blob: e375ce4bd93e335678c6152313f9259ecaae9ce9 [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
Ingo Molnar5a505082012-12-02 19:56:46 +000027 * anon_vma->rwsem
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 *
Ingo Molnar5a505082012-12-02 19:56:46 +000040 * anon_vma->rwsem,mapping->i_mutex (memory_failure, collect_procs_anon)
Peter Zijlstra9b679322011-06-27 16:18:09 -070041 * ->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 /*
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +000090 * Synchronize against page_lock_anon_vma_read() such that
Peter Zijlstra88c22082011-05-24 17:12:13 -070091 * 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
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +000096 * down_read_trylock() from page_lock_anon_vma_read(). This orders:
Peter Zijlstra88c22082011-05-24 17:12:13 -070097 *
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +000098 * page_lock_anon_vma_read() VS put_anon_vma()
99 * down_read_trylock() atomic_dec_and_test()
Peter Zijlstra88c22082011-05-24 17:12:13 -0700100 * LOCK MB
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000101 * atomic_read() rwsem_is_locked()
Peter Zijlstra88c22082011-05-24 17:12:13 -0700102 *
103 * LOCK should suffice since the actual taking of the lock must
104 * happen _before_ what follows.
105 */
Hugh Dickins7f39dda2014-06-04 16:05:33 -0700106 might_sleep();
Ingo Molnar5a505082012-12-02 19:56:46 +0000107 if (rwsem_is_locked(&anon_vma->root->rwsem)) {
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000108 anon_vma_lock_write(anon_vma);
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800109 anon_vma_unlock_write(anon_vma);
Peter Zijlstra88c22082011-05-24 17:12:13 -0700110 }
111
Adrian Bunkfdd2e5f2008-10-18 20:28:38 -0700112 kmem_cache_free(anon_vma_cachep, anon_vma);
113}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114
Linus Torvaldsdd347392011-06-17 19:05:36 -0700115static inline struct anon_vma_chain *anon_vma_chain_alloc(gfp_t gfp)
Rik van Riel5beb4932010-03-05 13:42:07 -0800116{
Linus Torvaldsdd347392011-06-17 19:05:36 -0700117 return kmem_cache_alloc(anon_vma_chain_cachep, gfp);
Rik van Riel5beb4932010-03-05 13:42:07 -0800118}
119
Namhyung Kime574b5f2010-10-26 14:22:02 -0700120static void anon_vma_chain_free(struct anon_vma_chain *anon_vma_chain)
Rik van Riel5beb4932010-03-05 13:42:07 -0800121{
122 kmem_cache_free(anon_vma_chain_cachep, anon_vma_chain);
123}
124
Kautuk Consul6583a842012-03-21 16:34:01 -0700125static void anon_vma_chain_link(struct vm_area_struct *vma,
126 struct anon_vma_chain *avc,
127 struct anon_vma *anon_vma)
128{
129 avc->vma = vma;
130 avc->anon_vma = anon_vma;
131 list_add(&avc->same_vma, &vma->anon_vma_chain);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700132 anon_vma_interval_tree_insert(avc, &anon_vma->rb_root);
Kautuk Consul6583a842012-03-21 16:34:01 -0700133}
134
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700135/**
136 * anon_vma_prepare - attach an anon_vma to a memory region
137 * @vma: the memory region in question
138 *
139 * This makes sure the memory mapping described by 'vma' has
140 * an 'anon_vma' attached to it, so that we can associate the
141 * anonymous pages mapped into it with that anon_vma.
142 *
143 * The common case will be that we already have one, but if
Figo.zhang23a07902010-12-27 15:14:06 +0100144 * not we either need to find an adjacent mapping that we
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700145 * can re-use the anon_vma from (very common when the only
146 * reason for splitting a vma has been mprotect()), or we
147 * allocate a new one.
148 *
149 * Anon-vma allocations are very subtle, because we may have
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000150 * optimistically looked up an anon_vma in page_lock_anon_vma_read()
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700151 * and that may actually touch the spinlock even in the newly
152 * allocated vma (it depends on RCU to make sure that the
153 * anon_vma isn't actually destroyed).
