s390/mm: implement software dirty bits
[deliverable/linux.git] / mm / rmap.c
CommitLineData
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)
3d48ae45 26 * mapping->i_mmap_mutex
5a505085 27 * anon_vma->rwsem
82591e6e
NP
28 * mm->page_table_lock or pte_lock
29 * zone->lru_lock (in mark_page_accessed, isolate_lru_page)
30 * swap_lock (in swap_duplicate, swap_info_get)
31 * mmlist_lock (in mmput, drain_mmlist and others)
32 * mapping->private_lock (in __set_page_dirty_buffers)
250df6ed 33 * inode->i_lock (in set_page_dirty's __mark_inode_dirty)
f758eeab 34 * bdi.wb->list_lock (in set_page_dirty's __mark_inode_dirty)
82591e6e
NP
35 * sb_lock (within inode_lock in fs/fs-writeback.c)
36 * mapping->tree_lock (widely used, in set_page_dirty,
37 * in arch-dependent flush_dcache_mmap_lock,
f758eeab 38 * within bdi.wb->list_lock in __sync_single_inode)
6a46079c 39 *
5a505085 40 * anon_vma->rwsem,mapping->i_mutex (memory_failure, collect_procs_anon)
9b679320 41 * ->tasklist_lock
6a46079c 42 * pte map lock
1da177e4
LT
43 */
44
45#include <linux/mm.h>
46#include <linux/pagemap.h>
47#include <linux/swap.h>
48#include <linux/swapops.h>
49#include <linux/slab.h>
50#include <linux/init.h>
5ad64688 51#include <linux/ksm.h>
1da177e4
LT
52#include <linux/rmap.h>
53#include <linux/rcupdate.h>
b95f1b31 54#include <linux/export.h>
8a9f3ccd 55#include <linux/memcontrol.h>
cddb8a5c 56#include <linux/mmu_notifier.h>
64cdd548 57#include <linux/migrate.h>
0fe6e20b 58#include <linux/hugetlb.h>
ef5d437f 59#include <linux/backing-dev.h>
1da177e4
LT
60
61#include <asm/tlbflush.h>
62
b291f000
NP
63#include "internal.h"
64
fdd2e5f8 65static struct kmem_cache *anon_vma_cachep;
5beb4930 66static struct kmem_cache *anon_vma_chain_cachep;
fdd2e5f8
AB
67
68static inline struct anon_vma *anon_vma_alloc(void)
69{
01d8b20d
PZ
70 struct anon_vma *anon_vma;
71
72 anon_vma = kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL);
73 if (anon_vma) {
74 atomic_set(&anon_vma->refcount, 1);
75 /*
76 * Initialise the anon_vma root to point to itself. If called
77 * from fork, the root will be reset to the parents anon_vma.
78 */
79 anon_vma->root = anon_vma;
80 }
81
82 return anon_vma;
fdd2e5f8
AB
83}
84
01d8b20d 85static inline void anon_vma_free(struct anon_vma *anon_vma)
fdd2e5f8 86{
01d8b20d 87 VM_BUG_ON(atomic_read(&anon_vma->refcount));
88c22088
PZ
88
89 /*
4fc3f1d6 90 * Synchronize against page_lock_anon_vma_read() such that
88c22088
PZ
91 * we can safely hold the lock without the anon_vma getting
92 * freed.
93 *
94 * Relies on the full mb implied by the atomic_dec_and_test() from
95 * put_anon_vma() against the acquire barrier implied by
4fc3f1d6 96 * down_read_trylock() from page_lock_anon_vma_read(). This orders:
88c22088 97 *
4fc3f1d6
IM
98 * page_lock_anon_vma_read() VS put_anon_vma()
99 * down_read_trylock() atomic_dec_and_test()
88c22088 100 * LOCK MB
4fc3f1d6 101 * atomic_read() rwsem_is_locked()
88c22088
PZ
102 *
103 * LOCK should suffice since the actual taking of the lock must
104 * happen _before_ what follows.
105 */
5a505085 106 if (rwsem_is_locked(&anon_vma->root->rwsem)) {
4fc3f1d6 107 anon_vma_lock_write(anon_vma);
88c22088
PZ
108 anon_vma_unlock(anon_vma);
109 }
110
fdd2e5f8
AB
111 kmem_cache_free(anon_vma_cachep, anon_vma);
112}
1da177e4 113
dd34739c 114static inline struct anon_vma_chain *anon_vma_chain_alloc(gfp_t gfp)
5beb4930 115{
dd34739c 116 return kmem_cache_alloc(anon_vma_chain_cachep, gfp);
5beb4930
RR
117}
118
e574b5fd 119static void anon_vma_chain_free(struct anon_vma_chain *anon_vma_chain)
5beb4930
RR
120{
121 kmem_cache_free(anon_vma_chain_cachep, anon_vma_chain);
122}
123
6583a843
KC
124static void anon_vma_chain_link(struct vm_area_struct *vma,
125 struct anon_vma_chain *avc,
126 struct anon_vma *anon_vma)
127{
128 avc->vma = vma;
129 avc->anon_vma = anon_vma;
130 list_add(&avc->same_vma, &vma->anon_vma_chain);
bf181b9f 131 anon_vma_interval_tree_insert(avc, &anon_vma->rb_root);
6583a843
KC
132}
133
d9d332e0
LT
134/**
135 * anon_vma_prepare - attach an anon_vma to a memory region
136 * @vma: the memory region in question
137 *
138 * This makes sure the memory mapping described by 'vma' has
139 * an 'anon_vma' attached to it, so that we can associate the
140 * anonymous pages mapped into it with that anon_vma.
141 *
142 * The common case will be that we already have one, but if
23a0790a 143 * not we either need to find an adjacent mapping that we
d9d332e0
LT
144 * can re-use the anon_vma from (very common when the only
145 * reason for splitting a vma has been mprotect()), or we
146 * allocate a new one.
147 *
148 * Anon-vma allocations are very subtle, because we may have
4fc3f1d6 149 * optimistically looked up an anon_vma in page_lock_anon_vma_read()
d9d332e0
LT
150 * and that may actually touch the spinlock even in the newly
151 * allocated vma (it depends on RCU to make sure that the
152 * anon_vma isn't actually destroyed).
153 *
154 * As a result, we need to do proper anon_vma locking even
155 * for the new allocation. At the same time, we do not want
156 * to do any locking for the common case of already having
157 * an anon_vma.
158 *
159 * This must be called with the mmap_sem held for reading.
160 */
1da177e4
LT
161int anon_vma_prepare(struct vm_area_struct *vma)
162{
163 struct anon_vma *anon_vma = vma->anon_vma;
5beb4930 164 struct anon_vma_chain *avc;
1da177e4
LT
165
166 might_sleep();
167 if (unlikely(!anon_vma)) {
168 struct mm_struct *mm = vma->vm_mm;
d9d332e0 169 struct anon_vma *allocated;
1da177e4 170
dd34739c 171 avc = anon_vma_chain_alloc(GFP_KERNEL);
5beb4930
RR
172 if (!avc)
173 goto out_enomem;
174
1da177e4 175 anon_vma = find_mergeable_anon_vma(vma);
d9d332e0
LT
176 allocated = NULL;
177 if (!anon_vma) {
1da177e4
LT
178 anon_vma = anon_vma_alloc();
179 if (unlikely(!anon_vma))
5beb4930 180 goto out_enomem_free_avc;
1da177e4 181 allocated = anon_vma;
1da177e4
LT
182 }
183
4fc3f1d6 184 anon_vma_lock_write(anon_vma);
1da177e4
LT
185 /* page_table_lock to protect against threads */
186 spin_lock(&mm->page_table_lock);
187 if (likely(!vma->anon_vma)) {
188 vma->anon_vma = anon_vma;
6583a843 189 anon_vma_chain_link(vma, avc, anon_vma);
1da177e4 190 allocated = NULL;
31f2b0eb 191 avc = NULL;
1da177e4
LT
192 }
193 spin_unlock(&mm->page_table_lock);
cba48b98 194 anon_vma_unlock(anon_vma);
31f2b0eb
ON
195
196 if (unlikely(allocated))
01d8b20d 197 put_anon_vma(allocated);
31f2b0eb 198 if (unlikely(avc))
5beb4930 199 anon_vma_chain_free(avc);
1da177e4
LT
200 }
201 return 0;
5beb4930
RR
202
203 out_enomem_free_avc:
204 anon_vma_chain_free(avc);
205 out_enomem:
206 return -ENOMEM;
1da177e4
LT
207}
208
bb4aa396
LT
209/*
210 * This is a useful helper function for locking the anon_vma root as
211 * we traverse the vma->anon_vma_chain, looping over anon_vma's that
212 * have the same vma.
213 *
214 * Such anon_vma's should have the same root, so you'd expect to see
215 * just a single mutex_lock for the whole traversal.
216 */
217static inline struct anon_vma *lock_anon_vma_root(struct anon_vma *root, struct anon_vma *anon_vma)
218{
219 struct anon_vma *new_root = anon_vma->root;
220 if (new_root != root) {
221 if (WARN_ON_ONCE(root))
5a505085 222 up_write(&root->rwsem);
bb4aa396 223 root = new_root;
5a505085 224 down_write(&root->rwsem);
bb4aa396
LT
225 }
226 return root;
227}
228
229static inline void unlock_anon_vma_root(struct anon_vma *root)
230{
231 if (root)
5a505085 232 up_write(&root->rwsem);
bb4aa396
LT
233}
234
5beb4930
RR
235/*
236 * Attach the anon_vmas from src to dst.
