn(*pvmw.pte); | |
1320 | address = pvmw.address; | |
1da177e4 | 1321 | |
c7ab0d2f KS |
1322 | /* Unexpected PMD-mapped THP? */ |
1323 | VM_BUG_ON_PAGE(!pvmw.pte, page); | |
1324 | ||
1325 | /* | |
1326 | * If the page is mlock()d, we cannot swap it out. | |
1327 | * If it's recently referenced (perhaps page_referenced | |
1328 | * skipped over this mm) then we should reactivate it. | |
1329 | */ | |
1330 | if (!(flags & TTU_IGNORE_MLOCK)) { | |
1331 | if (vma->vm_flags & VM_LOCKED) { | |
1332 | /* PTE-mapped THP are never mlocked */ | |
1333 | if (!PageTransCompound(page)) { | |
1334 | /* | |
1335 | * Holding pte lock, we do *not* need | |
1336 | * mmap_sem here | |
1337 | */ | |
1338 | mlock_vma_page(page); | |
1339 | } | |
1340 | ret = SWAP_MLOCK; | |
1341 | page_vma_mapped_walk_done(&pvmw); | |
1342 | break; | |
9a73f61b | 1343 | } |
c7ab0d2f KS |
1344 | if (flags & TTU_MUNLOCK) |
1345 | continue; | |
b87537d9 | 1346 | } |
c7ab0d2f KS |
1347 | |
1348 | if (!(flags & TTU_IGNORE_ACCESS)) { | |
1349 | if (ptep_clear_flush_young_notify(vma, address, | |
1350 | pvmw.pte)) { | |
1351 | ret = SWAP_FAIL; | |
1352 | page_vma_mapped_walk_done(&pvmw); | |
1353 | break; | |
1354 | } | |
b291f000 | 1355 | } |
1da177e4 | 1356 | |
c7ab0d2f KS |
1357 | /* Nuke the page table entry. */ |
1358 | flush_cache_page(vma, address, pte_pfn(*pvmw.pte)); | |
1359 | if (should_defer_flush(mm, flags)) { | |
1360 | /* | |
1361 | * We clear the PTE but do not flush so potentially | |
1362 | * a remote CPU could still be writing to the page. | |
1363 | * If the entry was previously clean then the | |
1364 | * architecture must guarantee that a clear->dirty | |
1365 | * transition on a cached TLB entry is written through | |
1366 | * and traps if the PTE is unmapped. | |
1367 | */ | |
1368 | pteval = ptep_get_and_clear(mm, address, pvmw.pte); | |
1369 | ||
1370 | set_tlb_ubc_flush_pending(mm, pte_dirty(pteval)); | |
1371 | } else { | |
1372 | pteval = ptep_clear_flush(vma, address, pvmw.pte); | |
1373 | } | |
72b252ae | 1374 | |
c7ab0d2f KS |
1375 | /* Move the dirty bit to the page. Now the pte is gone. */ |
1376 | if (pte_dirty(pteval)) | |
1377 | set_page_dirty(page); | |
1da177e4 | 1378 | |
c7ab0d2f KS |
1379 | /* Update high watermark before we lower rss */ |
1380 | update_hiwater_rss(mm); | |
1da177e4 | 1381 | |
c7ab0d2f KS |
1382 | if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) { |
1383 | if (PageHuge(page)) { | |
1384 | int nr = 1 << compound_order(page); | |
1385 | hugetlb_count_sub(nr, mm); | |
1386 | } else { | |
1387 | dec_mm_counter(mm, mm_counter(page)); | |
1388 | } | |
365e9c87 | 1389 | |
c7ab0d2f KS |
1390 | pteval = swp_entry_to_pte(make_hwpoison_entry(subpage)); |
1391 | set_pte_at(mm, address, pvmw.pte, pteval); | |
1392 | } else if (pte_unused(pteval)) { | |
1393 | /* | |
1394 | * The guest indicated that the page content is of no | |
1395 | * interest anymore. Simply discard the pte, vmscan | |
1396 | * will take care of the rest. | |
1397 | */ | |
eca56ff9 | 1398 | dec_mm_counter(mm, mm_counter(page)); |
c7ab0d2f KS |
1399 | } else if (IS_ENABLED(CONFIG_MIGRATION) && |
