]>
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 | |
f4b7e272 | 30 | * pgdat->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) | |
b93b0163 | 35 | * i_pages lock (widely used) |
88f306b6 KS |
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) | |
b93b0163 | 39 | * i_pages lock (widely used, in set_page_dirty, |
88f306b6 KS |
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> |
444f84fd | 64 | #include <linux/huge_mm.h> |
ef5d437f | 65 | #include <linux/backing-dev.h> |
33c3fc71 | 66 | #include <linux/page_idle.h> |
a5430dda | 67 | #include <linux/memremap.h> |
bce73e48 | 68 | #include <linux/userfaultfd_k.h> |
1da177e4 LT |
69 | |
70 | #include <asm/tlbflush.h> | |
71 | ||
72b252ae MG |
72 | #include <trace/events/tlb.h> |
73 | ||
b291f000 NP |
74 | #include "internal.h" |
75 | ||
fdd2e5f8 | 76 | static struct kmem_cache *anon_vma_cachep; |
5beb4930 | 77 | static struct kmem_cache *anon_vma_chain_cachep; |
fdd2e5f8 AB |
78 | |
79 | static inline struct anon_vma *anon_vma_alloc(void) | |
80 | { | |
01d8b20d PZ |
81 | struct anon_vma *anon_vma; |
82 | ||
83 | anon_vma = kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL); | |
84 | if (anon_vma) { | |
85 | atomic_set(&anon_vma->refcount, 1); | |
7a3ef208 KK |
86 | anon_vma->degree = 1; /* Reference for first vma */ |
87 | anon_vma->parent = anon_vma; | |
01d8b20d PZ |
88 | /* |
89 | * Initialise the anon_vma root to point to itself. If called | |
90 | * from fork, the root will be reset to the parents anon_vma. | |
91 | */ | |
92 | anon_vma->root = anon_vma; | |
93 | } | |
94 | ||
95 | return anon_vma; | |
fdd2e5f8 AB |
96 | } |
97 | ||
01d8b20d | 98 | static inline void anon_vma_free(struct anon_vma *anon_vma) |
fdd2e5f8 | 99 | { |
01d8b20d | 100 | VM_BUG_ON(atomic_read(&anon_vma->refcount)); |
88c22088 PZ |
101 | |
102 | /* | |
4fc3f1d6 | 103 | * Synchronize against page_lock_anon_vma_read() such that |
88c22088 PZ |
104 | * we can safely hold the lock without the anon_vma getting |
105 | * freed. | |
106 | * | |
107 | * Relies on the full mb implied by the atomic_dec_and_test() from | |
108 | * put_anon_vma() against the acquire barrier implied by | |
4fc3f1d6 | 109 | * down_read_trylock() from page_lock_anon_vma_read(). This orders: |
88c22088 | 110 | * |
4fc3f1d6 IM |
111 | * page_lock_anon_vma_read() VS put_anon_vma() |
112 | * down_read_trylock() atomic_dec_and_test() | |
88c22088 | 113 | * LOCK MB |
4fc3f1d6 | 114 | * atomic_read() rwsem_is_locked() |
88c22088 PZ |
115 | * |
116 | * LOCK should suffice since the actual taking of the lock must | |
117 | * happen _before_ what follows. | |
118 | */ | |
7f39dda9 | 119 | might_sleep(); |
5a505085 | 120 | if (rwsem_is_locked(&anon_vma->root->rwsem)) { |
4fc3f1d6 | 121 | anon_vma_lock_write(anon_vma); |
08b52706 | 122 | anon_vma_unlock_write(anon_vma); |
88c22088 PZ |
123 | } |
124 | ||
fdd2e5f8 AB |
125 | kmem_cache_free(anon_vma_cachep, anon_vma); |
126 | } | |
1da177e4 | 127 | |
dd34739c | 128 | static inline struct anon_vma_chain *anon_vma_chain_alloc(gfp_t gfp) |
5beb4930 | 129 | { |
dd34739c | 130 | return kmem_cache_alloc(anon_vma_chain_cachep, gfp); |
5beb4930 RR |
131 | } |
132 | ||
e574b5fd | 133 | static void anon_vma_chain_free(struct anon_vma_chain *anon_vma_chain) |
5beb4930 RR |
134 | { |
135 | kmem_cache_free(anon_vma_chain_cachep, anon_vma_chain); | |
136 | } | |
137 | ||
6583a843 KC |
138 | static void anon_vma_chain_link(struct vm_area_struct *vma, |
139 | struct anon_vma_chain *avc, | |
140 | struct anon_vma *anon_vma) | |
141 | { | |
142 | avc->vma = vma; | |
143 | avc->anon_vma = anon_vma; | |
144 | list_add(&avc->same_vma, &vma->anon_vma_chain); | |
bf181b9f | 145 | anon_vma_interval_tree_insert(avc, &anon_vma->rb_root); |
6583a843 KC |
146 | } |
147 | ||
d9d332e0 | 148 | /** |
d5a187da | 149 | * __anon_vma_prepare - attach an anon_vma to a memory region |
d9d332e0 LT |
150 | * @vma: the memory region in question |
151 | * | |
152 | * This makes sure the memory mapping described by 'vma' has | |
153 | * an 'anon_vma' attached to it, so that we can associate the | |
154 | * anonymous pages mapped into it with that anon_vma. | |
155 | * | |
d5a187da VB |
156 | * The common case will be that we already have one, which |
157 | * is handled inline by anon_vma_prepare(). But if | |
23a0790a | 158 | * not we either need to find an adjacent mapping that we |
d9d332e0 LT |
159 | * can re-use the anon_vma from (very common when the only |
160 | * reason for splitting a vma has been mprotect()), or we | |
161 | * allocate a new one. | |
162 | * | |
163 | * Anon-vma allocations are very subtle, because we may have | |
4fc3f1d6 | 164 | * optimistically looked up an anon_vma in page_lock_anon_vma_read() |
d9d332e0 LT |
165 | * and that may actually touch the spinlock even in the newly |
166 | * allocated vma (it depends on RCU to make sure that the | |
167 | * anon_vma isn't actually destroyed). | |
168 | * | |
169 | * As a result, we need to do proper anon_vma locking even | |
170 | * for the new allocation. At the same time, we do not want | |
171 | * to do any locking for the common case of already having | |
172 | * an anon_vma. | |
173 | * | |
174 | * This must be called with the mmap_sem held for reading. | |
175 | */ | |
d5a187da | 176 | int __anon_vma_prepare(struct vm_area_struct *vma) |
1da177e4 | 177 | { |
d5a187da VB |
178 | struct mm_struct *mm = vma->vm_mm; |
179 | struct anon_vma *anon_vma, *allocated; | |
5beb4930 | 180 | struct anon_vma_chain *avc; |
1da177e4 LT |
181 | |
182 | might_sleep(); | |
1da177e4 | 183 | |
d5a187da VB |
184 | avc = anon_vma_chain_alloc(GFP_KERNEL); |
185 | if (!avc) | |
186 | goto out_enomem; | |
187 | ||
188 | anon_vma = find_mergeable_anon_vma(vma); | |
189 | allocated = NULL; | |
190 | if (!anon_vma) { | |
191 | anon_vma = anon_vma_alloc(); | |
192 | if (unlikely(!anon_vma)) | |
193 | goto out_enomem_free_avc; | |
194 | allocated = anon_vma; | |
195 | } | |
5beb4930 | 196 | |
d5a187da VB |
197 | anon_vma_lock_write(anon_vma); |
198 | /* page_table_lock to protect against threads */ | |
199 | spin_lock(&mm->page_table_lock); | |
200 | if (likely(!vma->anon_vma)) { | |
201 | vma->anon_vma = anon_vma; | |
202 | anon_vma_chain_link(vma, avc, anon_vma); | |
203 | /* vma reference or self-parent link for new root */ | |
204 | anon_vma->degree++; | |
d9d332e0 | 205 | allocated = NULL; |
d5a187da VB |
206 | avc = NULL; |
207 | } | |
208 | spin_unlock(&mm->page_table_lock); | |
209 | anon_vma_unlock_write(anon_vma); | |
1da177e4 | 210 | |
d5a187da VB |
211 | if (unlikely(allocated)) |
212 | put_anon_vma(allocated); | |
213 | if (unlikely(avc)) | |
214 | anon_vma_chain_free(avc); | |
31f2b0eb | 215 | |
1da177e4 | 216 | return 0; |
5beb4930 RR |
217 | |
218 | out_enomem_free_avc: | |
219 | anon_vma_chain_free(avc); | |
220 | out_enomem: | |
221 | return -ENOMEM; | |
1da177e4 LT |
222 | } |
223 | ||
bb4aa396 LT |
224 | /* |
225 | * This is a useful helper function for locking the anon_vma root as | |
226 | * we traverse the vma->anon_vma_chain, looping over anon_vma's that | |
227 | * have the same vma. | |
228 | * | |
229 | * Such anon_vma's should have the same root, so you'd expect to see | |
230 | * just a single mutex_lock for the whole traversal. | |
231 | */ | |
232 | static inline struct anon_vma *lock_anon_vma_root(struct anon_vma *root, struct anon_vma *anon_vma) | |
233 | { | |
234 | struct anon_vma *new_root = anon_vma->root; | |
235 | if (new_root != root) { | |
236 | if (WARN_ON_ONCE(root)) | |
5a505085 | 237 | up_write(&root->rwsem); |
bb4aa396 | 238 | root = new_root; |
5a505085 | 239 | down_write(&root->rwsem); |
bb4aa396 LT |
240 | } |
241 | return root; | |
242 | } | |
243 | ||
244 | static inline void unlock_anon_vma_root(struct anon_vma *root) | |
245 | { | |
246 | if (root) | |
5a505085 | 247 | up_write(&root->rwsem); |
bb4aa396 LT |
248 | } |
249 | ||
5beb4930 RR |
250 | /* |
251 | * Attach the anon_vmas from src to dst. | |
252 | * Returns 0 on success, -ENOMEM on failure. | |
7a3ef208 KK |
253 | * |
254 | * If dst->anon_vma is NULL this function tries to find and reuse existing | |
255 | * anon_vma which has no vmas and only one child anon_vma. This prevents | |
256 | * degradation of anon_vma hierarchy to endless linear chain in case of | |
257 | * constantly forking task. On the other hand, an anon_vma with more than one | |
258 | * child isn't reused even if there was no alive vma, thus rmap walker has a | |
259 | * good chance of avoiding scanning the whole hierarchy when it searches where | |
260 | * page is mapped. | |
5beb4930 RR |
261 | */ |
262 | int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src) | |
1da177e4 | 263 | { |
5beb4930 | 264 | struct anon_vma_chain *avc, *pavc; |
bb4aa396 | 265 | struct anon_vma *root = NULL; |
5beb4930 | 266 | |
646d87b4 | 267 | list_for_each_entry_reverse(pavc, &src->anon_vma_chain, same_vma) { |
bb4aa396 LT |
268 | struct anon_vma *anon_vma; |
269 | ||
dd34739c LT |
270 | avc = anon_vma_chain_alloc(GFP_NOWAIT | __GFP_NOWARN); |
271 | if (unlikely(!avc)) { | |
272 | unlock_anon_vma_root(root); | |
273 | root = NULL; | |
274 | avc = anon_vma_chain_alloc(GFP_KERNEL); | |
275 | if (!avc) | |
276 | goto enomem_failure; | |
277 | } | |
bb4aa396 LT |
278 | anon_vma = pavc->anon_vma; |
279 | root = lock_anon_vma_root(root, anon_vma); | |
280 | anon_vma_chain_link(dst, avc, anon_vma); | |
7a3ef208 KK |
281 | |
282 | /* | |
283 | * Reuse existing anon_vma if its degree lower than two, | |
284 | * that means it has no vma and only one anon_vma child. | |
285 | * | |
286 | * Do not chose parent anon_vma, otherwise first child | |
287 | * will always reuse it. Root anon_vma is never reused: | |
