blob: cf7e99a87c32865c4e6cf5b6ed4f48f7f05aa9e0 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
Hugh Dickins98f32602009-05-21 20:33:58 +010017 * Contributions by Hugh Dickins 2003, 2004
Linus Torvalds1da177e2005-04-16 15:20:36 -070018 */
19
20/*
21 * Lock ordering in mm:
22 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080023 * inode->i_mutex (while writing or truncating, not reading or faulting)
Nick Piggin82591e62006-10-19 23:29:10 -070024 * mm->mmap_sem
25 * page->flags PG_locked (lock_page)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -070026 * mapping->i_mmap_mutex
Peter Zijlstra2b575eb2011-05-24 17:12:11 -070027 * anon_vma->mutex
Nick Piggin82591e62006-10-19 23:29:10 -070028 * 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)
Dave Chinner250df6e2011-03-22 22:23:36 +110033 * inode->i_lock (in set_page_dirty's __mark_inode_dirty)
Christoph Hellwigf758eea2011-04-21 18:19:44 -060034 * bdi.wb->list_lock (in set_page_dirty's __mark_inode_dirty)
Nick Piggin82591e62006-10-19 23:29:10 -070035 * 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,
Christoph Hellwigf758eea2011-04-21 18:19:44 -060038 * within bdi.wb->list_lock in __sync_single_inode)
Andi Kleen6a460792009-09-16 11:50:15 +020039 *
Peter Zijlstra9b679322011-06-27 16:18:09 -070040 * anon_vma->mutex,mapping->i_mutex (memory_failure, collect_procs_anon)
41 * ->tasklist_lock
Andi Kleen6a460792009-09-16 11:50:15 +020042 * pte map lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070043 */
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>
Hugh Dickins5ad64682009-12-14 17:59:24 -080051#include <linux/ksm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#include <linux/rmap.h>
53#include <linux/rcupdate.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040054#include <linux/export.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080055#include <linux/memcontrol.h>
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070056#include <linux/mmu_notifier.h>
KOSAKI Motohiro64cdd542009-01-06 14:39:16 -080057#include <linux/migrate.h>
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +090058#include <linux/hugetlb.h>
Jan Karaef5d4372012-10-25 13:37:31 -070059#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
61#include <asm/tlbflush.h>
62
Nick Pigginb291f002008-10-18 20:26:44 -070063#include "internal.h"
64
Adrian Bunkfdd2e5f2008-10-18 20:28:38 -070065static struct kmem_cache *anon_vma_cachep;
Rik van Riel5beb4932010-03-05 13:42:07 -080066static struct kmem_cache *anon_vma_chain_cachep;
Adrian Bunkfdd2e5f2008-10-18 20:28:38 -070067
68static inline struct anon_vma *anon_vma_alloc(void)
69{
Peter Zijlstra01d8b202011-03-22 16:32:49 -070070 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;
Adrian Bunkfdd2e5f2008-10-18 20:28:38 -070083}
84
Peter Zijlstra01d8b202011-03-22 16:32:49 -070085static inline void anon_vma_free(struct anon_vma *anon_vma)
Adrian Bunkfdd2e5f2008-10-18 20:28:38 -070086{
Peter Zijlstra01d8b202011-03-22 16:32:49 -070087 VM_BUG_ON(atomic_read(&anon_vma->refcount));
Peter Zijlstra88c22082011-05-24 17:12:13 -070088
89 /*
90 * Synchronize against page_lock_anon_vma() such that
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
96 * mutex_trylock() from page_lock_anon_vma(). This orders:
97 *
98 * page_lock_anon_vma() VS put_anon_vma()
99 * mutex_trylock() atomic_dec_and_test()
100 * LOCK MB
101 * atomic_read() mutex_is_locked()
102 *
103 * LOCK should suffice since the actual taking of the lock must
104 * happen _before_ what follows.
105 */
106 if (mutex_is_locked(&anon_vma->root->mutex)) {
107 anon_vma_lock(anon_vma);
108 anon_vma_unlock(anon_vma);
109 }
110
Adrian Bunkfdd2e5f2008-10-18 20:28:38 -0700111 kmem_cache_free(anon_vma_cachep, anon_vma);
112}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113
Linus Torvaldsdd347392011-06-17 19:05:36 -0700114static inline struct anon_vma_chain *anon_vma_chain_alloc(gfp_t gfp)
Rik van Riel5beb4932010-03-05 13:42:07 -0800115{
Linus Torvaldsdd347392011-06-17 19:05:36 -0700116 return kmem_cache_alloc(anon_vma_chain_cachep, gfp);
Rik van Riel5beb4932010-03-05 13:42:07 -0800117}
118
Namhyung Kime574b5f2010-10-26 14:22:02 -0700119static void anon_vma_chain_free(struct anon_vma_chain *anon_vma_chain)
Rik van Riel5beb4932010-03-05 13:42:07 -0800120{
121 kmem_cache_free(anon_vma_chain_cachep, anon_vma_chain);
122}
123
Kautuk Consul6583a842012-03-21 16:34:01 -0700124static 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);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700131 anon_vma_interval_tree_insert(avc, &anon_vma->rb_root);
Kautuk Consul6583a842012-03-21 16:34:01 -0700132}
133
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700134/**
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
Figo.zhang23a07902010-12-27 15:14:06 +0100143 * not we either need to find an adjacent mapping that we
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700144 * 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
149 * optimistically looked up an anon_vma in page_lock_anon_vma()
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 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161int anon_vma_prepare(struct vm_area_struct *vma)
162{
163 struct anon_vma *anon_vma = vma->anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -0800164 struct anon_vma_chain *avc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165
166 might_sleep();
167 if (unlikely(!anon_vma)) {
168 struct mm_struct *mm = vma->vm_mm;
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700169 struct anon_vma *allocated;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170
Linus Torvaldsdd347392011-06-17 19:05:36 -0700171 avc = anon_vma_chain_alloc(GFP_KERNEL);
Rik van Riel5beb4932010-03-05 13:42:07 -0800172 if (!avc)
173 goto out_enomem;
174
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 anon_vma = find_mergeable_anon_vma(vma);
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700176 allocated = NULL;
177 if (!anon_vma) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 anon_vma = anon_vma_alloc();
179 if (unlikely(!anon_vma))
Rik van Riel5beb4932010-03-05 13:42:07 -0800180 goto out_enomem_free_avc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 allocated = anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 }
183
Rik van Rielcba48b92010-08-09 17:18:38 -0700184 anon_vma_lock(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 /* 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;
Kautuk Consul6583a842012-03-21 16:34:01 -0700189 anon_vma_chain_link(vma, avc, anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 allocated = NULL;
Oleg Nesterov31f2b0e2010-04-23 13:18:01 -0400191 avc = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 }
193 spin_unlock(&mm->page_table_lock);
Rik van Rielcba48b92010-08-09 17:18:38 -0700194 anon_vma_unlock(anon_vma);
Oleg Nesterov31f2b0e2010-04-23 13:18:01 -0400195
196 if (unlikely(allocated))
Peter Zijlstra01d8b202011-03-22 16:32:49 -0700197 put_anon_vma(allocated);
Oleg Nesterov31f2b0e2010-04-23 13:18:01 -0400198 if (unlikely(avc))
Rik van Riel5beb4932010-03-05 13:42:07 -0800199 anon_vma_chain_free(avc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 }
201 return 0;
Rik van Riel5beb4932010-03-05 13:42:07 -0800202
203 out_enomem_free_avc:
204 anon_vma_chain_free(avc);
205 out_enomem:
206 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207}
208
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700209/*
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))
222 mutex_unlock(&root->mutex);
223 root = new_root;
224 mutex_lock(&root->mutex);
225 }
226 return root;
227}
228
229static inline void unlock_anon_vma_root(struct anon_vma *root)
230{
231 if (root)
232 mutex_unlock(&root->mutex);
233}
234
Rik van Riel5beb4932010-03-05 13:42:07 -0800235/*
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)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240{
Rik van Riel5beb4932010-03-05 13:42:07 -0800241 struct anon_vma_chain *avc, *pavc;
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700242 struct anon_vma *root = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243
Linus Torvalds646d87b2010-04-11 17:15:03 -0700244 list_for_each_entry_reverse(pavc, &src->anon_vma_chain, same_vma) {
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700245 struct anon_vma *anon_vma;
246
Linus Torvaldsdd347392011-06-17 19:05:36 -0700247 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 }
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700255 anon_vma = pavc->anon_vma;
256 root = lock_anon_vma_root(root, anon_vma);
257 anon_vma_chain_link(dst, avc, anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 }
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700259 unlock_anon_vma_root(root);
Rik van Riel5beb4932010-03-05 13:42:07 -0800260 return 0;
261
262 enomem_failure:
263 unlink_anon_vmas(dst);
264 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265}
266
Rik van Riel5beb4932010-03-05 13:42:07 -0800267/*
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)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273{
Rik van Riel5beb4932010-03-05 13:42:07 -0800274 struct anon_vma_chain *avc;
275 struct anon_vma *anon_vma;
276
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;
Linus Torvaldsdd347392011-06-17 19:05:36 -0700292 avc = anon_vma_chain_alloc(GFP_KERNEL);
Rik van Riel5beb4932010-03-05 13:42:07 -0800293 if (!avc)
294 goto out_error_free_anon_vma;
Rik van Riel5c341ee12010-08-09 17:18:39 -0700295
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;
Rik van Riel76545062010-08-09 17:18:41 -0700301 /*
Peter Zijlstra01d8b202011-03-22 16:32:49 -0700302 * 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.