1400 | (flags & TTU_MIGRATION)) { | |
1401 | swp_entry_t entry; | |
1402 | pte_t swp_pte; | |
1403 | /* | |
1404 | * Store the pfn of the page in a special migration | |
1405 | * pte. do_swap_page() will wait until the migration | |
1406 | * pte is removed and then restart fault handling. | |
1407 | */ | |
1408 | entry = make_migration_entry(subpage, | |
1409 | pte_write(pteval)); | |
1410 | swp_pte = swp_entry_to_pte(entry); | |
1411 | if (pte_soft_dirty(pteval)) | |
1412 | swp_pte = pte_swp_mksoft_dirty(swp_pte); | |
1413 | set_pte_at(mm, address, pvmw.pte, swp_pte); | |
1414 | } else if (PageAnon(page)) { | |
1415 | swp_entry_t entry = { .val = page_private(subpage) }; | |
1416 | pte_t swp_pte; | |
1417 | /* | |
1418 | * Store the swap location in the pte. | |
1419 | * See handle_pte_fault() ... | |
1420 | */ | |
1421 | VM_BUG_ON_PAGE(!PageSwapCache(page), page); | |
1422 | ||
1423 | if (!PageDirty(page) && (flags & TTU_LZFREE)) { | |
1424 | /* It's a freeable page by MADV_FREE */ | |
1425 | dec_mm_counter(mm, MM_ANONPAGES); | |
1426 | rp->lazyfreed++; | |
1427 | goto discard; | |
1428 | } | |
854e9ed0 | 1429 | |
c7ab0d2f KS |
1430 | if (swap_duplicate(entry) < 0) { |
1431 | set_pte_at(mm, address, pvmw.pte, pteval); | |
1432 | ret = SWAP_FAIL; | |
1433 | page_vma_mapped_walk_done(&pvmw); | |
1434 | break; | |
1435 | } | |
1436 | if (list_empty(&mm->mmlist)) { | |
1437 | spin_lock(&mmlist_lock); | |
1438 | if (list_empty(&mm->mmlist)) | |
1439 | list_add(&mm->mmlist, &init_mm.mmlist); | |
1440 | spin_unlock(&mmlist_lock); | |
1441 | } | |
854e9ed0 | 1442 | dec_mm_counter(mm, MM_ANONPAGES); |
c7ab0d2f KS |
1443 | inc_mm_counter(mm, MM_SWAPENTS); |
1444 | swp_pte = swp_entry_to_pte(entry); | |
1445 | if (pte_soft_dirty(pteval)) | |
1446 | swp_pte = pte_swp_mksoft_dirty(swp_pte); | |
1447 | set_pte_at(mm, address, pvmw.pte, swp_pte); | |
1448 | } else | |
1449 | dec_mm_counter(mm, mm_counter_file(page)); | |
854e9ed0 | 1450 | discard: |
c7ab0d2f KS |
1451 | page_remove_rmap(subpage, PageHuge(page)); |
1452 | put_page(page); | |
2ec74c3e | 1453 | mmu_notifier_invalidate_page(mm, address); |
c7ab0d2f | 1454 | } |
caed0f48 | 1455 | return ret; |
1da177e4 LT |
1456 | } |
1457 | ||
71e3aac0 | 1458 | bool is_vma_temporary_stack(struct vm_area_struct *vma) |
a8bef8ff MG |
1459 | { |
1460 | int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP); | |
1461 | ||
1462 | if (!maybe_stack) | |
1463 | return false; | |
1464 | ||
1465 | if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) == | |
1466 | VM_STACK_INCOMPLETE_SETUP) | |
1467 | return true; | |
1468 | ||
1469 | return false; | |
1470 | } | |
1471 | ||
52629506 JK |
1472 | static bool invalid_migration_vma(struct vm_area_struct *vma, void *arg) |
1473 | { | |
1474 | return is_vma_temporary_stack(vma); | |
1475 | } | |
1476 | ||
2a52bcbc | 1477 | static int page_mapcount_is_zero(struct page *page) |
52629506 | 1478 | { |
c7ab0d2f | 1479 | return !total_mapcount(page); |
2a52bcbc | 1480 | } |
52629506 | 1481 | |
1da177e4 LT |
1482 | /** |
1483 | * try_to_unmap - try to remove all page table mappings to a page | |