288 | * it has self-parent reference and at least one child. | |
289 | */ | |
290 | if (!dst->anon_vma && anon_vma != src->anon_vma && | |
291 | anon_vma->degree < 2) | |
292 | dst->anon_vma = anon_vma; | |
5beb4930 | 293 | } |
7a3ef208 KK |
294 | if (dst->anon_vma) |
295 | dst->anon_vma->degree++; | |
bb4aa396 | 296 | unlock_anon_vma_root(root); |
5beb4930 | 297 | return 0; |
1da177e4 | 298 | |
5beb4930 | 299 | enomem_failure: |
3fe89b3e LY |
300 | /* |
301 | * dst->anon_vma is dropped here otherwise its degree can be incorrectly | |
302 | * decremented in unlink_anon_vmas(). | |
303 | * We can safely do this because callers of anon_vma_clone() don't care | |
304 | * about dst->anon_vma if anon_vma_clone() failed. | |
305 | */ | |
306 | dst->anon_vma = NULL; | |
5beb4930 RR |
307 | unlink_anon_vmas(dst); |
308 | return -ENOMEM; | |
1da177e4 LT |
309 | } |
310 | ||
5beb4930 RR |
311 | /* |
312 | * Attach vma to its own anon_vma, as well as to the anon_vmas that | |
313 | * the corresponding VMA in the parent process is attached to. | |
314 | * Returns 0 on success, non-zero on failure. | |
315 | */ | |
316 | int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma) | |
1da177e4 | 317 | { |
5beb4930 RR |
318 | struct anon_vma_chain *avc; |
319 | struct anon_vma *anon_vma; | |
c4ea95d7 | 320 | int error; |
1da177e4 | 321 | |
5beb4930 RR |
322 | /* Don't bother if the parent process has no anon_vma here. */ |
323 | if (!pvma->anon_vma) | |
324 | return 0; | |
325 | ||
7a3ef208 KK |
326 | /* Drop inherited anon_vma, we'll reuse existing or allocate new. */ |
327 | vma->anon_vma = NULL; | |
328 | ||
5beb4930 RR |
329 | /* |
330 | * First, attach the new VMA to the parent VMA's anon_vmas, | |
331 | * so rmap can find non-COWed pages in child processes. | |
332 | */ | |
c4ea95d7 DF |
333 | error = anon_vma_clone(vma, pvma); |
334 | if (error) | |
335 | return error; | |
5beb4930 | 336 | |
7a3ef208 KK |
337 | /* An existing anon_vma has been reused, all done then. */ |
338 | if (vma->anon_vma) | |
339 | return 0; | |
340 | ||
5beb4930 RR |
341 | /* Then add our own anon_vma. */ |
342 | anon_vma = anon_vma_alloc(); | |
343 | if (!anon_vma) | |
344 | goto out_error; | |
dd34739c | 345 | avc = anon_vma_chain_alloc(GFP_KERNEL); |
5beb4930 RR |
346 | if (!avc) |
347 | goto out_error_free_anon_vma; | |
5c341ee1 RR |
348 | |
349 | /* | |
350 | * The root anon_vma's spinlock is the lock actually used when we | |
351 | * lock any of the anon_vmas in this anon_vma tree. | |
352 | */ | |
353 | anon_vma->root = pvma->anon_vma->root; | |
7a3ef208 | 354 | anon_vma->parent = pvma->anon_vma; |
76545066 | 355 | /* |
01d8b20d PZ |
356 | * With refcounts, an anon_vma can stay around longer than the |
357 | * process it belongs to. The root anon_vma needs to be pinned until | |
358 | * this anon_vma is freed, because the lock lives in the root. | |
76545066 RR |
359 | */ |
360 | get_anon_vma(anon_vma->root); | |
5beb4930 RR |
361 | /* Mark this anon_vma as the one where our new (COWed) pages go. */ |
362 | vma->anon_vma = anon_vma; | |
4fc3f1d6 | 363 | anon_vma_lock_write(anon_vma); |
5c341ee1 | 364 | anon_vma_chain_link(vma, avc, anon_vma); |
7a3ef208 | 365 | anon_vma->parent->degree++; |
08b52706 | 366 | anon_vma_unlock_write(anon_vma); |
5beb4930 RR |
367 | |
368 | return 0; | |
369 | ||
370 | out_error_free_anon_vma: | |
01d8b20d | 371 | put_anon_vma(anon_vma); |
5beb4930 | 372 | out_error: |
4946d54c | 373 | unlink_anon_vmas(vma); |
5beb4930 | 374 | return -ENOMEM; |
1da177e4 LT |
375 | } |
376 | ||
5beb4930 RR |
377 | void unlink_anon_vmas(struct vm_area_struct *vma) |
378 | { | |
379 | struct anon_vma_chain *avc, *next; | |
eee2acba | 380 | struct anon_vma *root = NULL; |
5beb4930 | 381 | |
5c341ee1 RR |
382 | /* |
383 | * Unlink each anon_vma chained to the VMA. This list is ordered | |
384 | * from newest to oldest, ensuring the root anon_vma gets freed last. | |
385 | */ | |
5beb4930 | 386 | list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) { |
eee2acba PZ |
387 | struct anon_vma *anon_vma = avc->anon_vma; |
388 | ||
389 | root = lock_anon_vma_root(root, anon_vma); | |
bf181b9f | 390 | anon_vma_interval_tree_remove(avc, &anon_vma->rb_root); |
eee2acba PZ |
391 | |
392 | /* | |
393 | * Leave empty anon_vmas on the list - we'll need | |
394 | * to free them outside the lock. | |
395 | */ | |
f808c13f | 396 | if (RB_EMPTY_ROOT(&anon_vma->rb_root.rb_root)) { |
7a3ef208 | 397 | anon_vma->parent->degree--; |
eee2acba | 398 | continue; |
7a3ef208 | 399 | } |
eee2acba PZ |
400 | |
401 | list_del(&avc->same_vma); | |
402 | anon_vma_chain_free(avc); | |
403 | } | |
7a3ef208 KK |
404 | if (vma->anon_vma) |
405 | vma->anon_vma->degree--; | |
eee2acba PZ |
406 | unlock_anon_vma_root(root); |
407 | ||
408 | /* | |
409 | * Iterate the list once more, it now only contains empty and unlinked | |
410 | * anon_vmas, destroy them. Could not do before due to __put_anon_vma() | |
5a505085 | 411 | * needing to write-acquire the anon_vma->root->rwsem. |
eee2acba PZ |
412 | */ |
413 | list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) { | |
414 | struct anon_vma *anon_vma = avc->anon_vma; | |
415 | ||
e4c5800a | 416 | VM_WARN_ON(anon_vma->degree); |
eee2acba PZ |
417 | put_anon_vma(anon_vma); |
418 | ||
5beb4930 RR |
419 | list_del(&avc->same_vma); |
420 | anon_vma_chain_free(avc); | |
421 | } | |
422 | } | |
423 | ||
51cc5068 | 424 | static void anon_vma_ctor(void *data) |
1da177e4 | 425 | { |
a35afb83 | 426 | struct anon_vma *anon_vma = data; |
1da177e4 | 427 | |
5a505085 | 428 | init_rwsem(&anon_vma->rwsem); |
83813267 | 429 | atomic_set(&anon_vma->refcount, 0); |
f808c13f | 430 | anon_vma->rb_root = RB_ROOT_CACHED; |
1da177e4 LT |
431 | } |
432 | ||
433 | void __init anon_vma_init(void) | |
434 | { | |
435 | anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma), | |
5f0d5a3a | 436 | 0, SLAB_TYPESAFE_BY_RCU|SLAB_PANIC|SLAB_ACCOUNT, |
5d097056 VD |
437 | anon_vma_ctor); |
438 | anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain, | |
439 | SLAB_PANIC|SLAB_ACCOUNT); | |
1da177e4 LT |
440 | } |
441 | ||
442 | /* | |
6111e4ca PZ |
443 | * Getting a lock on a stable anon_vma from a page off the LRU is tricky! |
444 | * | |
445 | * Since there is no serialization what so ever against page_remove_rmap() | |
446 | * the best this function can do is return a locked anon_vma that might | |
447 | * have been relevant to this page. | |
448 | * | |
449 | * The page might have been remapped to a different anon_vma or the anon_vma | |
450 | * returned may already be freed (and even reused). | |
451 | * | |
bc658c96 PZ |
452 | * In case it was remapped to a different anon_vma, the new anon_vma will be a |
453 | * child of the old anon_vma, and the anon_vma lifetime rules will therefore | |
454 | * ensure that any anon_vma obtained from the page will still be valid for as | |
455 | * long as we observe page_mapped() [ hence all those page_mapped() tests ]. | |
456 | * | |
6111e4ca PZ |
457 | * All users of this function must be very careful when walking the anon_vma |
458 | * chain and verify that the page in question is indeed mapped in it | |
459 | * [ something equivalent to page_mapped_in_vma() ]. | |
460 | * | |
461 | * Since anon_vma's slab is DESTROY_BY_RCU and we know from page_remove_rmap() | |
462 | * that the anon_vma pointer from page->mapping is valid if there is a | |
463 | * mapcount, we can dereference the anon_vma after observing those. | |
1da177e4 | 464 | */ |
746b18d4 | 465 | struct anon_vma *page_get_anon_vma(struct page *page) |
1da177e4 | 466 | { |
746b18d4 | 467 | struct anon_vma *anon_vma = NULL; |
1da177e4 LT |
468 | unsigned long anon_mapping; |
469 | ||
470 | rcu_read_lock(); | |
4db0c3c2 | 471 | anon_mapping = (unsigned long)READ_ONCE(page->mapping); |
3ca7b3c5 | 472 | if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON) |
1da177e4 LT |
473 | goto out; |
474 | if (!page_mapped(page)) | |
475 | goto out; | |
476 | ||
477 | anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON); | |
746b18d4 PZ |
478 | if (!atomic_inc_not_zero(&anon_vma->refcount)) { |
479 | anon_vma = NULL; | |
480 | goto out; | |
481 | } | |
f1819427 HD |
482 | |
483 | /* | |
484 | * If this page is still mapped, then its anon_vma cannot have been | |
746b18d4 PZ |
485 | * freed. But if it has been unmapped, we have no security against the |
486 | * anon_vma structure being freed and reused (for another anon_vma: | |
5f0d5a3a | 487 | * SLAB_TYPESAFE_BY_RCU guarantees that - so the atomic_inc_not_zero() |
746b18d4 | 488 | * above cannot corrupt). |
f1819427 | 489 | */ |
746b18d4 | 490 | if (!page_mapped(page)) { |
7f39dda9 | 491 | rcu_read_unlock(); |
746b18d4 | 492 | put_anon_vma(anon_vma); |
7f39dda9 | 493 | return NULL; |
746b18d4 | 494 | } |
1da177e4 LT |
495 | out: |
496 | rcu_read_unlock(); | |
746b18d4 PZ |
497 | |
498 | return anon_vma; | |
499 | } | |
500 | ||
88c22088 PZ |
501 | /* |
502 | * Similar to page_get_anon_vma() except it locks the anon_vma. | |
503 | * | |
504 | * Its a little more complex as it tries to keep the fast path to a single | |
505 | * atomic op -- the trylock. If we fail the trylock, we fall back to getting a | |
506 | * reference like with page_get_anon_vma() and then block on the mutex. | |
507 | */ | |
4fc3f1d6 | 508 | struct anon_vma *page_lock_anon_vma_read(struct page *page) |
746b18d4 | 509 | { |
88c22088 | 510 | struct anon_vma *anon_vma = NULL; |
eee0f252 | 511 | struct anon_vma *root_anon_vma; |
88c22088 | 512 | unsigned long anon_mapping; |
746b18d4 | 513 | |
88c22088 | 514 | rcu_read_lock(); |
4db0c3c2 | 515 | anon_mapping = (unsigned long)READ_ONCE(page->mapping); |
88c22088 PZ |
516 | if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON) |