Rik van Riel76545062010-08-09 17:18:41 -0700305 */
306 get_anon_vma(anon_vma->root);
Rik van Riel5beb4932010-03-05 13:42:07 -0800307 /* Mark this anon_vma as the one where our new (COWed) pages go. */
308 vma->anon_vma = anon_vma;
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700309 anon_vma_lock(anon_vma);
Rik van Riel5c341ee12010-08-09 17:18:39 -0700310 anon_vma_chain_link(vma, avc, anon_vma);
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700311 anon_vma_unlock(anon_vma);
Rik van Riel5beb4932010-03-05 13:42:07 -0800312
313 return 0;
314
315 out_error_free_anon_vma:
Peter Zijlstra01d8b202011-03-22 16:32:49 -0700316 put_anon_vma(anon_vma);
Rik van Riel5beb4932010-03-05 13:42:07 -0800317 out_error:
Rik van Riel4946d542010-04-05 12:13:33 -0400318 unlink_anon_vmas(vma);
Rik van Riel5beb4932010-03-05 13:42:07 -0800319 return -ENOMEM;
320}
321
Rik van Riel5beb4932010-03-05 13:42:07 -0800322void unlink_anon_vmas(struct vm_area_struct *vma)
323{
324 struct anon_vma_chain *avc, *next;
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200325 struct anon_vma *root = NULL;
Rik van Riel5beb4932010-03-05 13:42:07 -0800326
Rik van Riel5c341ee12010-08-09 17:18:39 -0700327 /*
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 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800331 list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) {
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200332 struct anon_vma *anon_vma = avc->anon_vma;
333
334 root = lock_anon_vma_root(root, anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700335 anon_vma_interval_tree_remove(avc, &anon_vma->rb_root);
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200336
337 /*
338 * Leave empty anon_vmas on the list - we'll need
339 * to free them outside the lock.
340 */
Michel Lespinassebf181b92012-10-08 16:31:39 -0700341 if (RB_EMPTY_ROOT(&anon_vma->rb_root))
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200342 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()
352 * needing to acquire the anon_vma->root->mutex.
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
Rik van Riel5beb4932010-03-05 13:42:07 -0800359 list_del(&avc->same_vma);
360 anon_vma_chain_free(avc);
361 }
362}
363
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700364static void anon_vma_ctor(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365{
Christoph Lametera35afb82007-05-16 22:10:57 -0700366 struct anon_vma *anon_vma = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367
Peter Zijlstra2b575eb2011-05-24 17:12:11 -0700368 mutex_init(&anon_vma->mutex);
Peter Zijlstra83813262011-03-22 16:32:48 -0700369 atomic_set(&anon_vma->refcount, 0);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700370 anon_vma->rb_root = RB_ROOT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371}
372
373void __init anon_vma_init(void)
374{
375 anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma),
Paul Mundt20c2df82007-07-20 10:11:58 +0900376 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor);
Rik van Riel5beb4932010-03-05 13:42:07 -0800377 anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain, SLAB_PANIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378}
379
380/*
Peter Zijlstra6111e4c2011-05-24 17:12:08 -0700381 * 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 *
Peter Zijlstrabc658c92011-05-29 10:33:44 +0200390 * 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 *
Peter Zijlstra6111e4c2011-05-24 17:12:08 -0700395 * 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.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 */
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700403struct anon_vma *page_get_anon_vma(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404{
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700405 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 unsigned long anon_mapping;
407
408 rcu_read_lock();
Hugh Dickins80e148222009-12-14 17:59:29 -0800409 anon_mapping = (unsigned long) ACCESS_ONCE(page->mapping);
Hugh Dickins3ca7b3c2009-12-14 17:58:57 -0800410 if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 goto out;
412 if (!page_mapped(page))
413 goto out;
414
415 anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700416 if (!atomic_inc_not_zero(&anon_vma->refcount)) {
417 anon_vma = NULL;
418 goto out;
419 }
Hugh Dickinsf18194272010-08-25 23:12:54 -0700420
421 /*
422 * If this page is still mapped, then its anon_vma cannot have been
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700423 * 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).
Hugh Dickinsf18194272010-08-25 23:12:54 -0700427 */
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700428 if (!page_mapped(page)) {
429 put_anon_vma(anon_vma);
430 anon_vma = NULL;
431 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432out:
433 rcu_read_unlock();
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700434
435 return anon_vma;
436}
437
Peter Zijlstra88c22082011-05-24 17:12:13 -0700438/*
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 */
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700445struct anon_vma *page_lock_anon_vma(struct page *page)
446{
Peter Zijlstra88c22082011-05-24 17:12:13 -0700447 struct anon_vma *anon_vma = NULL;
Hugh Dickinseee0f252011-05-28 13:20:21 -0700448 struct anon_vma *root_anon_vma;
Peter Zijlstra88c22082011-05-24 17:12:13 -0700449 unsigned long anon_mapping;
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700450
Peter Zijlstra88c22082011-05-24 17:12:13 -0700451 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;
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700457
Peter Zijlstra88c22082011-05-24 17:12:13 -0700458 anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
Hugh Dickinseee0f252011-05-28 13:20:21 -0700459 root_anon_vma = ACCESS_ONCE(anon_vma->root);
460 if (mutex_trylock(&root_anon_vma->mutex)) {
Peter Zijlstra88c22082011-05-24 17:12:13 -0700461 /*
Hugh Dickinseee0f252011-05-28 13:20:21 -0700462 * If the page is still mapped, then this anon_vma is still
463 * its anon_vma, and holding the mutex ensures that it will
Peter Zijlstrabc658c92011-05-29 10:33:44 +0200464 * not go away, see anon_vma_free().
Peter Zijlstra88c22082011-05-24 17:12:13 -0700465 */
Hugh Dickinseee0f252011-05-28 13:20:21 -0700466 if (!page_mapped(page)) {
467 mutex_unlock(&root_anon_vma->mutex);
Peter Zijlstra88c22082011-05-24 17:12:13 -0700468 anon_vma = NULL;
469 }
470 goto out;
471 }
472
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();
487 anon_vma_lock(anon_vma);
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
493 * we'll deadlock on the anon_vma_lock() recursion.
494 */
495 anon_vma_unlock(anon_vma);
496 __put_anon_vma(anon_vma);
497 anon_vma = NULL;
498 }
499
500 return anon_vma;
501
502out:
503 rcu_read_unlock();
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700504 return anon_vma;
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800505}
506
Andi Kleen10be22d2009-09-16 11:50:04 +0200507void page_unlock_anon_vma(struct anon_vma *anon_vma)
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800508{
Rik van Rielcba48b92010-08-09 17:18:38 -0700509 anon_vma_unlock(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510}
511
512/*
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -0800513 * At what user virtual address is page expected in @vma?
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 */
Michel Lespinasse86c2ad12012-10-08 16:31:42 -0700515static inline unsigned long
516__vma_address(struct page *page, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517{
518 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +0900520 if (unlikely(is_vm_hugetlb_page(vma)))
521 pgoff = page->index << huge_page_order(page_hstate(page));
Michel Lespinasse86c2ad12012-10-08 16:31:42 -0700522
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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 return address;
535}
536
537/*
Huang Shijiebf89c8c2009-10-01 15:44:04 -0700538 * At what user virtual address is page expected in vma?
Naoya Horiguchiab941e02010-05-11 14:06:55 -0700539 * Caller should check the page is actually part of the vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 */
541unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma)
542{
Michel Lespinasse86c2ad12012-10-08 16:31:42 -0700543 unsigned long address;
Andrea Arcangeli21d0d442010-08-09 17:19:10 -0700544 if (PageAnon(page)) {
Hugh Dickins4829b902010-10-02 17:46:06 -0700545 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)
Andrea Arcangeli21d0d442010-08-09 17:19:10 -0700552 return -EFAULT;
553 } else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) {
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800554 if (!vma->vm_file ||
555 vma->vm_file->f_mapping != page->mapping)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 return -EFAULT;
557 } else
558 return -EFAULT;
Michel Lespinasse86c2ad12012-10-08 16:31:42 -0700559 address = __vma_address(page, vma);
560 if (unlikely(address < vma->vm_start || address >= vma->vm_end))
561 return -EFAULT;
562 return address;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563}
564
Bob Liu62190492012-12-11 16:00:37 -0800565pmd_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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586/*
Nikita Danilov81b40822005-05-01 08:58:36 -0700587 * Check that @page is mapped at @address into @mm.
