]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - mm/rmap.c
staging: fsl-dpaa2/eth: Only store bpid in priv struct
[mirror_ubuntu-bionic-kernel.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)
88f306b6
KS
26 * hugetlbfs_i_mmap_rwsem_key (in huge_pmd_share)
27 * mapping->i_mmap_rwsem
28 * anon_vma->rwsem
29 * mm->page_table_lock or pte_lock
a52633d8 30 * zone_lru_lock (in mark_page_accessed, isolate_lru_page)
88f306b6
KS
31 * swap_lock (in swap_duplicate, swap_info_get)
32 * mmlist_lock (in mmput, drain_mmlist and others)
33 * mapping->private_lock (in __set_page_dirty_buffers)
34 * mem_cgroup_{begin,end}_page_stat (memcg->move_lock)
35 * mapping->tree_lock (widely used)
36 * inode->i_lock (in set_page_dirty's __mark_inode_dirty)
37 * bdi.wb->list_lock (in set_page_dirty's __mark_inode_dirty)
38 * sb_lock (within inode_lock in fs/fs-writeback.c)
39 * mapping->tree_lock (widely used, in set_page_dirty,
40 * in arch-dependent flush_dcache_mmap_lock,
41 * within bdi.wb->list_lock in __sync_single_inode)
6a46079c 42 *
5a505085 43 * anon_vma->rwsem,mapping->i_mutex (memory_failure, collect_procs_anon)
9b679320 44 * ->tasklist_lock
6a46079c 45 * pte map lock
1da177e4
LT
46 */
47
48#include <linux/mm.h>
6e84f315 49#include <linux/sched/mm.h>
29930025 50#include <linux/sched/task.h>
1da177e4
LT
51#include <linux/pagemap.h>
52#include <linux/swap.h>
53#include <linux/swapops.h>
54#include <linux/slab.h>
55#include <linux/init.h>
5ad64688 56#include <linux/ksm.h>
1da177e4
LT
57#include <linux/rmap.h>
58#include <linux/rcupdate.h>
b95f1b31 59#include <linux/export.h>
8a9f3ccd 60#include <linux/memcontrol.h>
cddb8a5c 61#include <linux/mmu_notifier.h>
64cdd548 62#include <linux/migrate.h>
0fe6e20b 63#include <linux/hugetlb.h>
ef5d437f 64#include <linux/backing-dev.h>
33c3fc71 65#include <linux/page_idle.h>
1da177e4
LT
66
67#include <asm/tlbflush.h>
68
72b252ae
MG
69#include <trace/events/tlb.h>
70
b291f000
NP
71#include "internal.h"
72
fdd2e5f8 73static struct kmem_cache *anon_vma_cachep;
5beb4930 74static struct kmem_cache *anon_vma_chain_cachep;
fdd2e5f8
AB
75
76static inline struct anon_vma *anon_vma_alloc(void)
77{
01d8b20d
PZ
78 struct anon_vma *anon_vma;
79
80 anon_vma = kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL);
81 if (anon_vma) {
82 atomic_set(&anon_vma->refcount, 1);
7a3ef208
KK
83 anon_vma->degree = 1; /* Reference for first vma */
84 anon_vma->parent = anon_vma;
01d8b20d
PZ
85 /*
86 * Initialise the anon_vma root to point to itself. If called
87 * from fork, the root will be reset to the parents anon_vma.
88 */
89 anon_vma->root = anon_vma;
90 }
91
92 return anon_vma;
fdd2e5f8
AB
93}
94
01d8b20d 95static inline void anon_vma_free(struct anon_vma *anon_vma)
fdd2e5f8 96{
01d8b20d 97 VM_BUG_ON(atomic_read(&anon_vma->refcount));
88c22088
PZ
98
99 /*
4fc3f1d6 100 * Synchronize against page_lock_anon_vma_read() such that
88c22088
PZ
101 * we can safely hold the lock without the anon_vma getting
102 * freed.
103 *
104 * Relies on the full mb implied by the atomic_dec_and_test() from
105 * put_anon_vma() against the acquire barrier implied by
4fc3f1d6 106 * down_read_trylock() from page_lock_anon_vma_read(). This orders:
88c22088 107 *
4fc3f1d6
IM
108 * page_lock_anon_vma_read() VS put_anon_vma()
109 * down_read_trylock() atomic_dec_and_test()
88c22088 110 * LOCK MB
4fc3f1d6 111 * atomic_read() rwsem_is_locked()
88c22088
PZ
112 *
113 * LOCK should suffice since the actual taking of the lock must
114 * happen _before_ what follows.
115 */
7f39dda9 116 might_sleep();
5a505085 117 if (rwsem_is_locked(&anon_vma->root->rwsem)) {
4fc3f1d6 118 anon_vma_lock_write(anon_vma);
08b52706 119 anon_vma_unlock_write(anon_vma);
88c22088
PZ
120 }
121
fdd2e5f8
AB
122 kmem_cache_free(anon_vma_cachep, anon_vma);
123}
1da177e4 124
dd34739c 125static inline struct anon_vma_chain *anon_vma_chain_alloc(gfp_t gfp)
5beb4930 126{
dd34739c 127 return kmem_cache_alloc(anon_vma_chain_cachep, gfp);
5beb4930
RR
128}
129
e574b5fd 130static void anon_vma_chain_free(struct anon_vma_chain *anon_vma_chain)
5beb4930
RR
131{
132 kmem_cache_free(anon_vma_chain_cachep, anon_vma_chain);
133}
134
6583a843
KC
135static void anon_vma_chain_link(struct vm_area_struct *vma,
136 struct anon_vma_chain *avc,
137 struct anon_vma *anon_vma)
138{
139 avc->vma = vma;
140 avc->anon_vma = anon_vma;
141 list_add(&avc->same_vma, &vma->anon_vma_chain);
bf181b9f 142 anon_vma_interval_tree_insert(avc, &anon_vma->rb_root);
6583a843
KC
143}
144
d9d332e0 145/**
d5a187da 146 * __anon_vma_prepare - attach an anon_vma to a memory region
d9d332e0
LT
147 * @vma: the memory region in question
148 *
149 * This makes sure the memory mapping described by 'vma' has
150 * an 'anon_vma' attached to it, so that we can associate the
151 * anonymous pages mapped into it with that anon_vma.
152 *
d5a187da
VB
153 * The common case will be that we already have one, which
154 * is handled inline by anon_vma_prepare(). But if
23a0790a 155 * not we either need to find an adjacent mapping that we
d9d332e0
LT
156 * can re-use the anon_vma from (very common when the only
157 * reason for splitting a vma has been mprotect()), or we
158 * allocate a new one.
159 *
160 * Anon-vma allocations are very subtle, because we may have
4fc3f1d6 161 * optimistically looked up an anon_vma in page_lock_anon_vma_read()
d9d332e0
LT
162 * and that may actually touch the spinlock even in the newly
163 * allocated vma (it depends on RCU to make sure that the
164 * anon_vma isn't actually destroyed).
165 *
166 * As a result, we need to do proper anon_vma locking even
167 * for the new allocation. At the same time, we do not want
168 * to do any locking for the common case of already having
169 * an anon_vma.
170 *
171 * This must be called with the mmap_sem held for reading.
172 */
d5a187da 173int __anon_vma_prepare(struct vm_area_struct *vma)
1da177e4 174{
d5a187da
VB
175 struct mm_struct *mm = vma->vm_mm;
176 struct anon_vma *anon_vma, *allocated;
5beb4930 177 struct anon_vma_chain *avc;
1da177e4
LT
178
179 might_sleep();
1da177e4 180
d5a187da
VB
181 avc = anon_vma_chain_alloc(GFP_KERNEL);
182 if (!avc)
183 goto out_enomem;
184
185 anon_vma = find_mergeable_anon_vma(vma);
186 allocated = NULL;
187 if (!anon_vma) {
188 anon_vma = anon_vma_alloc();
189 if (unlikely(!anon_vma))
190 goto out_enomem_free_avc;
191 allocated = anon_vma;
192 }
5beb4930 193
d5a187da
VB
194 anon_vma_lock_write(anon_vma);
195 /* page_table_lock to protect against threads */
196 spin_lock(&mm->page_table_lock);
197 if (likely(!vma->anon_vma)) {
198 vma->anon_vma = anon_vma;
199 anon_vma_chain_link(vma, avc, anon_vma);
200 /* vma reference or self-parent link for new root */
201 anon_vma->degree++;
d9d332e0 202 allocated = NULL;
d5a187da
VB
203 avc = NULL;
204 }
205 spin_unlock(&mm->page_table_lock);
206 anon_vma_unlock_write(anon_vma);
1da177e4 207
d5a187da
VB
208 if (unlikely(allocated))
209 put_anon_vma(allocated);
210 if (unlikely(avc))
211 anon_vma_chain_free(avc);
31f2b0eb 212
1da177e4 213 return 0;
5beb4930
RR
214
215 out_enomem_free_avc:
216 anon_vma_chain_free(avc);
217 out_enomem:
218 return -ENOMEM;
1da177e4
LT
219}
220
bb4aa396
LT
221/*
222 * This is a useful helper function for locking the anon_vma root as
223 * we traverse the vma->anon_vma_chain, looping over anon_vma's that
224 * have the same vma.