517 | goto out; | |
518 | if (!page_mapped(page)) | |
519 | goto out; | |
520 | ||
521 | anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON); | |
4db0c3c2 | 522 | root_anon_vma = READ_ONCE(anon_vma->root); |
4fc3f1d6 | 523 | if (down_read_trylock(&root_anon_vma->rwsem)) { |
88c22088 | 524 | /* |
eee0f252 HD |
525 | * If the page is still mapped, then this anon_vma is still |
526 | * its anon_vma, and holding the mutex ensures that it will | |
bc658c96 | 527 | * not go away, see anon_vma_free(). |
88c22088 | 528 | */ |
eee0f252 | 529 | if (!page_mapped(page)) { |
4fc3f1d6 | 530 | up_read(&root_anon_vma->rwsem); |
88c22088 PZ |
531 | anon_vma = NULL; |
532 | } | |
533 | goto out; | |
534 | } | |
746b18d4 | 535 | |
88c22088 PZ |
536 | /* trylock failed, we got to sleep */ |
537 | if (!atomic_inc_not_zero(&anon_vma->refcount)) { | |
538 | anon_vma = NULL; | |
539 | goto out; | |
540 | } | |
541 | ||
542 | if (!page_mapped(page)) { | |
7f39dda9 | 543 | rcu_read_unlock(); |
88c22088 | 544 | put_anon_vma(anon_vma); |
7f39dda9 | 545 | return NULL; |
88c22088 PZ |
546 | } |
547 | ||
548 | /* we pinned the anon_vma, its safe to sleep */ | |
549 | rcu_read_unlock(); | |
4fc3f1d6 | 550 | anon_vma_lock_read(anon_vma); |
88c22088 PZ |
551 | |
552 | if (atomic_dec_and_test(&anon_vma->refcount)) { | |
553 | /* | |
554 | * Oops, we held the last refcount, release the lock | |
555 | * and bail -- can't simply use put_anon_vma() because | |
4fc3f1d6 | 556 | * we'll deadlock on the anon_vma_lock_write() recursion. |
88c22088 | 557 | */ |
4fc3f1d6 | 558 | anon_vma_unlock_read(anon_vma); |
88c22088 PZ |
559 | __put_anon_vma(anon_vma); |
560 | anon_vma = NULL; | |
561 | } | |
562 | ||
563 | return anon_vma; | |
564 | ||
565 | out: | |
566 | rcu_read_unlock(); | |
746b18d4 | 567 | return anon_vma; |
34bbd704 ON |
568 | } |
569 | ||
4fc3f1d6 | 570 | void page_unlock_anon_vma_read(struct anon_vma *anon_vma) |
34bbd704 | 571 | { |
4fc3f1d6 | 572 | anon_vma_unlock_read(anon_vma); |
1da177e4 LT |
573 | } |
574 | ||
72b252ae | 575 | #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH |
72b252ae MG |
576 | /* |
577 | * Flush TLB entries for recently unmapped pages from remote CPUs. It is | |
578 | * important if a PTE was dirty when it was unmapped that it's flushed | |
579 | * before any IO is initiated on the page to prevent lost writes. Similarly, | |
580 | * it must be flushed before freeing to prevent data leakage. | |
581 | */ | |
582 | void try_to_unmap_flush(void) | |
583 | { | |
584 | struct tlbflush_unmap_batch *tlb_ubc = ¤t->tlb_ubc; | |
72b252ae MG |
585 | |
586 | if (!tlb_ubc->flush_required) | |
587 | return; | |
588 | ||
e73ad5ff | 589 | arch_tlbbatch_flush(&tlb_ubc->arch); |
72b252ae | 590 | tlb_ubc->flush_required = false; |
d950c947 | 591 | tlb_ubc->writable = false; |
72b252ae MG |
592 | } |
593 | ||
d950c947 MG |
594 | /* Flush iff there are potentially writable TLB entries that can race with IO */ |
595 | void try_to_unmap_flush_dirty(void) | |
596 | { | |
597 | struct tlbflush_unmap_batch *tlb_ubc = ¤t->tlb_ubc; | |
598 | ||
599 | if (tlb_ubc->writable) | |
600 | try_to_unmap_flush(); | |
601 | } | |
602 | ||
c7ab0d2f | 603 | static void set_tlb_ubc_flush_pending(struct mm_struct *mm, bool writable) |
72b252ae MG |
604 | { |
605 | struct tlbflush_unmap_batch *tlb_ubc = ¤t->tlb_ubc; | |
606 | ||
e73ad5ff | 607 | arch_tlbbatch_add_mm(&tlb_ubc->arch, mm); |
72b252ae | 608 | tlb_ubc->flush_required = true; |
d950c947 | 609 | |
3ea27719 MG |
610 | /* |
611 | * Ensure compiler does not re-order the setting of tlb_flush_batched | |
612 | * before the PTE is cleared. | |
613 | */ | |
614 | barrier(); | |
615 | mm->tlb_flush_batched = true; | |
616 | ||
d950c947 MG |
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 | } | |
3ea27719 MG |
644 | |
645 | /* | |
646 | * Reclaim unmaps pages under the PTL but do not flush the TLB prior to | |
647 | * releasing the PTL if TLB flushes are batched. It's possible for a parallel | |
648 | * operation such as mprotect or munmap to race between reclaim unmapping | |
649 | * the page and flushing the page. If this race occurs, it potentially allows | |
650 | * access to data via a stale TLB entry. Tracking all mm's that have TLB | |
651 | * batching in flight would be expensive during reclaim so instead track | |
652 | * whether TLB batching occurred in the past and if so then do a flush here | |
653 | * if required. This will cost one additional flush per reclaim cycle paid | |
654 | * by the first operation at risk such as mprotect and mumap. | |
655 | * | |
656 | * This must be called under the PTL so that an access to tlb_flush_batched | |
657 | * that is potentially a "reclaim vs mprotect/munmap/etc" race will synchronise | |
658 | * via the PTL. | |
659 | */ | |
660 | void flush_tlb_batched_pending(struct mm_struct *mm) | |
661 | { | |
662 | if (mm->tlb_flush_batched) { | |
663 | flush_tlb_mm(mm); | |
664 | ||
665 | /* | |
666 | * Do not allow the compiler to re-order the clearing of | |
667 | * tlb_flush_batched before the tlb is flushed. | |
668 | */ | |
669 | barrier(); | |
670 | mm->tlb_flush_batched = false; | |
671 | } | |
672 | } | |
72b252ae | 673 | #else |
c7ab0d2f | 674 | static void set_tlb_ubc_flush_pending(struct mm_struct *mm, bool writable) |
72b252ae MG |
675 | { |
676 | } | |
677 | ||
678 | static bool should_defer_flush(struct mm_struct *mm, enum ttu_flags flags) | |
679 | { | |
680 | return false; | |
681 | } | |
682 | #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */ | |
683 | ||
1da177e4 | 684 | /* |
bf89c8c8 | 685 | * At what user virtual address is page expected in vma? |
ab941e0f | 686 | * Caller should check the page is actually part of the vma. |
1da177e4 LT |
687 | */ |
688 | unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma) | |
689 | { | |
21d0d443 | 690 | if (PageAnon(page)) { |
4829b906 HD |
691 | struct anon_vma *page__anon_vma = page_anon_vma(page); |
692 | /* | |
693 | * Note: swapoff's unuse_vma() is more efficient with this | |
694 | * check, and needs it to match anon_vma when KSM is active. | |
695 | */ | |
696 | if (!vma->anon_vma || !page__anon_vma || | |
697 | vma->anon_vma->root != page__anon_vma->root) | |
21d0d443 | 698 | return -EFAULT; |
5ab5ad93 JW |
699 | } else if (!vma->vm_file) { |
700 | return -EFAULT; | |
701 | } else if (vma->vm_file->f_mapping != compound_head(page)->mapping) { | |
1da177e4 | 702 | return -EFAULT; |
5ab5ad93 | 703 | } |
5bbd1b12 HD |
704 | |
705 | return vma_address(page, vma); | |
1da177e4 LT |
706 | } |
707 | ||
6219049a BL |
708 | pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address) |
709 | { | |
710 | pgd_t *pgd; | |
c2febafc | 711 | p4d_t *p4d; |
6219049a BL |
712 | pud_t *pud; |
713 | pmd_t *pmd = NULL; | |
f72e7dcd | 714 | pmd_t pmde; |
6219049a BL |
715 | |
716 | pgd = pgd_offset(mm, address); | |
717 | if (!pgd_present(*pgd)) | |
718 | goto out; | |
719 | ||
c2febafc KS |
720 | p4d = p4d_offset(pgd, address); |
721 | if (!p4d_present(*p4d)) | |
722 | goto out; | |
723 | ||
724 | pud = pud_offset(p4d, address); | |
6219049a BL |
725 | if (!pud_present(*pud)) |
726 | goto out; | |
727 | ||
728 | pmd = pmd_offset(pud, address); | |
f72e7dcd | 729 | /* |
8809aa2d | 730 | * Some THP functions use the sequence pmdp_huge_clear_flush(), set_pmd_at() |
f72e7dcd HD |
731 | * without holding anon_vma lock for write. So when looking for a |
732 | * genuine pmde (in which to find pte), test present and !THP together. | |
733 | */ | |
e37c6982 CB |
734 | pmde = *pmd; |
735 | barrier(); | |
f72e7dcd | 736 | if (!pmd_present(pmde) || pmd_trans_huge(pmde)) |
6219049a BL |
737 | pmd = NULL; |
738 | out: | |
739 | return pmd; | |
740 | } | |
741 | ||
8749cfea VD |
742 | struct page_referenced_arg { |
743 | int mapcount; | |
744 | int referenced; | |
745 | unsigned long vm_flags; | |
746 | struct mem_cgroup *memcg; | |
747 | }; | |
748 | /* | |
749 | * arg: page_referenced_arg will be passed | |
750 | */ | |
e4b82222 | 751 | static bool page_referenced_one(struct page *page, struct vm_area_struct *vma, |
8749cfea VD |
752 | unsigned long address, void *arg) |
753 | { | |
8749cfea | 754 | struct page_referenced_arg *pra = arg; |
8eaedede KS |
755 | struct page_vma_mapped_walk pvmw = { |
756 | .page = page, | |
757 | .vma = vma, | |
758 | .address = address, | |
759 | }; | |
8749cfea VD |
760 | int referenced = 0; |
761 | ||
8eaedede KS |
762 | while (page_vma_mapped_walk(&pvmw)) { |
763 | address = pvmw.address; | |
b20ce5e0 | 764 | |
8eaedede KS |
765 | if (vma->vm_flags & VM_LOCKED) { |
766 | page_vma_mapped_walk_done(&pvmw); | |
767 | pra->vm_flags |= VM_LOCKED; | |
e4b82222 | 768 | return false; /* To break the loop */ |
8eaedede | 769 | } |
71e3aac0 | 770 | |
8eaedede KS |
771 | if (pvmw.pte) { |
772 | if (ptep_clear_flush_young_notify(vma, address, | |
773 | pvmw.pte)) { | |
774 | /* | |
775 | * Don't treat a reference through | |
776 | * a sequentially read mapping as such. | |
777 | * If the page has been used in another mapping, | |
778 | * we will catch it; if this other mapping is | |
779 | * already gone, the unmap path will have set | |
780 | * PG_referenced or activated the page. | |
781 | */ | |
782 | if (likely(!(vma->vm_flags & VM_SEQ_READ))) | |
783 | referenced++; | |
784 | } | |
785 | } else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) { | |
786 | if (pmdp_clear_flush_young_notify(vma, address, | |
787 | pvmw.pmd)) | |
8749cfea | 788 | referenced++; |
8eaedede KS |
789 | } else { |
790 | /* unexpected pmd-mapped page? */ | |
791 | WARN_ON_ONCE(1); | |
8749cfea | 792 | } |
8eaedede KS |
793 | |
794 | pra->mapcount--; | |
b20ce5e0 | 795 | } |
b20ce5e0 | 796 | |
33c3fc71 VD |