588 *
Nick Piggin479db0b2008-08-20 14:09:18 -0700589 * 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 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -0700593 * On success returns with pte mapped and locked.
Nikita Danilov81b40822005-05-01 08:58:36 -0700594 */
Namhyung Kime9a81a82010-10-26 14:22:01 -0700595pte_t *__page_check_address(struct page *page, struct mm_struct *mm,
Nick Piggin479db0b2008-08-20 14:09:18 -0700596 unsigned long address, spinlock_t **ptlp, int sync)
Nikita Danilov81b40822005-05-01 08:58:36 -0700597{
Nikita Danilov81b40822005-05-01 08:58:36 -0700598 pmd_t *pmd;
599 pte_t *pte;
Hugh Dickinsc0718802005-10-29 18:16:31 -0700600 spinlock_t *ptl;
Nikita Danilov81b40822005-05-01 08:58:36 -0700601
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +0900602 if (unlikely(PageHuge(page))) {
603 pte = huge_pte_offset(mm, address);
604 ptl = &mm->page_table_lock;
605 goto check;
606 }
607
Bob Liu62190492012-12-11 16:00:37 -0800608 pmd = mm_find_pmd(mm, address);
609 if (!pmd)
Hugh Dickinsc0718802005-10-29 18:16:31 -0700610 return NULL;
611
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800612 if (pmd_trans_huge(*pmd))
613 return NULL;
Hugh Dickinsc0718802005-10-29 18:16:31 -0700614
615 pte = pte_offset_map(pmd, address);
616 /* Make a quick check before getting the lock */
Nick Piggin479db0b2008-08-20 14:09:18 -0700617 if (!sync && !pte_present(*pte)) {
Hugh Dickinsc0718802005-10-29 18:16:31 -0700618 pte_unmap(pte);
619 return NULL;
Nikita Danilov81b40822005-05-01 08:58:36 -0700620 }
Hugh Dickinsc0718802005-10-29 18:16:31 -0700621
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700622 ptl = pte_lockptr(mm, pmd);
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +0900623check:
Hugh Dickinsc0718802005-10-29 18:16:31 -0700624 spin_lock(ptl);
625 if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) {
626 *ptlp = ptl;
627 return pte;
628 }
629 pte_unmap_unlock(pte, ptl);
630 return NULL;
Nikita Danilov81b40822005-05-01 08:58:36 -0700631}
632
Nick Pigginb291f002008-10-18 20:26:44 -0700633/**
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 */
Andi Kleen6a460792009-09-16 11:50:15 +0200642int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma)
Nick Pigginb291f002008-10-18 20:26:44 -0700643{
644 unsigned long address;
645 pte_t *pte;
646 spinlock_t *ptl;
647
Michel Lespinasse86c2ad12012-10-08 16:31:42 -0700648 address = __vma_address(page, vma);
649 if (unlikely(address < vma->vm_start || address >= vma->vm_end))
Nick Pigginb291f002008-10-18 20:26:44 -0700650 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
Nikita Danilov81b40822005-05-01 08:58:36 -0700659/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 * Subfunctions of page_referenced: page_referenced_one called
661 * repeatedly from either page_referenced_anon or page_referenced_file.
662 */
Hugh Dickins5ad64682009-12-14 17:59:24 -0800663int page_referenced_one(struct page *page, struct vm_area_struct *vma,
664 unsigned long address, unsigned int *mapcount,
665 unsigned long *vm_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666{
667 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 int referenced = 0;
669
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800670 if (unlikely(PageTransHuge(page))) {
671 pmd_t *pmd;
672
673 spin_lock(&mm->page_table_lock);
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800674 /*
675 * rmap might return false positives; we must filter
676 * these out using page_check_address_pmd().
677 */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800678 pmd = page_check_address_pmd(page, mm, address,
679 PAGE_CHECK_ADDRESS_PMD_FLAG);
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800680 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))
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800694 referenced++;
695 spin_unlock(&mm->page_table_lock);
696 } else {
697 pte_t *pte;
698 spinlock_t *ptl;
699
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800700 /*
701 * rmap might return false positives; we must filter
702 * these out using page_check_address().
703 */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800704 pte = page_check_address(page, mm, address, &ptl, 0);
705 if (!pte)
706 goto out;
707
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800708 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
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800715 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
Hugh Dickinsc0718802005-10-29 18:16:31 -0700729 (*mapcount)--;
Huang Shijie273f0472009-12-14 17:58:51 -0800730
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700731 if (referenced)
732 *vm_flags |= vma->vm_flags;
Huang Shijie273f0472009-12-14 17:58:51 -0800733out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 return referenced;
735}
736
Balbir Singhbed71612008-02-07 00:14:01 -0800737static int page_referenced_anon(struct page *page,
Johannes Weiner72835c82012-01-12 17:18:32 -0800738 struct mem_cgroup *memcg,
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700739 unsigned long *vm_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740{
741 unsigned int mapcount;
742 struct anon_vma *anon_vma;
Michel Lespinassebf181b92012-10-08 16:31:39 -0700743 pgoff_t pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -0800744 struct anon_vma_chain *avc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 int referenced = 0;
746
747 anon_vma = page_lock_anon_vma(page);
748 if (!anon_vma)
749 return referenced;
750
751 mapcount = page_mapcount(page);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700752 pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
753 anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) {
Rik van Riel5beb4932010-03-05 13:42:07 -0800754 struct vm_area_struct *vma = avc->vma;
Hugh Dickins1cb17292009-12-14 17:59:01 -0800755 unsigned long address = vma_address(page, vma);
Balbir Singhbed71612008-02-07 00:14:01 -0800756 /*
757 * If we are reclaiming on behalf of a cgroup, skip
758 * counting on behalf of references from different
759 * cgroups
760 */
Johannes Weiner72835c82012-01-12 17:18:32 -0800761 if (memcg && !mm_match_cgroup(vma->vm_mm, memcg))
Balbir Singhbed71612008-02-07 00:14:01 -0800762 continue;
Hugh Dickins1cb17292009-12-14 17:59:01 -0800763 referenced += page_referenced_one(page, vma, address,
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700764 &mapcount, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 if (!mapcount)
766 break;
767 }
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800768
769 page_unlock_anon_vma(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 return referenced;
771}
772
773/**
774 * page_referenced_file - referenced check for object-based rmap
775 * @page: the page we're checking references on.
Johannes Weiner72835c82012-01-12 17:18:32 -0800776 * @memcg: target memory control group
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700777 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 *
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 */
Balbir Singhbed71612008-02-07 00:14:01 -0800786static int page_referenced_file(struct page *page,
Johannes Weiner72835c82012-01-12 17:18:32 -0800787 struct mem_cgroup *memcg,
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700788 unsigned long *vm_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789{
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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 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,
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700807 * so we can safely take mapping->i_mmap_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 */
809 BUG_ON(!PageLocked(page));
810
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700811 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812
813 /*
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700814 * i_mmap_mutex does not stabilize mapcount at all, but mapcount
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 * is more likely to be accurate if we note it after spinning.
816 */
817 mapcount = page_mapcount(page);
818
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700819 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
Hugh Dickins1cb17292009-12-14 17:59:01 -0800820 unsigned long address = vma_address(page, vma);
Balbir Singhbed71612008-02-07 00:14:01 -0800821 /*
822 * If we are reclaiming on behalf of a cgroup, skip
823 * counting on behalf of references from different
824 * cgroups
825 */
Johannes Weiner72835c82012-01-12 17:18:32 -0800826 if (memcg && !mm_match_cgroup(vma->vm_mm, memcg))
Balbir Singhbed71612008-02-07 00:14:01 -0800827 continue;
Hugh Dickins1cb17292009-12-14 17:59:01 -0800828 referenced += page_referenced_one(page, vma, address,
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700829 &mapcount, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 if (!mapcount)
831 break;
832 }
833
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700834 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 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
Johannes Weiner72835c82012-01-12 17:18:32 -0800842 * @memcg: target memory cgroup
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700843 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 *
845 * Quick test_and_clear_referenced for all mappings to a page,
846 * returns the number of ptes which referenced the page.