225 *
226 * Such anon_vma's should have the same root, so you'd expect to see
227 * just a single mutex_lock for the whole traversal.
228 */
229static inline struct anon_vma *lock_anon_vma_root(struct anon_vma *root, struct anon_vma *anon_vma)
230{
231 struct anon_vma *new_root = anon_vma->root;
232 if (new_root != root) {
233 if (WARN_ON_ONCE(root))
5a505085 234 up_write(&root->rwsem);
bb4aa396 235 root = new_root;
5a505085 236 down_write(&root->rwsem);
bb4aa396
LT
237 }
238 return root;
239}
240
241static inline void unlock_anon_vma_root(struct anon_vma *root)
242{
243 if (root)
5a505085 244 up_write(&root->rwsem);
bb4aa396
LT
245}
246
5beb4930
RR
247/*
248 * Attach the anon_vmas from src to dst.
249 * Returns 0 on success, -ENOMEM on failure.
7a3ef208
KK
250 *
251 * If dst->anon_vma is NULL this function tries to find and reuse existing
252 * anon_vma which has no vmas and only one child anon_vma. This prevents
253 * degradation of anon_vma hierarchy to endless linear chain in case of
254 * constantly forking task. On the other hand, an anon_vma with more than one
255 * child isn't reused even if there was no alive vma, thus rmap walker has a
256 * good chance of avoiding scanning the whole hierarchy when it searches where
257 * page is mapped.
5beb4930
RR
258 */
259int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src)
1da177e4 260{
5beb4930 261 struct anon_vma_chain *avc, *pavc;
bb4aa396 262 struct anon_vma *root = NULL;
5beb4930 263
646d87b4 264 list_for_each_entry_reverse(pavc, &src->anon_vma_chain, same_vma) {
bb4aa396
LT
265 struct anon_vma *anon_vma;
266
dd34739c
LT
267 avc = anon_vma_chain_alloc(GFP_NOWAIT | __GFP_NOWARN);
268 if (unlikely(!avc)) {
269 unlock_anon_vma_root(root);
270 root = NULL;
271 avc = anon_vma_chain_alloc(GFP_KERNEL);
272 if (!avc)
273 goto enomem_failure;
274 }
bb4aa396
LT
275 anon_vma = pavc->anon_vma;
276 root = lock_anon_vma_root(root, anon_vma);
277 anon_vma_chain_link(dst, avc, anon_vma);
7a3ef208
KK
278
279 /*
280 * Reuse existing anon_vma if its degree lower than two,
281 * that means it has no vma and only one anon_vma child.
282 *
283 * Do not chose parent anon_vma, otherwise first child
284 * will always reuse it. Root anon_vma is never reused:
285 * it has self-parent reference and at least one child.
286 */
287 if (!dst->anon_vma && anon_vma != src->anon_vma &&
288 anon_vma->degree < 2)
289 dst->anon_vma = anon_vma;
5beb4930 290 }
7a3ef208
KK
291 if (dst->anon_vma)
292 dst->anon_vma->degree++;
bb4aa396 293 unlock_anon_vma_root(root);
5beb4930 294 return 0;
1da177e4 295
5beb4930 296 enomem_failure:
3fe89b3e
LY
297 /*
298 * dst->anon_vma is dropped here otherwise its degree can be incorrectly
299 * decremented in unlink_anon_vmas().
300 * We can safely do this because callers of anon_vma_clone() don't care
301 * about dst->anon_vma if anon_vma_clone() failed.
302 */
303 dst->anon_vma = NULL;
5beb4930
RR
304 unlink_anon_vmas(dst);
305 return -ENOMEM;
1da177e4
LT
306}
307
5beb4930
RR
308/*
309 * Attach vma to its own anon_vma, as well as to the anon_vmas that
310 * the corresponding VMA in the parent process is attached to.
311 * Returns 0 on success, non-zero on failure.
312 */
313int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma)
1da177e4 314{
5beb4930
RR
315 struct anon_vma_chain *avc;
316 struct anon_vma *anon_vma;
c4ea95d7 317 int error;
1da177e4 318
5beb4930
RR
319 /* Don't bother if the parent process has no anon_vma here. */
320 if (!pvma->anon_vma)
321 return 0;
322
7a3ef208
KK
323 /* Drop inherited anon_vma, we'll reuse existing or allocate new. */
324 vma->anon_vma = NULL;
325
5beb4930
RR
326 /*
327 * First, attach the new VMA to the parent VMA's anon_vmas,
328 * so rmap can find non-COWed pages in child processes.
329 */
c4ea95d7
DF
330 error = anon_vma_clone(vma, pvma);
331 if (error)
332 return error;
5beb4930 333
7a3ef208
KK
334 /* An existing anon_vma has been reused, all done then. */
335 if (vma->anon_vma)
336 return 0;
337
5beb4930
RR
338 /* Then add our own anon_vma. */
339 anon_vma = anon_vma_alloc();
340 if (!anon_vma)
341 goto out_error;
dd34739c 342 avc = anon_vma_chain_alloc(GFP_KERNEL);
5beb4930
RR
343 if (!avc)
344 goto out_error_free_anon_vma;
5c341ee1
RR
345
346 /*
347 * The root anon_vma's spinlock is the lock actually used when we
348 * lock any of the anon_vmas in this anon_vma tree.
349 */
350 anon_vma->root = pvma->anon_vma->root;
7a3ef208 351 anon_vma->parent = pvma->anon_vma;
76545066 352 /*
01d8b20d
PZ
353 * With refcounts, an anon_vma can stay around longer than the
354 * process it belongs to. The root anon_vma needs to be pinned until
355 * this anon_vma is freed, because the lock lives in the root.
76545066
RR
356 */
357 get_anon_vma(anon_vma->root);
5beb4930
RR
358 /* Mark this anon_vma as the one where our new (COWed) pages go. */
359 vma->anon_vma = anon_vma;
4fc3f1d6 360 anon_vma_lock_write(anon_vma);
5c341ee1 361 anon_vma_chain_link(vma, avc, anon_vma);
7a3ef208 362 anon_vma->parent->degree++;
08b52706 363 anon_vma_unlock_write(anon_vma);
5beb4930
RR
364
365 return 0;
366
367 out_error_free_anon_vma:
01d8b20d 368 put_anon_vma(anon_vma);
5beb4930 369 out_error:
4946d54c 370 unlink_anon_vmas(vma);
5beb4930 371 return -ENOMEM;
1da177e4
LT
372}
373
5beb4930
RR
374void unlink_anon_vmas(struct vm_area_struct *vma)
375{
376 struct anon_vma_chain *avc, *next;
eee2acba 377 struct anon_vma *root = NULL;
5beb4930 378
5c341ee1
RR
379 /*
380 * Unlink each anon_vma chained to the VMA. This list is ordered
381 * from newest to oldest, ensuring the root anon_vma gets freed last.
382 */
5beb4930 383 list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) {
eee2acba
PZ
384 struct anon_vma *anon_vma = avc->anon_vma;
385
386 root = lock_anon_vma_root(root, anon_vma);
bf181b9f 387 anon_vma_interval_tree_remove(avc, &anon_vma->rb_root);
eee2acba
PZ
388
389 /*
390 * Leave empty anon_vmas on the list - we'll need
391 * to free them outside the lock.
392 */
7a3ef208
KK
393 if (RB_EMPTY_ROOT(&anon_vma->rb_root)) {
394 anon_vma->parent->degree--;
eee2acba 395 continue;
7a3ef208 396 }
eee2acba
PZ
397
398 list_del(&avc->same_vma);
399 anon_vma_chain_free(avc);
400 }
7a3ef208
KK
401 if (vma->anon_vma)
402 vma->anon_vma->degree--;
eee2acba
PZ
403 unlock_anon_vma_root(root);
404
405 /*
406 * Iterate the list once more, it now only contains empty and unlinked
407 * anon_vmas, destroy them. Could not do before due to __put_anon_vma()
5a505085 408 * needing to write-acquire the anon_vma->root->rwsem.
eee2acba
PZ
409 */
410 list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) {
411 struct anon_vma *anon_vma = avc->anon_vma;
412
e4c5800a 413 VM_WARN_ON(anon_vma->degree);
eee2acba
PZ
414 put_anon_vma(anon_vma);
415
5beb4930
RR
416 list_del(&avc->same_vma);
417 anon_vma_chain_free(avc);
418 }
419}
420
51cc5068 421static void anon_vma_ctor(void *data)
1da177e4 422{
a35afb83 423 struct anon_vma *anon_vma = data;
1da177e4 424
5a505085 425 init_rwsem(&anon_vma->rwsem);
83813267 426 atomic_set(&anon_vma->refcount, 0);
bf181b9f 427 anon_vma->rb_root = RB_ROOT;
1da177e4
LT
428}
429
430void __init anon_vma_init(void)
431{
432 anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma),
5f0d5a3a 433 0, SLAB_TYPESAFE_BY_RCU|SLAB_PANIC|SLAB_ACCOUNT,
5d097056
VD
434 anon_vma_ctor);
435 anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain,
436 SLAB_PANIC|SLAB_ACCOUNT);
1da177e4
LT
437}
438
439/*
6111e4ca
PZ
440 * Getting a lock on a stable anon_vma from a page off the LRU is tricky!