797 | if (referenced) |
798 | clear_page_idle(page); | |
799 | if (test_and_clear_page_young(page)) | |
800 | referenced++; | |
801 | ||
9f32624b JK |
802 | if (referenced) { |
803 | pra->referenced++; | |
804 | pra->vm_flags |= vma->vm_flags; | |
1da177e4 | 805 | } |
34bbd704 | 806 | |
9f32624b | 807 | if (!pra->mapcount) |
e4b82222 | 808 | return false; /* To break the loop */ |
9f32624b | 809 | |
e4b82222 | 810 | return true; |
1da177e4 LT |
811 | } |
812 | ||
9f32624b | 813 | static bool invalid_page_referenced_vma(struct vm_area_struct *vma, void *arg) |
1da177e4 | 814 | { |
9f32624b JK |
815 | struct page_referenced_arg *pra = arg; |
816 | struct mem_cgroup *memcg = pra->memcg; | |
1da177e4 | 817 | |
9f32624b JK |
818 | if (!mm_match_cgroup(vma->vm_mm, memcg)) |
819 | return true; | |
1da177e4 | 820 | |
9f32624b | 821 | return false; |
1da177e4 LT |
822 | } |
823 | ||
824 | /** | |
825 | * page_referenced - test if the page was referenced | |
826 | * @page: the page to test | |
827 | * @is_locked: caller holds lock on the page | |
72835c86 | 828 | * @memcg: target memory cgroup |
6fe6b7e3 | 829 | * @vm_flags: collect encountered vma->vm_flags who actually referenced the page |
1da177e4 LT |
830 | * |
831 | * Quick test_and_clear_referenced for all mappings to a page, | |
832 | * returns the number of ptes which referenced the page. | |
833 | */ | |
6fe6b7e3 WF |
834 | int page_referenced(struct page *page, |
835 | int is_locked, | |
72835c86 | 836 | struct mem_cgroup *memcg, |
6fe6b7e3 | 837 | unsigned long *vm_flags) |
1da177e4 | 838 | { |
5ad64688 | 839 | int we_locked = 0; |
9f32624b | 840 | struct page_referenced_arg pra = { |
b20ce5e0 | 841 | .mapcount = total_mapcount(page), |
9f32624b JK |
842 | .memcg = memcg, |
843 | }; | |
844 | struct rmap_walk_control rwc = { | |
845 | .rmap_one = page_referenced_one, | |
846 | .arg = (void *)&pra, | |
847 | .anon_lock = page_lock_anon_vma_read, | |
848 | }; | |
1da177e4 | 849 | |
6fe6b7e3 | 850 | *vm_flags = 0; |
059d8442 | 851 | if (!pra.mapcount) |
9f32624b JK |
852 | return 0; |
853 | ||
854 | if (!page_rmapping(page)) | |
855 | return 0; | |
856 | ||
857 | if (!is_locked && (!PageAnon(page) || PageKsm(page))) { | |
858 | we_locked = trylock_page(page); | |
859 | if (!we_locked) | |
860 | return 1; | |
1da177e4 | 861 | } |
9f32624b JK |
862 | |
863 | /* | |
864 | * If we are reclaiming on behalf of a cgroup, skip | |
865 | * counting on behalf of references from different | |
866 | * cgroups | |
867 | */ | |
868 | if (memcg) { | |
869 | rwc.invalid_vma = invalid_page_referenced_vma; | |
870 | } | |
871 | ||
c24f386c | 872 | rmap_walk(page, &rwc); |
9f32624b JK |
873 | *vm_flags = pra.vm_flags; |
874 | ||
875 | if (we_locked) | |
876 | unlock_page(page); | |
877 | ||
878 | return pra.referenced; | |
1da177e4 LT |
879 | } |
880 | ||
e4b82222 | 881 | static bool page_mkclean_one(struct page *page, struct vm_area_struct *vma, |
9853a407 | 882 | unsigned long address, void *arg) |
d08b3851 | 883 | { |
f27176cf KS |
884 | struct page_vma_mapped_walk pvmw = { |
885 | .page = page, | |
886 | .vma = vma, | |
887 | .address = address, | |
888 | .flags = PVMW_SYNC, | |
889 | }; | |
ac46d4f3 | 890 | struct mmu_notifier_range range; |
9853a407 | 891 | int *cleaned = arg; |
d08b3851 | 892 | |
369ea824 JG |
893 | /* |
894 | * We have to assume the worse case ie pmd for invalidation. Note that | |
895 | * the page can not be free from this function. | |
896 | */ | |
7269f999 JG |
897 | mmu_notifier_range_init(&range, MMU_NOTIFY_PROTECTION_PAGE, |
898 | 0, vma, vma->vm_mm, address, | |
5bbd1b12 | 899 | vma_address_end(page, vma)); |
ac46d4f3 | 900 | mmu_notifier_invalidate_range_start(&range); |
369ea824 | 901 | |
f27176cf KS |
902 | while (page_vma_mapped_walk(&pvmw)) { |
903 | int ret = 0; | |
369ea824 | 904 | |
1f18b296 | 905 | address = pvmw.address; |
f27176cf KS |
906 | if (pvmw.pte) { |
907 | pte_t entry; | |
908 | pte_t *pte = pvmw.pte; | |
909 | ||
910 | if (!pte_dirty(*pte) && !pte_write(*pte)) | |
911 | continue; | |
912 | ||
785373b4 LT |
913 | flush_cache_page(vma, address, pte_pfn(*pte)); |
914 | entry = ptep_clear_flush(vma, address, pte); | |
f27176cf KS |
915 | entry = pte_wrprotect(entry); |
916 | entry = pte_mkclean(entry); | |
785373b4 | 917 | set_pte_at(vma->vm_mm, address, pte, entry); |
f27176cf KS |
918 | ret = 1; |
919 | } else { | |
920 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE | |
921 | pmd_t *pmd = pvmw.pmd; | |
922 | pmd_t entry; | |
923 | ||
924 | if (!pmd_dirty(*pmd) && !pmd_write(*pmd)) | |
925 | continue; | |
926 | ||
785373b4 | 927 | flush_cache_page(vma, address, page_to_pfn(page)); |
024eee0e | 928 | entry = pmdp_invalidate(vma, address, pmd); |
f27176cf KS |
929 | entry = pmd_wrprotect(entry); |
930 | entry = pmd_mkclean(entry); | |
785373b4 | 931 | set_pmd_at(vma->vm_mm, address, pmd, entry); |
f27176cf KS |
932 | ret = 1; |
933 | #else | |
934 | /* unexpected pmd-mapped page? */ | |
935 | WARN_ON_ONCE(1); | |
936 | #endif | |
937 | } | |
d08b3851 | 938 | |
0f10851e JG |
939 | /* |
940 | * No need to call mmu_notifier_invalidate_range() as we are | |
941 | * downgrading page table protection not changing it to point | |
942 | * to a new page. | |
943 | * | |
ad56b738 | 944 | * See Documentation/vm/mmu_notifier.rst |
0f10851e JG |
945 | */ |
946 | if (ret) | |
f27176cf | 947 | (*cleaned)++; |
c2fda5fe | 948 | } |
d08b3851 | 949 | |
ac46d4f3 | 950 | mmu_notifier_invalidate_range_end(&range); |
369ea824 | 951 | |
e4b82222 | 952 | return true; |
d08b3851 PZ |
953 | } |
954 | ||
9853a407 | 955 | static bool invalid_mkclean_vma(struct vm_area_struct *vma, void *arg) |
d08b3851 | 956 | { |
9853a407 | 957 | if (vma->vm_flags & VM_SHARED) |
871beb8c | 958 | return false; |
d08b3851 | 959 | |
871beb8c | 960 | return true; |
d08b3851 PZ |
961 | } |
962 | ||
963 | int page_mkclean(struct page *page) | |
964 | { | |
9853a407 JK |
965 | int cleaned = 0; |
966 | struct address_space *mapping; | |
967 | struct rmap_walk_control rwc = { | |
968 | .arg = (void *)&cleaned, | |
969 | .rmap_one = page_mkclean_one, | |
970 | .invalid_vma = invalid_mkclean_vma, | |
971 | }; | |
d08b3851 PZ |
972 | |
973 | BUG_ON(!PageLocked(page)); | |
974 | ||
9853a407 JK |
975 | if (!page_mapped(page)) |
976 | return 0; | |
977 | ||
978 | mapping = page_mapping(page); | |
979 | if (!mapping) | |
980 | return 0; | |
981 | ||
982 | rmap_walk(page, &rwc); | |
d08b3851 | 983 | |
9853a407 | 984 | return cleaned; |
d08b3851 | 985 | } |
60b59bea | 986 | EXPORT_SYMBOL_GPL(page_mkclean); |
d08b3851 | 987 | |
c44b6743 RR |
988 | /** |
989 | * page_move_anon_rmap - move a page to our anon_vma | |
990 | * @page: the page to move to our anon_vma | |
991 | * @vma: the vma the page belongs to | |
c44b6743 RR |
992 | * |
993 | * When a page belongs exclusively to one process after a COW event, | |
994 | * that page can be moved into the anon_vma that belongs to just that | |
995 | * process, so the rmap code will not search the parent or sibling | |
996 | * processes. | |
997 | */ | |
5a49973d | 998 | void page_move_anon_rmap(struct page *page, struct vm_area_struct *vma) |
c44b6743 RR |
999 | { |
1000 | struct anon_vma *anon_vma = vma->anon_vma; | |
1001 | ||
5a49973d HD |
1002 | page = compound_head(page); |
1003 | ||
309381fe | 1004 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
81d1b09c | 1005 | VM_BUG_ON_VMA(!anon_vma, vma); |
c44b6743 RR |
1006 | |
1007 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; | |
414e2fb8 VD |
1008 | /* |
1009 | * Ensure that anon_vma and the PAGE_MAPPING_ANON bit are written | |
1010 | * simultaneously, so a concurrent reader (eg page_referenced()'s | |
1011 | * PageAnon()) will not see one without the other. | |
1012 | */ | |
1013 | WRITE_ONCE(page->mapping, (struct address_space *) anon_vma); | |
c44b6743 RR |
1014 | } |
1015 | ||
9617d95e | 1016 | /** |
4e1c1975 | 1017 | * __page_set_anon_rmap - set up new anonymous rmap |
451b9514 | 1018 | * @page: Page or Hugepage to add to rmap |
4e1c1975 AK |
1019 | * @vma: VM area to add page to. |
1020 | * @address: User virtual address of the mapping | |
e8a03feb | 1021 | * @exclusive: the page is exclusively owned by the current process |
9617d95e NP |
1022 | */ |
1023 | static void __page_set_anon_rmap(struct page *page, | |
e8a03feb | 1024 | struct vm_area_struct *vma, unsigned long address, int exclusive) |
9617d95e | 1025 | { |
e8a03feb | 1026 | struct anon_vma *anon_vma = vma->anon_vma; |
ea90002b | 1027 | |
e8a03feb | 1028 | BUG_ON(!anon_vma); |
ea90002b | 1029 | |
4e1c1975 AK |
1030 | if (PageAnon(page)) |
1031 | return; | |
1032 | ||
ea90002b | 1033 | /* |
e8a03feb RR |
1034 | * If the page isn't exclusively mapped into this vma, |
1035 | * we must use the _oldest_ possible anon_vma for the | |
1036 | * page mapping! | |
ea90002b | 1037 | */ |
4e1c1975 | 1038 | if (!exclusive) |
288468c3 | 1039 | anon_vma = anon_vma->root; |
9617d95e | 1040 | |
9617d95e NP |
1041 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; |
1042 | page->mapping = (struct address_space *) anon_vma; | |
9617d95e | 1043 | page->index = linear_page_index(vma, address); |
9617d95e NP |
1044 | } |
1045 | ||
c97a9e10 | 1046 | /** |
43d8eac4 | 1047 | * __page_check_anon_rmap - sanity check anonymous rmap addition |
c97a9e10 NP |
1048 | * @page: the page to add the mapping to |
1049 | * @vma: the vm area in which the mapping is added | |
1050 | * @address: the user virtual address mapped | |
1051 | */ | |
1052 | static void __page_check_anon_rmap(struct page *page, | |
1053 | struct vm_area_struct *vma, unsigned long address) | |
1054 | { | |
1055 | #ifdef CONFIG_DEBUG_VM | |
1056 | /* | |
1057 | * The page's anon-rmap details (mapping and index) are guaranteed to | |