847 */
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700848int page_referenced(struct page *page,
849 int is_locked,
Johannes Weiner72835c82012-01-12 17:18:32 -0800850 struct mem_cgroup *memcg,
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700851 unsigned long *vm_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852{
853 int referenced = 0;
Hugh Dickins5ad64682009-12-14 17:59:24 -0800854 int we_locked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700856 *vm_flags = 0;
Hugh Dickins3ca7b3c2009-12-14 17:58:57 -0800857 if (page_mapped(page) && page_rmapping(page)) {
Hugh Dickins5ad64682009-12-14 17:59:24 -0800858 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)))
Johannes Weiner72835c82012-01-12 17:18:32 -0800866 referenced += page_referenced_ksm(page, memcg,
Hugh Dickins5ad64682009-12-14 17:59:24 -0800867 vm_flags);
868 else if (PageAnon(page))
Johannes Weiner72835c82012-01-12 17:18:32 -0800869 referenced += page_referenced_anon(page, memcg,
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700870 vm_flags);
Hugh Dickins5ad64682009-12-14 17:59:24 -0800871 else if (page->mapping)
Johannes Weiner72835c82012-01-12 17:18:32 -0800872 referenced += page_referenced_file(page, memcg,
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700873 vm_flags);
Hugh Dickins5ad64682009-12-14 17:59:24 -0800874 if (we_locked)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 unlock_page(page);
Martin Schwidefsky50a15982011-07-24 10:47:59 +0200876
877 if (page_test_and_clear_young(page_to_pfn(page)))
878 referenced++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 }
Hugh Dickins5ad64682009-12-14 17:59:24 -0800880out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 return referenced;
882}
883
Hugh Dickins1cb17292009-12-14 17:59:01 -0800884static int page_mkclean_one(struct page *page, struct vm_area_struct *vma,
885 unsigned long address)
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700886{
887 struct mm_struct *mm = vma->vm_mm;
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100888 pte_t *pte;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700889 spinlock_t *ptl;
890 int ret = 0;
891
Nick Piggin479db0b2008-08-20 14:09:18 -0700892 pte = page_check_address(page, mm, address, &ptl, 1);
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700893 if (!pte)
894 goto out;
895
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100896 if (pte_dirty(*pte) || pte_write(*pte)) {
897 pte_t entry;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700898
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100899 flush_cache_page(vma, address, pte_pfn(*pte));
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700900 entry = ptep_clear_flush(vma, address, pte);
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100901 entry = pte_wrprotect(entry);
902 entry = pte_mkclean(entry);
Al Virod6e88e62006-12-29 16:48:35 -0800903 set_pte_at(mm, address, pte, entry);
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100904 ret = 1;
905 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700906
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700907 pte_unmap_unlock(pte, ptl);
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700908
909 if (ret)
910 mmu_notifier_invalidate_page(mm, address);
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700911out:
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;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700919 int ret = 0;
920
921 BUG_ON(PageAnon(page));
922
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700923 mutex_lock(&mapping->i_mmap_mutex);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700924 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
Hugh Dickins1cb17292009-12-14 17:59:01 -0800925 if (vma->vm_flags & VM_SHARED) {
926 unsigned long address = vma_address(page, vma);
Hugh Dickins1cb17292009-12-14 17:59:01 -0800927 ret += page_mkclean_one(page, vma, address);
928 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700929 }
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700930 mutex_unlock(&mapping->i_mmap_mutex);
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700931 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);
Jan Karaef5d4372012-10-25 13:37:31 -0700942 if (mapping)
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700943 ret = page_mkclean_file(mapping, page);
944 }
945
946 return ret;
947}
Jaya Kumar60b59be2007-05-08 00:37:37 -0700948EXPORT_SYMBOL_GPL(page_mkclean);
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700949
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950/**
Rik van Rielc44b6742010-03-05 13:42:09 -0800951 * 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
974/**
Andi Kleen4e1c1972010-09-22 12:43:56 +0200975 * __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
Rik van Riele8a03fe2010-04-14 17:59:28 -0400979 * @exclusive: the page is exclusively owned by the current process
Nick Piggin9617d952006-01-06 00:11:12 -0800980 */
981static void __page_set_anon_rmap(struct page *page,
Rik van Riele8a03fe2010-04-14 17:59:28 -0400982 struct vm_area_struct *vma, unsigned long address, int exclusive)
Nick Piggin9617d952006-01-06 00:11:12 -0800983{
Rik van Riele8a03fe2010-04-14 17:59:28 -0400984 struct anon_vma *anon_vma = vma->anon_vma;
Nick Piggin9617d952006-01-06 00:11:12 -0800985
Rik van Riele8a03fe2010-04-14 17:59:28 -0400986 BUG_ON(!anon_vma);
Linus Torvaldsea900022010-04-12 12:44:29 -0700987
Andi Kleen4e1c1972010-09-22 12:43:56 +0200988 if (PageAnon(page))
989 return;
990
Linus Torvaldsea900022010-04-12 12:44:29 -0700991 /*
Rik van Riele8a03fe2010-04-14 17:59:28 -0400992 * If the page isn't exclusively mapped into this vma,
993 * we must use the _oldest_ possible anon_vma for the
994 * page mapping!
Linus Torvaldsea900022010-04-12 12:44:29 -0700995 */
Andi Kleen4e1c1972010-09-22 12:43:56 +0200996 if (!exclusive)
Andrea Arcangeli288468c2010-08-09 17:19:09 -0700997 anon_vma = anon_vma->root;
Linus Torvaldsea900022010-04-12 12:44:29 -0700998
Nick Piggin9617d952006-01-06 00:11:12 -0800999 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
1000 page->mapping = (struct address_space *) anon_vma;
Nick Piggin9617d952006-01-06 00:11:12 -08001001 page->index = linear_page_index(vma, address);
Nick Piggin9617d952006-01-06 00:11:12 -08001002}
1003
1004/**
Randy Dunlap43d8eac2008-03-19 17:00:43 -07001005 * __page_check_anon_rmap - sanity check anonymous rmap addition
Nick Pigginc97a9e12007-05-16 22:11:21 -07001006 * @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 */
Andrea Arcangeli44ab57a2010-08-09 17:19:10 -07001026 BUG_ON(page_anon_vma(page)->root != vma->anon_vma->root);
Nick Pigginc97a9e12007-05-16 22:11:21 -07001027 BUG_ON(page->index != linear_page_index(vma, address));
1028#endif
1029}
1030
1031/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 * 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 *
Hugh Dickins5ad64682009-12-14 17:59:24 -08001037 * The caller needs to hold the pte lock, and the page must be locked in
Hugh Dickins80e148222009-12-14 17:59:29 -08001038 * 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).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 */
1042void page_add_anon_rmap(struct page *page,
1043 struct vm_area_struct *vma, unsigned long address)
1044{
Rik van Rielad8c2ee2010-08-09 17:19:48 -07001045 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)
1055{
Hugh Dickins5ad64682009-12-14 17:59:24 -08001056 int first = atomic_inc_and_test(&page->_mapcount);
Andrea Arcangeli79134172011-01-13 15:46:58 -08001057 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 }
Hugh Dickins5ad64682009-12-14 17:59:24 -08001064 if (unlikely(PageKsm(page)))
1065 return;
1066
Nick Pigginc97a9e12007-05-16 22:11:21 -07001067 VM_BUG_ON(!PageLocked(page));
Hugh Dickins5dbe0af2011-05-28 13:17:04 -07001068 /* address might be in next vma when migration races vma_adjust */
Hugh Dickins5ad64682009-12-14 17:59:24 -08001069 if (first)
Rik van Rielad8c2ee2010-08-09 17:19:48 -07001070 __page_set_anon_rmap(page, vma, address, exclusive);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001071 else
Nick Pigginc97a9e12007-05-16 22:11:21 -07001072 __page_check_anon_rmap(page, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073}
1074
Randy Dunlap43d8eac2008-03-19 17:00:43 -07001075/**
Nick Piggin9617d952006-01-06 00:11:12 -08001076 * 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.
Nick Pigginc97a9e12007-05-16 22:11:21 -07001083 * Page does not have to be locked.