237 * Returns 0 on success, -ENOMEM on failure.
238 */
239int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src)
1da177e4 240{
5beb4930 241 struct anon_vma_chain *avc, *pavc;
bb4aa396 242 struct anon_vma *root = NULL;
5beb4930 243
646d87b4 244 list_for_each_entry_reverse(pavc, &src->anon_vma_chain, same_vma) {
bb4aa396
LT
245 struct anon_vma *anon_vma;
246
dd34739c
LT
247 avc = anon_vma_chain_alloc(GFP_NOWAIT | __GFP_NOWARN);
248 if (unlikely(!avc)) {
249 unlock_anon_vma_root(root);
250 root = NULL;
251 avc = anon_vma_chain_alloc(GFP_KERNEL);
252 if (!avc)
253 goto enomem_failure;
254 }
bb4aa396
LT
255 anon_vma = pavc->anon_vma;
256 root = lock_anon_vma_root(root, anon_vma);
257 anon_vma_chain_link(dst, avc, anon_vma);
5beb4930 258 }
bb4aa396 259 unlock_anon_vma_root(root);
5beb4930 260 return 0;
1da177e4 261
5beb4930
RR
262 enomem_failure:
263 unlink_anon_vmas(dst);
264 return -ENOMEM;
1da177e4
LT
265}
266
5beb4930
RR
267/*
268 * Attach vma to its own anon_vma, as well as to the anon_vmas that
269 * the corresponding VMA in the parent process is attached to.
270 * Returns 0 on success, non-zero on failure.
271 */
272int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma)
1da177e4 273{
5beb4930
RR
274 struct anon_vma_chain *avc;
275 struct anon_vma *anon_vma;
1da177e4 276
5beb4930
RR
277 /* Don't bother if the parent process has no anon_vma here. */
278 if (!pvma->anon_vma)
279 return 0;
280
281 /*
282 * First, attach the new VMA to the parent VMA's anon_vmas,
283 * so rmap can find non-COWed pages in child processes.
284 */
285 if (anon_vma_clone(vma, pvma))
286 return -ENOMEM;
287
288 /* Then add our own anon_vma. */
289 anon_vma = anon_vma_alloc();
290 if (!anon_vma)
291 goto out_error;
dd34739c 292 avc = anon_vma_chain_alloc(GFP_KERNEL);
5beb4930
RR
293 if (!avc)
294 goto out_error_free_anon_vma;
5c341ee1
RR
295
296 /*
297 * The root anon_vma's spinlock is the lock actually used when we
298 * lock any of the anon_vmas in this anon_vma tree.
299 */
300 anon_vma->root = pvma->anon_vma->root;
76545066 301 /*
01d8b20d
PZ
302 * With refcounts, an anon_vma can stay around longer than the
303 * process it belongs to. The root anon_vma needs to be pinned until
304 * this anon_vma is freed, because the lock lives in the root.
76545066
RR
305 */
306 get_anon_vma(anon_vma->root);
5beb4930
RR
307 /* Mark this anon_vma as the one where our new (COWed) pages go. */
308 vma->anon_vma = anon_vma;
4fc3f1d6 309 anon_vma_lock_write(anon_vma);
5c341ee1 310 anon_vma_chain_link(vma, avc, anon_vma);
bb4aa396 311 anon_vma_unlock(anon_vma);
5beb4930
RR
312
313 return 0;
314
315 out_error_free_anon_vma:
01d8b20d 316 put_anon_vma(anon_vma);
5beb4930 317 out_error:
4946d54c 318 unlink_anon_vmas(vma);
5beb4930 319 return -ENOMEM;
1da177e4
LT
320}
321
5beb4930
RR
322void unlink_anon_vmas(struct vm_area_struct *vma)
323{
324 struct anon_vma_chain *avc, *next;
eee2acba 325 struct anon_vma *root = NULL;
5beb4930 326
5c341ee1
RR
327 /*
328 * Unlink each anon_vma chained to the VMA. This list is ordered
329 * from newest to oldest, ensuring the root anon_vma gets freed last.
330 */
5beb4930 331 list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) {
eee2acba
PZ
332 struct anon_vma *anon_vma = avc->anon_vma;
333
334 root = lock_anon_vma_root(root, anon_vma);
bf181b9f 335 anon_vma_interval_tree_remove(avc, &anon_vma->rb_root);
eee2acba
PZ
336
337 /*
338 * Leave empty anon_vmas on the list - we'll need
339 * to free them outside the lock.
340 */
bf181b9f 341 if (RB_EMPTY_ROOT(&anon_vma->rb_root))
eee2acba
PZ
342 continue;
343
344 list_del(&avc->same_vma);
345 anon_vma_chain_free(avc);
346 }
347 unlock_anon_vma_root(root);
348
349 /*
350 * Iterate the list once more, it now only contains empty and unlinked
351 * anon_vmas, destroy them. Could not do before due to __put_anon_vma()
5a505085 352 * needing to write-acquire the anon_vma->root->rwsem.
eee2acba
PZ
353 */
354 list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) {
355 struct anon_vma *anon_vma = avc->anon_vma;
356
357 put_anon_vma(anon_vma);
358
5beb4930
RR
359 list_del(&avc->same_vma);
360 anon_vma_chain_free(avc);
361 }
362}
363
51cc5068 364static void anon_vma_ctor(void *data)
1da177e4 365{
a35afb83 366 struct anon_vma *anon_vma = data;
1da177e4 367
5a505085 368 init_rwsem(&anon_vma->rwsem);
83813267 369 atomic_set(&anon_vma->refcount, 0);
bf181b9f 370 anon_vma->rb_root = RB_ROOT;
1da177e4
LT
371}
372
373void __init anon_vma_init(void)
374{
375 anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma),
20c2df83 376 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor);
5beb4930 377 anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain, SLAB_PANIC);
1da177e4
LT
378}
379
380/*
6111e4ca
PZ
381 * Getting a lock on a stable anon_vma from a page off the LRU is tricky!
382 *
383 * Since there is no serialization what so ever against page_remove_rmap()
384 * the best this function can do is return a locked anon_vma that might
385 * have been relevant to this page.
386 *
387 * The page might have been remapped to a different anon_vma or the anon_vma
388 * returned may already be freed (and even reused).
389 *
bc658c96
PZ
390 * In case it was remapped to a different anon_vma, the new anon_vma will be a
391 * child of the old anon_vma, and the anon_vma lifetime rules will therefore
392 * ensure that any anon_vma obtained from the page will still be valid for as
393 * long as we observe page_mapped() [ hence all those page_mapped() tests ].
394 *
6111e4ca
PZ
395 * All users of this function must be very careful when walking the anon_vma
396 * chain and verify that the page in question is indeed mapped in it
397 * [ something equivalent to page_mapped_in_vma() ].
398 *
399 * Since anon_vma's slab is DESTROY_BY_RCU and we know from page_remove_rmap()
400 * that the anon_vma pointer from page->mapping is valid if there is a
401 * mapcount, we can dereference the anon_vma after observing those.
1da177e4 402 */
746b18d4 403struct anon_vma *page_get_anon_vma(struct page *page)
1da177e4 404{
746b18d4 405 struct anon_vma *anon_vma = NULL;
1da177e4
LT
406 unsigned long anon_mapping;
407
408 rcu_read_lock();
80e14822 409 anon_mapping = (unsigned long) ACCESS_ONCE(page->mapping);
3ca7b3c5 410 if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
1da177e4
LT
411 goto out;
412 if (!page_mapped(page))
413 goto out;
414
415 anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
746b18d4
PZ
416 if (!atomic_inc_not_zero(&anon_vma->refcount)) {
417 anon_vma = NULL;
418 goto out;
419 }
f1819427
HD
420
421 /*
422 * If this page is still mapped, then its anon_vma cannot have been
746b18d4
PZ
423 * freed. But if it has been unmapped, we have no security against the
424 * anon_vma structure being freed and reused (for another anon_vma:
425 * SLAB_DESTROY_BY_RCU guarantees that - so the atomic_inc_not_zero()
426 * above cannot corrupt).
f1819427 427 */
746b18d4
PZ
428 if (!page_mapped(page)) {
429 put_anon_vma(anon_vma);
430 anon_vma = NULL;
431 }
1da177e4
LT
432out:
433 rcu_read_unlock();
746b18d4
PZ
434
435 return anon_vma;
436}
437
88c22088
PZ
438/*
439 * Similar to page_get_anon_vma() except it locks the anon_vma.
440 *
441 * Its a little more complex as it tries to keep the fast path to a single
442 * atomic op -- the trylock. If we fail the trylock, we fall back to getting a
443 * reference like with page_get_anon_vma() and then block on the mutex.
444 */
4fc3f1d6 445struct anon_vma *page_lock_anon_vma_read(struct page *page)
746b18d4 446{
88c22088 447 struct anon_vma *anon_vma = NULL;
eee0f252 448 struct anon_vma *root_anon_vma;
88c22088 449 unsigned long anon_mapping;
746b18d4 450
88c22088
PZ
451 rcu_read_lock();
452 anon_mapping = (unsigned long) ACCESS_ONCE(page->mapping);
453 if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
454 goto out;
455 if (!page_mapped(page))
456 goto out;
457
458 anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
eee0f252 459 root_anon_vma = ACCESS_ONCE(anon_vma->root);
4fc3f1d6 460 if (down_read_trylock(&root_anon_vma->rwsem)) {
88c22088 461 /*
eee0f252
HD
462 * If the page is still mapped, then this anon_vma is still
463 * its anon_vma, and holding the mutex ensures that it will
bc658c96 464 * not go away, see anon_vma_free().