154 *
155 * As a result, we need to do proper anon_vma locking even
156 * for the new allocation. At the same time, we do not want
157 * to do any locking for the common case of already having
158 * an anon_vma.
159 *
160 * This must be called with the mmap_sem held for reading.
161 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162int anon_vma_prepare(struct vm_area_struct *vma)
163{
164 struct anon_vma *anon_vma = vma->anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -0800165 struct anon_vma_chain *avc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166
167 might_sleep();
168 if (unlikely(!anon_vma)) {
169 struct mm_struct *mm = vma->vm_mm;
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700170 struct anon_vma *allocated;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171
Linus Torvaldsdd347392011-06-17 19:05:36 -0700172 avc = anon_vma_chain_alloc(GFP_KERNEL);
Rik van Riel5beb4932010-03-05 13:42:07 -0800173 if (!avc)
174 goto out_enomem;
175
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 anon_vma = find_mergeable_anon_vma(vma);
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700177 allocated = NULL;
178 if (!anon_vma) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 anon_vma = anon_vma_alloc();
180 if (unlikely(!anon_vma))
Rik van Riel5beb4932010-03-05 13:42:07 -0800181 goto out_enomem_free_avc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 allocated = anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 }
184
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000185 anon_vma_lock_write(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 /* page_table_lock to protect against threads */
187 spin_lock(&mm->page_table_lock);
188 if (likely(!vma->anon_vma)) {
189 vma->anon_vma = anon_vma;
Kautuk Consul6583a842012-03-21 16:34:01 -0700190 anon_vma_chain_link(vma, avc, anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 allocated = NULL;
Oleg Nesterov31f2b0e2010-04-23 13:18:01 -0400192 avc = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 }
194 spin_unlock(&mm->page_table_lock);
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800195 anon_vma_unlock_write(anon_vma);
Oleg Nesterov31f2b0e2010-04-23 13:18:01 -0400196
197 if (unlikely(allocated))
Peter Zijlstra01d8b202011-03-22 16:32:49 -0700198 put_anon_vma(allocated);
Oleg Nesterov31f2b0e2010-04-23 13:18:01 -0400199 if (unlikely(avc))
Rik van Riel5beb4932010-03-05 13:42:07 -0800200 anon_vma_chain_free(avc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 }
202 return 0;
Rik van Riel5beb4932010-03-05 13:42:07 -0800203
204 out_enomem_free_avc:
205 anon_vma_chain_free(avc);
206 out_enomem:
207 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208}
209
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700210/*
211 * This is a useful helper function for locking the anon_vma root as
212 * we traverse the vma->anon_vma_chain, looping over anon_vma's that
213 * have the same vma.
214 *
215 * Such anon_vma's should have the same root, so you'd expect to see
216 * just a single mutex_lock for the whole traversal.
217 */
218static inline struct anon_vma *lock_anon_vma_root(struct anon_vma *root, struct anon_vma *anon_vma)
219{
220 struct anon_vma *new_root = anon_vma->root;
221 if (new_root != root) {
222 if (WARN_ON_ONCE(root))
Ingo Molnar5a505082012-12-02 19:56:46 +0000223 up_write(&root->rwsem);
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700224 root = new_root;
Ingo Molnar5a505082012-12-02 19:56:46 +0000225 down_write(&root->rwsem);
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700226 }
227 return root;
228}
229
230static inline void unlock_anon_vma_root(struct anon_vma *root)
231{
232 if (root)
Ingo Molnar5a505082012-12-02 19:56:46 +0000233 up_write(&root->rwsem);
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700234}
235
Rik van Riel5beb4932010-03-05 13:42:07 -0800236/*
237 * Attach the anon_vmas from src to dst.
238 * Returns 0 on success, -ENOMEM on failure.