1484 | * @page: the page to get unmapped | |
14fa31b8 | 1485 | * @flags: action and flags |
1da177e4 LT |
1486 | * |
1487 | * Tries to remove all the page table entries which are mapping this | |
1488 | * page, used in the pageout path. Caller must hold the page lock. | |
1489 | * Return values are: | |
1490 | * | |
1491 | * SWAP_SUCCESS - we succeeded in removing all mappings | |
1492 | * SWAP_AGAIN - we missed a mapping, try again later | |
1493 | * SWAP_FAIL - the page is unswappable | |
b291f000 | 1494 | * SWAP_MLOCK - page is mlocked. |
1da177e4 | 1495 | */ |
14fa31b8 | 1496 | int try_to_unmap(struct page *page, enum ttu_flags flags) |
1da177e4 LT |
1497 | { |
1498 | int ret; | |
854e9ed0 MK |
1499 | struct rmap_private rp = { |
1500 | .flags = flags, | |
1501 | .lazyfreed = 0, | |
1502 | }; | |
1503 | ||
52629506 JK |
1504 | struct rmap_walk_control rwc = { |
1505 | .rmap_one = try_to_unmap_one, | |
854e9ed0 | 1506 | .arg = &rp, |
2a52bcbc | 1507 | .done = page_mapcount_is_zero, |
52629506 JK |
1508 | .anon_lock = page_lock_anon_vma_read, |
1509 | }; | |
1da177e4 | 1510 | |
52629506 JK |
1511 | /* |
1512 | * During exec, a temporary VMA is setup and later moved. | |
1513 | * The VMA is moved under the anon_vma lock but not the | |
1514 | * page tables leading to a race where migration cannot | |
1515 | * find the migration ptes. Rather than increasing the | |
1516 | * locking requirements of exec(), migration skips | |
1517 | * temporary VMAs until after exec() completes. | |
1518 | */ | |
daa5ba76 | 1519 | if ((flags & TTU_MIGRATION) && !PageKsm(page) && PageAnon(page)) |
52629506 JK |
1520 | rwc.invalid_vma = invalid_migration_vma; |
1521 | ||
2a52bcbc KS |
1522 | if (flags & TTU_RMAP_LOCKED) |
1523 | ret = rmap_walk_locked(page, &rwc); | |
1524 | else | |
1525 | ret = rmap_walk(page, &rwc); | |
52629506 | 1526 | |
2a52bcbc | 1527 | if (ret != SWAP_MLOCK && !page_mapcount(page)) { |
1da177e4 | 1528 | ret = SWAP_SUCCESS; |
854e9ed0 MK |
1529 | if (rp.lazyfreed && !PageDirty(page)) |
1530 | ret = SWAP_LZFREE; | |
1531 | } | |
1da177e4 LT |
1532 | return ret; |
1533 | } | |
81b4082d | 1534 | |
2a52bcbc KS |
1535 | static int page_not_mapped(struct page *page) |
1536 | { | |
1537 | return !page_mapped(page); | |
1538 | }; | |
1539 | ||
b291f000 NP |
1540 | /** |
1541 | * try_to_munlock - try to munlock a page | |
1542 | * @page: the page to be munlocked | |
1543 | * | |
1544 | * Called from munlock code. Checks all of the VMAs mapping the page | |
1545 | * to make sure nobody else has this page mlocked. The page will be | |
1546 | * returned with PG_mlocked cleared if no other vmas have it mlocked. | |
1547 | * | |
1548 | * Return values are: | |
1549 | * | |
53f79acb | 1550 | * SWAP_AGAIN - no vma is holding page mlocked, or, |
b291f000 | 1551 | * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem |
5ad64688 | 1552 | * SWAP_FAIL - page cannot be located at present |
b291f000 NP |
1553 | * SWAP_MLOCK - page is now mlocked. |
1554 | */ | |
1555 | int try_to_munlock(struct page *page) | |
1556 | { | |
e8351ac9 | 1557 | int ret; |
854e9ed0 MK |
1558 | struct rmap_private rp = { |