Nick Piggin9617d952006-01-06 00:11:12 -08001084 */
1085void page_add_new_anon_rmap(struct page *page,
1086 struct vm_area_struct *vma, unsigned long address)
1087{
Hugh Dickinsb5934c52009-01-06 14:39:25 -08001088 VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end);
Hugh Dickinscbf84b72009-01-06 14:39:27 -08001089 SetPageSwapBacked(page);
1090 atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */
Andrea Arcangeli79134172011-01-13 15:46:58 -08001091 if (!PageTransHuge(page))
1092 __inc_zone_page_state(page, NR_ANON_PAGES);
1093 else
1094 __inc_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES);
Rik van Riele8a03fe2010-04-14 17:59:28 -04001095 __page_set_anon_rmap(page, vma, address, 1);
Hugh Dickins39b5f292012-10-08 16:33:18 -07001096 if (!mlocked_vma_newpage(vma, page))
Hugh Dickinscbf84b72009-01-06 14:39:27 -08001097 lru_cache_add_lru(page, LRU_ACTIVE_ANON);
Hugh Dickinsb5934c52009-01-06 14:39:25 -08001098 else
1099 add_page_to_unevictable_list(page);
Nick Piggin9617d952006-01-06 00:11:12 -08001100}
1101
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102/**
1103 * page_add_file_rmap - add pte mapping to a file page
1104 * @page: the page to add the mapping to
1105 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -07001106 * The caller needs to hold the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 */
1108void page_add_file_rmap(struct page *page)
1109{
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001110 bool locked;
1111 unsigned long flags;
1112
1113 mem_cgroup_begin_update_page_stat(page, &locked, &flags);
Balbir Singhd69b0422009-06-17 16:26:34 -07001114 if (atomic_inc_and_test(&page->_mapcount)) {
Christoph Lameter65ba55f2006-06-30 01:55:34 -07001115 __inc_zone_page_state(page, NR_FILE_MAPPED);
Greg Thelen2a7106f2011-01-13 15:47:37 -08001116 mem_cgroup_inc_page_stat(page, MEMCG_NR_FILE_MAPPED);
Balbir Singhd69b0422009-06-17 16:26:34 -07001117 }
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001118 mem_cgroup_end_update_page_stat(page, &locked, &flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119}
1120
1121/**
1122 * page_remove_rmap - take down pte mapping from a page
1123 * @page: page to remove mapping from
1124 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -07001125 * The caller needs to hold the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 */
Hugh Dickinsedc315f2009-01-06 14:40:11 -08001127void page_remove_rmap(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128{
Jan Karaef5d4372012-10-25 13:37:31 -07001129 struct address_space *mapping = page_mapping(page);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001130 bool anon = PageAnon(page);
1131 bool locked;
1132 unsigned long flags;
1133
1134 /*
1135 * The anon case has no mem_cgroup page_stat to update; but may
1136 * uncharge_page() below, where the lock ordering can deadlock if
1137 * we hold the lock against page_stat move: so avoid it on anon.
1138 */
1139 if (!anon)
1140 mem_cgroup_begin_update_page_stat(page, &locked, &flags);
1141
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001142 /* page still mapped by someone else? */
1143 if (!atomic_add_negative(-1, &page->_mapcount))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001144 goto out;
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001145
1146 /*
1147 * Now that the last pte has gone, s390 must transfer dirty
1148 * flag from storage key to struct page. We can usually skip
1149 * this if the page is anon, so about to be freed; but perhaps
1150 * not if it's in swapcache - there might be another pte slot
1151 * containing the swap entry, but page not yet written to swap.
Jan Karaef5d4372012-10-25 13:37:31 -07001152 *
1153 * And we can skip it on file pages, so long as the filesystem
Jan Karae749eb92012-12-11 16:01:09 -08001154 * participates in dirty tracking (note that this is not only an
1155 * optimization but also solves problems caused by dirty flag in
1156 * storage key getting set by a write from inside kernel); but need to
1157 * catch shm and tmpfs and ramfs pages which have been modified since
1158 * creation by read fault.
Jan Karaef5d4372012-10-25 13:37:31 -07001159 *
1160 * Note that mapping must be decided above, before decrementing
1161 * mapcount (which luckily provides a barrier): once page is unmapped,
1162 * it could be truncated and page->mapping reset to NULL at any moment.
1163 * Note also that we are relying on page_mapping(page) to set mapping
1164 * to &swapper_space when PageSwapCache(page).
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001165 */
Jan Karaef5d4372012-10-25 13:37:31 -07001166 if (mapping && !mapping_cap_account_dirty(mapping) &&
Martin Schwidefsky2d425522011-05-23 10:24:39 +02001167 page_test_and_clear_dirty(page_to_pfn(page), 1))
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001168 set_page_dirty(page);
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001169 /*
1170 * Hugepages are not counted in NR_ANON_PAGES nor NR_FILE_MAPPED
1171 * and not charged by memcg for now.
1172 */
1173 if (unlikely(PageHuge(page)))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001174 goto out;
1175 if (anon) {
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001176 mem_cgroup_uncharge_page(page);
Andrea Arcangeli79134172011-01-13 15:46:58 -08001177 if (!PageTransHuge(page))
1178 __dec_zone_page_state(page, NR_ANON_PAGES);
1179 else
1180 __dec_zone_page_state(page,
1181 NR_ANON_TRANSPARENT_HUGEPAGES);
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001182 } else {
1183 __dec_zone_page_state(page, NR_FILE_MAPPED);
Greg Thelen2a7106f2011-01-13 15:47:37 -08001184 mem_cgroup_dec_page_stat(page, MEMCG_NR_FILE_MAPPED);
Hugh Dickinse6c509f2012-10-08 16:33:19 -07001185 mem_cgroup_end_update_page_stat(page, &locked, &flags);
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001186 }
Hugh Dickinse6c509f2012-10-08 16:33:19 -07001187 if (unlikely(PageMlocked(page)))
1188 clear_page_mlock(page);
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001189 /*
1190 * It would be tidy to reset the PageAnon mapping here,
1191 * but that might overwrite a racing page_add_anon_rmap
1192 * which increments mapcount after us but sets mapping
1193 * before us: so leave the reset to free_hot_cold_page,
1194 * and remember that it's only reliable while mapped.
1195 * Leaving it set also helps swapoff to reinstate ptes
1196 * faster for those pages still in swapcache.
1197 */
Hugh Dickinse6c509f2012-10-08 16:33:19 -07001198 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001199out:
1200 if (!anon)
1201 mem_cgroup_end_update_page_stat(page, &locked, &flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202}
1203
1204/*
1205 * Subfunctions of try_to_unmap: try_to_unmap_one called
Wanlong Gao99ef0312011-10-31 17:08:51 -07001206 * repeatedly from try_to_unmap_ksm, try_to_unmap_anon or try_to_unmap_file.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 */
Hugh Dickins5ad64682009-12-14 17:59:24 -08001208int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
1209 unsigned long address, enum ttu_flags flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210{
1211 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 pte_t *pte;
1213 pte_t pteval;
Hugh Dickinsc0718802005-10-29 18:16:31 -07001214 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 int ret = SWAP_AGAIN;
1216
Nick Piggin479db0b2008-08-20 14:09:18 -07001217 pte = page_check_address(page, mm, address, &ptl, 0);
Hugh Dickinsc0718802005-10-29 18:16:31 -07001218 if (!pte)
Nikita Danilov81b40822005-05-01 08:58:36 -07001219 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220
1221 /*
1222 * If the page is mlock()d, we cannot swap it out.
1223 * If it's recently referenced (perhaps page_referenced
1224 * skipped over this mm) then we should reactivate it.
1225 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001226 if (!(flags & TTU_IGNORE_MLOCK)) {
KOSAKI Motohirocaed0f482009-12-14 17:59:45 -08001227 if (vma->vm_flags & VM_LOCKED)
1228 goto out_mlock;
1229
Hugh Dickinsaf8e3352009-12-14 17:58:59 -08001230 if (TTU_ACTION(flags) == TTU_MUNLOCK)
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001231 goto out_unmap;
Andi Kleen14fa31b2009-09-16 11:50:10 +02001232 }
1233 if (!(flags & TTU_IGNORE_ACCESS)) {
Nick Pigginb291f002008-10-18 20:26:44 -07001234 if (ptep_clear_flush_young_notify(vma, address, pte)) {
1235 ret = SWAP_FAIL;
1236 goto out_unmap;
1237 }
1238 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 /* Nuke the page table entry. */
1241 flush_cache_page(vma, address, page_to_pfn(page));
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001242 pteval = ptep_clear_flush(vma, address, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243
1244 /* Move the dirty bit to the physical page now the pte is gone. */
1245 if (pte_dirty(pteval))
1246 set_page_dirty(page);
1247
Hugh Dickins365e9c872005-10-29 18:16:18 -07001248 /* Update high watermark before we lower rss */
1249 update_hiwater_rss(mm);
1250
Andi Kleen888b9f72009-09-16 11:50:11 +02001251 if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) {
1252 if (PageAnon(page))
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001253 dec_mm_counter(mm, MM_ANONPAGES);
Andi Kleen888b9f72009-09-16 11:50:11 +02001254 else
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001255 dec_mm_counter(mm, MM_FILEPAGES);
Andi Kleen888b9f72009-09-16 11:50:11 +02001256 set_pte_at(mm, address, pte,
1257 swp_entry_to_pte(make_hwpoison_entry(page)));
1258 } else if (PageAnon(page)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001259 swp_entry_t entry = { .val = page_private(page) };
Christoph Lameter06972122006-06-23 02:03:35 -07001260
1261 if (PageSwapCache(page)) {
1262 /*
1263 * Store the swap location in the pte.