441 *
442 * Since there is no serialization what so ever against page_remove_rmap()
443 * the best this function can do is return a locked anon_vma that might
444 * have been relevant to this page.
445 *
446 * The page might have been remapped to a different anon_vma or the anon_vma
447 * returned may already be freed (and even reused).
448 *
bc658c96
PZ
449 * In case it was remapped to a different anon_vma, the new anon_vma will be a
450 * child of the old anon_vma, and the anon_vma lifetime rules will therefore
451 * ensure that any anon_vma obtained from the page will still be valid for as
452 * long as we observe page_mapped() [ hence all those page_mapped() tests ].
453 *
6111e4ca
PZ
454 * All users of this function must be very careful when walking the anon_vma
455 * chain and verify that the page in question is indeed mapped in it
456 * [ something equivalent to page_mapped_in_vma() ].
457 *
458 * Since anon_vma's slab is DESTROY_BY_RCU and we know from page_remove_rmap()
459 * that the anon_vma pointer from page->mapping is valid if there is a
460 * mapcount, we can dereference the anon_vma after observing those.
1da177e4 461 */
746b18d4 462struct anon_vma *page_get_anon_vma(struct page *page)
1da177e4 463{
746b18d4 464 struct anon_vma *anon_vma = NULL;
1da177e4
LT
465 unsigned long anon_mapping;
466
467 rcu_read_lock();
4db0c3c2 468 anon_mapping = (unsigned long)READ_ONCE(page->mapping);
3ca7b3c5 469 if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
1da177e4
LT
470 goto out;
471 if (!page_mapped(page))
472 goto out;
473
474 anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
746b18d4
PZ
475 if (!atomic_inc_not_zero(&anon_vma->refcount)) {
476 anon_vma = NULL;
477 goto out;
478 }
f1819427
HD
479
480 /*
481 * If this page is still mapped, then its anon_vma cannot have been
746b18d4
PZ
482 * freed. But if it has been unmapped, we have no security against the
483 * anon_vma structure being freed and reused (for another anon_vma:
5f0d5a3a 484 * SLAB_TYPESAFE_BY_RCU guarantees that - so the atomic_inc_not_zero()
746b18d4 485 * above cannot corrupt).
f1819427 486 */
746b18d4 487 if (!page_mapped(page)) {
7f39dda9 488 rcu_read_unlock();
746b18d4 489 put_anon_vma(anon_vma);
7f39dda9 490 return NULL;
746b18d4 491 }
1da177e4
LT
492out:
493 rcu_read_unlock();
746b18d4
PZ
494
495 return anon_vma;
496}
497
88c22088
PZ
498/*
499 * Similar to page_get_anon_vma() except it locks the anon_vma.
500 *
501 * Its a little more complex as it tries to keep the fast path to a single
502 * atomic op -- the trylock. If we fail the trylock, we fall back to getting a
503 * reference like with page_get_anon_vma() and then block on the mutex.
504 */
4fc3f1d6 505struct anon_vma *page_lock_anon_vma_read(struct page *page)
746b18d4 506{
88c22088 507 struct anon_vma *anon_vma = NULL;
eee0f252 508 struct anon_vma *root_anon_vma;
88c22088 509 unsigned long anon_mapping;
746b18d4 510
88c22088 511 rcu_read_lock();
4db0c3c2 512 anon_mapping = (unsigned long)READ_ONCE(page->mapping);
88c22088
PZ
513 if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
514 goto out;
515 if (!page_mapped(page))
516 goto out;
517
518 anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
4db0c3c2 519 root_anon_vma = READ_ONCE(anon_vma->root);
4fc3f1d6 520 if (down_read_trylock(&root_anon_vma->rwsem)) {
88c22088 521 /*
eee0f252
HD
522 * If the page is still mapped, then this anon_vma is still
523 * its anon_vma, and holding the mutex ensures that it will
bc658c96 524 * not go away, see anon_vma_free().
88c22088 525 */
eee0f252 526 if (!page_mapped(page)) {
4fc3f1d6 527 up_read(&root_anon_vma->rwsem);
88c22088
PZ
528 anon_vma = NULL;
529 }
530 goto out;
531 }
746b18d4 532
88c22088
PZ
533 /* trylock failed, we got to sleep */
534 if (!atomic_inc_not_zero(&anon_vma->refcount)) {
535 anon_vma = NULL;
536 goto out;
537 }
538
539 if (!page_mapped(page)) {
7f39dda9 540 rcu_read_unlock();
88c22088 541 put_anon_vma(anon_vma);
7f39dda9 542 return NULL;
88c22088
PZ
543 }
544
545 /* we pinned the anon_vma, its safe to sleep */
546 rcu_read_unlock();
4fc3f1d6 547 anon_vma_lock_read(anon_vma);
88c22088
PZ
548
549 if (atomic_dec_and_test(&anon_vma->refcount)) {
550 /*
551 * Oops, we held the last refcount, release the lock
552 * and bail -- can't simply use put_anon_vma() because
4fc3f1d6 553 * we'll deadlock on the anon_vma_lock_write() recursion.
88c22088 554 */
4fc3f1d6 555 anon_vma_unlock_read(anon_vma);
88c22088
PZ
556 __put_anon_vma(anon_vma);
557 anon_vma = NULL;
558 }
559
560 return anon_vma;
561
562out:
563 rcu_read_unlock();
746b18d4 564 return anon_vma;
34bbd704
ON
565}
566
4fc3f1d6 567void page_unlock_anon_vma_read(struct anon_vma *anon_vma)
34bbd704 568{
4fc3f1d6 569 anon_vma_unlock_read(anon_vma);
1da177e4
LT
570}
571
72b252ae 572#ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
72b252ae
MG
573/*
574 * Flush TLB entries for recently unmapped pages from remote CPUs. It is
575 * important if a PTE was dirty when it was unmapped that it's flushed
576 * before any IO is initiated on the page to prevent lost writes. Similarly,
577 * it must be flushed before freeing to prevent data leakage.
578 */
579void try_to_unmap_flush(void)
580{
581 struct tlbflush_unmap_batch *tlb_ubc = &current->tlb_ubc;
582 int cpu;
583
584 if (!tlb_ubc->flush_required)
585 return;
586
587 cpu = get_cpu();
588
858eaaa7
NA
589 if (cpumask_test_cpu(cpu, &tlb_ubc->cpumask)) {
590 count_vm_tlb_event(NR_TLB_LOCAL_FLUSH_ALL);
591 local_flush_tlb();
592 trace_tlb_flush(TLB_LOCAL_SHOOTDOWN, TLB_FLUSH_ALL);
72b252ae 593 }
858eaaa7
NA
594
595 if (cpumask_any_but(&tlb_ubc->cpumask, cpu) < nr_cpu_ids)
596 flush_tlb_others(&tlb_ubc->cpumask, NULL, 0, TLB_FLUSH_ALL);
72b252ae
MG
597 cpumask_clear(&tlb_ubc->cpumask);
598 tlb_ubc->flush_required = false;
d950c947 599 tlb_ubc->writable = false;
72b252ae
MG
600 put_cpu();
601}
602
d950c947
MG
603/* Flush iff there are potentially writable TLB entries that can race with IO */
604void try_to_unmap_flush_dirty(void)
605{
606 struct tlbflush_unmap_batch *tlb_ubc = &current->tlb_ubc;
607
608 if (tlb_ubc->writable)
609 try_to_unmap_flush();
610}
611
c7ab0d2f 612static void set_tlb_ubc_flush_pending(struct mm_struct *mm, bool writable)
72b252ae
MG
613{
614 struct tlbflush_unmap_batch *tlb_ubc = &current->tlb_ubc;
615
616 cpumask_or(&tlb_ubc->cpumask, &tlb_ubc->cpumask, mm_cpumask(mm));
617 tlb_ubc->flush_required = true;
d950c947
MG
618
619 /*
620 * If the PTE was dirty then it's best to assume it's writable. The
621 * caller must use try_to_unmap_flush_dirty() or try_to_unmap_flush()
622 * before the page is queued for IO.
623 */
624 if (writable)
625 tlb_ubc->writable = true;
72b252ae
MG
626}
627
628/*
629 * Returns true if the TLB flush should be deferred to the end of a batch of
630 * unmap operations to reduce IPIs.
631 */
632static bool should_defer_flush(struct mm_struct *mm, enum ttu_flags flags)
633{
634 bool should_defer = false;
635
636 if (!(flags & TTU_BATCH_FLUSH))
637 return false;
638
639 /* If remote CPUs need to be flushed then defer batch the flush */
640 if (cpumask_any_but(mm_cpumask(mm), get_cpu()) < nr_cpu_ids)
641 should_defer = true;
642 put_cpu();
643
644 return should_defer;
645}
646#else
c7ab0d2f 647static void set_tlb_ubc_flush_pending(struct mm_struct *mm, bool writable)
72b252ae
MG
648{
649}
650
651static bool should_defer_flush(struct mm_struct *mm, enum ttu_flags flags)
652{
653 return false;
654}
655#endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
656
1da177e4 657/*
bf89c8c8 658 * At what user virtual address is page expected in vma?
ab941e0f 659 * Caller should check the page is actually part of the vma.