1559 | .flags = TTU_MUNLOCK, | |
1560 | .lazyfreed = 0, | |
1561 | }; | |
1562 | ||
e8351ac9 JK |
1563 | struct rmap_walk_control rwc = { |
1564 | .rmap_one = try_to_unmap_one, | |
854e9ed0 | 1565 | .arg = &rp, |
e8351ac9 | 1566 | .done = page_not_mapped, |
e8351ac9 JK |
1567 | .anon_lock = page_lock_anon_vma_read, |
1568 | ||
1569 | }; | |
1570 | ||
309381fe | 1571 | VM_BUG_ON_PAGE(!PageLocked(page) || PageLRU(page), page); |
b291f000 | 1572 | |
e8351ac9 JK |
1573 | ret = rmap_walk(page, &rwc); |
1574 | return ret; | |
b291f000 | 1575 | } |
e9995ef9 | 1576 | |
01d8b20d | 1577 | void __put_anon_vma(struct anon_vma *anon_vma) |
76545066 | 1578 | { |
01d8b20d | 1579 | struct anon_vma *root = anon_vma->root; |
76545066 | 1580 | |
624483f3 | 1581 | anon_vma_free(anon_vma); |
01d8b20d PZ |
1582 | if (root != anon_vma && atomic_dec_and_test(&root->refcount)) |
1583 | anon_vma_free(root); | |
76545066 | 1584 | } |
76545066 | 1585 | |
0dd1c7bb JK |
1586 | static struct anon_vma *rmap_walk_anon_lock(struct page *page, |
1587 | struct rmap_walk_control *rwc) | |
faecd8dd JK |
1588 | { |
1589 | struct anon_vma *anon_vma; | |
1590 | ||
0dd1c7bb JK |
1591 | if (rwc->anon_lock) |
1592 | return rwc->anon_lock(page); | |
1593 | ||
faecd8dd JK |
1594 | /* |
1595 | * Note: remove_migration_ptes() cannot use page_lock_anon_vma_read() | |
1596 | * because that depends on page_mapped(); but not all its usages | |
1597 | * are holding mmap_sem. Users without mmap_sem are required to | |
1598 | * take a reference count to prevent the anon_vma disappearing | |
1599 | */ | |
1600 | anon_vma = page_anon_vma(page); | |
1601 | if (!anon_vma) | |
1602 | return NULL; | |
1603 | ||
1604 | anon_vma_lock_read(anon_vma); | |
1605 | return anon_vma; | |
1606 | } | |
1607 | ||
e9995ef9 | 1608 | /* |
e8351ac9 JK |
1609 | * rmap_walk_anon - do something to anonymous page using the object-based |
1610 | * rmap method | |
1611 | * @page: the page to be handled | |
1612 | * @rwc: control variable according to each walk type | |
1613 | * | |
1614 | * Find all the mappings of a page using the mapping pointer and the vma chains | |
1615 | * contained in the anon_vma struct it points to. | |
1616 | * | |
1617 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma | |
1618 | * where the page was found will be held for write. So, we won't recheck | |
1619 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be | |
1620 | * LOCKED. | |
e9995ef9 | 1621 | */ |
b9773199 KS |
1622 | static int rmap_walk_anon(struct page *page, struct rmap_walk_control *rwc, |
1623 | bool locked) | |
e9995ef9 HD |
1624 | { |
1625 | struct anon_vma *anon_vma; | |
a8fa41ad | 1626 | pgoff_t pgoff_start, pgoff_end; |
5beb4930 | 1627 | struct anon_vma_chain *avc; |
e9995ef9 HD |
1628 | int ret = SWAP_AGAIN; |
1629 | ||
b9773199 KS |
1630 | if (locked) { |
1631 | anon_vma = page_anon_vma(page); | |
1632 | /* anon_vma disappear under us? */ | |
1633 | VM_BUG_ON_PAGE(!anon_vma, page); | |
1634 | } else { | |
1635 | anon_vma = rmap_walk_anon_lock(page, rwc); | |
1636 | } | |
e9995ef9 HD |
1637 | if (!anon_vma) |
1638 | return ret; | |
faecd8dd | 1639 | |