239 */
240int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241{
Rik van Riel5beb4932010-03-05 13:42:07 -0800242 struct anon_vma_chain *avc, *pavc;
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700243 struct anon_vma *root = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244
Linus Torvalds646d87b2010-04-11 17:15:03 -0700245 list_for_each_entry_reverse(pavc, &src->anon_vma_chain, same_vma) {
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700246 struct anon_vma *anon_vma;
247
Linus Torvaldsdd347392011-06-17 19:05:36 -0700248 avc = anon_vma_chain_alloc(GFP_NOWAIT | __GFP_NOWARN);
249 if (unlikely(!avc)) {
250 unlock_anon_vma_root(root);
251 root = NULL;
252 avc = anon_vma_chain_alloc(GFP_KERNEL);
253 if (!avc)
254 goto enomem_failure;
255 }
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700256 anon_vma = pavc->anon_vma;
257 root = lock_anon_vma_root(root, anon_vma);
258 anon_vma_chain_link(dst, avc, anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 }
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700260 unlock_anon_vma_root(root);
Rik van Riel5beb4932010-03-05 13:42:07 -0800261 return 0;
262
263 enomem_failure:
264 unlink_anon_vmas(dst);
265 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266}
267
Rik van Riel5beb4932010-03-05 13:42:07 -0800268/*
269 * Attach vma to its own anon_vma, as well as to the anon_vmas that
270 * the corresponding VMA in the parent process is attached to.
271 * Returns 0 on success, non-zero on failure.
272 */
273int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274{
Rik van Riel5beb4932010-03-05 13:42:07 -0800275 struct anon_vma_chain *avc;
276 struct anon_vma *anon_vma;
277
278 /* Don't bother if the parent process has no anon_vma here. */
279 if (!pvma->anon_vma)
280 return 0;
281
282 /*
283 * First, attach the new VMA to the parent VMA's anon_vmas,
284 * so rmap can find non-COWed pages in child processes.
285 */
286 if (anon_vma_clone(vma, pvma))
287 return -ENOMEM;
288
289 /* Then add our own anon_vma. */
290 anon_vma = anon_vma_alloc();
291 if (!anon_vma)
292 goto out_error;
Linus Torvaldsdd347392011-06-17 19:05:36 -0700293 avc = anon_vma_chain_alloc(GFP_KERNEL);
Rik van Riel5beb4932010-03-05 13:42:07 -0800294 if (!avc)
295 goto out_error_free_anon_vma;
Rik van Riel5c341ee12010-08-09 17:18:39 -0700296
297 /*
298 * The root anon_vma's spinlock is the lock actually used when we
299 * lock any of the anon_vmas in this anon_vma tree.
300 */
301 anon_vma->root = pvma->anon_vma->root;
Rik van Riel76545062010-08-09 17:18:41 -0700302 /*
Peter Zijlstra01d8b202011-03-22 16:32:49 -0700303 * With refcounts, an anon_vma can stay around longer than the
304 * process it belongs to. The root anon_vma needs to be pinned until
305 * this anon_vma is freed, because the lock lives in the root.
Rik van Riel76545062010-08-09 17:18:41 -0700306 */
307 get_anon_vma(anon_vma->root);
Rik van Riel5beb4932010-03-05 13:42:07 -0800308 /* Mark this anon_vma as the one where our new (COWed) pages go. */
309 vma->anon_vma = anon_vma;
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000310 anon_vma_lock_write(anon_vma);
Rik van Riel5c341ee12010-08-09 17:18:39 -0700311 anon_vma_chain_link(vma, avc, anon_vma);
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800312 anon_vma_unlock_write(anon_vma);
Rik van Riel5beb4932010-03-05 13:42:07 -0800313
314 return 0;
315
316 out_error_free_anon_vma:
Peter Zijlstra01d8b202011-03-22 16:32:49 -0700317 put_anon_vma(anon_vma);
Rik van Riel5beb4932010-03-05 13:42:07 -0800318 out_error:
Rik van Riel4946d542010-04-05 12:13:33 -0400319 unlink_anon_vmas(vma);
Rik van Riel5beb4932010-03-05 13:42:07 -0800320 return -ENOMEM;
321}
322
Rik van Riel5beb4932010-03-05 13:42:07 -0800323void unlink_anon_vmas(struct vm_area_struct *vma)
324{
325 struct anon_vma_chain *avc, *next;
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200326 struct anon_vma *root = NULL;
Rik van Riel5beb4932010-03-05 13:42:07 -0800327
Rik van Riel5c341ee12010-08-09 17:18:39 -0700328 /*
329 * Unlink each anon_vma chained to the VMA. This list is ordered
330 * from newest to oldest, ensuring the root anon_vma gets freed last.