1da177e4
LT
660 */
661unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma)
662{
86c2ad19 663 unsigned long address;
21d0d443 664 if (PageAnon(page)) {
4829b906
HD
665 struct anon_vma *page__anon_vma = page_anon_vma(page);
666 /*
667 * Note: swapoff's unuse_vma() is more efficient with this
668 * check, and needs it to match anon_vma when KSM is active.
669 */
670 if (!vma->anon_vma || !page__anon_vma ||
671 vma->anon_vma->root != page__anon_vma->root)
21d0d443 672 return -EFAULT;
27ba0644
KS
673 } else if (page->mapping) {
674 if (!vma->vm_file || vma->vm_file->f_mapping != page->mapping)
1da177e4
LT
675 return -EFAULT;
676 } else
677 return -EFAULT;
86c2ad19
ML
678 address = __vma_address(page, vma);
679 if (unlikely(address < vma->vm_start || address >= vma->vm_end))
680 return -EFAULT;
681 return address;
1da177e4
LT
682}
683
6219049a
BL
684pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address)
685{
686 pgd_t *pgd;
c2febafc 687 p4d_t *p4d;
6219049a
BL
688 pud_t *pud;
689 pmd_t *pmd = NULL;
f72e7dcd 690 pmd_t pmde;
6219049a
BL
691
692 pgd = pgd_offset(mm, address);
693 if (!pgd_present(*pgd))
694 goto out;
695
c2febafc
KS
696 p4d = p4d_offset(pgd, address);
697 if (!p4d_present(*p4d))
698 goto out;
699
700 pud = pud_offset(p4d, address);
6219049a
BL
701 if (!pud_present(*pud))
702 goto out;
703
704 pmd = pmd_offset(pud, address);
f72e7dcd 705 /*
8809aa2d 706 * Some THP functions use the sequence pmdp_huge_clear_flush(), set_pmd_at()
f72e7dcd
HD
707 * without holding anon_vma lock for write. So when looking for a
708 * genuine pmde (in which to find pte), test present and !THP together.
709 */
e37c6982
CB
710 pmde = *pmd;
711 barrier();
f72e7dcd 712 if (!pmd_present(pmde) || pmd_trans_huge(pmde))
6219049a
BL
713 pmd = NULL;
714out:
715 return pmd;
716}
717
8749cfea
VD
718struct page_referenced_arg {
719 int mapcount;
720 int referenced;
721 unsigned long vm_flags;
722 struct mem_cgroup *memcg;
723};
724/*
725 * arg: page_referenced_arg will be passed
726 */
e4b82222 727static bool page_referenced_one(struct page *page, struct vm_area_struct *vma,
8749cfea
VD
728 unsigned long address, void *arg)
729{
8749cfea 730 struct page_referenced_arg *pra = arg;
8eaedede
KS
731 struct page_vma_mapped_walk pvmw = {
732 .page = page,
733 .vma = vma,
734 .address = address,
735 };
8749cfea
VD
736 int referenced = 0;
737
8eaedede
KS
738 while (page_vma_mapped_walk(&pvmw)) {
739 address = pvmw.address;
b20ce5e0 740
8eaedede
KS
741 if (vma->vm_flags & VM_LOCKED) {
742 page_vma_mapped_walk_done(&pvmw);
743 pra->vm_flags |= VM_LOCKED;
e4b82222 744 return false; /* To break the loop */
8eaedede 745 }
71e3aac0 746
8eaedede
KS
747 if (pvmw.pte) {
748 if (ptep_clear_flush_young_notify(vma, address,
749 pvmw.pte)) {
750 /*
751 * Don't treat a reference through
752 * a sequentially read mapping as such.
753 * If the page has been used in another mapping,
754 * we will catch it; if this other mapping is
755 * already gone, the unmap path will have set
756 * PG_referenced or activated the page.
757 */
758 if (likely(!(vma->vm_flags & VM_SEQ_READ)))
759 referenced++;
760 }
761 } else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) {
762 if (pmdp_clear_flush_young_notify(vma, address,
763 pvmw.pmd))
8749cfea 764 referenced++;
8eaedede
KS
765 } else {
766 /* unexpected pmd-mapped page? */
767 WARN_ON_ONCE(1);
8749cfea 768 }
8eaedede
KS
769
770 pra->mapcount--;
b20ce5e0 771 }
b20ce5e0 772
33c3fc71
VD
773 if (referenced)
774 clear_page_idle(page);
775 if (test_and_clear_page_young(page))
776 referenced++;
777
9f32624b
JK
778 if (referenced) {
779 pra->referenced++;
780 pra->vm_flags |= vma->vm_flags;
1da177e4 781 }
34bbd704 782
9f32624b 783 if (!pra->mapcount)
e4b82222 784 return false; /* To break the loop */
9f32624b 785
e4b82222 786 return true;
1da177e4
LT
787}
788
9f32624b 789static bool invalid_page_referenced_vma(struct vm_area_struct *vma, void *arg)
1da177e4 790{
9f32624b
JK
791 struct page_referenced_arg *pra = arg;
792 struct mem_cgroup *memcg = pra->memcg;
1da177e4 793
9f32624b
JK
794 if (!mm_match_cgroup(vma->vm_mm, memcg))
795 return true;
1da177e4 796
9f32624b 797 return false;
1da177e4
LT
798}
799
800/**
801 * page_referenced - test if the page was referenced
802 * @page: the page to test
803 * @is_locked: caller holds lock on the page
72835c86 804 * @memcg: target memory cgroup
6fe6b7e3 805 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
1da177e4
LT
806 *
807 * Quick test_and_clear_referenced for all mappings to a page,
808 * returns the number of ptes which referenced the page.
809 */
6fe6b7e3
WF
810int page_referenced(struct page *page,
811 int is_locked,
72835c86 812 struct mem_cgroup *memcg,
6fe6b7e3 813 unsigned long *vm_flags)
1da177e4 814{
5ad64688 815 int we_locked = 0;
9f32624b 816 struct page_referenced_arg pra = {
b20ce5e0 817 .mapcount = total_mapcount(page),
9f32624b
JK
818 .memcg = memcg,
819 };
820 struct rmap_walk_control rwc = {
821 .rmap_one = page_referenced_one,
822 .arg = (void *)&pra,
823 .anon_lock = page_lock_anon_vma_read,
824 };
1da177e4 825
6fe6b7e3 826 *vm_flags = 0;
9f32624b
JK
827 if (!page_mapped(page))
828 return 0;
829
830 if (!page_rmapping(page))
831 return 0;
832
833 if (!is_locked && (!PageAnon(page) || PageKsm(page))) {
834 we_locked = trylock_page(page);
835 if (!we_locked)
836 return 1;
1da177e4 837 }
9f32624b
JK
838
839 /*
840 * If we are reclaiming on behalf of a cgroup, skip
841 * counting on behalf of references from different
842 * cgroups
843 */
844 if (memcg) {
845 rwc.invalid_vma = invalid_page_referenced_vma;
846 }
847
c24f386c 848 rmap_walk(page, &rwc);
9f32624b
JK
849 *vm_flags = pra.vm_flags;
850
851 if (we_locked)
852 unlock_page(page);
853
854 return pra.referenced;
1da177e4
LT
855}
856
e4b82222 857static bool page_mkclean_one(struct page *page, struct vm_area_struct *vma,
9853a407 858 unsigned long address, void *arg)
d08b3851 859{
f27176cf
KS
860 struct page_vma_mapped_walk pvmw = {
861 .page = page,
862 .vma = vma,
863 .address = address,
864 .flags = PVMW_SYNC,
865 };
9853a407 866 int *cleaned = arg;
d08b3851 867
f27176cf
KS
868 while (page_vma_mapped_walk(&pvmw)) {
869 int ret = 0;
870 address = pvmw.address;
871 if (pvmw.pte) {
872 pte_t entry;
873 pte_t *pte = pvmw.pte;
874
875 if (!pte_dirty(*pte) && !pte_write(*pte))
876 continue;
877
878 flush_cache_page(vma, address, pte_pfn(*pte));
879 entry = ptep_clear_flush(vma, address, pte);
880 entry = pte_wrprotect(entry);
881 entry = pte_mkclean(entry);
882 set_pte_at(vma->vm_mm, address, pte, entry);
883 ret = 1;
884 } else {
885#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
886 pmd_t *pmd = pvmw.pmd;
887 pmd_t entry;
888
889 if (!pmd_dirty(*pmd) && !pmd_write(*pmd))
890 continue;
891
892 flush_cache_page(vma, address, page_to_pfn(page));
893 entry = pmdp_huge_clear_flush(vma, address, pmd);
894 entry = pmd_wrprotect(entry);
895 entry = pmd_mkclean(entry);
896 set_pmd_at(vma->vm_mm, address, pmd, entry);
897 ret = 1;
898#else
899 /* unexpected pmd-mapped page? */
900 WARN_ON_ONCE(1);
901#endif
902 }
d08b3851 903
f27176cf
KS
904 if (ret) {
905 mmu_notifier_invalidate_page(vma->vm_mm, address);
906 (*cleaned)++;
907 }
c2fda5fe 908 }
d08b3851 909
e4b82222 910 return true;
d08b3851
PZ
911}
912
9853a407 913static bool invalid_mkclean_vma(struct vm_area_struct *vma, void *arg)
d08b3851 914{
9853a407 915 if (vma->vm_flags & VM_SHARED)
871beb8c 916 return false;
d08b3851 917
871beb8c 918 return true;
d08b3851
PZ
919}
920
921int page_mkclean(struct page *page)
922{
9853a407
JK
923 int cleaned = 0;
924 struct address_space *mapping;
925 struct rmap_walk_control rwc = {
926 .arg = (void *)&cleaned,
927 .rmap_one = page_mkclean_one,
928 .invalid_vma = invalid_mkclean_vma,
929 };
d08b3851
PZ
930
931 BUG_ON(!PageLocked(page));
932
9853a407
JK
933 if (!page_mapped(page))
934 return 0;
935
936 mapping = page_mapping(page);
937 if (!mapping)
938 return 0;
939
940 rmap_walk(page, &rwc);
d08b3851 941
9853a407 942 return cleaned;
d08b3851 943}
60b59bea 944EXPORT_SYMBOL_GPL(page_mkclean);
d08b3851 945
c44b6743
RR
946/**
947 * page_move_anon_rmap - move a page to our anon_vma
948 * @page: the page to move to our anon_vma
949 * @vma: the vma the page belongs to
c44b6743
RR
950 *
951 * When a page belongs exclusively to one process after a COW event,
952 * that page can be moved into the anon_vma that belongs to just that
953 * process, so the rmap code will not search the parent or sibling
954 * processes.