1058 | * be set up correctly at this point. | |
1059 | * | |
1060 | * We have exclusion against page_add_anon_rmap because the caller | |
1061 | * always holds the page locked, except if called from page_dup_rmap, | |
1062 | * in which case the page is already known to be setup. | |
1063 | * | |
1064 | * We have exclusion against page_add_new_anon_rmap because those pages | |
1065 | * are initially only visible via the pagetables, and the pte is locked | |
1066 | * over the call to page_add_new_anon_rmap. | |
1067 | */ | |
44ab57a0 | 1068 | BUG_ON(page_anon_vma(page)->root != vma->anon_vma->root); |
53f9263b | 1069 | BUG_ON(page_to_pgoff(page) != linear_page_index(vma, address)); |
c97a9e10 NP |
1070 | #endif |
1071 | } | |
1072 | ||
1da177e4 LT |
1073 | /** |
1074 | * page_add_anon_rmap - add pte mapping to an anonymous page | |
1075 | * @page: the page to add the mapping to | |
1076 | * @vma: the vm area in which the mapping is added | |
1077 | * @address: the user virtual address mapped | |
d281ee61 | 1078 | * @compound: charge the page as compound or small page |
1da177e4 | 1079 | * |
5ad64688 | 1080 | * The caller needs to hold the pte lock, and the page must be locked in |
80e14822 HD |
1081 | * the anon_vma case: to serialize mapping,index checking after setting, |
1082 | * and to ensure that PageAnon is not being upgraded racily to PageKsm | |
1083 | * (but PageKsm is never downgraded to PageAnon). | |
1da177e4 LT |
1084 | */ |
1085 | void page_add_anon_rmap(struct page *page, | |
d281ee61 | 1086 | struct vm_area_struct *vma, unsigned long address, bool compound) |
ad8c2ee8 | 1087 | { |
d281ee61 | 1088 | do_page_add_anon_rmap(page, vma, address, compound ? RMAP_COMPOUND : 0); |
ad8c2ee8 RR |
1089 | } |
1090 | ||
1091 | /* | |
1092 | * Special version of the above for do_swap_page, which often runs | |
1093 | * into pages that are exclusively owned by the current process. | |
1094 | * Everybody else should continue to use page_add_anon_rmap above. | |
1095 | */ | |
1096 | void do_page_add_anon_rmap(struct page *page, | |
d281ee61 | 1097 | struct vm_area_struct *vma, unsigned long address, int flags) |
1da177e4 | 1098 | { |
53f9263b KS |
1099 | bool compound = flags & RMAP_COMPOUND; |
1100 | bool first; | |
1101 | ||
e9b61f19 KS |
1102 | if (compound) { |
1103 | atomic_t *mapcount; | |
53f9263b | 1104 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
e9b61f19 KS |
1105 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
1106 | mapcount = compound_mapcount_ptr(page); | |
1107 | first = atomic_inc_and_test(mapcount); | |
53f9263b KS |
1108 | } else { |
1109 | first = atomic_inc_and_test(&page->_mapcount); | |
1110 | } | |
1111 | ||
79134171 | 1112 | if (first) { |
d281ee61 | 1113 | int nr = compound ? hpage_nr_pages(page) : 1; |
bea04b07 JZ |
1114 | /* |
1115 | * We use the irq-unsafe __{inc|mod}_zone_page_stat because | |
1116 | * these counters are not modified in interrupt context, and | |
1117 | * pte lock(a spinlock) is held, which implies preemption | |
1118 | * disabled. | |
1119 | */ | |
65c45377 | 1120 | if (compound) |
11fb9989 | 1121 | __inc_node_page_state(page, NR_ANON_THPS); |
4b9d0fab | 1122 | __mod_node_page_state(page_pgdat(page), NR_ANON_MAPPED, nr); |
79134171 | 1123 | } |
5ad64688 HD |
1124 | if (unlikely(PageKsm(page))) |
1125 | return; | |
1126 | ||
309381fe | 1127 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
53f9263b | 1128 | |
5dbe0af4 | 1129 | /* address might be in next vma when migration races vma_adjust */ |
5ad64688 | 1130 | if (first) |
d281ee61 KS |
1131 | __page_set_anon_rmap(page, vma, address, |
1132 | flags & RMAP_EXCLUSIVE); | |
69029cd5 | 1133 | else |
c97a9e10 | 1134 | __page_check_anon_rmap(page, vma, address); |
1da177e4 LT |
1135 | } |
1136 | ||
43d8eac4 | 1137 | /** |
9617d95e NP |
1138 | * page_add_new_anon_rmap - add pte mapping to a new anonymous page |
1139 | * @page: the page to add the mapping to | |
1140 | * @vma: the vm area in which the mapping is added | |
1141 | * @address: the user virtual address mapped | |
d281ee61 | 1142 | * @compound: charge the page as compound or small page |
9617d95e NP |
1143 | * |
1144 | * Same as page_add_anon_rmap but must only be called on *new* pages. | |
1145 | * This means the inc-and-test can be bypassed. | |
c97a9e10 | 1146 | * Page does not have to be locked. |
9617d95e NP |
1147 | */ |
1148 | void page_add_new_anon_rmap(struct page *page, | |
d281ee61 | 1149 | struct vm_area_struct *vma, unsigned long address, bool compound) |
9617d95e | 1150 | { |
d281ee61 KS |
1151 | int nr = compound ? hpage_nr_pages(page) : 1; |
1152 | ||
81d1b09c | 1153 | VM_BUG_ON_VMA(address < vma->vm_start || address >= vma->vm_end, vma); |
fa9949da | 1154 | __SetPageSwapBacked(page); |
d281ee61 KS |
1155 | if (compound) { |
1156 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
53f9263b KS |
1157 | /* increment count (starts at -1) */ |
1158 | atomic_set(compound_mapcount_ptr(page), 0); | |
11fb9989 | 1159 | __inc_node_page_state(page, NR_ANON_THPS); |
53f9263b KS |
1160 | } else { |
1161 | /* Anon THP always mapped first with PMD */ | |
1162 | VM_BUG_ON_PAGE(PageTransCompound(page), page); | |
1163 | /* increment count (starts at -1) */ | |
1164 | atomic_set(&page->_mapcount, 0); | |
d281ee61 | 1165 | } |
4b9d0fab | 1166 | __mod_node_page_state(page_pgdat(page), NR_ANON_MAPPED, nr); |
e8a03feb | 1167 | __page_set_anon_rmap(page, vma, address, 1); |
9617d95e NP |
1168 | } |
1169 | ||
1da177e4 LT |
1170 | /** |
1171 | * page_add_file_rmap - add pte mapping to a file page | |
1172 | * @page: the page to add the mapping to | |
e8b098fc | 1173 | * @compound: charge the page as compound or small page |
1da177e4 | 1174 | * |
b8072f09 | 1175 | * The caller needs to hold the pte lock. |
1da177e4 | 1176 | */ |
dd78fedd | 1177 | void page_add_file_rmap(struct page *page, bool compound) |
1da177e4 | 1178 | { |
dd78fedd KS |
1179 | int i, nr = 1; |
1180 | ||
1181 | VM_BUG_ON_PAGE(compound && !PageTransHuge(page), page); | |
62cccb8c | 1182 | lock_page_memcg(page); |
dd78fedd KS |
1183 | if (compound && PageTransHuge(page)) { |
1184 | for (i = 0, nr = 0; i < HPAGE_PMD_NR; i++) { | |
1185 | if (atomic_inc_and_test(&page[i]._mapcount)) | |
1186 | nr++; | |
1187 | } | |
1188 | if (!atomic_inc_and_test(compound_mapcount_ptr(page))) | |
1189 | goto out; | |
99cb0dbd SL |
1190 | if (PageSwapBacked(page)) |
1191 | __inc_node_page_state(page, NR_SHMEM_PMDMAPPED); | |
1192 | else | |
1193 | __inc_node_page_state(page, NR_FILE_PMDMAPPED); | |
dd78fedd | 1194 | } else { |
c8efc390 KS |
1195 | if (PageTransCompound(page) && page_mapping(page)) { |
1196 | VM_WARN_ON_ONCE(!PageLocked(page)); | |
1197 | ||
9a73f61b KS |
1198 | SetPageDoubleMap(compound_head(page)); |
1199 | if (PageMlocked(page)) | |
1200 | clear_page_mlock(compound_head(page)); | |
1201 | } | |
dd78fedd KS |
1202 | if (!atomic_inc_and_test(&page->_mapcount)) |
1203 | goto out; | |
d69b042f | 1204 | } |
00f3ca2c | 1205 | __mod_lruvec_page_state(page, NR_FILE_MAPPED, nr); |
dd78fedd | 1206 | out: |
62cccb8c | 1207 | unlock_page_memcg(page); |
1da177e4 LT |
1208 | } |
1209 | ||
dd78fedd | 1210 | static void page_remove_file_rmap(struct page *page, bool compound) |
8186eb6a | 1211 | { |
dd78fedd KS |
1212 | int i, nr = 1; |
1213 | ||
57dea93a | 1214 | VM_BUG_ON_PAGE(compound && !PageHead(page), page); |
62cccb8c | 1215 | lock_page_memcg(page); |
8186eb6a | 1216 | |
53f9263b KS |
1217 | /* Hugepages are not counted in NR_FILE_MAPPED for now. */ |
1218 | if (unlikely(PageHuge(page))) { | |
1219 | /* hugetlb pages are always mapped with pmds */ | |
1220 | atomic_dec(compound_mapcount_ptr(page)); | |
8186eb6a | 1221 | goto out; |
53f9263b | 1222 | } |
8186eb6a | 1223 | |
53f9263b | 1224 | /* page still mapped by someone else? */ |
dd78fedd KS |
1225 | if (compound && PageTransHuge(page)) { |
1226 | for (i = 0, nr = 0; i < HPAGE_PMD_NR; i++) { | |
1227 | if (atomic_add_negative(-1, &page[i]._mapcount)) | |
1228 | nr++; | |
1229 | } | |
1230 | if (!atomic_add_negative(-1, compound_mapcount_ptr(page))) | |
1231 | goto out; | |
99cb0dbd SL |
1232 | if (PageSwapBacked(page)) |
1233 | __dec_node_page_state(page, NR_SHMEM_PMDMAPPED); | |
1234 | else | |
1235 | __dec_node_page_state(page, NR_FILE_PMDMAPPED); | |
dd78fedd KS |
1236 | } else { |
1237 | if (!atomic_add_negative(-1, &page->_mapcount)) | |
1238 | goto out; | |
1239 | } | |
8186eb6a JW |
1240 | |
1241 | /* | |
00f3ca2c | 1242 | * We use the irq-unsafe __{inc|mod}_lruvec_page_state because |
8186eb6a JW |
1243 | * these counters are not modified in interrupt context, and |
1244 | * pte lock(a spinlock) is held, which implies preemption disabled. | |
1245 | */ | |
00f3ca2c | 1246 | __mod_lruvec_page_state(page, NR_FILE_MAPPED, -nr); |
8186eb6a JW |
1247 | |
1248 | if (unlikely(PageMlocked(page))) | |
1249 | clear_page_mlock(page); | |
1250 | out: | |
62cccb8c | 1251 | unlock_page_memcg(page); |
8186eb6a JW |
1252 | } |
1253 | ||
53f9263b KS |
1254 | static void page_remove_anon_compound_rmap(struct page *page) |
1255 | { | |
1256 | int i, nr; | |
1257 | ||
1258 | if (!atomic_add_negative(-1, compound_mapcount_ptr(page))) | |
1259 | return; | |
1260 | ||
1261 | /* Hugepages are not counted in NR_ANON_PAGES for now. */ | |
1262 | if (unlikely(PageHuge(page))) | |
1263 | return; | |
1264 | ||
1265 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) | |
1266 | return; | |
1267 | ||
11fb9989 | 1268 | __dec_node_page_state(page, NR_ANON_THPS); |
53f9263b KS |
1269 | |
1270 | if (TestClearPageDoubleMap(page)) { | |
1271 | /* | |
1272 | * Subpages can be mapped with PTEs too. Check how many of | |
1273 | * themi are still mapped. | |
1274 | */ | |