331 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800332 list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) {
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200333 struct anon_vma *anon_vma = avc->anon_vma;
334
335 root = lock_anon_vma_root(root, anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700336 anon_vma_interval_tree_remove(avc, &anon_vma->rb_root);
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200337
338 /*
339 * Leave empty anon_vmas on the list - we'll need
340 * to free them outside the lock.
341 */
Michel Lespinassebf181b92012-10-08 16:31:39 -0700342 if (RB_EMPTY_ROOT(&anon_vma->rb_root))
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200343 continue;
344
345 list_del(&avc->same_vma);
346 anon_vma_chain_free(avc);
347 }
348 unlock_anon_vma_root(root);
349
350 /*
351 * Iterate the list once more, it now only contains empty and unlinked
352 * anon_vmas, destroy them. Could not do before due to __put_anon_vma()
Ingo Molnar5a505082012-12-02 19:56:46 +0000353 * needing to write-acquire the anon_vma->root->rwsem.
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200354 */
355 list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) {
356 struct anon_vma *anon_vma = avc->anon_vma;
357
358 put_anon_vma(anon_vma);
359
Rik van Riel5beb4932010-03-05 13:42:07 -0800360 list_del(&avc->same_vma);
361 anon_vma_chain_free(avc);
362 }
363}
364
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700365static void anon_vma_ctor(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366{
Christoph Lametera35afb82007-05-16 22:10:57 -0700367 struct anon_vma *anon_vma = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368
Ingo Molnar5a505082012-12-02 19:56:46 +0000369 init_rwsem(&anon_vma->rwsem);
Peter Zijlstra83813262011-03-22 16:32:48 -0700370 atomic_set(&anon_vma->refcount, 0);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700371 anon_vma->rb_root = RB_ROOT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372}
373
374void __init anon_vma_init(void)
375{
376 anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma),
Paul Mundt20c2df82007-07-20 10:11:58 +0900377 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor);
Rik van Riel5beb4932010-03-05 13:42:07 -0800378 anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain, SLAB_PANIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379}
380
381/*
Peter Zijlstra6111e4c2011-05-24 17:12:08 -0700382 * Getting a lock on a stable anon_vma from a page off the LRU is tricky!
383 *
384 * Since there is no serialization what so ever against page_remove_rmap()
385 * the best this function can do is return a locked anon_vma that might
386 * have been relevant to this page.
387 *
388 * The page might have been remapped to a different anon_vma or the anon_vma
389 * returned may already be freed (and even reused).
390 *
Peter Zijlstrabc658c92011-05-29 10:33:44 +0200391 * In case it was remapped to a different anon_vma, the new anon_vma will be a
392 * child of the old anon_vma, and the anon_vma lifetime rules will therefore
393 * ensure that any anon_vma obtained from the page will still be valid for as
394 * long as we observe page_mapped() [ hence all those page_mapped() tests ].
395 *
Peter Zijlstra6111e4c2011-05-24 17:12:08 -0700396 * All users of this function must be very careful when walking the anon_vma
397 * chain and verify that the page in question is indeed mapped in it
398 * [ something equivalent to page_mapped_in_vma() ].