a8fa41ad KS |
1640 | pgoff_start = page_to_pgoff(page); |
1641 | pgoff_end = pgoff_start + hpage_nr_pages(page) - 1; | |
1642 | anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, | |
1643 | pgoff_start, pgoff_end) { | |
5beb4930 | 1644 | struct vm_area_struct *vma = avc->vma; |
e9995ef9 | 1645 | unsigned long address = vma_address(page, vma); |
0dd1c7bb | 1646 | |
ad12695f AA |
1647 | cond_resched(); |
1648 | ||
0dd1c7bb JK |
1649 | if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg)) |
1650 | continue; | |
1651 | ||
051ac83a | 1652 | ret = rwc->rmap_one(page, vma, address, rwc->arg); |
e9995ef9 HD |
1653 | if (ret != SWAP_AGAIN) |
1654 | break; | |
0dd1c7bb JK |
1655 | if (rwc->done && rwc->done(page)) |
1656 | break; | |
e9995ef9 | 1657 | } |
b9773199 KS |
1658 | |
1659 | if (!locked) | |
1660 | anon_vma_unlock_read(anon_vma); | |
e9995ef9 HD |
1661 | return ret; |
1662 | } | |
1663 | ||
e8351ac9 JK |
1664 | /* |
1665 | * rmap_walk_file - do something to file page using the object-based rmap method | |
1666 | * @page: the page to be handled | |
1667 | * @rwc: control variable according to each walk type | |
1668 | * | |
1669 | * Find all the mappings of a page using the mapping pointer and the vma chains | |
1670 | * contained in the address_space struct it points to. | |
1671 | * | |
1672 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma | |
1673 | * where the page was found will be held for write. So, we won't recheck | |
1674 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be | |
1675 | * LOCKED. | |
1676 | */ | |
b9773199 KS |
1677 | static int rmap_walk_file(struct page *page, struct rmap_walk_control *rwc, |
1678 | bool locked) | |
e9995ef9 | 1679 | { |
b9773199 | 1680 | struct address_space *mapping = page_mapping(page); |
a8fa41ad | 1681 | pgoff_t pgoff_start, pgoff_end; |
e9995ef9 | 1682 | struct vm_area_struct *vma; |
e9995ef9 HD |
1683 | int ret = SWAP_AGAIN; |
1684 | ||
9f32624b JK |
1685 | /* |
1686 | * The page lock not only makes sure that page->mapping cannot | |
1687 | * suddenly be NULLified by truncation, it makes sure that the | |
1688 | * structure at mapping cannot be freed and reused yet, | |
c8c06efa | 1689 | * so we can safely take mapping->i_mmap_rwsem. |
9f32624b | 1690 | */ |
81d1b09c | 1691 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
9f32624b | 1692 | |
e9995ef9 HD |
1693 | if (!mapping) |
1694 | return ret; | |
3dec0ba0 | 1695 | |
a8fa41ad KS |
1696 | pgoff_start = page_to_pgoff(page); |
1697 | pgoff_end = pgoff_start + hpage_nr_pages(page) - 1; | |
b9773199 KS |
1698 | if (!locked) |
1699 | i_mmap_lock_read(mapping); | |
a8fa41ad KS |
1700 | vma_interval_tree_foreach(vma, &mapping->i_mmap, |
1701 | pgoff_start, pgoff_end) { | |
e9995ef9 | 1702 | unsigned long address = vma_address(page, vma); |
0dd1c7bb | 1703 | |
ad12695f AA |
1704 | cond_resched(); |
1705 | ||
0dd1c7bb JK |
1706 | if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg)) |
1707 | continue; | |
1708 | ||
051ac83a | 1709 | ret = rwc->rmap_one(page, vma, address, rwc->arg); |
e9995ef9 | 1710 | if (ret != SWAP_AGAIN) |
0dd1c7bb JK |
1711 | goto done; |