1275 | for (i = 0, nr = 0; i < HPAGE_PMD_NR; i++) { | |
1276 | if (atomic_add_negative(-1, &page[i]._mapcount)) | |
1277 | nr++; | |
1278 | } | |
1279 | } else { | |
1280 | nr = HPAGE_PMD_NR; | |
1281 | } | |
1282 | ||
e90309c9 KS |
1283 | if (unlikely(PageMlocked(page))) |
1284 | clear_page_mlock(page); | |
1285 | ||
9a982250 | 1286 | if (nr) { |
4b9d0fab | 1287 | __mod_node_page_state(page_pgdat(page), NR_ANON_MAPPED, -nr); |
9a982250 KS |
1288 | deferred_split_huge_page(page); |
1289 | } | |
53f9263b KS |
1290 | } |
1291 | ||
1da177e4 LT |
1292 | /** |
1293 | * page_remove_rmap - take down pte mapping from a page | |
d281ee61 KS |
1294 | * @page: page to remove mapping from |
1295 | * @compound: uncharge the page as compound or small page | |
1da177e4 | 1296 | * |
b8072f09 | 1297 | * The caller needs to hold the pte lock. |
1da177e4 | 1298 | */ |
d281ee61 | 1299 | void page_remove_rmap(struct page *page, bool compound) |
1da177e4 | 1300 | { |
dd78fedd KS |
1301 | if (!PageAnon(page)) |
1302 | return page_remove_file_rmap(page, compound); | |
89c06bd5 | 1303 | |
53f9263b KS |
1304 | if (compound) |
1305 | return page_remove_anon_compound_rmap(page); | |
1306 | ||
b904dcfe KM |
1307 | /* page still mapped by someone else? */ |
1308 | if (!atomic_add_negative(-1, &page->_mapcount)) | |
8186eb6a JW |
1309 | return; |
1310 | ||
0fe6e20b | 1311 | /* |
bea04b07 JZ |
1312 | * We use the irq-unsafe __{inc|mod}_zone_page_stat because |
1313 | * these counters are not modified in interrupt context, and | |
bea04b07 | 1314 | * pte lock(a spinlock) is held, which implies preemption disabled. |
0fe6e20b | 1315 | */ |
4b9d0fab | 1316 | __dec_node_page_state(page, NR_ANON_MAPPED); |
8186eb6a | 1317 | |
e6c509f8 HD |
1318 | if (unlikely(PageMlocked(page))) |
1319 | clear_page_mlock(page); | |
8186eb6a | 1320 | |
9a982250 KS |
1321 | if (PageTransCompound(page)) |
1322 | deferred_split_huge_page(compound_head(page)); | |
1323 | ||
b904dcfe KM |
1324 | /* |
1325 | * It would be tidy to reset the PageAnon mapping here, | |
1326 | * but that might overwrite a racing page_add_anon_rmap | |
1327 | * which increments mapcount after us but sets mapping | |
2d4894b5 | 1328 | * before us: so leave the reset to free_unref_page, |
b904dcfe KM |
1329 | * and remember that it's only reliable while mapped. |
1330 | * Leaving it set also helps swapoff to reinstate ptes | |
1331 | * faster for those pages still in swapcache. | |
1332 | */ | |
1da177e4 LT |
1333 | } |
1334 | ||
1335 | /* | |
52629506 | 1336 | * @arg: enum ttu_flags will be passed to this argument |
1da177e4 | 1337 | */ |
e4b82222 | 1338 | static bool try_to_unmap_one(struct page *page, struct vm_area_struct *vma, |
52629506 | 1339 | unsigned long address, void *arg) |
1da177e4 LT |
1340 | { |
1341 | struct mm_struct *mm = vma->vm_mm; | |
c7ab0d2f KS |
1342 | struct page_vma_mapped_walk pvmw = { |
1343 | .page = page, | |
1344 | .vma = vma, | |
1345 | .address = address, | |
1346 | }; | |
1da177e4 | 1347 | pte_t pteval; |
c7ab0d2f | 1348 | struct page *subpage; |
785373b4 | 1349 | bool ret = true; |
ac46d4f3 | 1350 | struct mmu_notifier_range range; |
802a3a92 | 1351 | enum ttu_flags flags = (enum ttu_flags)arg; |
1da177e4 | 1352 | |
10f78390 HD |
1353 | /* |
1354 | * When racing against e.g. zap_pte_range() on another cpu, | |
1355 | * in between its ptep_get_and_clear_full() and page_remove_rmap(), | |
1356 | * try_to_unmap() may return false when it is about to become true, | |
1357 | * if page table locking is skipped: use TTU_SYNC to wait for that. | |
1358 | */ | |
1359 | if (flags & TTU_SYNC) | |
1360 | pvmw.flags = PVMW_SYNC; | |
1361 | ||
b87537d9 HD |
1362 | /* munlock has nothing to gain from examining un-locked vmas */ |
1363 | if ((flags & TTU_MUNLOCK) && !(vma->vm_flags & VM_LOCKED)) | |
e4b82222 | 1364 | return true; |
b87537d9 | 1365 | |
a5430dda JG |
1366 | if (IS_ENABLED(CONFIG_MIGRATION) && (flags & TTU_MIGRATION) && |
1367 | is_zone_device_page(page) && !is_device_private_page(page)) | |
1368 | return true; | |
1369 | ||
fec89c10 KS |
1370 | if (flags & TTU_SPLIT_HUGE_PMD) { |
1371 | split_huge_pmd_address(vma, address, | |
b5ff8161 | 1372 | flags & TTU_SPLIT_FREEZE, page); |
fec89c10 KS |
1373 | } |
1374 | ||
369ea824 | 1375 | /* |
017b1660 MK |
1376 | * For THP, we have to assume the worse case ie pmd for invalidation. |
1377 | * For hugetlb, it could be much worse if we need to do pud | |
1378 | * invalidation in the case of pmd sharing. | |
1379 | * | |
1380 | * Note that the page can not be free in this function as call of | |
1381 | * try_to_unmap() must hold a reference on the page. | |
369ea824 | 1382 | */ |
5bbd1b12 HD |
1383 | range.end = PageKsm(page) ? |
1384 | address + PAGE_SIZE : vma_address_end(page, vma); | |
7269f999 | 1385 | mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma, vma->vm_mm, |
5bbd1b12 | 1386 | address, range.end); |
017b1660 MK |
1387 | if (PageHuge(page)) { |
1388 | /* | |
1389 | * If sharing is possible, start and end will be adjusted | |
1390 | * accordingly. | |
1391 | */ | |
ac46d4f3 JG |
1392 | adjust_range_if_pmd_sharing_possible(vma, &range.start, |
1393 | &range.end); | |
017b1660 | 1394 | } |
ac46d4f3 | 1395 | mmu_notifier_invalidate_range_start(&range); |
369ea824 | 1396 | |
c7ab0d2f | 1397 | while (page_vma_mapped_walk(&pvmw)) { |
616b8371 ZY |
1398 | #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION |
1399 | /* PMD-mapped THP migration entry */ | |
1400 | if (!pvmw.pte && (flags & TTU_MIGRATION)) { | |
1401 | VM_BUG_ON_PAGE(PageHuge(page) || !PageTransCompound(page), page); | |
1402 | ||
616b8371 ZY |
1403 | set_pmd_migration_entry(&pvmw, page); |
1404 | continue; | |
1405 | } | |
1406 | #endif | |
1407 | ||
c7ab0d2f KS |
1408 | /* |
1409 | * If the page is mlock()d, we cannot swap it out. | |
1410 | * If it's recently referenced (perhaps page_referenced | |
1411 | * skipped over this mm) then we should reactivate it. | |
1412 | */ | |
1413 | if (!(flags & TTU_IGNORE_MLOCK)) { | |
1414 | if (vma->vm_flags & VM_LOCKED) { | |
1415 | /* PTE-mapped THP are never mlocked */ | |
1416 | if (!PageTransCompound(page)) { | |
1417 | /* | |
1418 | * Holding pte lock, we do *not* need | |
1419 | * mmap_sem here | |
1420 | */ | |
1421 | mlock_vma_page(page); | |
1422 | } | |
e4b82222 | 1423 | ret = false; |
c7ab0d2f KS |
1424 | page_vma_mapped_walk_done(&pvmw); |
1425 | break; | |
9a73f61b | 1426 | } |
c7ab0d2f KS |
1427 | if (flags & TTU_MUNLOCK) |
1428 | continue; | |
b87537d9 | 1429 | } |
c7ab0d2f | 1430 | |
8346242a KS |
1431 | /* Unexpected PMD-mapped THP? */ |
1432 | VM_BUG_ON_PAGE(!pvmw.pte, page); | |
1433 | ||
1434 | subpage = page - page_to_pfn(page) + pte_pfn(*pvmw.pte); | |
785373b4 LT |
1435 | address = pvmw.address; |
1436 | ||
017b1660 MK |
1437 | if (PageHuge(page)) { |
1438 | if (huge_pmd_unshare(mm, &address, pvmw.pte)) { | |
1439 | /* | |
1440 | * huge_pmd_unshare unmapped an entire PMD | |
1441 | * page. There is no way of knowing exactly | |
1442 | * which PMDs may be cached for this mm, so | |
1443 | * we must flush them all. start/end were | |
1444 | * already adjusted above to cover this range. | |
1445 | */ | |
ac46d4f3 JG |
1446 | flush_cache_range(vma, range.start, range.end); |
1447 | flush_tlb_range(vma, range.start, range.end); | |
1448 | mmu_notifier_invalidate_range(mm, range.start, | |
1449 | range.end); | |
017b1660 MK |
1450 | |
1451 | /* | |
1452 | * The ref count of the PMD page was dropped | |
1453 | * which is part of the way map counting | |
1454 | * is done for shared PMDs. Return 'true' | |
1455 | * here. When there is no other sharing, | |
1456 | * huge_pmd_unshare returns false and we will | |
1457 | * unmap the actual page and drop map count | |
1458 | * to zero. | |
1459 | */ | |
1460 | page_vma_mapped_walk_done(&pvmw); | |
1461 | break; | |
1462 | } | |
1463 | } | |
8346242a | 1464 | |
a5430dda JG |
1465 | if (IS_ENABLED(CONFIG_MIGRATION) && |
1466 | (flags & TTU_MIGRATION) && | |
1467 | is_zone_device_page(page)) { | |
1468 | swp_entry_t entry; | |
1469 | pte_t swp_pte; | |
1470 | ||
1471 | pteval = ptep_get_and_clear(mm, pvmw.address, pvmw.pte); | |
1472 | ||
1473 | /* | |
1474 | * Store the pfn of the page in a special migration | |
1475 | * pte. do_swap_page() will wait until the migration | |
1476 | * pte is removed and then restart fault handling. | |
1477 | */ | |
1478 | entry = make_migration_entry(page, 0); | |
1479 | swp_pte = swp_entry_to_pte(entry); | |
1480 | if (pte_soft_dirty(pteval)) | |
1481 | swp_pte = pte_swp_mksoft_dirty(swp_pte); | |
1482 | set_pte_at(mm, pvmw.address, pvmw.pte, swp_pte); | |
0f10851e JG |
1483 | /* |
1484 | * No need to invalidate here it will synchronize on | |
1485 | * against the special swap migration pte. | |
1de13ee5 RC |
1486 | * |
1487 | * The assignment to subpage above was computed from a | |
1488 | * swap PTE which results in an invalid pointer. | |
1489 | * Since only PAGE_SIZE pages can currently be | |
1490 | * migrated, just set it to page. This will need to be | |
1491 | * changed when hugepage migrations to device private | |
1492 | * memory are supported. | |
0f10851e | 1493 | */ |
1de13ee5 | 1494 | subpage = page; |
a5430dda JG |
1495 | goto discard; |
1496 | } | |
1497 | ||
c7ab0d2f | 1498 | if (!(flags & TTU_IGNORE_ACCESS)) { |
785373b4 | 1499 | if (ptep_clear_flush_young_notify(vma, address, |
c7ab0d2f | 1500 | pvmw.pte)) { |
e4b82222 | 1501 | ret = false; |
c7ab0d2f KS |
1502 | page_vma_mapped_walk_done(&pvmw); |
1503 | break; | |
1504 | } | |
b291f000 | 1505 | } |
1da177e4 | 1506 | |
c7ab0d2f | 1507 | /* Nuke the page table entry. */ |
785373b4 | 1508 | flush_cache_page(vma, address, pte_pfn(*pvmw.pte)); |
c7ab0d2f KS |
1509 | if (should_defer_flush(mm, flags)) { |