399 *
400 * Since anon_vma's slab is DESTROY_BY_RCU and we know from page_remove_rmap()
401 * that the anon_vma pointer from page->mapping is valid if there is a
402 * mapcount, we can dereference the anon_vma after observing those.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 */
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700404struct anon_vma *page_get_anon_vma(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405{
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700406 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 unsigned long anon_mapping;
408
409 rcu_read_lock();
Hugh Dickins80e148222009-12-14 17:59:29 -0800410 anon_mapping = (unsigned long) ACCESS_ONCE(page->mapping);
Hugh Dickins3ca7b3c2009-12-14 17:58:57 -0800411 if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 goto out;
413 if (!page_mapped(page))
414 goto out;
415
416 anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700417 if (!atomic_inc_not_zero(&anon_vma->refcount)) {
418 anon_vma = NULL;
419 goto out;
420 }
Hugh Dickinsf18194272010-08-25 23:12:54 -0700421
422 /*
423 * If this page is still mapped, then its anon_vma cannot have been
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700424 * freed. But if it has been unmapped, we have no security against the
425 * anon_vma structure being freed and reused (for another anon_vma:
426 * SLAB_DESTROY_BY_RCU guarantees that - so the atomic_inc_not_zero()
427 * above cannot corrupt).
Hugh Dickinsf18194272010-08-25 23:12:54 -0700428 */
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700429 if (!page_mapped(page)) {
Hugh Dickins7f39dda2014-06-04 16:05:33 -0700430 rcu_read_unlock();
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700431 put_anon_vma(anon_vma);
Hugh Dickins7f39dda2014-06-04 16:05:33 -0700432 return NULL;
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700433 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434out:
435 rcu_read_unlock();
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700436
437 return anon_vma;
438}
439
Peter Zijlstra88c22082011-05-24 17:12:13 -0700440/*
441 * Similar to page_get_anon_vma() except it locks the anon_vma.
442 *
443 * Its a little more complex as it tries to keep the fast path to a single
444 * atomic op -- the trylock. If we fail the trylock, we fall back to getting a
445 * reference like with page_get_anon_vma() and then block on the mutex.
446 */
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000447struct anon_vma *page_lock_anon_vma_read(struct page *page)
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700448{
Peter Zijlstra88c22082011-05-24 17:12:13 -0700449 struct anon_vma *anon_vma = NULL;
Hugh Dickinseee0f252011-05-28 13:20:21 -0700450 struct anon_vma *root_anon_vma;
Peter Zijlstra88c22082011-05-24 17:12:13 -0700451 unsigned long anon_mapping;
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700452
Peter Zijlstra88c22082011-05-24 17:12:13 -0700453 rcu_read_lock();
454 anon_mapping = (unsigned long) ACCESS_ONCE(page->mapping);
455 if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
456 goto out;
457 if (!page_mapped(page))
458 goto out;
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700459
Peter Zijlstra88c22082011-05-24 17:12:13 -0700460 anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
Hugh Dickinseee0f252011-05-28 13:20:21 -0700461 root_anon_vma = ACCESS_ONCE(anon_vma->root);
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000462 if (down_read_trylock(&root_anon_vma->rwsem)) {
Peter Zijlstra88c22082011-05-24 17:12:13 -0700463 /*
Hugh Dickinseee0f252011-05-28 13:20:21 -0700464 * If the page is still mapped, then this anon_vma is still
465 * its anon_vma, and holding the mutex ensures that it will
Peter Zijlstrabc658c92011-05-29 10:33:44 +0200466 * not go away, see anon_vma_free().
Peter Zijlstra88c22082011-05-24 17:12:13 -0700467 */
Hugh Dickinseee0f252011-05-28 13:20:21 -0700468 if (!page_mapped(page)) {
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000469 up_read(&root_anon_vma->rwsem);
Peter Zijlstra88c22082011-05-24 17:12:13 -0700470 anon_vma = NULL;
471 }
472 goto out;
473 }
474
475 /* trylock failed, we got to sleep */
476 if (!atomic_inc_not_zero(&anon_vma->refcount)) {
477 anon_vma = NULL;
478 goto out;
479 }
480
481 if (!page_mapped(page)) {
Hugh Dickins7f39dda2014-06-04 16:05:33 -0700482 rcu_read_unlock();
Peter Zijlstra88c22082011-05-24 17:12:13 -0700483 put_anon_vma(anon_vma);
Hugh Dickins7f39dda2014-06-04 16:05:33 -0700484 return NULL;
Peter Zijlstra88c22082011-05-24 17:12:13 -0700485 }
486
487 /* we pinned the anon_vma, its safe to sleep */
488 rcu_read_unlock();
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000489 anon_vma_lock_read(anon_vma);
Peter Zijlstra88c22082011-05-24 17:12:13 -0700490
491 if (atomic_dec_and_test(&anon_vma->refcount)) {
492 /*
493 * Oops, we held the last refcount, release the lock
494 * and bail -- can't simply use put_anon_vma() because
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000495 * we'll deadlock on the anon_vma_lock_write() recursion.