88c22088 465 */
eee0f252 466 if (!page_mapped(page)) {
4fc3f1d6 467 up_read(&root_anon_vma->rwsem);
88c22088
PZ
468 anon_vma = NULL;
469 }
470 goto out;
471 }
746b18d4 472
88c22088
PZ
473 /* trylock failed, we got to sleep */
474 if (!atomic_inc_not_zero(&anon_vma->refcount)) {
475 anon_vma = NULL;
476 goto out;
477 }
478
479 if (!page_mapped(page)) {
480 put_anon_vma(anon_vma);
481 anon_vma = NULL;
482 goto out;
483 }
484
485 /* we pinned the anon_vma, its safe to sleep */
486 rcu_read_unlock();
4fc3f1d6 487 anon_vma_lock_read(anon_vma);
88c22088
PZ
488
489 if (atomic_dec_and_test(&anon_vma->refcount)) {
490 /*
491 * Oops, we held the last refcount, release the lock
492 * and bail -- can't simply use put_anon_vma() because
4fc3f1d6 493 * we'll deadlock on the anon_vma_lock_write() recursion.
88c22088 494 */
4fc3f1d6 495 anon_vma_unlock_read(anon_vma);
88c22088
PZ
496 __put_anon_vma(anon_vma);
497 anon_vma = NULL;
498 }
499
500 return anon_vma;
501
502out:
503 rcu_read_unlock();
746b18d4 504 return anon_vma;
34bbd704
ON
505}
506
4fc3f1d6 507void page_unlock_anon_vma_read(struct anon_vma *anon_vma)
34bbd704 508{
4fc3f1d6 509 anon_vma_unlock_read(anon_vma);
1da177e4
LT
510}
511
512/*
3ad33b24 513 * At what user virtual address is page expected in @vma?
1da177e4 514 */
86c2ad19
ML
515static inline unsigned long
516__vma_address(struct page *page, struct vm_area_struct *vma)
1da177e4
LT
517{
518 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
1da177e4 519
0fe6e20b
NH
520 if (unlikely(is_vm_hugetlb_page(vma)))
521 pgoff = page->index << huge_page_order(page_hstate(page));
86c2ad19
ML
522
523 return vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
524}
525
526inline unsigned long
527vma_address(struct page *page, struct vm_area_struct *vma)
528{
529 unsigned long address = __vma_address(page, vma);
530
531 /* page should be within @vma mapping range */
532 VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end);
533
1da177e4
LT
534 return address;
535}
536
537/*
bf89c8c8 538 * At what user virtual address is page expected in vma?
ab941e0f 539 * Caller should check the page is actually part of the vma.
1da177e4
LT
540 */
541unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma)
542{
86c2ad19 543 unsigned long address;
21d0d443 544 if (PageAnon(page)) {
4829b906
HD
545 struct anon_vma *page__anon_vma = page_anon_vma(page);
546 /*
547 * Note: swapoff's unuse_vma() is more efficient with this
548 * check, and needs it to match anon_vma when KSM is active.
549 */
550 if (!vma->anon_vma || !page__anon_vma ||
551 vma->anon_vma->root != page__anon_vma->root)
21d0d443
AA
552 return -EFAULT;
553 } else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) {
ee498ed7
HD
554 if (!vma->vm_file ||
555 vma->vm_file->f_mapping != page->mapping)
1da177e4
LT
556 return -EFAULT;
557 } else
558 return -EFAULT;
86c2ad19
ML
559 address = __vma_address(page, vma);
560 if (unlikely(address < vma->vm_start || address >= vma->vm_end))
561 return -EFAULT;
562 return address;
1da177e4
LT
563}
564
6219049a
BL
565pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address)
566{
567 pgd_t *pgd;
568 pud_t *pud;
569 pmd_t *pmd = NULL;
570
571 pgd = pgd_offset(mm, address);
572 if (!pgd_present(*pgd))
573 goto out;
574
575 pud = pud_offset(pgd, address);
576 if (!pud_present(*pud))
577 goto out;
578
579 pmd = pmd_offset(pud, address);
580 if (!pmd_present(*pmd))
581 pmd = NULL;
582out:
583 return pmd;
584}
585
81b4082d
ND
586/*
587 * Check that @page is mapped at @address into @mm.
588 *
479db0bf
NP
589 * If @sync is false, page_check_address may perform a racy check to avoid
590 * the page table lock when the pte is not present (helpful when reclaiming
591 * highly shared pages).
592 *
b8072f09 593 * On success returns with pte mapped and locked.
81b4082d 594 */
e9a81a82 595pte_t *__page_check_address(struct page *page, struct mm_struct *mm,
479db0bf 596 unsigned long address, spinlock_t **ptlp, int sync)
81b4082d 597{
81b4082d
ND
598 pmd_t *pmd;
599 pte_t *pte;
c0718806 600 spinlock_t *ptl;
81b4082d 601
0fe6e20b
NH
602 if (unlikely(PageHuge(page))) {
603 pte = huge_pte_offset(mm, address);
604 ptl = &mm->page_table_lock;
605 goto check;
606 }
607
6219049a
BL
608 pmd = mm_find_pmd(mm, address);
609 if (!pmd)
c0718806
HD
610 return NULL;
611
71e3aac0
AA
612 if (pmd_trans_huge(*pmd))
613 return NULL;
c0718806
HD
614
615 pte = pte_offset_map(pmd, address);
616 /* Make a quick check before getting the lock */
479db0bf 617 if (!sync && !pte_present(*pte)) {
c0718806
HD
618 pte_unmap(pte);
619 return NULL;
620 }
621
4c21e2f2 622 ptl = pte_lockptr(mm, pmd);
0fe6e20b 623check:
c0718806
HD
624 spin_lock(ptl);
625 if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) {
626 *ptlp = ptl;
627 return pte;
81b4082d 628 }
c0718806
HD
629 pte_unmap_unlock(pte, ptl);
630 return NULL;
81b4082d
ND
631}
632
b291f000
NP
633/**
634 * page_mapped_in_vma - check whether a page is really mapped in a VMA
635 * @page: the page to test
636 * @vma: the VMA to test
637 *
638 * Returns 1 if the page is mapped into the page tables of the VMA, 0
639 * if the page is not mapped into the page tables of this VMA. Only
640 * valid for normal file or anonymous VMAs.
641 */
6a46079c 642int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma)
b291f000
NP
643{
644 unsigned long address;
645 pte_t *pte;
646 spinlock_t *ptl;
647
86c2ad19
ML
648 address = __vma_address(page, vma);
649 if (unlikely(address < vma->vm_start || address >= vma->vm_end))
b291f000
NP
650 return 0;
651 pte = page_check_address(page, vma->vm_mm, address, &ptl, 1);
652 if (!pte) /* the page is not in this mm */
653 return 0;
654 pte_unmap_unlock(pte, ptl);
655
656 return 1;
657}
658
1da177e4
LT
659/*
660 * Subfunctions of page_referenced: page_referenced_one called
661 * repeatedly from either page_referenced_anon or page_referenced_file.
662 */
5ad64688
HD
663int page_referenced_one(struct page *page, struct vm_area_struct *vma,
664 unsigned long address, unsigned int *mapcount,
665 unsigned long *vm_flags)
1da177e4
LT
666{
667 struct mm_struct *mm = vma->vm_mm;
1da177e4
LT
668 int referenced = 0;
669
71e3aac0
AA
670 if (unlikely(PageTransHuge(page))) {
671 pmd_t *pmd;
672
673 spin_lock(&mm->page_table_lock);
2da28bfd
AA
674 /*
675 * rmap might return false positives; we must filter
676 * these out using page_check_address_pmd().
677 */
71e3aac0
AA
678 pmd = page_check_address_pmd(page, mm, address,
679 PAGE_CHECK_ADDRESS_PMD_FLAG);
2da28bfd
AA
680 if (!pmd) {
681 spin_unlock(&mm->page_table_lock);
682 goto out;
683 }
684
685 if (vma->vm_flags & VM_LOCKED) {
686 spin_unlock(&mm->page_table_lock);
687 *mapcount = 0; /* break early from loop */
688 *vm_flags |= VM_LOCKED;
689 goto out;
690 }
691
692 /* go ahead even if the pmd is pmd_trans_splitting() */
693 if (pmdp_clear_flush_young_notify(vma, address, pmd))
71e3aac0
AA
694 referenced++;
695 spin_unlock(&mm->page_table_lock);
696 } else {
697 pte_t *pte;
698 spinlock_t *ptl;
699
2da28bfd
AA
700 /*
701 * rmap might return false positives; we must filter
702 * these out using page_check_address().
703 */
71e3aac0
AA
704 pte = page_check_address(page, mm, address, &ptl, 0);
705 if (!pte)
706 goto out;
707
2da28bfd
AA
708 if (vma->vm_flags & VM_LOCKED) {
709 pte_unmap_unlock(pte, ptl);
710 *mapcount = 0; /* break early from loop */
711 *vm_flags |= VM_LOCKED;
712 goto out;
713 }
714
71e3aac0
AA
715 if (ptep_clear_flush_young_notify(vma, address, pte)) {
716 /*
717 * Don't treat a reference through a sequentially read
718 * mapping as such. If the page has been used in
719 * another mapping, we will catch it; if this other
720 * mapping is already gone, the unmap path will have
721 * set PG_referenced or activated the page.