1510 | /* | |
1511 | * We clear the PTE but do not flush so potentially | |
1512 | * a remote CPU could still be writing to the page. | |
1513 | * If the entry was previously clean then the | |
1514 | * architecture must guarantee that a clear->dirty | |
1515 | * transition on a cached TLB entry is written through | |
1516 | * and traps if the PTE is unmapped. | |
1517 | */ | |
785373b4 | 1518 | pteval = ptep_get_and_clear(mm, address, pvmw.pte); |
c7ab0d2f KS |
1519 | |
1520 | set_tlb_ubc_flush_pending(mm, pte_dirty(pteval)); | |
1521 | } else { | |
785373b4 | 1522 | pteval = ptep_clear_flush(vma, address, pvmw.pte); |
c7ab0d2f | 1523 | } |
72b252ae | 1524 | |
c7ab0d2f KS |
1525 | /* Move the dirty bit to the page. Now the pte is gone. */ |
1526 | if (pte_dirty(pteval)) | |
1527 | set_page_dirty(page); | |
1da177e4 | 1528 | |
c7ab0d2f KS |
1529 | /* Update high watermark before we lower rss */ |
1530 | update_hiwater_rss(mm); | |
1da177e4 | 1531 | |
c7ab0d2f | 1532 | if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) { |
5fd27b8e | 1533 | pteval = swp_entry_to_pte(make_hwpoison_entry(subpage)); |
c7ab0d2f | 1534 | if (PageHuge(page)) { |
d8c6546b | 1535 | hugetlb_count_sub(compound_nr(page), mm); |
785373b4 | 1536 | set_huge_swap_pte_at(mm, address, |
5fd27b8e PA |
1537 | pvmw.pte, pteval, |
1538 | vma_mmu_pagesize(vma)); | |
c7ab0d2f KS |
1539 | } else { |
1540 | dec_mm_counter(mm, mm_counter(page)); | |
785373b4 | 1541 | set_pte_at(mm, address, pvmw.pte, pteval); |
c7ab0d2f | 1542 | } |
365e9c87 | 1543 | |
bce73e48 | 1544 | } else if (pte_unused(pteval) && !userfaultfd_armed(vma)) { |
c7ab0d2f KS |
1545 | /* |
1546 | * The guest indicated that the page content is of no | |
1547 | * interest anymore. Simply discard the pte, vmscan | |
1548 | * will take care of the rest. | |
bce73e48 CB |
1549 | * A future reference will then fault in a new zero |
1550 | * page. When userfaultfd is active, we must not drop | |
1551 | * this page though, as its main user (postcopy | |
1552 | * migration) will not expect userfaults on already | |
1553 | * copied pages. | |
c7ab0d2f | 1554 | */ |
eca56ff9 | 1555 | dec_mm_counter(mm, mm_counter(page)); |
0f10851e JG |
1556 | /* We have to invalidate as we cleared the pte */ |
1557 | mmu_notifier_invalidate_range(mm, address, | |
1558 | address + PAGE_SIZE); | |
c7ab0d2f | 1559 | } else if (IS_ENABLED(CONFIG_MIGRATION) && |
b5ff8161 | 1560 | (flags & (TTU_MIGRATION|TTU_SPLIT_FREEZE))) { |
c7ab0d2f KS |
1561 | swp_entry_t entry; |
1562 | pte_t swp_pte; | |
ca827d55 KA |
1563 | |
1564 | if (arch_unmap_one(mm, vma, address, pteval) < 0) { | |
1565 | set_pte_at(mm, address, pvmw.pte, pteval); | |
1566 | ret = false; | |
1567 | page_vma_mapped_walk_done(&pvmw); | |
1568 | break; | |
1569 | } | |
1570 | ||
c7ab0d2f KS |
1571 | /* |
1572 | * Store the pfn of the page in a special migration | |
1573 | * pte. do_swap_page() will wait until the migration | |
1574 | * pte is removed and then restart fault handling. | |
1575 | */ | |
1576 | entry = make_migration_entry(subpage, | |
1577 | pte_write(pteval)); | |
1578 | swp_pte = swp_entry_to_pte(entry); | |
1579 | if (pte_soft_dirty(pteval)) | |
1580 | swp_pte = pte_swp_mksoft_dirty(swp_pte); | |
785373b4 | 1581 | set_pte_at(mm, address, pvmw.pte, swp_pte); |
0f10851e JG |
1582 | /* |
1583 | * No need to invalidate here it will synchronize on | |
1584 | * against the special swap migration pte. | |
1585 | */ | |
c7ab0d2f KS |
1586 | } else if (PageAnon(page)) { |
1587 | swp_entry_t entry = { .val = page_private(subpage) }; | |
1588 | pte_t swp_pte; | |
1589 | /* | |
1590 | * Store the swap location in the pte. | |
1591 | * See handle_pte_fault() ... | |
1592 | */ | |
eb94a878 MK |
1593 | if (unlikely(PageSwapBacked(page) != PageSwapCache(page))) { |
1594 | WARN_ON_ONCE(1); | |
83612a94 | 1595 | ret = false; |
369ea824 | 1596 | /* We have to invalidate as we cleared the pte */ |
0f10851e JG |
1597 | mmu_notifier_invalidate_range(mm, address, |
1598 | address + PAGE_SIZE); | |
eb94a878 MK |
1599 | page_vma_mapped_walk_done(&pvmw); |
1600 | break; | |
1601 | } | |
c7ab0d2f | 1602 | |
802a3a92 SL |
1603 | /* MADV_FREE page check */ |
1604 | if (!PageSwapBacked(page)) { | |
1605 | if (!PageDirty(page)) { | |
0f10851e JG |
1606 | /* Invalidate as we cleared the pte */ |
1607 | mmu_notifier_invalidate_range(mm, | |
1608 | address, address + PAGE_SIZE); | |
802a3a92 SL |
1609 | dec_mm_counter(mm, MM_ANONPAGES); |
1610 | goto discard; | |
1611 | } | |
1612 | ||
1613 | /* | |
1614 | * If the page was redirtied, it cannot be | |
1615 | * discarded. Remap the page to page table. | |
1616 | */ | |
785373b4 | 1617 | set_pte_at(mm, address, pvmw.pte, pteval); |
18863d3a | 1618 | SetPageSwapBacked(page); |
e4b82222 | 1619 | ret = false; |
802a3a92 SL |
1620 | page_vma_mapped_walk_done(&pvmw); |
1621 | break; | |
c7ab0d2f | 1622 | } |
854e9ed0 | 1623 | |
c7ab0d2f | 1624 | if (swap_duplicate(entry) < 0) { |
785373b4 | 1625 | set_pte_at(mm, address, pvmw.pte, pteval); |
e4b82222 | 1626 | ret = false; |
c7ab0d2f KS |
1627 | page_vma_mapped_walk_done(&pvmw); |
1628 | break; | |
1629 | } | |
ca827d55 KA |
1630 | if (arch_unmap_one(mm, vma, address, pteval) < 0) { |
1631 | set_pte_at(mm, address, pvmw.pte, pteval); | |
1632 | ret = false; | |
1633 | page_vma_mapped_walk_done(&pvmw); | |
1634 | break; | |
1635 | } | |
c7ab0d2f KS |
1636 | if (list_empty(&mm->mmlist)) { |
1637 | spin_lock(&mmlist_lock); | |
1638 | if (list_empty(&mm->mmlist)) | |
1639 | list_add(&mm->mmlist, &init_mm.mmlist); | |
1640 | spin_unlock(&mmlist_lock); | |
1641 | } | |
854e9ed0 | 1642 | dec_mm_counter(mm, MM_ANONPAGES); |
c7ab0d2f KS |
1643 | inc_mm_counter(mm, MM_SWAPENTS); |
1644 | swp_pte = swp_entry_to_pte(entry); | |
1645 | if (pte_soft_dirty(pteval)) | |
1646 | swp_pte = pte_swp_mksoft_dirty(swp_pte); | |
785373b4 | 1647 | set_pte_at(mm, address, pvmw.pte, swp_pte); |
0f10851e JG |
1648 | /* Invalidate as we cleared the pte */ |
1649 | mmu_notifier_invalidate_range(mm, address, | |
1650 | address + PAGE_SIZE); | |
1651 | } else { | |
1652 | /* | |
906f9cdf HD |
1653 | * This is a locked file-backed page, thus it cannot |
1654 | * be removed from the page cache and replaced by a new | |
1655 | * page before mmu_notifier_invalidate_range_end, so no | |
0f10851e JG |
1656 | * concurrent thread might update its page table to |
1657 | * point at new page while a device still is using this | |
1658 | * page. | |
1659 | * | |
ad56b738 | 1660 | * See Documentation/vm/mmu_notifier.rst |
0f10851e | 1661 | */ |
c7ab0d2f | 1662 | dec_mm_counter(mm, mm_counter_file(page)); |
0f10851e | 1663 | } |
854e9ed0 | 1664 | discard: |
0f10851e JG |
1665 | /* |
1666 | * No need to call mmu_notifier_invalidate_range() it has be | |
1667 | * done above for all cases requiring it to happen under page | |
1668 | * table lock before mmu_notifier_invalidate_range_end() | |
1669 | * | |
ad56b738 | 1670 | * See Documentation/vm/mmu_notifier.rst |
0f10851e | 1671 | */ |
c7ab0d2f KS |
1672 | page_remove_rmap(subpage, PageHuge(page)); |
1673 | put_page(page); | |
c7ab0d2f | 1674 | } |
369ea824 | 1675 | |
ac46d4f3 | 1676 | mmu_notifier_invalidate_range_end(&range); |
369ea824 | 1677 | |
caed0f48 | 1678 | return ret; |
1da177e4 LT |
1679 | } |
1680 | ||
71e3aac0 | 1681 | bool is_vma_temporary_stack(struct vm_area_struct *vma) |
a8bef8ff MG |
1682 | { |
1683 | int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP); | |
1684 | ||
1685 | if (!maybe_stack) | |
1686 | return false; | |
1687 | ||
1688 | if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) == | |
1689 | VM_STACK_INCOMPLETE_SETUP) | |
1690 | return true; | |
1691 | ||
1692 | return false; | |
1693 | } | |
1694 | ||
52629506 JK |
1695 | static bool invalid_migration_vma(struct vm_area_struct *vma, void *arg) |
1696 | { | |
1697 | return is_vma_temporary_stack(vma); | |
1698 | } | |
1699 | ||
42c1bbdc | 1700 | static int page_not_mapped(struct page *page) |
52629506 | 1701 | { |
42c1bbdc | 1702 | return !page_mapped(page); |
2a52bcbc | 1703 | } |
52629506 | 1704 | |
1da177e4 LT |
1705 | /** |
1706 | * try_to_unmap - try to remove all page table mappings to a page | |
1707 | * @page: the page to get unmapped | |
14fa31b8 | 1708 | * @flags: action and flags |
1da177e4 LT |
1709 | * |
1710 | * Tries to remove all the page table entries which are mapping this | |
1711 | * page, used in the pageout path. Caller must hold the page lock. | |
1da177e4 | 1712 | * |
666e5a40 | 1713 | * If unmap is successful, return true. Otherwise, false. |
1da177e4 | 1714 | */ |
666e5a40 | 1715 | bool try_to_unmap(struct page *page, enum ttu_flags flags) |
1da177e4 | 1716 | { |
52629506 JK |
1717 | struct rmap_walk_control rwc = { |
1718 | .rmap_one = try_to_unmap_one, | |
802a3a92 | 1719 | .arg = (void *)flags, |
42c1bbdc | 1720 | .done = page_not_mapped, |
52629506 JK |
1721 | .anon_lock = page_lock_anon_vma_read, |
1722 | }; | |
1da177e4 | 1723 | |
52629506 JK |
1724 | /* |
1725 | * During exec, a temporary VMA is setup and later moved. | |
1726 | * The VMA is moved under the anon_vma lock but not the | |
1727 | * page tables leading to a race where migration cannot | |
1728 | * find the migration ptes. Rather than increasing the | |
1729 | * locking requirements of exec(), migration skips | |
1730 | * temporary VMAs until after exec() completes. | |
1731 | */ | |
b5ff8161 NH |
1732 | if ((flags & (TTU_MIGRATION|TTU_SPLIT_FREEZE)) |
1733 | && !PageKsm(page) && PageAnon(page)) | |
52629506 JK |
1734 | rwc.invalid_vma = invalid_migration_vma; |
1735 | ||
2a52bcbc | 1736 | if (flags & TTU_RMAP_LOCKED) |
33fc80e2 | 1737 | rmap_walk_locked(page, &rwc); |
2a52bcbc | 1738 | else |