955 */
5a49973d 956void page_move_anon_rmap(struct page *page, struct vm_area_struct *vma)
c44b6743
RR
957{
958 struct anon_vma *anon_vma = vma->anon_vma;
959
5a49973d
HD
960 page = compound_head(page);
961
309381fe 962 VM_BUG_ON_PAGE(!PageLocked(page), page);
81d1b09c 963 VM_BUG_ON_VMA(!anon_vma, vma);
c44b6743
RR
964
965 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
414e2fb8
VD
966 /*
967 * Ensure that anon_vma and the PAGE_MAPPING_ANON bit are written
968 * simultaneously, so a concurrent reader (eg page_referenced()'s
969 * PageAnon()) will not see one without the other.
970 */
971 WRITE_ONCE(page->mapping, (struct address_space *) anon_vma);
c44b6743
RR
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);
53f9263b 1027 BUG_ON(page_to_pgoff(page) != linear_page_index(vma, address));
c97a9e10
NP
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
d281ee61 1036 * @compound: charge the page as compound or small page
1da177e4 1037 *
5ad64688 1038 * The caller needs to hold the pte lock, and the page must be locked in
80e14822
HD
1039 * the anon_vma case: to serialize mapping,index checking after setting,
1040 * and to ensure that PageAnon is not being upgraded racily to PageKsm
1041 * (but PageKsm is never downgraded to PageAnon).
1da177e4
LT
1042 */
1043void page_add_anon_rmap(struct page *page,
d281ee61 1044 struct vm_area_struct *vma, unsigned long address, bool compound)
ad8c2ee8 1045{
d281ee61 1046 do_page_add_anon_rmap(page, vma, address, compound ? RMAP_COMPOUND : 0);
ad8c2ee8
RR
1047}
1048
1049/*
1050 * Special version of the above for do_swap_page, which often runs
1051 * into pages that are exclusively owned by the current process.
1052 * Everybody else should continue to use page_add_anon_rmap above.
1053 */
1054void do_page_add_anon_rmap(struct page *page,
d281ee61 1055 struct vm_area_struct *vma, unsigned long address, int flags)
1da177e4 1056{
53f9263b
KS
1057 bool compound = flags & RMAP_COMPOUND;
1058 bool first;
1059
e9b61f19
KS
1060 if (compound) {
1061 atomic_t *mapcount;
53f9263b 1062 VM_BUG_ON_PAGE(!PageLocked(page), page);
e9b61f19
KS
1063 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
1064 mapcount = compound_mapcount_ptr(page);
1065 first = atomic_inc_and_test(mapcount);
53f9263b
KS
1066 } else {
1067 first = atomic_inc_and_test(&page->_mapcount);
1068 }
1069
79134171 1070 if (first) {
d281ee61 1071 int nr = compound ? hpage_nr_pages(page) : 1;
bea04b07
JZ
1072 /*
1073 * We use the irq-unsafe __{inc|mod}_zone_page_stat because
1074 * these counters are not modified in interrupt context, and
1075 * pte lock(a spinlock) is held, which implies preemption
1076 * disabled.
1077 */
65c45377 1078 if (compound)
11fb9989 1079 __inc_node_page_state(page, NR_ANON_THPS);
4b9d0fab 1080 __mod_node_page_state(page_pgdat(page), NR_ANON_MAPPED, nr);
79134171 1081 }
5ad64688
HD
1082 if (unlikely(PageKsm(page)))
1083 return;
1084
309381fe 1085 VM_BUG_ON_PAGE(!PageLocked(page), page);
53f9263b 1086
5dbe0af4 1087 /* address might be in next vma when migration races vma_adjust */
5ad64688 1088 if (first)
d281ee61
KS
1089 __page_set_anon_rmap(page, vma, address,
1090 flags & RMAP_EXCLUSIVE);
69029cd5 1091 else
c97a9e10 1092 __page_check_anon_rmap(page, vma, address);
1da177e4
LT
1093}
1094
43d8eac4 1095/**
9617d95e
NP
1096 * page_add_new_anon_rmap - add pte mapping to a new anonymous page
1097 * @page: the page to add the mapping to
1098 * @vma: the vm area in which the mapping is added
1099 * @address: the user virtual address mapped
d281ee61 1100 * @compound: charge the page as compound or small page
9617d95e
NP
1101 *
1102 * Same as page_add_anon_rmap but must only be called on *new* pages.
1103 * This means the inc-and-test can be bypassed.
c97a9e10 1104 * Page does not have to be locked.
9617d95e
NP
1105 */
1106void page_add_new_anon_rmap(struct page *page,
d281ee61 1107 struct vm_area_struct *vma, unsigned long address, bool compound)
9617d95e 1108{
d281ee61
KS
1109 int nr = compound ? hpage_nr_pages(page) : 1;
1110
81d1b09c 1111 VM_BUG_ON_VMA(address < vma->vm_start || address >= vma->vm_end, vma);
fa9949da 1112 __SetPageSwapBacked(page);
d281ee61
KS
1113 if (compound) {
1114 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
53f9263b
KS
1115 /* increment count (starts at -1) */
1116 atomic_set(compound_mapcount_ptr(page), 0);
11fb9989 1117 __inc_node_page_state(page, NR_ANON_THPS);
53f9263b
KS
1118 } else {
1119 /* Anon THP always mapped first with PMD */
1120 VM_BUG_ON_PAGE(PageTransCompound(page), page);
1121 /* increment count (starts at -1) */
1122 atomic_set(&page->_mapcount, 0);
d281ee61 1123 }
4b9d0fab 1124 __mod_node_page_state(page_pgdat(page), NR_ANON_MAPPED, nr);
e8a03feb 1125 __page_set_anon_rmap(page, vma, address, 1);
9617d95e
NP
1126}
1127
1da177e4
LT
1128/**
1129 * page_add_file_rmap - add pte mapping to a file page
1130 * @page: the page to add the mapping to
1131 *
b8072f09 1132 * The caller needs to hold the pte lock.