722 */
723 if (likely(!VM_SequentialReadHint(vma)))
724 referenced++;
725 }
726 pte_unmap_unlock(pte, ptl);
727 }
728
c0718806 729 (*mapcount)--;
273f047e 730
6fe6b7e3
WF
731 if (referenced)
732 *vm_flags |= vma->vm_flags;
273f047e 733out:
1da177e4
LT
734 return referenced;
735}
736
bed7161a 737static int page_referenced_anon(struct page *page,
72835c86 738 struct mem_cgroup *memcg,
6fe6b7e3 739 unsigned long *vm_flags)
1da177e4
LT
740{
741 unsigned int mapcount;
742 struct anon_vma *anon_vma;
bf181b9f 743 pgoff_t pgoff;
5beb4930 744 struct anon_vma_chain *avc;
1da177e4
LT
745 int referenced = 0;
746
4fc3f1d6 747 anon_vma = page_lock_anon_vma_read(page);
1da177e4
LT
748 if (!anon_vma)
749 return referenced;
750
751 mapcount = page_mapcount(page);
bf181b9f
ML
752 pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
753 anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) {
5beb4930 754 struct vm_area_struct *vma = avc->vma;
1cb1729b 755 unsigned long address = vma_address(page, vma);
bed7161a
BS
756 /*
757 * If we are reclaiming on behalf of a cgroup, skip
758 * counting on behalf of references from different
759 * cgroups
760 */
72835c86 761 if (memcg && !mm_match_cgroup(vma->vm_mm, memcg))
bed7161a 762 continue;
1cb1729b 763 referenced += page_referenced_one(page, vma, address,
6fe6b7e3 764 &mapcount, vm_flags);
1da177e4
LT
765 if (!mapcount)
766 break;
767 }
34bbd704 768
4fc3f1d6 769 page_unlock_anon_vma_read(anon_vma);
1da177e4
LT
770 return referenced;
771}
772
773/**
774 * page_referenced_file - referenced check for object-based rmap
775 * @page: the page we're checking references on.
72835c86 776 * @memcg: target memory control group
6fe6b7e3 777 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
1da177e4
LT
778 *
779 * For an object-based mapped page, find all the places it is mapped and
780 * check/clear the referenced flag. This is done by following the page->mapping
781 * pointer, then walking the chain of vmas it holds. It returns the number
782 * of references it found.
783 *
784 * This function is only called from page_referenced for object-based pages.
785 */
bed7161a 786static int page_referenced_file(struct page *page,
72835c86 787 struct mem_cgroup *memcg,
6fe6b7e3 788 unsigned long *vm_flags)
1da177e4
LT
789{
790 unsigned int mapcount;
791 struct address_space *mapping = page->mapping;
792 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
793 struct vm_area_struct *vma;
1da177e4
LT
794 int referenced = 0;
795
796 /*
797 * The caller's checks on page->mapping and !PageAnon have made
798 * sure that this is a file page: the check for page->mapping
799 * excludes the case just before it gets set on an anon page.
800 */
801 BUG_ON(PageAnon(page));
802
803 /*
804 * The page lock not only makes sure that page->mapping cannot
805 * suddenly be NULLified by truncation, it makes sure that the
806 * structure at mapping cannot be freed and reused yet,
3d48ae45 807 * so we can safely take mapping->i_mmap_mutex.
1da177e4
LT
808 */
809 BUG_ON(!PageLocked(page));
810
3d48ae45 811 mutex_lock(&mapping->i_mmap_mutex);
1da177e4
LT
812
813 /*
3d48ae45 814 * i_mmap_mutex does not stabilize mapcount at all, but mapcount
1da177e4
LT
815 * is more likely to be accurate if we note it after spinning.
816 */
817 mapcount = page_mapcount(page);
818
6b2dbba8 819 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
1cb1729b 820 unsigned long address = vma_address(page, vma);
bed7161a
BS
821 /*
822 * If we are reclaiming on behalf of a cgroup, skip
823 * counting on behalf of references from different
824 * cgroups
825 */
72835c86 826 if (memcg && !mm_match_cgroup(vma->vm_mm, memcg))
bed7161a 827 continue;
1cb1729b 828 referenced += page_referenced_one(page, vma, address,
6fe6b7e3 829 &mapcount, vm_flags);
1da177e4
LT
830 if (!mapcount)
831 break;
832 }
833
3d48ae45 834 mutex_unlock(&mapping->i_mmap_mutex);
1da177e4
LT
835 return referenced;
836}
837
838/**
839 * page_referenced - test if the page was referenced
840 * @page: the page to test
841 * @is_locked: caller holds lock on the page
72835c86 842 * @memcg: target memory cgroup
6fe6b7e3 843 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
1da177e4
LT
844 *
845 * Quick test_and_clear_referenced for all mappings to a page,
846 * returns the number of ptes which referenced the page.
847 */
6fe6b7e3
WF
848int page_referenced(struct page *page,
849 int is_locked,
72835c86 850 struct mem_cgroup *memcg,
6fe6b7e3 851 unsigned long *vm_flags)
1da177e4
LT
852{
853 int referenced = 0;
5ad64688 854 int we_locked = 0;
1da177e4 855
6fe6b7e3 856 *vm_flags = 0;
3ca7b3c5 857 if (page_mapped(page) && page_rmapping(page)) {
5ad64688
HD
858 if (!is_locked && (!PageAnon(page) || PageKsm(page))) {
859 we_locked = trylock_page(page);
860 if (!we_locked) {
861 referenced++;
862 goto out;
863 }
864 }
865 if (unlikely(PageKsm(page)))
72835c86 866 referenced += page_referenced_ksm(page, memcg,
5ad64688
HD
867 vm_flags);
868 else if (PageAnon(page))
72835c86 869 referenced += page_referenced_anon(page, memcg,
6fe6b7e3 870 vm_flags);
5ad64688 871 else if (page->mapping)
72835c86 872 referenced += page_referenced_file(page, memcg,
6fe6b7e3 873 vm_flags);
5ad64688 874 if (we_locked)
1da177e4 875 unlock_page(page);
50a15981
MS
876
877 if (page_test_and_clear_young(page_to_pfn(page)))
878 referenced++;
1da177e4 879 }
5ad64688 880out:
1da177e4
LT
881 return referenced;
882}
883
1cb1729b
HD
884static int page_mkclean_one(struct page *page, struct vm_area_struct *vma,
885 unsigned long address)
d08b3851
PZ
886{
887 struct mm_struct *mm = vma->vm_mm;
c2fda5fe 888 pte_t *pte;
d08b3851
PZ
889 spinlock_t *ptl;
890 int ret = 0;
891
479db0bf 892 pte = page_check_address(page, mm, address, &ptl, 1);
d08b3851
PZ
893 if (!pte)
894 goto out;
895
c2fda5fe
PZ
896 if (pte_dirty(*pte) || pte_write(*pte)) {
897 pte_t entry;
d08b3851 898
c2fda5fe 899 flush_cache_page(vma, address, pte_pfn(*pte));
2ec74c3e 900 entry = ptep_clear_flush(vma, address, pte);
c2fda5fe
PZ
901 entry = pte_wrprotect(entry);
902 entry = pte_mkclean(entry);
d6e88e67 903 set_pte_at(mm, address, pte, entry);
c2fda5fe
PZ
904 ret = 1;
905 }
d08b3851 906
d08b3851 907 pte_unmap_unlock(pte, ptl);
2ec74c3e
SG
908
909 if (ret)
910 mmu_notifier_invalidate_page(mm, address);
d08b3851
PZ
911out:
912 return ret;
913}
914
915static int page_mkclean_file(struct address_space *mapping, struct page *page)
916{
917 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
918 struct vm_area_struct *vma;
d08b3851
PZ
919 int ret = 0;
920
921 BUG_ON(PageAnon(page));
922
3d48ae45 923 mutex_lock(&mapping->i_mmap_mutex);
6b2dbba8 924 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
1cb1729b
HD
925 if (vma->vm_flags & VM_SHARED) {
926 unsigned long address = vma_address(page, vma);
1cb1729b
HD
927 ret += page_mkclean_one(page, vma, address);
928 }
d08b3851 929 }
3d48ae45 930 mutex_unlock(&mapping->i_mmap_mutex);
d08b3851
PZ
931 return ret;
932}
933
934int page_mkclean(struct page *page)
935{
936 int ret = 0;
937
938 BUG_ON(!PageLocked(page));
939
940 if (page_mapped(page)) {
941 struct address_space *mapping = page_mapping(page);
ef5d437f 942 if (mapping)
d08b3851
PZ
943 ret = page_mkclean_file(mapping, page);
944 }
945
946 return ret;
947}
60b59bea 948EXPORT_SYMBOL_GPL(page_mkclean);
d08b3851 949
c44b6743
RR
950/**
951 * page_move_anon_rmap - move a page to our anon_vma
952 * @page: the page to move to our anon_vma
953 * @vma: the vma the page belongs to
954 * @address: the user virtual address mapped
955 *
956 * When a page belongs exclusively to one process after a COW event,
957 * that page can be moved into the anon_vma that belongs to just that
958 * process, so the rmap code will not search the parent or sibling
959 * processes.