33fc80e2 | 1739 | rmap_walk(page, &rwc); |
52629506 | 1740 | |
10f78390 HD |
1741 | /* |
1742 | * When racing against e.g. zap_pte_range() on another cpu, | |
1743 | * in between its ptep_get_and_clear_full() and page_remove_rmap(), | |
1744 | * try_to_unmap() may return false when it is about to become true, | |
1745 | * if page table locking is skipped: use TTU_SYNC to wait for that. | |
1746 | */ | |
1747 | return !page_mapcount(page); | |
1da177e4 | 1748 | } |
81b4082d | 1749 | |
b291f000 NP |
1750 | /** |
1751 | * try_to_munlock - try to munlock a page | |
1752 | * @page: the page to be munlocked | |
1753 | * | |
1754 | * Called from munlock code. Checks all of the VMAs mapping the page | |
1755 | * to make sure nobody else has this page mlocked. The page will be | |
1756 | * returned with PG_mlocked cleared if no other vmas have it mlocked. | |
b291f000 | 1757 | */ |
854e9ed0 | 1758 | |
192d7232 MK |
1759 | void try_to_munlock(struct page *page) |
1760 | { | |
e8351ac9 JK |
1761 | struct rmap_walk_control rwc = { |
1762 | .rmap_one = try_to_unmap_one, | |
802a3a92 | 1763 | .arg = (void *)TTU_MUNLOCK, |
e8351ac9 | 1764 | .done = page_not_mapped, |
e8351ac9 JK |
1765 | .anon_lock = page_lock_anon_vma_read, |
1766 | ||
1767 | }; | |
1768 | ||
309381fe | 1769 | VM_BUG_ON_PAGE(!PageLocked(page) || PageLRU(page), page); |
192d7232 | 1770 | VM_BUG_ON_PAGE(PageCompound(page) && PageDoubleMap(page), page); |
b291f000 | 1771 | |
192d7232 | 1772 | rmap_walk(page, &rwc); |
b291f000 | 1773 | } |
e9995ef9 | 1774 | |
01d8b20d | 1775 | void __put_anon_vma(struct anon_vma *anon_vma) |
76545066 | 1776 | { |
01d8b20d | 1777 | struct anon_vma *root = anon_vma->root; |
76545066 | 1778 | |
624483f3 | 1779 | anon_vma_free(anon_vma); |
01d8b20d PZ |
1780 | if (root != anon_vma && atomic_dec_and_test(&root->refcount)) |
1781 | anon_vma_free(root); | |
76545066 | 1782 | } |
76545066 | 1783 | |
0dd1c7bb JK |
1784 | static struct anon_vma *rmap_walk_anon_lock(struct page *page, |
1785 | struct rmap_walk_control *rwc) | |
faecd8dd JK |
1786 | { |
1787 | struct anon_vma *anon_vma; | |
1788 | ||
0dd1c7bb JK |
1789 | if (rwc->anon_lock) |
1790 | return rwc->anon_lock(page); | |
1791 | ||
faecd8dd JK |
1792 | /* |
1793 | * Note: remove_migration_ptes() cannot use page_lock_anon_vma_read() | |
1794 | * because that depends on page_mapped(); but not all its usages | |
1795 | * are holding mmap_sem. Users without mmap_sem are required to | |
1796 | * take a reference count to prevent the anon_vma disappearing | |
1797 | */ | |
1798 | anon_vma = page_anon_vma(page); | |
1799 | if (!anon_vma) | |
1800 | return NULL; | |
1801 | ||
1802 | anon_vma_lock_read(anon_vma); | |
1803 | return anon_vma; | |
1804 | } | |
1805 | ||
e9995ef9 | 1806 | /* |
e8351ac9 JK |
1807 | * rmap_walk_anon - do something to anonymous page using the object-based |
1808 | * rmap method | |
1809 | * @page: the page to be handled | |
1810 | * @rwc: control variable according to each walk type | |
1811 | * | |
1812 | * Find all the mappings of a page using the mapping pointer and the vma chains | |
1813 | * contained in the anon_vma struct it points to. | |
1814 | * | |
1815 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma | |
1816 | * where the page was found will be held for write. So, we won't recheck | |
1817 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be | |
1818 | * LOCKED. | |
e9995ef9 | 1819 | */ |
1df631ae | 1820 | static void rmap_walk_anon(struct page *page, struct rmap_walk_control *rwc, |
b9773199 | 1821 | bool locked) |
e9995ef9 HD |
1822 | { |
1823 | struct anon_vma *anon_vma; | |
a8fa41ad | 1824 | pgoff_t pgoff_start, pgoff_end; |
5beb4930 | 1825 | struct anon_vma_chain *avc; |
e9995ef9 | 1826 | |
b9773199 KS |
1827 | if (locked) { |
1828 | anon_vma = page_anon_vma(page); | |
1829 | /* anon_vma disappear under us? */ | |
1830 | VM_BUG_ON_PAGE(!anon_vma, page); | |
1831 | } else { | |
1832 | anon_vma = rmap_walk_anon_lock(page, rwc); | |
1833 | } | |
e9995ef9 | 1834 | if (!anon_vma) |
1df631ae | 1835 | return; |
faecd8dd | 1836 | |
a8fa41ad KS |
1837 | pgoff_start = page_to_pgoff(page); |
1838 | pgoff_end = pgoff_start + hpage_nr_pages(page) - 1; | |
1839 | anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, | |
1840 | pgoff_start, pgoff_end) { | |
5beb4930 | 1841 | struct vm_area_struct *vma = avc->vma; |
e9995ef9 | 1842 | unsigned long address = vma_address(page, vma); |
0dd1c7bb | 1843 | |
5bbd1b12 | 1844 | VM_BUG_ON_VMA(address == -EFAULT, vma); |
ad12695f AA |
1845 | cond_resched(); |
1846 | ||
0dd1c7bb JK |
1847 | if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg)) |
1848 | continue; | |
1849 | ||
e4b82222 | 1850 | if (!rwc->rmap_one(page, vma, address, rwc->arg)) |
e9995ef9 | 1851 | break; |
0dd1c7bb JK |
1852 | if (rwc->done && rwc->done(page)) |
1853 | break; | |
e9995ef9 | 1854 | } |
b9773199 KS |
1855 | |
1856 | if (!locked) | |
1857 | anon_vma_unlock_read(anon_vma); | |
e9995ef9 HD |
1858 | } |
1859 | ||
e8351ac9 JK |
1860 | /* |
1861 | * rmap_walk_file - do something to file page using the object-based rmap method | |
1862 | * @page: the page to be handled | |
1863 | * @rwc: control variable according to each walk type | |
1864 | * | |
1865 | * Find all the mappings of a page using the mapping pointer and the vma chains | |
1866 | * contained in the address_space struct it points to. | |
1867 | * | |
1868 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma | |
1869 | * where the page was found will be held for write. So, we won't recheck | |
1870 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be | |
1871 | * LOCKED. | |
1872 | */ | |
1df631ae | 1873 | static void rmap_walk_file(struct page *page, struct rmap_walk_control *rwc, |
b9773199 | 1874 | bool locked) |
e9995ef9 | 1875 | { |
b9773199 | 1876 | struct address_space *mapping = page_mapping(page); |
a8fa41ad | 1877 | pgoff_t pgoff_start, pgoff_end; |
e9995ef9 | 1878 | struct vm_area_struct *vma; |
e9995ef9 | 1879 | |
9f32624b JK |
1880 | /* |
1881 | * The page lock not only makes sure that page->mapping cannot | |
1882 | * suddenly be NULLified by truncation, it makes sure that the | |
1883 | * structure at mapping cannot be freed and reused yet, | |
c8c06efa | 1884 | * so we can safely take mapping->i_mmap_rwsem. |
9f32624b | 1885 | */ |
81d1b09c | 1886 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
9f32624b | 1887 | |
e9995ef9 | 1888 | if (!mapping) |
1df631ae | 1889 | return; |
3dec0ba0 | 1890 | |
a8fa41ad KS |
1891 | pgoff_start = page_to_pgoff(page); |
1892 | pgoff_end = pgoff_start + hpage_nr_pages(page) - 1; | |
b9773199 KS |
1893 | if (!locked) |
1894 | i_mmap_lock_read(mapping); | |
a8fa41ad KS |
1895 | vma_interval_tree_foreach(vma, &mapping->i_mmap, |
1896 | pgoff_start, pgoff_end) { | |
e9995ef9 | 1897 | unsigned long address = vma_address(page, vma); |
0dd1c7bb | 1898 | |
5bbd1b12 | 1899 | VM_BUG_ON_VMA(address == -EFAULT, vma); |
ad12695f AA |
1900 | cond_resched(); |
1901 | ||
0dd1c7bb JK |
1902 | if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg)) |
1903 | continue; | |
1904 | ||
e4b82222 | 1905 | if (!rwc->rmap_one(page, vma, address, rwc->arg)) |
0dd1c7bb JK |
1906 | goto done; |
1907 | if (rwc->done && rwc->done(page)) | |
1908 | goto done; | |
e9995ef9 | 1909 | } |
0dd1c7bb | 1910 | |
0dd1c7bb | 1911 | done: |
b9773199 KS |
1912 | if (!locked) |
1913 | i_mmap_unlock_read(mapping); | |
e9995ef9 HD |
1914 | } |
1915 | ||
1df631ae | 1916 | void rmap_walk(struct page *page, struct rmap_walk_control *rwc) |
e9995ef9 | 1917 | { |
e9995ef9 | 1918 | if (unlikely(PageKsm(page))) |
1df631ae | 1919 | rmap_walk_ksm(page, rwc); |
e9995ef9 | 1920 | else if (PageAnon(page)) |
1df631ae | 1921 | rmap_walk_anon(page, rwc, false); |
b9773199 | 1922 | else |
1df631ae | 1923 | rmap_walk_file(page, rwc, false); |
b9773199 KS |
1924 | } |
1925 | ||
1926 | /* Like rmap_walk, but caller holds relevant rmap lock */ | |
1df631ae | 1927 | void rmap_walk_locked(struct page *page, struct rmap_walk_control *rwc) |
b9773199 KS |
1928 | { |
1929 | /* no ksm support for now */ | |
1930 | VM_BUG_ON_PAGE(PageKsm(page), page); | |
1931 | if (PageAnon(page)) | |
1df631ae | 1932 | rmap_walk_anon(page, rwc, true); |
e9995ef9 | 1933 | else |
1df631ae | 1934 | rmap_walk_file(page, rwc, true); |
e9995ef9 | 1935 | } |
0fe6e20b | 1936 | |
e3390f67 | 1937 | #ifdef CONFIG_HUGETLB_PAGE |
0fe6e20b | 1938 | /* |
451b9514 | 1939 | * The following two functions are for anonymous (private mapped) hugepages. |
0fe6e20b NH |
1940 | * Unlike common anonymous pages, anonymous hugepages have no accounting code |
1941 | * and no lru code, because we handle hugepages differently from common pages. | |
1942 | */ | |
0fe6e20b NH |
1943 | void hugepage_add_anon_rmap(struct page *page, |
1944 | struct vm_area_struct *vma, unsigned long address) | |
1945 | { | |
1946 | struct anon_vma *anon_vma = vma->anon_vma; | |
1947 | int first; | |
a850ea30 NH |
1948 | |
1949 | BUG_ON(!PageLocked(page)); | |
0fe6e20b | 1950 | BUG_ON(!anon_vma); |
5dbe0af4 | 1951 | /* address might be in next vma when migration races vma_adjust */ |
53f9263b | 1952 | first = atomic_inc_and_test(compound_mapcount_ptr(page)); |
0fe6e20b | 1953 | if (first) |
451b9514 | 1954 | __page_set_anon_rmap(page, vma, address, 0); |
0fe6e20b NH |
1955 | } |
1956 | ||
1957 | void hugepage_add_new_anon_rmap(struct page *page, | |
1958 | struct vm_area_struct *vma, unsigned long address) | |
1959 | { | |
1960 | BUG_ON(address < vma->vm_start || address >= vma->vm_end); | |
53f9263b | 1961 | atomic_set(compound_mapcount_ptr(page), 0); |
451b9514 | 1962 | __page_set_anon_rmap(page, vma, address, 1); |
0fe6e20b | 1963 | } |
e3390f67 | 1964 | #endif /* CONFIG_HUGETLB_PAGE */ |