1264 * See handle_pte_fault() ...
1265 */
Hugh Dickins570a335b2009-12-14 17:58:46 -08001266 if (swap_duplicate(entry) < 0) {
1267 set_pte_at(mm, address, pte, pteval);
1268 ret = SWAP_FAIL;
1269 goto out_unmap;
1270 }
Christoph Lameter06972122006-06-23 02:03:35 -07001271 if (list_empty(&mm->mmlist)) {
1272 spin_lock(&mmlist_lock);
1273 if (list_empty(&mm->mmlist))
1274 list_add(&mm->mmlist, &init_mm.mmlist);
1275 spin_unlock(&mmlist_lock);
1276 }
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001277 dec_mm_counter(mm, MM_ANONPAGES);
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08001278 inc_mm_counter(mm, MM_SWAPENTS);
Konstantin Khlebnikovce1744f2012-03-21 16:33:59 -07001279 } else if (IS_ENABLED(CONFIG_MIGRATION)) {
Christoph Lameter06972122006-06-23 02:03:35 -07001280 /*
1281 * Store the pfn of the page in a special migration
1282 * pte. do_swap_page() will wait until the migration
1283 * pte is removed and then restart fault handling.
1284 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001285 BUG_ON(TTU_ACTION(flags) != TTU_MIGRATION);
Christoph Lameter06972122006-06-23 02:03:35 -07001286 entry = make_migration_entry(page, pte_write(pteval));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 }
1288 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
1289 BUG_ON(pte_file(*pte));
Konstantin Khlebnikovce1744f2012-03-21 16:33:59 -07001290 } else if (IS_ENABLED(CONFIG_MIGRATION) &&
1291 (TTU_ACTION(flags) == TTU_MIGRATION)) {
Christoph Lameter04e62a22006-06-23 02:03:38 -07001292 /* Establish migration entry for a file page */
1293 swp_entry_t entry;
1294 entry = make_migration_entry(page, pte_write(pteval));
1295 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
1296 } else
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001297 dec_mm_counter(mm, MM_FILEPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298
Hugh Dickinsedc315f2009-01-06 14:40:11 -08001299 page_remove_rmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 page_cache_release(page);
1301
1302out_unmap:
Hugh Dickinsc0718802005-10-29 18:16:31 -07001303 pte_unmap_unlock(pte, ptl);
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001304 if (ret != SWAP_FAIL)
1305 mmu_notifier_invalidate_page(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306out:
1307 return ret;
KOSAKI Motohirocaed0f482009-12-14 17:59:45 -08001308
1309out_mlock:
1310 pte_unmap_unlock(pte, ptl);
1311
1312
1313 /*
1314 * We need mmap_sem locking, Otherwise VM_LOCKED check makes
1315 * unstable result and race. Plus, We can't wait here because
Peter Zijlstra2b575eb2011-05-24 17:12:11 -07001316 * we now hold anon_vma->mutex or mapping->i_mmap_mutex.
KOSAKI Motohirocaed0f482009-12-14 17:59:45 -08001317 * if trylock failed, the page remain in evictable lru and later
1318 * vmscan could retry to move the page to unevictable lru if the
1319 * page is actually mlocked.
1320 */
1321 if (down_read_trylock(&vma->vm_mm->mmap_sem)) {
1322 if (vma->vm_flags & VM_LOCKED) {
1323 mlock_vma_page(page);
1324 ret = SWAP_MLOCK;
1325 }
1326 up_read(&vma->vm_mm->mmap_sem);
1327 }
1328 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329}
1330
1331/*
1332 * objrmap doesn't work for nonlinear VMAs because the assumption that
1333 * offset-into-file correlates with offset-into-virtual-addresses does not hold.
1334 * Consequently, given a particular page and its ->index, we cannot locate the
1335 * ptes which are mapping that page without an exhaustive linear search.
1336 *
1337 * So what this code does is a mini "virtual scan" of each nonlinear VMA which
1338 * maps the file to which the target page belongs. The ->vm_private_data field
1339 * holds the current cursor into that scan. Successive searches will circulate
1340 * around the vma's virtual address space.
1341 *
1342 * So as more replacement pressure is applied to the pages in a nonlinear VMA,
1343 * more scanning pressure is placed against them as well. Eventually pages
1344 * will become fully unmapped and are eligible for eviction.
1345 *
1346 * For very sparsely populated VMAs this is a little inefficient - chances are
1347 * there there won't be many ptes located within the scan cluster. In this case
1348 * maybe we could scan further - to the end of the pte page, perhaps.
Nick Pigginb291f002008-10-18 20:26:44 -07001349 *
1350 * Mlocked pages: check VM_LOCKED under mmap_sem held for read, if we can
1351 * acquire it without blocking. If vma locked, mlock the pages in the cluster,
1352 * rather than unmapping them. If we encounter the "check_page" that vmscan is
1353 * trying to unmap, return SWAP_MLOCK, else default SWAP_AGAIN.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 */
1355#define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE)
1356#define CLUSTER_MASK (~(CLUSTER_SIZE - 1))
1357
Nick Pigginb291f002008-10-18 20:26:44 -07001358static int try_to_unmap_cluster(unsigned long cursor, unsigned int *mapcount,
1359 struct vm_area_struct *vma, struct page *check_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360{
1361 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 pmd_t *pmd;
Hugh Dickinsc0718802005-10-29 18:16:31 -07001363 pte_t *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 pte_t pteval;
Hugh Dickinsc0718802005-10-29 18:16:31 -07001365 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 struct page *page;
1367 unsigned long address;
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001368 unsigned long mmun_start; /* For mmu_notifiers */
1369 unsigned long mmun_end; /* For mmu_notifiers */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 unsigned long end;
Nick Pigginb291f002008-10-18 20:26:44 -07001371 int ret = SWAP_AGAIN;
1372 int locked_vma = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 address = (vma->vm_start + cursor) & CLUSTER_MASK;
1375 end = address + CLUSTER_SIZE;
1376 if (address < vma->vm_start)
1377 address = vma->vm_start;
1378 if (end > vma->vm_end)
1379 end = vma->vm_end;
1380
Bob Liu62190492012-12-11 16:00:37 -08001381 pmd = mm_find_pmd(mm, address);
1382 if (!pmd)
Nick Pigginb291f002008-10-18 20:26:44 -07001383 return ret;
1384
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001385 mmun_start = address;
1386 mmun_end = end;
1387 mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
1388
Nick Pigginb291f002008-10-18 20:26:44 -07001389 /*
Hugh Dickinsaf8e3352009-12-14 17:58:59 -08001390 * If we can acquire the mmap_sem for read, and vma is VM_LOCKED,
Nick Pigginb291f002008-10-18 20:26:44 -07001391 * keep the sem while scanning the cluster for mlocking pages.
1392 */
Hugh Dickinsaf8e3352009-12-14 17:58:59 -08001393 if (down_read_trylock(&vma->vm_mm->mmap_sem)) {
Nick Pigginb291f002008-10-18 20:26:44 -07001394 locked_vma = (vma->vm_flags & VM_LOCKED);
1395 if (!locked_vma)
1396 up_read(&vma->vm_mm->mmap_sem); /* don't need it */
1397 }
Hugh Dickinsc0718802005-10-29 18:16:31 -07001398
1399 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400
Hugh Dickins365e9c872005-10-29 18:16:18 -07001401 /* Update high watermark before we lower rss */
1402 update_hiwater_rss(mm);
1403
Hugh Dickinsc0718802005-10-29 18:16:31 -07001404 for (; address < end; pte++, address += PAGE_SIZE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 if (!pte_present(*pte))
1406 continue;
Linus Torvalds6aab3412005-11-28 14:34:23 -08001407 page = vm_normal_page(vma, address, *pte);
1408 BUG_ON(!page || PageAnon(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409
Nick Pigginb291f002008-10-18 20:26:44 -07001410 if (locked_vma) {
1411 mlock_vma_page(page); /* no-op if already mlocked */
1412 if (page == check_page)
1413 ret = SWAP_MLOCK;
1414 continue; /* don't unmap */
1415 }
1416
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001417 if (ptep_clear_flush_young_notify(vma, address, pte))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 continue;
1419
1420 /* Nuke the page table entry. */
Ben Collinseca35132005-11-29 11:45:26 -08001421 flush_cache_page(vma, address, pte_pfn(*pte));
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001422 pteval = ptep_clear_flush(vma, address, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423
1424 /* If nonlinear, store the file page offset in the pte. */
1425 if (page->index != linear_page_index(vma, address))
1426 set_pte_at(mm, address, pte, pgoff_to_pte(page->index));
1427
1428 /* Move the dirty bit to the physical page now the pte is gone. */
1429 if (pte_dirty(pteval))
1430 set_page_dirty(page);
1431
Hugh Dickinsedc315f2009-01-06 14:40:11 -08001432 page_remove_rmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 page_cache_release(page);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001434 dec_mm_counter(mm, MM_FILEPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 (*mapcount)--;
1436 }
Hugh Dickinsc0718802005-10-29 18:16:31 -07001437 pte_unmap_unlock(pte - 1, ptl);
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001438 mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
Nick Pigginb291f002008-10-18 20:26:44 -07001439 if (locked_vma)
1440 up_read(&vma->vm_mm->mmap_sem);
1441 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442}
1443
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001444bool is_vma_temporary_stack(struct vm_area_struct *vma)
Mel Gormana8bef8ff2010-05-24 14:32:24 -07001445{
1446 int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP);
1447
1448 if (!maybe_stack)
1449 return false;
1450
1451 if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) ==
1452 VM_STACK_INCOMPLETE_SETUP)
1453 return true;
1454
1455 return false;
1456}
1457
Nick Pigginb291f002008-10-18 20:26:44 -07001458/**
1459 * try_to_unmap_anon - unmap or unlock anonymous page using the object-based
1460 * rmap method
1461 * @page: the page to unmap/unlock
Huang Shijie8051be52009-12-14 17:58:50 -08001462 * @flags: action and flags
Nick Pigginb291f002008-10-18 20:26:44 -07001463 *
1464 * Find all the mappings of a page using the mapping pointer and the vma chains
1465 * contained in the anon_vma struct it points to.