960 */
961void page_move_anon_rmap(struct page *page,
962 struct vm_area_struct *vma, unsigned long address)
963{
964 struct anon_vma *anon_vma = vma->anon_vma;
965
966 VM_BUG_ON(!PageLocked(page));
967 VM_BUG_ON(!anon_vma);
968 VM_BUG_ON(page->index != linear_page_index(vma, address));
969
970 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
971 page->mapping = (struct address_space *) anon_vma;
972}
973
9617d95e 974/**
4e1c1975
AK
975 * __page_set_anon_rmap - set up new anonymous rmap
976 * @page: Page to add to rmap
977 * @vma: VM area to add page to.
978 * @address: User virtual address of the mapping
e8a03feb 979 * @exclusive: the page is exclusively owned by the current process
9617d95e
NP
980 */
981static void __page_set_anon_rmap(struct page *page,
e8a03feb 982 struct vm_area_struct *vma, unsigned long address, int exclusive)
9617d95e 983{
e8a03feb 984 struct anon_vma *anon_vma = vma->anon_vma;
ea90002b 985
e8a03feb 986 BUG_ON(!anon_vma);
ea90002b 987
4e1c1975
AK
988 if (PageAnon(page))
989 return;
990
ea90002b 991 /*
e8a03feb
RR
992 * If the page isn't exclusively mapped into this vma,
993 * we must use the _oldest_ possible anon_vma for the
994 * page mapping!
ea90002b 995 */
4e1c1975 996 if (!exclusive)
288468c3 997 anon_vma = anon_vma->root;
9617d95e 998
9617d95e
NP
999 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
1000 page->mapping = (struct address_space *) anon_vma;
9617d95e 1001 page->index = linear_page_index(vma, address);
9617d95e
NP
1002}
1003
c97a9e10 1004/**
43d8eac4 1005 * __page_check_anon_rmap - sanity check anonymous rmap addition
c97a9e10
NP
1006 * @page: the page to add the mapping to
1007 * @vma: the vm area in which the mapping is added
1008 * @address: the user virtual address mapped
1009 */
1010static void __page_check_anon_rmap(struct page *page,
1011 struct vm_area_struct *vma, unsigned long address)
1012{
1013#ifdef CONFIG_DEBUG_VM
1014 /*
1015 * The page's anon-rmap details (mapping and index) are guaranteed to
1016 * be set up correctly at this point.
1017 *
1018 * We have exclusion against page_add_anon_rmap because the caller
1019 * always holds the page locked, except if called from page_dup_rmap,
1020 * in which case the page is already known to be setup.
1021 *
1022 * We have exclusion against page_add_new_anon_rmap because those pages
1023 * are initially only visible via the pagetables, and the pte is locked
1024 * over the call to page_add_new_anon_rmap.
1025 */
44ab57a0 1026 BUG_ON(page_anon_vma(page)->root != vma->anon_vma->root);
c97a9e10
NP
1027 BUG_ON(page->index != linear_page_index(vma, address));
1028#endif
1029}
1030
1da177e4
LT
1031/**
1032 * page_add_anon_rmap - add pte mapping to an anonymous page
1033 * @page: the page to add the mapping to
1034 * @vma: the vm area in which the mapping is added
1035 * @address: the user virtual address mapped
1036 *
5ad64688 1037 * The caller needs to hold the pte lock, and the page must be locked in
80e14822
HD
1038 * the anon_vma case: to serialize mapping,index checking after setting,
1039 * and to ensure that PageAnon is not being upgraded racily to PageKsm
1040 * (but PageKsm is never downgraded to PageAnon).
1da177e4
LT
1041 */
1042void page_add_anon_rmap(struct page *page,
1043 struct vm_area_struct *vma, unsigned long address)
ad8c2ee8
RR
1044{
1045 do_page_add_anon_rmap(page, vma, address, 0);
1046}
1047
1048/*
1049 * Special version of the above for do_swap_page, which often runs
1050 * into pages that are exclusively owned by the current process.
1051 * Everybody else should continue to use page_add_anon_rmap above.
1052 */
1053void do_page_add_anon_rmap(struct page *page,
1054 struct vm_area_struct *vma, unsigned long address, int exclusive)
1da177e4 1055{
5ad64688 1056 int first = atomic_inc_and_test(&page->_mapcount);
79134171
AA
1057 if (first) {
1058 if (!PageTransHuge(page))
1059 __inc_zone_page_state(page, NR_ANON_PAGES);
1060 else
1061 __inc_zone_page_state(page,
1062 NR_ANON_TRANSPARENT_HUGEPAGES);
1063 }
5ad64688
HD
1064 if (unlikely(PageKsm(page)))
1065 return;
1066
c97a9e10 1067 VM_BUG_ON(!PageLocked(page));
5dbe0af4 1068 /* address might be in next vma when migration races vma_adjust */
5ad64688 1069 if (first)
ad8c2ee8 1070 __page_set_anon_rmap(page, vma, address, exclusive);
69029cd5 1071 else
c97a9e10 1072 __page_check_anon_rmap(page, vma, address);
1da177e4
LT
1073}
1074
43d8eac4 1075/**
9617d95e
NP
1076 * page_add_new_anon_rmap - add pte mapping to a new anonymous page
1077 * @page: the page to add the mapping to
1078 * @vma: the vm area in which the mapping is added
1079 * @address: the user virtual address mapped
1080 *
1081 * Same as page_add_anon_rmap but must only be called on *new* pages.
1082 * This means the inc-and-test can be bypassed.
c97a9e10 1083 * Page does not have to be locked.
9617d95e
NP
1084 */
1085void page_add_new_anon_rmap(struct page *page,
1086 struct vm_area_struct *vma, unsigned long address)
1087{
b5934c53 1088 VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end);
cbf84b7a
HD
1089 SetPageSwapBacked(page);
1090 atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */
79134171
AA
1091 if (!PageTransHuge(page))
1092 __inc_zone_page_state(page, NR_ANON_PAGES);
1093 else
1094 __inc_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES);
e8a03feb 1095 __page_set_anon_rmap(page, vma, address, 1);
39b5f29a 1096 if (!mlocked_vma_newpage(vma, page))
cbf84b7a 1097 lru_cache_add_lru(page, LRU_ACTIVE_ANON);
b5934c53
HD
1098 else
1099 add_page_to_unevictable_list(page);
9617d95e
NP
1100}
1101
1da177e4
LT
1102/**
1103 * page_add_file_rmap - add pte mapping to a file page
1104 * @page: the page to add the mapping to
1105 *
b8072f09 1106 * The caller needs to hold the pte lock.
1da177e4
LT
1107 */
1108void page_add_file_rmap(struct page *page)
1109{
89c06bd5
KH
1110 bool locked;
1111 unsigned long flags;
1112
1113 mem_cgroup_begin_update_page_stat(page, &locked, &flags);
d69b042f 1114 if (atomic_inc_and_test(&page->_mapcount)) {
65ba55f5 1115 __inc_zone_page_state(page, NR_FILE_MAPPED);
2a7106f2 1116 mem_cgroup_inc_page_stat(page, MEMCG_NR_FILE_MAPPED);
d69b042f 1117 }
89c06bd5 1118 mem_cgroup_end_update_page_stat(page, &locked, &flags);
1da177e4
LT
1119}
1120
1121/**
1122 * page_remove_rmap - take down pte mapping from a page
1123 * @page: page to remove mapping from
1124 *
b8072f09 1125 * The caller needs to hold the pte lock.
1da177e4 1126 */
edc315fd 1127void page_remove_rmap(struct page *page)
1da177e4 1128{
89c06bd5
KH
1129 bool anon = PageAnon(page);
1130 bool locked;
1131 unsigned long flags;
1132
1133 /*
1134 * The anon case has no mem_cgroup page_stat to update; but may
1135 * uncharge_page() below, where the lock ordering can deadlock if
1136 * we hold the lock against page_stat move: so avoid it on anon.
1137 */
1138 if (!anon)
1139 mem_cgroup_begin_update_page_stat(page, &locked, &flags);
1140
b904dcfe
KM
1141 /* page still mapped by someone else? */
1142 if (!atomic_add_negative(-1, &page->_mapcount))
89c06bd5 1143 goto out;
b904dcfe 1144
0fe6e20b
NH
1145 /*
1146 * Hugepages are not counted in NR_ANON_PAGES nor NR_FILE_MAPPED
1147 * and not charged by memcg for now.
1148 */
1149 if (unlikely(PageHuge(page)))
89c06bd5
KH
1150 goto out;
1151 if (anon) {
b904dcfe 1152 mem_cgroup_uncharge_page(page);
79134171
AA
1153 if (!PageTransHuge(page))
1154 __dec_zone_page_state(page, NR_ANON_PAGES);
1155 else
1156 __dec_zone_page_state(page,
1157 NR_ANON_TRANSPARENT_HUGEPAGES);
b904dcfe
KM
1158 } else {
1159 __dec_zone_page_state(page, NR_FILE_MAPPED);
2a7106f2 1160 mem_cgroup_dec_page_stat(page, MEMCG_NR_FILE_MAPPED);
e6c509f8 1161 mem_cgroup_end_update_page_stat(page, &locked, &flags);
b904dcfe 1162 }
e6c509f8
HD
1163 if (unlikely(PageMlocked(page)))
1164 clear_page_mlock(page);
b904dcfe
KM
1165 /*
1166 * It would be tidy to reset the PageAnon mapping here,
1167 * but that might overwrite a racing page_add_anon_rmap
1168 * which increments mapcount after us but sets mapping
1169 * before us: so leave the reset to free_hot_cold_page,
1170 * and remember that it's only reliable while mapped.