1712 | if (rwc->done && rwc->done(page)) | |
1713 | goto done; | |
e9995ef9 | 1714 | } |
0dd1c7bb | 1715 | |
0dd1c7bb | 1716 | done: |
b9773199 KS |
1717 | if (!locked) |
1718 | i_mmap_unlock_read(mapping); | |
e9995ef9 HD |
1719 | return ret; |
1720 | } | |
1721 | ||
051ac83a | 1722 | int rmap_walk(struct page *page, struct rmap_walk_control *rwc) |
e9995ef9 | 1723 | { |
e9995ef9 | 1724 | if (unlikely(PageKsm(page))) |
051ac83a | 1725 | return rmap_walk_ksm(page, rwc); |
e9995ef9 | 1726 | else if (PageAnon(page)) |
b9773199 KS |
1727 | return rmap_walk_anon(page, rwc, false); |
1728 | else | |
1729 | return rmap_walk_file(page, rwc, false); | |
1730 | } | |
1731 | ||
1732 | /* Like rmap_walk, but caller holds relevant rmap lock */ | |
1733 | int rmap_walk_locked(struct page *page, struct rmap_walk_control *rwc) | |
1734 | { | |
1735 | /* no ksm support for now */ | |
1736 | VM_BUG_ON_PAGE(PageKsm(page), page); | |
1737 | if (PageAnon(page)) | |
1738 | return rmap_walk_anon(page, rwc, true); | |
e9995ef9 | 1739 | else |
b9773199 | 1740 | return rmap_walk_file(page, rwc, true); |
e9995ef9 | 1741 | } |
0fe6e20b | 1742 | |
e3390f67 | 1743 | #ifdef CONFIG_HUGETLB_PAGE |
0fe6e20b NH |
1744 | /* |
1745 | * The following three functions are for anonymous (private mapped) hugepages. | |
1746 | * Unlike common anonymous pages, anonymous hugepages have no accounting code | |
1747 | * and no lru code, because we handle hugepages differently from common pages. | |
1748 | */ | |
1749 | static void __hugepage_set_anon_rmap(struct page *page, | |
1750 | struct vm_area_struct *vma, unsigned long address, int exclusive) | |
1751 | { | |
1752 | struct anon_vma *anon_vma = vma->anon_vma; | |
433abed6 | 1753 | |
0fe6e20b | 1754 | BUG_ON(!anon_vma); |
433abed6 NH |
1755 | |
1756 | if (PageAnon(page)) | |
1757 | return; | |
1758 | if (!exclusive) | |
1759 | anon_vma = anon_vma->root; | |
1760 | ||
0fe6e20b NH |
1761 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; |
1762 | page->mapping = (struct address_space *) anon_vma; | |
1763 | page->index = linear_page_index(vma, address); | |
1764 | } | |
1765 | ||
1766 | void hugepage_add_anon_rmap(struct page *page, | |
1767 | struct vm_area_struct *vma, unsigned long address) | |
1768 | { | |
1769 | struct anon_vma *anon_vma = vma->anon_vma; | |
1770 | int first; | |
a850ea30 NH |
1771 | |
1772 | BUG_ON(!PageLocked(page)); | |
0fe6e20b | 1773 | BUG_ON(!anon_vma); |
5dbe0af4 | 1774 | /* address might be in next vma when migration races vma_adjust */ |
53f9263b | 1775 | first = atomic_inc_and_test(compound_mapcount_ptr(page)); |
0fe6e20b NH |
1776 | if (first) |
1777 | __hugepage_set_anon_rmap(page, vma, address, 0); | |
1778 | } | |
1779 | ||
1780 | void hugepage_add_new_anon_rmap(struct page *page, | |
1781 | struct vm_area_struct *vma, unsigned long address) | |
1782 | { | |
1783 | BUG_ON(address < vma->vm_start || address >= vma->vm_end); | |
53f9263b | 1784 | atomic_set(compound_mapcount_ptr(page), 0); |
0fe6e20b NH |
1785 | __hugepage_set_anon_rmap(page, vma, address, 1); |
1786 | } | |
e3390f67 | 1787 | #endif /* CONFIG_HUGETLB_PAGE */ |