1da177e4 1133 */
dd78fedd 1134void page_add_file_rmap(struct page *page, bool compound)
1da177e4 1135{
dd78fedd
KS
1136 int i, nr = 1;
1137
1138 VM_BUG_ON_PAGE(compound && !PageTransHuge(page), page);
62cccb8c 1139 lock_page_memcg(page);
dd78fedd
KS
1140 if (compound && PageTransHuge(page)) {
1141 for (i = 0, nr = 0; i < HPAGE_PMD_NR; i++) {
1142 if (atomic_inc_and_test(&page[i]._mapcount))
1143 nr++;
1144 }
1145 if (!atomic_inc_and_test(compound_mapcount_ptr(page)))
1146 goto out;
65c45377 1147 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
11fb9989 1148 __inc_node_page_state(page, NR_SHMEM_PMDMAPPED);
dd78fedd 1149 } else {
c8efc390
KS
1150 if (PageTransCompound(page) && page_mapping(page)) {
1151 VM_WARN_ON_ONCE(!PageLocked(page));
1152
9a73f61b
KS
1153 SetPageDoubleMap(compound_head(page));
1154 if (PageMlocked(page))
1155 clear_page_mlock(compound_head(page));
1156 }
dd78fedd
KS
1157 if (!atomic_inc_and_test(&page->_mapcount))
1158 goto out;
d69b042f 1159 }
50658e2e 1160 __mod_node_page_state(page_pgdat(page), NR_FILE_MAPPED, nr);
ccda7f43 1161 mod_memcg_page_state(page, NR_FILE_MAPPED, nr);
dd78fedd 1162out:
62cccb8c 1163 unlock_page_memcg(page);
1da177e4
LT
1164}
1165
dd78fedd 1166static void page_remove_file_rmap(struct page *page, bool compound)
8186eb6a 1167{
dd78fedd
KS
1168 int i, nr = 1;
1169
57dea93a 1170 VM_BUG_ON_PAGE(compound && !PageHead(page), page);
62cccb8c 1171 lock_page_memcg(page);
8186eb6a 1172
53f9263b
KS
1173 /* Hugepages are not counted in NR_FILE_MAPPED for now. */
1174 if (unlikely(PageHuge(page))) {
1175 /* hugetlb pages are always mapped with pmds */
1176 atomic_dec(compound_mapcount_ptr(page));
8186eb6a 1177 goto out;
53f9263b 1178 }
8186eb6a 1179
53f9263b 1180 /* page still mapped by someone else? */
dd78fedd
KS
1181 if (compound && PageTransHuge(page)) {
1182 for (i = 0, nr = 0; i < HPAGE_PMD_NR; i++) {
1183 if (atomic_add_negative(-1, &page[i]._mapcount))
1184 nr++;
1185 }
1186 if (!atomic_add_negative(-1, compound_mapcount_ptr(page)))
1187 goto out;
65c45377 1188 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
11fb9989 1189 __dec_node_page_state(page, NR_SHMEM_PMDMAPPED);
dd78fedd
KS
1190 } else {
1191 if (!atomic_add_negative(-1, &page->_mapcount))
1192 goto out;
1193 }
8186eb6a
JW
1194
1195 /*
50658e2e 1196 * We use the irq-unsafe __{inc|mod}_zone_page_state because
8186eb6a
JW
1197 * these counters are not modified in interrupt context, and
1198 * pte lock(a spinlock) is held, which implies preemption disabled.
1199 */
50658e2e 1200 __mod_node_page_state(page_pgdat(page), NR_FILE_MAPPED, -nr);
ccda7f43 1201 mod_memcg_page_state(page, NR_FILE_MAPPED, -nr);
8186eb6a
JW
1202
1203 if (unlikely(PageMlocked(page)))
1204 clear_page_mlock(page);
1205out:
62cccb8c 1206 unlock_page_memcg(page);
8186eb6a
JW
1207}
1208
53f9263b
KS
1209static void page_remove_anon_compound_rmap(struct page *page)
1210{
1211 int i, nr;
1212
1213 if (!atomic_add_negative(-1, compound_mapcount_ptr(page)))
1214 return;
1215
1216 /* Hugepages are not counted in NR_ANON_PAGES for now. */
1217 if (unlikely(PageHuge(page)))
1218 return;
1219
1220 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
1221 return;
1222
11fb9989 1223 __dec_node_page_state(page, NR_ANON_THPS);
53f9263b
KS
1224
1225 if (TestClearPageDoubleMap(page)) {
1226 /*
1227 * Subpages can be mapped with PTEs too. Check how many of
1228 * themi are still mapped.
1229 */
1230 for (i = 0, nr = 0; i < HPAGE_PMD_NR; i++) {
1231 if (atomic_add_negative(-1, &page[i]._mapcount))
1232 nr++;
1233 }
1234 } else {
1235 nr = HPAGE_PMD_NR;
1236 }
1237
e90309c9
KS
1238 if (unlikely(PageMlocked(page)))
1239 clear_page_mlock(page);
1240
9a982250 1241 if (nr) {
4b9d0fab 1242 __mod_node_page_state(page_pgdat(page), NR_ANON_MAPPED, -nr);
9a982250
KS
1243 deferred_split_huge_page(page);
1244 }
53f9263b
KS
1245}
1246
1da177e4
LT
1247/**
1248 * page_remove_rmap - take down pte mapping from a page
d281ee61
KS
1249 * @page: page to remove mapping from
1250 * @compound: uncharge the page as compound or small page
1da177e4 1251 *
b8072f09 1252 * The caller needs to hold the pte lock.
1da177e4 1253 */
d281ee61 1254void page_remove_rmap(struct page *page, bool compound)
1da177e4 1255{
dd78fedd
KS
1256 if (!PageAnon(page))
1257 return page_remove_file_rmap(page, compound);
89c06bd5 1258
53f9263b
KS
1259 if (compound)
1260 return page_remove_anon_compound_rmap(page);
1261
b904dcfe
KM
1262 /* page still mapped by someone else? */
1263 if (!atomic_add_negative(-1, &page->_mapcount))
8186eb6a
JW
1264 return;
1265
0fe6e20b 1266 /*
bea04b07
JZ
1267 * We use the irq-unsafe __{inc|mod}_zone_page_stat because
1268 * these counters are not modified in interrupt context, and
bea04b07 1269 * pte lock(a spinlock) is held, which implies preemption disabled.
0fe6e20b 1270 */
4b9d0fab 1271 __dec_node_page_state(page, NR_ANON_MAPPED);
8186eb6a 1272
e6c509f8
HD
1273 if (unlikely(PageMlocked(page)))
1274 clear_page_mlock(page);
8186eb6a 1275
9a982250
KS
1276 if (PageTransCompound(page))
1277 deferred_split_huge_page(compound_head(page));
1278
b904dcfe
KM
1279 /*
1280 * It would be tidy to reset the PageAnon mapping here,
1281 * but that might overwrite a racing page_add_anon_rmap
1282 * which increments mapcount after us but sets mapping
1283 * before us: so leave the reset to free_hot_cold_page,
1284 * and remember that it's only reliable while mapped.
1285 * Leaving it set also helps swapoff to reinstate ptes
1286 * faster for those pages still in swapcache.
1287 */
1da177e4
LT
1288}
1289
1290/*
52629506 1291 * @arg: enum ttu_flags will be passed to this argument
1da177e4 1292 */
e4b82222 1293static bool try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
52629506 1294 unsigned long address, void *arg)
1da177e4
LT
1295{
1296 struct mm_struct *mm = vma->vm_mm;
c7ab0d2f
KS
1297 struct page_vma_mapped_walk pvmw = {
1298 .page = page,
1299 .vma = vma,
1300 .address = address,
1301 };
1da177e4 1302 pte_t pteval;
c7ab0d2f 1303 struct page *subpage;
e4b82222 1304 bool ret = true;
802a3a92 1305 enum ttu_flags flags = (enum ttu_flags)arg;
1da177e4 1306
b87537d9
HD
1307 /* munlock has nothing to gain from examining un-locked vmas */
1308 if ((flags & TTU_MUNLOCK) && !(vma->vm_flags & VM_LOCKED))
e4b82222 1309 return true;
b87537d9 1310
fec89c10
KS
1311 if (flags & TTU_SPLIT_HUGE_PMD) {
1312 split_huge_pmd_address(vma, address,
1313 flags & TTU_MIGRATION, page);
fec89c10
KS
1314 }
1315
c7ab0d2f 1316 while (page_vma_mapped_walk(&pvmw)) {
c7ab0d2f
KS
1317 /*
1318 * If the page is mlock()d, we cannot swap it out.
1319 * If it's recently referenced (perhaps page_referenced
1320 * skipped over this mm) then we should reactivate it.
1321 */
1322 if (!(flags & TTU_IGNORE_MLOCK)) {
1323 if (vma->vm_flags & VM_LOCKED) {
1324 /* PTE-mapped THP are never mlocked */
1325 if (!PageTransCompound(page)) {
1326 /*
1327 * Holding pte lock, we do *not* need
1328 * mmap_sem here
1329 */
1330 mlock_vma_page(page);
1331 }
e4b82222 1332 ret = false;
c7ab0d2f
KS
1333 page_vma_mapped_walk_done(&pvmw);
1334 break;
9a73f61b 1335 }
c7ab0d2f
KS
1336 if (flags & TTU_MUNLOCK)
1337 continue;
b87537d9 1338 }
c7ab0d2f 1339
8346242a
KS
1340 /* Unexpected PMD-mapped THP? */
1341 VM_BUG_ON_PAGE(!pvmw.pte, page);
1342
1343 subpage = page - page_to_pfn(page) + pte_pfn(*pvmw.pte);
1344 address = pvmw.address;
1345
1346
c7ab0d2f
KS
1347 if (!(flags & TTU_IGNORE_ACCESS)) {
1348 if (ptep_clear_flush_young_notify(vma, address,
1349 pvmw.pte)) {
e4b82222 1350 ret = false;
c7ab0d2f
KS
1351 page_vma_mapped_walk_done(&pvmw);
1352 break;
1353 }
b291f000 1354 }
1da177e4 1355
c7ab0d2f
KS
1356 /* Nuke the page table entry. */
1357 flush_cache_page(vma, address, pte_pfn(*pvmw.pte));
1358 if (should_defer_flush(mm, flags)) {
1359 /*
1360 * We clear the PTE but do not flush so potentially
1361 * a remote CPU could still be writing to the page.