1171 * Leaving it set also helps swapoff to reinstate ptes
1172 * faster for those pages still in swapcache.
1173 */
e6c509f8 1174 return;
89c06bd5
KH
1175out:
1176 if (!anon)
1177 mem_cgroup_end_update_page_stat(page, &locked, &flags);
1da177e4
LT
1178}
1179
1180/*
1181 * Subfunctions of try_to_unmap: try_to_unmap_one called
99ef0315 1182 * repeatedly from try_to_unmap_ksm, try_to_unmap_anon or try_to_unmap_file.
1da177e4 1183 */
5ad64688
HD
1184int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
1185 unsigned long address, enum ttu_flags flags)
1da177e4
LT
1186{
1187 struct mm_struct *mm = vma->vm_mm;
1da177e4
LT
1188 pte_t *pte;
1189 pte_t pteval;
c0718806 1190 spinlock_t *ptl;
1da177e4
LT
1191 int ret = SWAP_AGAIN;
1192
479db0bf 1193 pte = page_check_address(page, mm, address, &ptl, 0);
c0718806 1194 if (!pte)
81b4082d 1195 goto out;
1da177e4
LT
1196
1197 /*
1198 * If the page is mlock()d, we cannot swap it out.
1199 * If it's recently referenced (perhaps page_referenced
1200 * skipped over this mm) then we should reactivate it.
1201 */
14fa31b8 1202 if (!(flags & TTU_IGNORE_MLOCK)) {
caed0f48
KM
1203 if (vma->vm_flags & VM_LOCKED)
1204 goto out_mlock;
1205
af8e3354 1206 if (TTU_ACTION(flags) == TTU_MUNLOCK)
53f79acb 1207 goto out_unmap;
14fa31b8
AK
1208 }
1209 if (!(flags & TTU_IGNORE_ACCESS)) {
b291f000
NP
1210 if (ptep_clear_flush_young_notify(vma, address, pte)) {
1211 ret = SWAP_FAIL;
1212 goto out_unmap;
1213 }
1214 }
1da177e4 1215
1da177e4
LT
1216 /* Nuke the page table entry. */
1217 flush_cache_page(vma, address, page_to_pfn(page));
2ec74c3e 1218 pteval = ptep_clear_flush(vma, address, pte);
1da177e4
LT
1219
1220 /* Move the dirty bit to the physical page now the pte is gone. */
1221 if (pte_dirty(pteval))
1222 set_page_dirty(page);
1223
365e9c87
HD
1224 /* Update high watermark before we lower rss */
1225 update_hiwater_rss(mm);
1226
888b9f7c 1227 if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) {
5f24ae58
NH
1228 if (!PageHuge(page)) {
1229 if (PageAnon(page))
1230 dec_mm_counter(mm, MM_ANONPAGES);
1231 else
1232 dec_mm_counter(mm, MM_FILEPAGES);
1233 }
888b9f7c 1234 set_pte_at(mm, address, pte,
5f24ae58 1235 swp_entry_to_pte(make_hwpoison_entry(page)));
888b9f7c 1236 } else if (PageAnon(page)) {
4c21e2f2 1237 swp_entry_t entry = { .val = page_private(page) };
0697212a
CL
1238
1239 if (PageSwapCache(page)) {
1240 /*
1241 * Store the swap location in the pte.
1242 * See handle_pte_fault() ...
1243 */
570a335b
HD
1244 if (swap_duplicate(entry) < 0) {
1245 set_pte_at(mm, address, pte, pteval);
1246 ret = SWAP_FAIL;
1247 goto out_unmap;
1248 }
0697212a
CL
1249 if (list_empty(&mm->mmlist)) {
1250 spin_lock(&mmlist_lock);
1251 if (list_empty(&mm->mmlist))
1252 list_add(&mm->mmlist, &init_mm.mmlist);
1253 spin_unlock(&mmlist_lock);
1254 }
d559db08 1255 dec_mm_counter(mm, MM_ANONPAGES);
b084d435 1256 inc_mm_counter(mm, MM_SWAPENTS);
ce1744f4 1257 } else if (IS_ENABLED(CONFIG_MIGRATION)) {
0697212a
CL
1258 /*
1259 * Store the pfn of the page in a special migration
1260 * pte. do_swap_page() will wait until the migration
1261 * pte is removed and then restart fault handling.
1262 */
14fa31b8 1263 BUG_ON(TTU_ACTION(flags) != TTU_MIGRATION);
0697212a 1264 entry = make_migration_entry(page, pte_write(pteval));
1da177e4
LT
1265 }
1266 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
1267 BUG_ON(pte_file(*pte));
ce1744f4
KK
1268 } else if (IS_ENABLED(CONFIG_MIGRATION) &&
1269 (TTU_ACTION(flags) == TTU_MIGRATION)) {
04e62a29
CL
1270 /* Establish migration entry for a file page */
1271 swp_entry_t entry;
1272 entry = make_migration_entry(page, pte_write(pteval));
1273 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
1274 } else
d559db08 1275 dec_mm_counter(mm, MM_FILEPAGES);
1da177e4 1276
edc315fd 1277 page_remove_rmap(page);
1da177e4
LT
1278 page_cache_release(page);
1279
1280out_unmap:
c0718806 1281 pte_unmap_unlock(pte, ptl);
2ec74c3e
SG
1282 if (ret != SWAP_FAIL)
1283 mmu_notifier_invalidate_page(mm, address);
caed0f48
KM
1284out:
1285 return ret;
53f79acb 1286
caed0f48
KM
1287out_mlock:
1288 pte_unmap_unlock(pte, ptl);
1289
1290
1291 /*
1292 * We need mmap_sem locking, Otherwise VM_LOCKED check makes
1293 * unstable result and race. Plus, We can't wait here because
5a505085 1294 * we now hold anon_vma->rwsem or mapping->i_mmap_mutex.
caed0f48
KM
1295 * if trylock failed, the page remain in evictable lru and later
1296 * vmscan could retry to move the page to unevictable lru if the
1297 * page is actually mlocked.
1298 */
1299 if (down_read_trylock(&vma->vm_mm->mmap_sem)) {
1300 if (vma->vm_flags & VM_LOCKED) {
1301 mlock_vma_page(page);
1302 ret = SWAP_MLOCK;
53f79acb 1303 }
caed0f48 1304 up_read(&vma->vm_mm->mmap_sem);
53f79acb 1305 }
1da177e4
LT
1306 return ret;
1307}
1308
1309/*
1310 * objrmap doesn't work for nonlinear VMAs because the assumption that
1311 * offset-into-file correlates with offset-into-virtual-addresses does not hold.
1312 * Consequently, given a particular page and its ->index, we cannot locate the
1313 * ptes which are mapping that page without an exhaustive linear search.
1314 *
1315 * So what this code does is a mini "virtual scan" of each nonlinear VMA which
1316 * maps the file to which the target page belongs. The ->vm_private_data field
1317 * holds the current cursor into that scan. Successive searches will circulate
1318 * around the vma's virtual address space.
1319 *
1320 * So as more replacement pressure is applied to the pages in a nonlinear VMA,
1321 * more scanning pressure is placed against them as well. Eventually pages
1322 * will become fully unmapped and are eligible for eviction.
1323 *
1324 * For very sparsely populated VMAs this is a little inefficient - chances are
1325 * there there won't be many ptes located within the scan cluster. In this case
1326 * maybe we could scan further - to the end of the pte page, perhaps.
b291f000
NP
1327 *
1328 * Mlocked pages: check VM_LOCKED under mmap_sem held for read, if we can
1329 * acquire it without blocking. If vma locked, mlock the pages in the cluster,
1330 * rather than unmapping them. If we encounter the "check_page" that vmscan is
1331 * trying to unmap, return SWAP_MLOCK, else default SWAP_AGAIN.
1da177e4
LT
1332 */
1333#define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE)
1334#define CLUSTER_MASK (~(CLUSTER_SIZE - 1))
1335
b291f000
NP
1336static int try_to_unmap_cluster(unsigned long cursor, unsigned int *mapcount,
1337 struct vm_area_struct *vma, struct page *check_page)
1da177e4
LT
1338{
1339 struct mm_struct *mm = vma->vm_mm;
1da177e4 1340 pmd_t *pmd;
c0718806 1341 pte_t *pte;
1da177e4 1342 pte_t pteval;
c0718806 1343 spinlock_t *ptl;
1da177e4
LT
1344 struct page *page;
1345 unsigned long address;
2ec74c3e
SG
1346 unsigned long mmun_start; /* For mmu_notifiers */
1347 unsigned long mmun_end; /* For mmu_notifiers */
1da177e4 1348 unsigned long end;
b291f000
NP
1349 int ret = SWAP_AGAIN;
1350 int locked_vma = 0;
1da177e4 1351
1da177e4
LT
1352 address = (vma->vm_start + cursor) & CLUSTER_MASK;
1353 end = address + CLUSTER_SIZE;
1354 if (address < vma->vm_start)
1355 address = vma->vm_start;
1356 if (end > vma->vm_end)
1357 end = vma->vm_end;
1358
6219049a
BL
1359 pmd = mm_find_pmd(mm, address);
1360 if (!pmd)
b291f000
NP
1361 return ret;
1362
2ec74c3e
SG
1363 mmun_start = address;
1364 mmun_end = end;
1365 mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
1366
b291f000 1367 /*
af8e3354 1368 * If we can acquire the mmap_sem for read, and vma is VM_LOCKED,
b291f000
NP
1369 * keep the sem while scanning the cluster for mlocking pages.