1466 *
1467 * This function is only called from try_to_unmap/try_to_munlock for
1468 * anonymous pages.
1469 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1470 * where the page was found will be held for write. So, we won't recheck
1471 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1472 * 'LOCKED.
1473 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001474static int try_to_unmap_anon(struct page *page, enum ttu_flags flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475{
1476 struct anon_vma *anon_vma;
Michel Lespinassebf181b92012-10-08 16:31:39 -07001477 pgoff_t pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001478 struct anon_vma_chain *avc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 int ret = SWAP_AGAIN;
Nick Pigginb291f002008-10-18 20:26:44 -07001480
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 anon_vma = page_lock_anon_vma(page);
1482 if (!anon_vma)
1483 return ret;
1484
Michel Lespinassebf181b92012-10-08 16:31:39 -07001485 pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
1486 anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) {
Rik van Riel5beb4932010-03-05 13:42:07 -08001487 struct vm_area_struct *vma = avc->vma;
Mel Gormana8bef8ff2010-05-24 14:32:24 -07001488 unsigned long address;
1489
1490 /*
1491 * During exec, a temporary VMA is setup and later moved.
1492 * The VMA is moved under the anon_vma lock but not the
1493 * page tables leading to a race where migration cannot
1494 * find the migration ptes. Rather than increasing the
1495 * locking requirements of exec(), migration skips
1496 * temporary VMAs until after exec() completes.
1497 */
Konstantin Khlebnikovce1744f2012-03-21 16:33:59 -07001498 if (IS_ENABLED(CONFIG_MIGRATION) && (flags & TTU_MIGRATION) &&
Mel Gormana8bef8ff2010-05-24 14:32:24 -07001499 is_vma_temporary_stack(vma))
1500 continue;
1501
1502 address = vma_address(page, vma);
Hugh Dickins1cb17292009-12-14 17:59:01 -08001503 ret = try_to_unmap_one(page, vma, address, flags);
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001504 if (ret != SWAP_AGAIN || !page_mapped(page))
1505 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 }
Oleg Nesterov34bbd702007-02-28 20:13:49 -08001507
1508 page_unlock_anon_vma(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 return ret;
1510}
1511
1512/**
Nick Pigginb291f002008-10-18 20:26:44 -07001513 * try_to_unmap_file - unmap/unlock file page using the object-based rmap method
1514 * @page: the page to unmap/unlock
Andi Kleen14fa31b2009-09-16 11:50:10 +02001515 * @flags: action and flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 *
1517 * Find all the mappings of a page using the mapping pointer and the vma chains
1518 * contained in the address_space struct it points to.
1519 *
Nick Pigginb291f002008-10-18 20:26:44 -07001520 * This function is only called from try_to_unmap/try_to_munlock for
1521 * object-based pages.
1522 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1523 * where the page was found will be held for write. So, we won't recheck
1524 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1525 * 'LOCKED.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001527static int try_to_unmap_file(struct page *page, enum ttu_flags flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528{
1529 struct address_space *mapping = page->mapping;
1530 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
1531 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 int ret = SWAP_AGAIN;
1533 unsigned long cursor;
1534 unsigned long max_nl_cursor = 0;
1535 unsigned long max_nl_size = 0;
1536 unsigned int mapcount;
1537
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07001538 mutex_lock(&mapping->i_mmap_mutex);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07001539 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
Hugh Dickins1cb17292009-12-14 17:59:01 -08001540 unsigned long address = vma_address(page, vma);
Hugh Dickins1cb17292009-12-14 17:59:01 -08001541 ret = try_to_unmap_one(page, vma, address, flags);
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001542 if (ret != SWAP_AGAIN || !page_mapped(page))
1543 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 }
1545
1546 if (list_empty(&mapping->i_mmap_nonlinear))
1547 goto out;
1548
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001549 /*
1550 * We don't bother to try to find the munlocked page in nonlinears.
1551 * It's costly. Instead, later, page reclaim logic may call
1552 * try_to_unmap(TTU_MUNLOCK) and recover PG_mlocked lazily.
1553 */
1554 if (TTU_ACTION(flags) == TTU_MUNLOCK)
1555 goto out;
1556
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07001558 shared.nonlinear) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 cursor = (unsigned long) vma->vm_private_data;
1560 if (cursor > max_nl_cursor)
1561 max_nl_cursor = cursor;
1562 cursor = vma->vm_end - vma->vm_start;
1563 if (cursor > max_nl_size)
1564 max_nl_size = cursor;
1565 }
1566
Nick Pigginb291f002008-10-18 20:26:44 -07001567 if (max_nl_size == 0) { /* all nonlinears locked or reserved ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568 ret = SWAP_FAIL;
1569 goto out;
1570 }
1571
1572 /*
1573 * We don't try to search for this page in the nonlinear vmas,
1574 * and page_referenced wouldn't have found it anyway. Instead
1575 * just walk the nonlinear vmas trying to age and unmap some.
1576 * The mapcount of the page we came in with is irrelevant,
1577 * but even so use it as a guide to how hard we should try?
1578 */
1579 mapcount = page_mapcount(page);
1580 if (!mapcount)
1581 goto out;
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07001582 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583
1584 max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK;
1585 if (max_nl_cursor == 0)
1586 max_nl_cursor = CLUSTER_SIZE;
1587
1588 do {
1589 list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07001590 shared.nonlinear) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 cursor = (unsigned long) vma->vm_private_data;
Hugh Dickins839b9682005-09-03 15:54:43 -07001592 while ( cursor < max_nl_cursor &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 cursor < vma->vm_end - vma->vm_start) {
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001594 if (try_to_unmap_cluster(cursor, &mapcount,
1595 vma, page) == SWAP_MLOCK)
1596 ret = SWAP_MLOCK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597 cursor += CLUSTER_SIZE;
1598 vma->vm_private_data = (void *) cursor;
1599 if ((int)mapcount <= 0)
1600 goto out;
1601 }
1602 vma->vm_private_data = (void *) max_nl_cursor;
1603 }
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07001604 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605 max_nl_cursor += CLUSTER_SIZE;
1606 } while (max_nl_cursor <= max_nl_size);
1607
1608 /*
1609 * Don't loop forever (perhaps all the remaining pages are
1610 * in locked vmas). Reset cursor on all unreserved nonlinear
1611 * vmas, now forgetting on which ones it had fallen behind.
1612 */
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07001613 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.nonlinear)
Hugh Dickins101d2be2005-11-21 21:32:16 -08001614 vma->vm_private_data = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615out:
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07001616 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617 return ret;
1618}
1619
1620/**
1621 * try_to_unmap - try to remove all page table mappings to a page
1622 * @page: the page to get unmapped
Andi Kleen14fa31b2009-09-16 11:50:10 +02001623 * @flags: action and flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 *
1625 * Tries to remove all the page table entries which are mapping this
1626 * page, used in the pageout path. Caller must hold the page lock.