Peter Zijlstra88c22082011-05-24 17:12:13 -0700496 */
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000497 anon_vma_unlock_read(anon_vma);
Peter Zijlstra88c22082011-05-24 17:12:13 -0700498 __put_anon_vma(anon_vma);
499 anon_vma = NULL;
500 }
501
502 return anon_vma;
503
504out:
505 rcu_read_unlock();
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700506 return anon_vma;
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800507}
508
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000509void page_unlock_anon_vma_read(struct anon_vma *anon_vma)
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800510{
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000511 anon_vma_unlock_read(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512}
513
514/*
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -0800515 * At what user virtual address is page expected in @vma?
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 */
Michel Lespinasse86c2ad12012-10-08 16:31:42 -0700517static inline unsigned long
518__vma_address(struct page *page, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519{
520 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +0900522 if (unlikely(is_vm_hugetlb_page(vma)))
523 pgoff = page->index << huge_page_order(page_hstate(page));
Michel Lespinasse86c2ad12012-10-08 16:31:42 -0700524
525 return vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
526}
527
528inline unsigned long
529vma_address(struct page *page, struct vm_area_struct *vma)
530{
531 unsigned long address = __vma_address(page, vma);
532
533 /* page should be within @vma mapping range */
534 VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end);
535
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 return address;
537}
538
539/*
Huang Shijiebf89c8c2009-10-01 15:44:04 -0700540 * At what user virtual address is page expected in vma?
Naoya Horiguchiab941e02010-05-11 14:06:55 -0700541 * Caller should check the page is actually part of the vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 */
543unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma)
544{
Michel Lespinasse86c2ad12012-10-08 16:31:42 -0700545 unsigned long address;
Andrea Arcangeli21d0d442010-08-09 17:19:10 -0700546 if (PageAnon(page)) {
Hugh Dickins4829b902010-10-02 17:46:06 -0700547 struct anon_vma *page__anon_vma = page_anon_vma(page);
548 /*
549 * Note: swapoff's unuse_vma() is more efficient with this
550 * check, and needs it to match anon_vma when KSM is active.
551 */
552 if (!vma->anon_vma || !page__anon_vma ||
553 vma->anon_vma->root != page__anon_vma->root)
Andrea Arcangeli21d0d442010-08-09 17:19:10 -0700554 return -EFAULT;
555 } else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) {
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800556 if (!vma->vm_file ||
557 vma->vm_file->f_mapping != page->mapping)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 return -EFAULT;
559 } else
560 return -EFAULT;
Michel Lespinasse86c2ad12012-10-08 16:31:42 -0700561 address = __vma_address(page, vma);
562 if (unlikely(address < vma->vm_start || address >= vma->vm_end))
563 return -EFAULT;
564 return address;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565}
566
Bob Liu62190492012-12-11 16:00:37 -0800567pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address)
568{
569 pgd_t *pgd;
570 pud_t *pud;
571 pmd_t *pmd = NULL;
572
573 pgd = pgd_offset(mm, address);
574 if (!pgd_present(*pgd))
575 goto out;
576
577 pud = pud_offset(pgd, address);
578 if (!pud_present(*pud))
579 goto out;
580
581 pmd = pmd_offset(pud, address);
582 if (!pmd_present(*pmd))
583 pmd = NULL;
584out:
585 return pmd;
586}
587
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588/*
Nikita Danilov81b40822005-05-01 08:58:36 -0700589 * Check that @page is mapped at @address into @mm.