1370 */
af8e3354 1371 if (down_read_trylock(&vma->vm_mm->mmap_sem)) {
b291f000
NP
1372 locked_vma = (vma->vm_flags & VM_LOCKED);
1373 if (!locked_vma)
1374 up_read(&vma->vm_mm->mmap_sem); /* don't need it */
1375 }
c0718806
HD
1376
1377 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
1da177e4 1378
365e9c87
HD
1379 /* Update high watermark before we lower rss */
1380 update_hiwater_rss(mm);
1381
c0718806 1382 for (; address < end; pte++, address += PAGE_SIZE) {
1da177e4
LT
1383 if (!pte_present(*pte))
1384 continue;
6aab341e
LT
1385 page = vm_normal_page(vma, address, *pte);
1386 BUG_ON(!page || PageAnon(page));
1da177e4 1387
b291f000
NP
1388 if (locked_vma) {
1389 mlock_vma_page(page); /* no-op if already mlocked */
1390 if (page == check_page)
1391 ret = SWAP_MLOCK;
1392 continue; /* don't unmap */
1393 }
1394
cddb8a5c 1395 if (ptep_clear_flush_young_notify(vma, address, pte))
1da177e4
LT
1396 continue;
1397
1398 /* Nuke the page table entry. */
eca35133 1399 flush_cache_page(vma, address, pte_pfn(*pte));
2ec74c3e 1400 pteval = ptep_clear_flush(vma, address, pte);
1da177e4
LT
1401
1402 /* If nonlinear, store the file page offset in the pte. */
1403 if (page->index != linear_page_index(vma, address))
1404 set_pte_at(mm, address, pte, pgoff_to_pte(page->index));
1405
1406 /* Move the dirty bit to the physical page now the pte is gone. */
1407 if (pte_dirty(pteval))
1408 set_page_dirty(page);
1409
edc315fd 1410 page_remove_rmap(page);
1da177e4 1411 page_cache_release(page);
d559db08 1412 dec_mm_counter(mm, MM_FILEPAGES);
1da177e4
LT
1413 (*mapcount)--;
1414 }
c0718806 1415 pte_unmap_unlock(pte - 1, ptl);
2ec74c3e 1416 mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
b291f000
NP
1417 if (locked_vma)
1418 up_read(&vma->vm_mm->mmap_sem);
1419 return ret;
1da177e4
LT
1420}
1421
71e3aac0 1422bool is_vma_temporary_stack(struct vm_area_struct *vma)
a8bef8ff
MG
1423{
1424 int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP);
1425
1426 if (!maybe_stack)
1427 return false;
1428
1429 if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) ==
1430 VM_STACK_INCOMPLETE_SETUP)
1431 return true;
1432
1433 return false;
1434}
1435
b291f000
NP
1436/**
1437 * try_to_unmap_anon - unmap or unlock anonymous page using the object-based
1438 * rmap method
1439 * @page: the page to unmap/unlock
8051be5e 1440 * @flags: action and flags
b291f000
NP
1441 *
1442 * Find all the mappings of a page using the mapping pointer and the vma chains
1443 * contained in the anon_vma struct it points to.
1444 *
1445 * This function is only called from try_to_unmap/try_to_munlock for
1446 * anonymous pages.
1447 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1448 * where the page was found will be held for write. So, we won't recheck
1449 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1450 * 'LOCKED.
1451 */
14fa31b8 1452static int try_to_unmap_anon(struct page *page, enum ttu_flags flags)
1da177e4
LT
1453{
1454 struct anon_vma *anon_vma;
bf181b9f 1455 pgoff_t pgoff;
5beb4930 1456 struct anon_vma_chain *avc;
1da177e4 1457 int ret = SWAP_AGAIN;
b291f000 1458
4fc3f1d6 1459 anon_vma = page_lock_anon_vma_read(page);
1da177e4
LT
1460 if (!anon_vma)
1461 return ret;
1462
bf181b9f
ML
1463 pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
1464 anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) {
5beb4930 1465 struct vm_area_struct *vma = avc->vma;
a8bef8ff
MG
1466 unsigned long address;
1467
1468 /*
1469 * During exec, a temporary VMA is setup and later moved.
1470 * The VMA is moved under the anon_vma lock but not the
1471 * page tables leading to a race where migration cannot
1472 * find the migration ptes. Rather than increasing the
1473 * locking requirements of exec(), migration skips
1474 * temporary VMAs until after exec() completes.
1475 */
ce1744f4 1476 if (IS_ENABLED(CONFIG_MIGRATION) && (flags & TTU_MIGRATION) &&
a8bef8ff
MG
1477 is_vma_temporary_stack(vma))
1478 continue;
1479
1480 address = vma_address(page, vma);
1cb1729b 1481 ret = try_to_unmap_one(page, vma, address, flags);
53f79acb
HD
1482 if (ret != SWAP_AGAIN || !page_mapped(page))
1483 break;
1da177e4 1484 }
34bbd704 1485
4fc3f1d6 1486 page_unlock_anon_vma_read(anon_vma);
1da177e4
LT
1487 return ret;
1488}
1489
1490/**
b291f000
NP
1491 * try_to_unmap_file - unmap/unlock file page using the object-based rmap method
1492 * @page: the page to unmap/unlock
14fa31b8 1493 * @flags: action and flags
1da177e4
LT
1494 *
1495 * Find all the mappings of a page using the mapping pointer and the vma chains
1496 * contained in the address_space struct it points to.
1497 *
b291f000
NP
1498 * This function is only called from try_to_unmap/try_to_munlock for
1499 * object-based pages.
1500 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1501 * where the page was found will be held for write. So, we won't recheck
1502 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1503 * 'LOCKED.
1da177e4 1504 */
14fa31b8 1505static int try_to_unmap_file(struct page *page, enum ttu_flags flags)
1da177e4
LT
1506{
1507 struct address_space *mapping = page->mapping;
1508 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
1509 struct vm_area_struct *vma;
1da177e4
LT
1510 int ret = SWAP_AGAIN;
1511 unsigned long cursor;
1512 unsigned long max_nl_cursor = 0;
1513 unsigned long max_nl_size = 0;
1514 unsigned int mapcount;
1515
3d48ae45 1516 mutex_lock(&mapping->i_mmap_mutex);
6b2dbba8 1517 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
1cb1729b 1518 unsigned long address = vma_address(page, vma);
1cb1729b 1519 ret = try_to_unmap_one(page, vma, address, flags);
53f79acb
HD
1520 if (ret != SWAP_AGAIN || !page_mapped(page))
1521 goto out;
1da177e4
LT
1522 }
1523
1524 if (list_empty(&mapping->i_mmap_nonlinear))
1525 goto out;
1526
53f79acb
HD
1527 /*
1528 * We don't bother to try to find the munlocked page in nonlinears.
1529 * It's costly. Instead, later, page reclaim logic may call
1530 * try_to_unmap(TTU_MUNLOCK) and recover PG_mlocked lazily.
1531 */
1532 if (TTU_ACTION(flags) == TTU_MUNLOCK)
1533 goto out;
1534
1da177e4 1535 list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
6b2dbba8 1536 shared.nonlinear) {
1da177e4
LT
1537 cursor = (unsigned long) vma->vm_private_data;
1538 if (cursor > max_nl_cursor)
1539 max_nl_cursor = cursor;
1540 cursor = vma->vm_end - vma->vm_start;
1541 if (cursor > max_nl_size)
1542 max_nl_size = cursor;
1543 }
1544
b291f000 1545 if (max_nl_size == 0) { /* all nonlinears locked or reserved ? */
1da177e4
LT
1546 ret = SWAP_FAIL;
1547 goto out;
1548 }
1549
1550 /*
1551 * We don't try to search for this page in the nonlinear vmas,
1552 * and page_referenced wouldn't have found it anyway. Instead
1553 * just walk the nonlinear vmas trying to age and unmap some.
1554 * The mapcount of the page we came in with is irrelevant,
1555 * but even so use it as a guide to how hard we should try?
1556 */
1557 mapcount = page_mapcount(page);
1558 if (!mapcount)
1559 goto out;
3d48ae45 1560 cond_resched();
1da177e4
LT
1561
1562 max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK;
1563 if (max_nl_cursor == 0)
1564 max_nl_cursor = CLUSTER_SIZE;
1565
1566 do {
1567 list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
6b2dbba8 1568 shared.nonlinear) {
1da177e4 1569 cursor = (unsigned long) vma->vm_private_data;
839b9685 1570 while ( cursor < max_nl_cursor &&
1da177e4 1571 cursor < vma->vm_end - vma->vm_start) {
53f79acb
HD
1572 if (try_to_unmap_cluster(cursor, &mapcount,
1573 vma, page) == SWAP_MLOCK)
1574 ret = SWAP_MLOCK;
1da177e4
LT
1575 cursor += CLUSTER_SIZE;
1576 vma->vm_private_data = (void *) cursor;
1577 if ((int)mapcount <= 0)
1578 goto out;
1579 }
1580 vma->vm_private_data = (void *) max_nl_cursor;
1581 }
3d48ae45 1582 cond_resched();
1da177e4
LT
1583 max_nl_cursor += CLUSTER_SIZE;
1584 } while (max_nl_cursor <= max_nl_size);
1585
1586 /*
1587 * Don't loop forever (perhaps all the remaining pages are
1588 * in locked vmas). Reset cursor on all unreserved nonlinear
1589 * vmas, now forgetting on which ones it had fallen behind.