1627 * Return values are:
1628 *
1629 * SWAP_SUCCESS - we succeeded in removing all mappings
1630 * SWAP_AGAIN - we missed a mapping, try again later
1631 * SWAP_FAIL - the page is unswappable
Nick Pigginb291f002008-10-18 20:26:44 -07001632 * SWAP_MLOCK - page is mlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001634int try_to_unmap(struct page *page, enum ttu_flags flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635{
1636 int ret;
1637
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 BUG_ON(!PageLocked(page));
Andrea Arcangeli91600e92011-01-13 15:47:16 -08001639 VM_BUG_ON(!PageHuge(page) && PageTransHuge(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640
Hugh Dickins5ad64682009-12-14 17:59:24 -08001641 if (unlikely(PageKsm(page)))
1642 ret = try_to_unmap_ksm(page, flags);
1643 else if (PageAnon(page))
Andi Kleen14fa31b2009-09-16 11:50:10 +02001644 ret = try_to_unmap_anon(page, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 else
Andi Kleen14fa31b2009-09-16 11:50:10 +02001646 ret = try_to_unmap_file(page, flags);
Nick Pigginb291f002008-10-18 20:26:44 -07001647 if (ret != SWAP_MLOCK && !page_mapped(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 ret = SWAP_SUCCESS;
1649 return ret;
1650}
Nikita Danilov81b40822005-05-01 08:58:36 -07001651
Nick Pigginb291f002008-10-18 20:26:44 -07001652/**
1653 * try_to_munlock - try to munlock a page
1654 * @page: the page to be munlocked
1655 *
1656 * Called from munlock code. Checks all of the VMAs mapping the page
1657 * to make sure nobody else has this page mlocked. The page will be
1658 * returned with PG_mlocked cleared if no other vmas have it mlocked.
1659 *
1660 * Return values are:
1661 *
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001662 * SWAP_AGAIN - no vma is holding page mlocked, or,
Nick Pigginb291f002008-10-18 20:26:44 -07001663 * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem
Hugh Dickins5ad64682009-12-14 17:59:24 -08001664 * SWAP_FAIL - page cannot be located at present
Nick Pigginb291f002008-10-18 20:26:44 -07001665 * SWAP_MLOCK - page is now mlocked.
1666 */
1667int try_to_munlock(struct page *page)
1668{
1669 VM_BUG_ON(!PageLocked(page) || PageLRU(page));
1670
Hugh Dickins5ad64682009-12-14 17:59:24 -08001671 if (unlikely(PageKsm(page)))
1672 return try_to_unmap_ksm(page, TTU_MUNLOCK);
1673 else if (PageAnon(page))
Andi Kleen14fa31b2009-09-16 11:50:10 +02001674 return try_to_unmap_anon(page, TTU_MUNLOCK);
Nick Pigginb291f002008-10-18 20:26:44 -07001675 else
Andi Kleen14fa31b2009-09-16 11:50:10 +02001676 return try_to_unmap_file(page, TTU_MUNLOCK);
Nick Pigginb291f002008-10-18 20:26:44 -07001677}
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001678
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001679void __put_anon_vma(struct anon_vma *anon_vma)
Rik van Riel76545062010-08-09 17:18:41 -07001680{
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001681 struct anon_vma *root = anon_vma->root;
Rik van Riel76545062010-08-09 17:18:41 -07001682
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001683 if (root != anon_vma && atomic_dec_and_test(&root->refcount))
1684 anon_vma_free(root);
Rik van Riel76545062010-08-09 17:18:41 -07001685
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001686 anon_vma_free(anon_vma);
Rik van Riel76545062010-08-09 17:18:41 -07001687}
Rik van Riel76545062010-08-09 17:18:41 -07001688
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001689#ifdef CONFIG_MIGRATION
1690/*
1691 * rmap_walk() and its helpers rmap_walk_anon() and rmap_walk_file():
1692 * Called by migrate.c to remove migration ptes, but might be used more later.
1693 */
1694static int rmap_walk_anon(struct page *page, int (*rmap_one)(struct page *,
1695 struct vm_area_struct *, unsigned long, void *), void *arg)
1696{
1697 struct anon_vma *anon_vma;
Michel Lespinassebf181b92012-10-08 16:31:39 -07001698 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
Rik van Riel5beb4932010-03-05 13:42:07 -08001699 struct anon_vma_chain *avc;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001700 int ret = SWAP_AGAIN;
1701
1702 /*
1703 * Note: remove_migration_ptes() cannot use page_lock_anon_vma()
1704 * because that depends on page_mapped(); but not all its usages
Mel Gorman3f6c8272010-05-24 14:32:17 -07001705 * are holding mmap_sem. Users without mmap_sem are required to
1706 * take a reference count to prevent the anon_vma disappearing
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001707 */
1708 anon_vma = page_anon_vma(page);
1709 if (!anon_vma)
1710 return ret;
Rik van Rielcba48b92010-08-09 17:18:38 -07001711 anon_vma_lock(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -07001712 anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) {
Rik van Riel5beb4932010-03-05 13:42:07 -08001713 struct vm_area_struct *vma = avc->vma;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001714 unsigned long address = vma_address(page, vma);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001715 ret = rmap_one(page, vma, address, arg);
1716 if (ret != SWAP_AGAIN)
1717 break;
1718 }
Rik van Rielcba48b92010-08-09 17:18:38 -07001719 anon_vma_unlock(anon_vma);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001720 return ret;
1721}
1722
1723static int rmap_walk_file(struct page *page, int (*rmap_one)(struct page *,
1724 struct vm_area_struct *, unsigned long, void *), void *arg)
1725{
1726 struct address_space *mapping = page->mapping;
1727 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
1728 struct vm_area_struct *vma;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001729 int ret = SWAP_AGAIN;
1730
1731 if (!mapping)
1732 return ret;
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07001733 mutex_lock(&mapping->i_mmap_mutex);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07001734 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001735 unsigned long address = vma_address(page, vma);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001736 ret = rmap_one(page, vma, address, arg);
1737 if (ret != SWAP_AGAIN)
1738 break;
1739 }
1740 /*
1741 * No nonlinear handling: being always shared, nonlinear vmas
1742 * never contain migration ptes. Decide what to do about this
1743 * limitation to linear when we need rmap_walk() on nonlinear.
1744 */
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07001745 mutex_unlock(&mapping->i_mmap_mutex);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001746 return ret;
1747}
1748
1749int rmap_walk(struct page *page, int (*rmap_one)(struct page *,
1750 struct vm_area_struct *, unsigned long, void *), void *arg)
1751{
1752 VM_BUG_ON(!PageLocked(page));
1753
1754 if (unlikely(PageKsm(page)))
1755 return rmap_walk_ksm(page, rmap_one, arg);
1756 else if (PageAnon(page))
1757 return rmap_walk_anon(page, rmap_one, arg);
1758 else
1759 return rmap_walk_file(page, rmap_one, arg);
1760}
1761#endif /* CONFIG_MIGRATION */
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001762
Naoya Horiguchie3390f62010-06-15 13:18:13 +09001763#ifdef CONFIG_HUGETLB_PAGE
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001764/*
1765 * The following three functions are for anonymous (private mapped) hugepages.
1766 * Unlike common anonymous pages, anonymous hugepages have no accounting code
1767 * and no lru code, because we handle hugepages differently from common pages.
1768 */
1769static void __hugepage_set_anon_rmap(struct page *page,
1770 struct vm_area_struct *vma, unsigned long address, int exclusive)
1771{
1772 struct anon_vma *anon_vma = vma->anon_vma;
Naoya Horiguchi433abed2010-09-10 13:23:03 +09001773
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001774 BUG_ON(!anon_vma);
Naoya Horiguchi433abed2010-09-10 13:23:03 +09001775
1776 if (PageAnon(page))
1777 return;
1778 if (!exclusive)
1779 anon_vma = anon_vma->root;
1780
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001781 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
1782 page->mapping = (struct address_space *) anon_vma;
1783 page->index = linear_page_index(vma, address);
1784}
1785
1786void hugepage_add_anon_rmap(struct page *page,
1787 struct vm_area_struct *vma, unsigned long address)
1788{
1789 struct anon_vma *anon_vma = vma->anon_vma;
1790 int first;
Naoya Horiguchia850ea32010-09-10 13:23:06 +09001791
1792 BUG_ON(!PageLocked(page));
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001793 BUG_ON(!anon_vma);
Hugh Dickins5dbe0af2011-05-28 13:17:04 -07001794 /* address might be in next vma when migration races vma_adjust */
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001795 first = atomic_inc_and_test(&page->_mapcount);
1796 if (first)
1797 __hugepage_set_anon_rmap(page, vma, address, 0);
1798}
1799
1800void hugepage_add_new_anon_rmap(struct page *page,
1801 struct vm_area_struct *vma, unsigned long address)
1802{
1803 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
1804 atomic_set(&page->_mapcount, 0);
1805 __hugepage_set_anon_rmap(page, vma, address, 1);
1806}
Naoya Horiguchie3390f62010-06-15 13:18:13 +09001807#endif /* CONFIG_HUGETLB_PAGE */