1362 * If the entry was previously clean then the
1363 * architecture must guarantee that a clear->dirty
1364 * transition on a cached TLB entry is written through
1365 * and traps if the PTE is unmapped.
1366 */
1367 pteval = ptep_get_and_clear(mm, address, pvmw.pte);
1368
1369 set_tlb_ubc_flush_pending(mm, pte_dirty(pteval));
1370 } else {
1371 pteval = ptep_clear_flush(vma, address, pvmw.pte);
1372 }
72b252ae 1373
c7ab0d2f
KS
1374 /* Move the dirty bit to the page. Now the pte is gone. */
1375 if (pte_dirty(pteval))
1376 set_page_dirty(page);
1da177e4 1377
c7ab0d2f
KS
1378 /* Update high watermark before we lower rss */
1379 update_hiwater_rss(mm);
1da177e4 1380
c7ab0d2f
KS
1381 if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) {
1382 if (PageHuge(page)) {
1383 int nr = 1 << compound_order(page);
1384 hugetlb_count_sub(nr, mm);
1385 } else {
1386 dec_mm_counter(mm, mm_counter(page));
1387 }
365e9c87 1388
c7ab0d2f
KS
1389 pteval = swp_entry_to_pte(make_hwpoison_entry(subpage));
1390 set_pte_at(mm, address, pvmw.pte, pteval);
1391 } else if (pte_unused(pteval)) {
1392 /*
1393 * The guest indicated that the page content is of no
1394 * interest anymore. Simply discard the pte, vmscan
1395 * will take care of the rest.
1396 */
eca56ff9 1397 dec_mm_counter(mm, mm_counter(page));
c7ab0d2f
KS
1398 } else if (IS_ENABLED(CONFIG_MIGRATION) &&
1399 (flags & TTU_MIGRATION)) {
1400 swp_entry_t entry;
1401 pte_t swp_pte;
1402 /*
1403 * Store the pfn of the page in a special migration
1404 * pte. do_swap_page() will wait until the migration
1405 * pte is removed and then restart fault handling.
1406 */
1407 entry = make_migration_entry(subpage,
1408 pte_write(pteval));
1409 swp_pte = swp_entry_to_pte(entry);
1410 if (pte_soft_dirty(pteval))
1411 swp_pte = pte_swp_mksoft_dirty(swp_pte);
1412 set_pte_at(mm, address, pvmw.pte, swp_pte);
1413 } else if (PageAnon(page)) {
1414 swp_entry_t entry = { .val = page_private(subpage) };
1415 pte_t swp_pte;
1416 /*
1417 * Store the swap location in the pte.
1418 * See handle_pte_fault() ...
1419 */
eb94a878
MK
1420 if (unlikely(PageSwapBacked(page) != PageSwapCache(page))) {
1421 WARN_ON_ONCE(1);
83612a94 1422 ret = false;
eb94a878
MK
1423 page_vma_mapped_walk_done(&pvmw);
1424 break;
1425 }
c7ab0d2f 1426
802a3a92
SL
1427 /* MADV_FREE page check */
1428 if (!PageSwapBacked(page)) {
1429 if (!PageDirty(page)) {
1430 dec_mm_counter(mm, MM_ANONPAGES);
1431 goto discard;
1432 }
1433
1434 /*
1435 * If the page was redirtied, it cannot be
1436 * discarded. Remap the page to page table.
1437 */
1438 set_pte_at(mm, address, pvmw.pte, pteval);
18863d3a 1439 SetPageSwapBacked(page);
e4b82222 1440 ret = false;
802a3a92
SL
1441 page_vma_mapped_walk_done(&pvmw);
1442 break;
c7ab0d2f 1443 }
854e9ed0 1444
c7ab0d2f
KS
1445 if (swap_duplicate(entry) < 0) {
1446 set_pte_at(mm, address, pvmw.pte, pteval);
e4b82222 1447 ret = false;
c7ab0d2f
KS
1448 page_vma_mapped_walk_done(&pvmw);
1449 break;
1450 }
1451 if (list_empty(&mm->mmlist)) {
1452 spin_lock(&mmlist_lock);
1453 if (list_empty(&mm->mmlist))
1454 list_add(&mm->mmlist, &init_mm.mmlist);
1455 spin_unlock(&mmlist_lock);
1456 }
854e9ed0 1457 dec_mm_counter(mm, MM_ANONPAGES);
c7ab0d2f
KS
1458 inc_mm_counter(mm, MM_SWAPENTS);
1459 swp_pte = swp_entry_to_pte(entry);
1460 if (pte_soft_dirty(pteval))
1461 swp_pte = pte_swp_mksoft_dirty(swp_pte);
1462 set_pte_at(mm, address, pvmw.pte, swp_pte);
1463 } else
1464 dec_mm_counter(mm, mm_counter_file(page));
854e9ed0 1465discard:
c7ab0d2f
KS
1466 page_remove_rmap(subpage, PageHuge(page));
1467 put_page(page);
2ec74c3e 1468 mmu_notifier_invalidate_page(mm, address);
c7ab0d2f 1469 }
caed0f48 1470 return ret;
1da177e4
LT
1471}
1472
71e3aac0 1473bool is_vma_temporary_stack(struct vm_area_struct *vma)
a8bef8ff
MG
1474{
1475 int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP);
1476
1477 if (!maybe_stack)
1478 return false;
1479
1480 if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) ==
1481 VM_STACK_INCOMPLETE_SETUP)
1482 return true;
1483
1484 return false;
1485}
1486
52629506
JK
1487static bool invalid_migration_vma(struct vm_area_struct *vma, void *arg)
1488{
1489 return is_vma_temporary_stack(vma);
1490}
1491
2a52bcbc 1492static int page_mapcount_is_zero(struct page *page)
52629506 1493{
c7ab0d2f 1494 return !total_mapcount(page);
2a52bcbc 1495}
52629506 1496
1da177e4
LT
1497/**
1498 * try_to_unmap - try to remove all page table mappings to a page
1499 * @page: the page to get unmapped
14fa31b8 1500 * @flags: action and flags
1da177e4
LT
1501 *
1502 * Tries to remove all the page table entries which are mapping this
1503 * page, used in the pageout path. Caller must hold the page lock.
1da177e4 1504 *
666e5a40 1505 * If unmap is successful, return true. Otherwise, false.
1da177e4 1506 */
666e5a40 1507bool try_to_unmap(struct page *page, enum ttu_flags flags)
1da177e4 1508{
52629506
JK
1509 struct rmap_walk_control rwc = {
1510 .rmap_one = try_to_unmap_one,
802a3a92 1511 .arg = (void *)flags,
2a52bcbc 1512 .done = page_mapcount_is_zero,
52629506
JK
1513 .anon_lock = page_lock_anon_vma_read,
1514 };
1da177e4 1515
52629506
JK
1516 /*
1517 * During exec, a temporary VMA is setup and later moved.
1518 * The VMA is moved under the anon_vma lock but not the
1519 * page tables leading to a race where migration cannot
1520 * find the migration ptes. Rather than increasing the
1521 * locking requirements of exec(), migration skips
1522 * temporary VMAs until after exec() completes.
1523 */
daa5ba76 1524 if ((flags & TTU_MIGRATION) && !PageKsm(page) && PageAnon(page))
52629506
JK
1525 rwc.invalid_vma = invalid_migration_vma;
1526
2a52bcbc 1527 if (flags & TTU_RMAP_LOCKED)
33fc80e2 1528 rmap_walk_locked(page, &rwc);
2a52bcbc 1529 else
33fc80e2 1530 rmap_walk(page, &rwc);
52629506 1531
666e5a40 1532 return !page_mapcount(page) ? true : false;
1da177e4 1533}
81b4082d 1534
2a52bcbc
KS
1535static int page_not_mapped(struct page *page)
1536{
1537 return !page_mapped(page);
1538};
1539
b291f000
NP
1540/**
1541 * try_to_munlock - try to munlock a page
1542 * @page: the page to be munlocked
1543 *
1544 * Called from munlock code. Checks all of the VMAs mapping the page
1545 * to make sure nobody else has this page mlocked. The page will be
1546 * returned with PG_mlocked cleared if no other vmas have it mlocked.