590 *
Nick Piggin479db0b2008-08-20 14:09:18 -0700591 * If @sync is false, page_check_address may perform a racy check to avoid
592 * the page table lock when the pte is not present (helpful when reclaiming
593 * highly shared pages).
594 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -0700595 * On success returns with pte mapped and locked.
Nikita Danilov81b40822005-05-01 08:58:36 -0700596 */
Namhyung Kime9a81a82010-10-26 14:22:01 -0700597pte_t *__page_check_address(struct page *page, struct mm_struct *mm,
Nick Piggin479db0b2008-08-20 14:09:18 -0700598 unsigned long address, spinlock_t **ptlp, int sync)
Nikita Danilov81b40822005-05-01 08:58:36 -0700599{
Nikita Danilov81b40822005-05-01 08:58:36 -0700600 pmd_t *pmd;
601 pte_t *pte;
Hugh Dickinsc0718802005-10-29 18:16:31 -0700602 spinlock_t *ptl;
Nikita Danilov81b40822005-05-01 08:58:36 -0700603
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +0900604 if (unlikely(PageHuge(page))) {
Jianguo Wu98398c32013-12-18 17:08:59 -0800605 /* when pud is not present, pte will be NULL */
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +0900606 pte = huge_pte_offset(mm, address);
Jianguo Wu98398c32013-12-18 17:08:59 -0800607 if (!pte)
608 return NULL;
609
Kirill A. Shutemovcb900f42013-11-14 14:31:02 -0800610 ptl = huge_pte_lockptr(page_hstate(page), mm, pte);
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +0900611 goto check;
612 }
613
Bob Liu62190492012-12-11 16:00:37 -0800614 pmd = mm_find_pmd(mm, address);
615 if (!pmd)
Hugh Dickinsc0718802005-10-29 18:16:31 -0700616 return NULL;
617
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800618 if (pmd_trans_huge(*pmd))
619 return NULL;
Hugh Dickinsc0718802005-10-29 18:16:31 -0700620
621 pte = pte_offset_map(pmd, address);
622 /* Make a quick check before getting the lock */
Nick Piggin479db0b2008-08-20 14:09:18 -0700623 if (!sync && !pte_present(*pte)) {
Hugh Dickinsc0718802005-10-29 18:16:31 -0700624 pte_unmap(pte);
625 return NULL;
Nikita Danilov81b40822005-05-01 08:58:36 -0700626 }
Hugh Dickinsc0718802005-10-29 18:16:31 -0700627
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700628 ptl = pte_lockptr(mm, pmd);
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +0900629check:
Hugh Dickinsc0718802005-10-29 18:16:31 -0700630 spin_lock(ptl);
631 if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) {
632 *ptlp = ptl;
633 return pte;
634 }
635 pte_unmap_unlock(pte, ptl);
636 return NULL;
Nikita Danilov81b40822005-05-01 08:58:36 -0700637}
638
Nick Pigginb291f002008-10-18 20:26:44 -0700639/**
640 * page_mapped_in_vma - check whether a page is really mapped in a VMA
641 * @page: the page to test
642 * @vma: the VMA to test
643 *
644 * Returns 1 if the page is mapped into the page tables of the VMA, 0
645 * if the page is not mapped into the page tables of this VMA. Only
646 * valid for normal file or anonymous VMAs.