1590 */
6b2dbba8 1591 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.nonlinear)
101d2be7 1592 vma->vm_private_data = NULL;
1da177e4 1593out:
3d48ae45 1594 mutex_unlock(&mapping->i_mmap_mutex);
1da177e4
LT
1595 return ret;
1596}
1597
1598/**
1599 * try_to_unmap - try to remove all page table mappings to a page
1600 * @page: the page to get unmapped
14fa31b8 1601 * @flags: action and flags
1da177e4
LT
1602 *
1603 * Tries to remove all the page table entries which are mapping this
1604 * page, used in the pageout path. Caller must hold the page lock.
1605 * Return values are:
1606 *
1607 * SWAP_SUCCESS - we succeeded in removing all mappings
1608 * SWAP_AGAIN - we missed a mapping, try again later
1609 * SWAP_FAIL - the page is unswappable
b291f000 1610 * SWAP_MLOCK - page is mlocked.
1da177e4 1611 */
14fa31b8 1612int try_to_unmap(struct page *page, enum ttu_flags flags)
1da177e4
LT
1613{
1614 int ret;
1615
1da177e4 1616 BUG_ON(!PageLocked(page));
91600e9e 1617 VM_BUG_ON(!PageHuge(page) && PageTransHuge(page));
1da177e4 1618
5ad64688
HD
1619 if (unlikely(PageKsm(page)))
1620 ret = try_to_unmap_ksm(page, flags);
1621 else if (PageAnon(page))
14fa31b8 1622 ret = try_to_unmap_anon(page, flags);
1da177e4 1623 else
14fa31b8 1624 ret = try_to_unmap_file(page, flags);
b291f000 1625 if (ret != SWAP_MLOCK && !page_mapped(page))
1da177e4
LT
1626 ret = SWAP_SUCCESS;
1627 return ret;
1628}
81b4082d 1629
b291f000
NP
1630/**
1631 * try_to_munlock - try to munlock a page
1632 * @page: the page to be munlocked
1633 *
1634 * Called from munlock code. Checks all of the VMAs mapping the page
1635 * to make sure nobody else has this page mlocked. The page will be
1636 * returned with PG_mlocked cleared if no other vmas have it mlocked.
1637 *
1638 * Return values are:
1639 *
53f79acb 1640 * SWAP_AGAIN - no vma is holding page mlocked, or,
b291f000 1641 * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem
5ad64688 1642 * SWAP_FAIL - page cannot be located at present
b291f000
NP
1643 * SWAP_MLOCK - page is now mlocked.
1644 */
1645int try_to_munlock(struct page *page)
1646{
1647 VM_BUG_ON(!PageLocked(page) || PageLRU(page));
1648
5ad64688
HD
1649 if (unlikely(PageKsm(page)))
1650 return try_to_unmap_ksm(page, TTU_MUNLOCK);
1651 else if (PageAnon(page))
14fa31b8 1652 return try_to_unmap_anon(page, TTU_MUNLOCK);
b291f000 1653 else
14fa31b8 1654 return try_to_unmap_file(page, TTU_MUNLOCK);
b291f000 1655}
e9995ef9 1656
01d8b20d 1657void __put_anon_vma(struct anon_vma *anon_vma)
76545066 1658{
01d8b20d 1659 struct anon_vma *root = anon_vma->root;
76545066 1660
01d8b20d
PZ
1661 if (root != anon_vma && atomic_dec_and_test(&root->refcount))
1662 anon_vma_free(root);
76545066 1663
01d8b20d 1664 anon_vma_free(anon_vma);
76545066 1665}
76545066 1666
e9995ef9
HD
1667#ifdef CONFIG_MIGRATION
1668/*
1669 * rmap_walk() and its helpers rmap_walk_anon() and rmap_walk_file():
1670 * Called by migrate.c to remove migration ptes, but might be used more later.
1671 */
1672static int rmap_walk_anon(struct page *page, int (*rmap_one)(struct page *,
1673 struct vm_area_struct *, unsigned long, void *), void *arg)
1674{
1675 struct anon_vma *anon_vma;
bf181b9f 1676 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
5beb4930 1677 struct anon_vma_chain *avc;
e9995ef9
HD
1678 int ret = SWAP_AGAIN;
1679
1680 /*
4fc3f1d6 1681 * Note: remove_migration_ptes() cannot use page_lock_anon_vma_read()
e9995ef9 1682 * because that depends on page_mapped(); but not all its usages
3f6c8272
MG
1683 * are holding mmap_sem. Users without mmap_sem are required to
1684 * take a reference count to prevent the anon_vma disappearing
e9995ef9
HD
1685 */
1686 anon_vma = page_anon_vma(page);
1687 if (!anon_vma)
1688 return ret;
4fc3f1d6 1689 anon_vma_lock_read(anon_vma);
bf181b9f 1690 anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) {
5beb4930 1691 struct vm_area_struct *vma = avc->vma;
e9995ef9 1692 unsigned long address = vma_address(page, vma);
e9995ef9
HD
1693 ret = rmap_one(page, vma, address, arg);
1694 if (ret != SWAP_AGAIN)
1695 break;
1696 }
4fc3f1d6 1697 anon_vma_unlock_read(anon_vma);
e9995ef9
HD
1698 return ret;
1699}
1700
1701static int rmap_walk_file(struct page *page, int (*rmap_one)(struct page *,
1702 struct vm_area_struct *, unsigned long, void *), void *arg)
1703{
1704 struct address_space *mapping = page->mapping;
1705 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
1706 struct vm_area_struct *vma;
e9995ef9
HD
1707 int ret = SWAP_AGAIN;
1708
1709 if (!mapping)
1710 return ret;
3d48ae45 1711 mutex_lock(&mapping->i_mmap_mutex);
6b2dbba8 1712 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
e9995ef9 1713 unsigned long address = vma_address(page, vma);
e9995ef9
HD
1714 ret = rmap_one(page, vma, address, arg);
1715 if (ret != SWAP_AGAIN)
1716 break;
1717 }
1718 /*
1719 * No nonlinear handling: being always shared, nonlinear vmas
1720 * never contain migration ptes. Decide what to do about this
1721 * limitation to linear when we need rmap_walk() on nonlinear.
1722 */
3d48ae45 1723 mutex_unlock(&mapping->i_mmap_mutex);
e9995ef9
HD
1724 return ret;
1725}
1726
1727int rmap_walk(struct page *page, int (*rmap_one)(struct page *,
1728 struct vm_area_struct *, unsigned long, void *), void *arg)
1729{
1730 VM_BUG_ON(!PageLocked(page));
1731
1732 if (unlikely(PageKsm(page)))
1733 return rmap_walk_ksm(page, rmap_one, arg);
1734 else if (PageAnon(page))
1735 return rmap_walk_anon(page, rmap_one, arg);
1736 else
1737 return rmap_walk_file(page, rmap_one, arg);
1738}
1739#endif /* CONFIG_MIGRATION */
0fe6e20b 1740
e3390f67 1741#ifdef CONFIG_HUGETLB_PAGE
0fe6e20b
NH
1742/*
1743 * The following three functions are for anonymous (private mapped) hugepages.
1744 * Unlike common anonymous pages, anonymous hugepages have no accounting code
1745 * and no lru code, because we handle hugepages differently from common pages.
1746 */
1747static void __hugepage_set_anon_rmap(struct page *page,
1748 struct vm_area_struct *vma, unsigned long address, int exclusive)
1749{
1750 struct anon_vma *anon_vma = vma->anon_vma;
433abed6 1751
0fe6e20b 1752 BUG_ON(!anon_vma);
433abed6
NH
1753
1754 if (PageAnon(page))
1755 return;
1756 if (!exclusive)
1757 anon_vma = anon_vma->root;
1758
0fe6e20b
NH
1759 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
1760 page->mapping = (struct address_space *) anon_vma;
1761 page->index = linear_page_index(vma, address);
1762}
1763
1764void hugepage_add_anon_rmap(struct page *page,
1765 struct vm_area_struct *vma, unsigned long address)
1766{
1767 struct anon_vma *anon_vma = vma->anon_vma;
1768 int first;
a850ea30
NH
1769
1770 BUG_ON(!PageLocked(page));
0fe6e20b 1771 BUG_ON(!anon_vma);
5dbe0af4 1772 /* address might be in next vma when migration races vma_adjust */
0fe6e20b
NH
1773 first = atomic_inc_and_test(&page->_mapcount);
1774 if (first)
1775 __hugepage_set_anon_rmap(page, vma, address, 0);
1776}
1777
1778void hugepage_add_new_anon_rmap(struct page *page,
1779 struct vm_area_struct *vma, unsigned long address)
1780{
1781 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
1782 atomic_set(&page->_mapcount, 0);
1783 __hugepage_set_anon_rmap(page, vma, address, 1);
1784}
e3390f67 1785#endif /* CONFIG_HUGETLB_PAGE */
This page took 0.843594 seconds and 5 git commands to generate.