b291f000 1547 */
854e9ed0 1548
192d7232
MK
1549void try_to_munlock(struct page *page)
1550{
e8351ac9
JK
1551 struct rmap_walk_control rwc = {
1552 .rmap_one = try_to_unmap_one,
802a3a92 1553 .arg = (void *)TTU_MUNLOCK,
e8351ac9 1554 .done = page_not_mapped,
e8351ac9
JK
1555 .anon_lock = page_lock_anon_vma_read,
1556
1557 };
1558
309381fe 1559 VM_BUG_ON_PAGE(!PageLocked(page) || PageLRU(page), page);
192d7232 1560 VM_BUG_ON_PAGE(PageCompound(page) && PageDoubleMap(page), page);
b291f000 1561
192d7232 1562 rmap_walk(page, &rwc);
b291f000 1563}
e9995ef9 1564
01d8b20d 1565void __put_anon_vma(struct anon_vma *anon_vma)
76545066 1566{
01d8b20d 1567 struct anon_vma *root = anon_vma->root;
76545066 1568
624483f3 1569 anon_vma_free(anon_vma);
01d8b20d
PZ
1570 if (root != anon_vma && atomic_dec_and_test(&root->refcount))
1571 anon_vma_free(root);
76545066 1572}
76545066 1573
0dd1c7bb
JK
1574static struct anon_vma *rmap_walk_anon_lock(struct page *page,
1575 struct rmap_walk_control *rwc)
faecd8dd
JK
1576{
1577 struct anon_vma *anon_vma;
1578
0dd1c7bb
JK
1579 if (rwc->anon_lock)
1580 return rwc->anon_lock(page);
1581
faecd8dd
JK
1582 /*
1583 * Note: remove_migration_ptes() cannot use page_lock_anon_vma_read()
1584 * because that depends on page_mapped(); but not all its usages
1585 * are holding mmap_sem. Users without mmap_sem are required to
1586 * take a reference count to prevent the anon_vma disappearing
1587 */
1588 anon_vma = page_anon_vma(page);
1589 if (!anon_vma)
1590 return NULL;
1591
1592 anon_vma_lock_read(anon_vma);
1593 return anon_vma;
1594}
1595
e9995ef9 1596/*
e8351ac9
JK
1597 * rmap_walk_anon - do something to anonymous page using the object-based
1598 * rmap method
1599 * @page: the page to be handled
1600 * @rwc: control variable according to each walk type
1601 *
1602 * Find all the mappings of a page using the mapping pointer and the vma chains
1603 * contained in the anon_vma struct it points to.
1604 *
1605 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1606 * where the page was found will be held for write. So, we won't recheck
1607 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1608 * LOCKED.
e9995ef9 1609 */
1df631ae 1610static void rmap_walk_anon(struct page *page, struct rmap_walk_control *rwc,
b9773199 1611 bool locked)
e9995ef9
HD
1612{
1613 struct anon_vma *anon_vma;
a8fa41ad 1614 pgoff_t pgoff_start, pgoff_end;
5beb4930 1615 struct anon_vma_chain *avc;
e9995ef9 1616
b9773199
KS
1617 if (locked) {
1618 anon_vma = page_anon_vma(page);
1619 /* anon_vma disappear under us? */
1620 VM_BUG_ON_PAGE(!anon_vma, page);
1621 } else {
1622 anon_vma = rmap_walk_anon_lock(page, rwc);
1623 }
e9995ef9 1624 if (!anon_vma)
1df631ae 1625 return;
faecd8dd 1626
a8fa41ad
KS
1627 pgoff_start = page_to_pgoff(page);
1628 pgoff_end = pgoff_start + hpage_nr_pages(page) - 1;
1629 anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root,
1630 pgoff_start, pgoff_end) {
5beb4930 1631 struct vm_area_struct *vma = avc->vma;
e9995ef9 1632 unsigned long address = vma_address(page, vma);
0dd1c7bb 1633
ad12695f
AA
1634 cond_resched();
1635
0dd1c7bb
JK
1636 if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg))
1637 continue;
1638
e4b82222 1639 if (!rwc->rmap_one(page, vma, address, rwc->arg))
e9995ef9 1640 break;
0dd1c7bb
JK
1641 if (rwc->done && rwc->done(page))
1642 break;
e9995ef9 1643 }
b9773199
KS
1644
1645 if (!locked)
1646 anon_vma_unlock_read(anon_vma);
e9995ef9
HD
1647}
1648
e8351ac9
JK
1649/*
1650 * rmap_walk_file - do something to file page using the object-based rmap method
1651 * @page: the page to be handled
1652 * @rwc: control variable according to each walk type
1653 *
1654 * Find all the mappings of a page using the mapping pointer and the vma chains
1655 * contained in the address_space struct it points to.
1656 *
1657 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1658 * where the page was found will be held for write. So, we won't recheck
1659 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1660 * LOCKED.
1661 */
1df631ae 1662static void rmap_walk_file(struct page *page, struct rmap_walk_control *rwc,
b9773199 1663 bool locked)
e9995ef9 1664{
b9773199 1665 struct address_space *mapping = page_mapping(page);
a8fa41ad 1666 pgoff_t pgoff_start, pgoff_end;
e9995ef9 1667 struct vm_area_struct *vma;
e9995ef9 1668
9f32624b
JK
1669 /*
1670 * The page lock not only makes sure that page->mapping cannot
1671 * suddenly be NULLified by truncation, it makes sure that the
1672 * structure at mapping cannot be freed and reused yet,
c8c06efa 1673 * so we can safely take mapping->i_mmap_rwsem.
9f32624b 1674 */
81d1b09c 1675 VM_BUG_ON_PAGE(!PageLocked(page), page);
9f32624b 1676
e9995ef9 1677 if (!mapping)
1df631ae 1678 return;
3dec0ba0 1679
a8fa41ad
KS
1680 pgoff_start = page_to_pgoff(page);
1681 pgoff_end = pgoff_start + hpage_nr_pages(page) - 1;
b9773199
KS
1682 if (!locked)
1683 i_mmap_lock_read(mapping);
a8fa41ad
KS
1684 vma_interval_tree_foreach(vma, &mapping->i_mmap,
1685 pgoff_start, pgoff_end) {
e9995ef9 1686 unsigned long address = vma_address(page, vma);
0dd1c7bb 1687
ad12695f
AA
1688 cond_resched();
1689
0dd1c7bb
JK
1690 if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg))
1691 continue;
1692
e4b82222 1693 if (!rwc->rmap_one(page, vma, address, rwc->arg))
0dd1c7bb
JK
1694 goto done;
1695 if (rwc->done && rwc->done(page))
1696 goto done;
e9995ef9 1697 }
0dd1c7bb 1698
0dd1c7bb 1699done:
b9773199
KS
1700 if (!locked)
1701 i_mmap_unlock_read(mapping);
e9995ef9
HD
1702}
1703
1df631ae 1704void rmap_walk(struct page *page, struct rmap_walk_control *rwc)
e9995ef9 1705{
e9995ef9 1706 if (unlikely(PageKsm(page)))
1df631ae 1707 rmap_walk_ksm(page, rwc);
e9995ef9 1708 else if (PageAnon(page))
1df631ae 1709 rmap_walk_anon(page, rwc, false);
b9773199 1710 else
1df631ae 1711 rmap_walk_file(page, rwc, false);
b9773199
KS
1712}
1713
1714/* Like rmap_walk, but caller holds relevant rmap lock */
1df631ae 1715void rmap_walk_locked(struct page *page, struct rmap_walk_control *rwc)
b9773199
KS
1716{
1717 /* no ksm support for now */
1718 VM_BUG_ON_PAGE(PageKsm(page), page);
1719 if (PageAnon(page))
1df631ae 1720 rmap_walk_anon(page, rwc, true);
e9995ef9 1721 else
1df631ae 1722 rmap_walk_file(page, rwc, true);
e9995ef9 1723}
0fe6e20b 1724
e3390f67 1725#ifdef CONFIG_HUGETLB_PAGE
0fe6e20b
NH
1726/*
1727 * The following three functions are for anonymous (private mapped) hugepages.
1728 * Unlike common anonymous pages, anonymous hugepages have no accounting code
1729 * and no lru code, because we handle hugepages differently from common pages.
1730 */
1731static void __hugepage_set_anon_rmap(struct page *page,
1732 struct vm_area_struct *vma, unsigned long address, int exclusive)
1733{
1734 struct anon_vma *anon_vma = vma->anon_vma;
433abed6 1735
0fe6e20b 1736 BUG_ON(!anon_vma);
433abed6
NH
1737
1738 if (PageAnon(page))
1739 return;
1740 if (!exclusive)
1741 anon_vma = anon_vma->root;
1742
0fe6e20b
NH
1743 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
1744 page->mapping = (struct address_space *) anon_vma;
1745 page->index = linear_page_index(vma, address);
1746}
1747
1748void hugepage_add_anon_rmap(struct page *page,
1749 struct vm_area_struct *vma, unsigned long address)
1750{
1751 struct anon_vma *anon_vma = vma->anon_vma;
1752 int first;
a850ea30
NH
1753
1754 BUG_ON(!PageLocked(page));
0fe6e20b 1755 BUG_ON(!anon_vma);
5dbe0af4 1756 /* address might be in next vma when migration races vma_adjust */
53f9263b 1757 first = atomic_inc_and_test(compound_mapcount_ptr(page));
0fe6e20b
NH
1758 if (first)
1759 __hugepage_set_anon_rmap(page, vma, address, 0);
1760}
1761
1762void hugepage_add_new_anon_rmap(struct page *page,
1763 struct vm_area_struct *vma, unsigned long address)
1764{
1765 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
53f9263b 1766 atomic_set(compound_mapcount_ptr(page), 0);
0fe6e20b
NH
1767 __hugepage_set_anon_rmap(page, vma, address, 1);
1768}
e3390f67 1769#endif /* CONFIG_HUGETLB_PAGE */