647 */
Andi Kleen6a460792009-09-16 11:50:15 +0200648int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma)
Nick Pigginb291f002008-10-18 20:26:44 -0700649{
650 unsigned long address;
651 pte_t *pte;
652 spinlock_t *ptl;
653
Michel Lespinasse86c2ad12012-10-08 16:31:42 -0700654 address = __vma_address(page, vma);
655 if (unlikely(address < vma->vm_start || address >= vma->vm_end))
Nick Pigginb291f002008-10-18 20:26:44 -0700656 return 0;
657 pte = page_check_address(page, vma->vm_mm, address, &ptl, 1);
658 if (!pte) /* the page is not in this mm */
659 return 0;
660 pte_unmap_unlock(pte, ptl);
661
662 return 1;
663}
664
Joonsoo Kim9f326242014-01-21 15:49:53 -0800665struct page_referenced_arg {
666 int mapcount;
667 int referenced;
668 unsigned long vm_flags;
669 struct mem_cgroup *memcg;
670};
Nikita Danilov81b40822005-05-01 08:58:36 -0700671/*
Joonsoo Kim9f326242014-01-21 15:49:53 -0800672 * arg: page_referenced_arg will be passed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 */
Kirill A. Shutemovac769502014-06-04 16:08:17 -0700674static int page_referenced_one(struct page *page, struct vm_area_struct *vma,
Joonsoo Kim9f326242014-01-21 15:49:53 -0800675 unsigned long address, void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676{
677 struct mm_struct *mm = vma->vm_mm;
Kirill A. Shutemov117b0792013-11-14 14:30:56 -0800678 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 int referenced = 0;
Joonsoo Kim9f326242014-01-21 15:49:53 -0800680 struct page_referenced_arg *pra = arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800682 if (unlikely(PageTransHuge(page))) {
683 pmd_t *pmd;
684
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800685 /*
686 * rmap might return false positives; we must filter
687 * these out using page_check_address_pmd().
688 */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800689 pmd = page_check_address_pmd(page, mm, address,
Kirill A. Shutemov117b0792013-11-14 14:30:56 -0800690 PAGE_CHECK_ADDRESS_PMD_FLAG, &ptl);
691 if (!pmd)
Joonsoo Kim9f326242014-01-21 15:49:53 -0800692 return SWAP_AGAIN;
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800693
694 if (vma->vm_flags & VM_LOCKED) {
Kirill A. Shutemov117b0792013-11-14 14:30:56 -0800695 spin_unlock(ptl);
Joonsoo Kim9f326242014-01-21 15:49:53 -0800696 pra->vm_flags |= VM_LOCKED;
697 return SWAP_FAIL; /* To break the loop */
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800698 }
699
700 /* go ahead even if the pmd is pmd_trans_splitting() */
701 if (pmdp_clear_flush_young_notify(vma, address, pmd))
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800702 referenced++;
Kirill A. Shutemov117b0792013-11-14 14:30:56 -0800703 spin_unlock(ptl);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800704 } else {
705 pte_t *pte;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800706
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800707 /*
708 * rmap might return false positives; we must filter
709 * these out using page_check_address().
710 */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800711 pte = page_check_address(page, mm, address, &ptl, 0);
712 if (!pte)
Joonsoo Kim9f326242014-01-21 15:49:53 -0800713 return SWAP_AGAIN;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800714
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800715 if (vma->vm_flags & VM_LOCKED) {
716 pte_unmap_unlock(pte, ptl);
Joonsoo Kim9f326242014-01-21 15:49:53 -0800717 pra->vm_flags |= VM_LOCKED;
718 return SWAP_FAIL; /* To break the loop */
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800719 }
720
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800721 if (ptep_clear_flush_young_notify(vma, address, pte)) {
722 /*
723 * Don't treat a reference through a sequentially read
724 * mapping as such. If the page has been used in
725 * another mapping, we will catch it; if this other
726 * mapping is already gone, the unmap path will have
727 * set PG_referenced or activated the page.
728 */
Joe Perches64363aa2013-07-08 16:00:18 -0700729 if (likely(!(vma->vm_flags & VM_SEQ_READ)))
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800730 referenced++;
731 }
732 pte_unmap_unlock(pte, ptl);
733 }
734
Joonsoo Kim9f326242014-01-21 15:49:53 -0800735 if (referenced) {
736 pra->referenced++;
737 pra->vm_flags |= vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 }
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800739
Joonsoo Kim9f326242014-01-21 15:49:53 -0800740 pra->mapcount--;
741 if (!pra->mapcount)
742 return SWAP_SUCCESS; /* To break the loop */
743
744 return SWAP_AGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745}
746
Joonsoo Kim9f326242014-01-21 15:49:53 -0800747static bool invalid_page_referenced_vma(struct vm_area_struct *vma, void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748{
Joonsoo Kim9f326242014-01-21 15:49:53 -0800749 struct page_referenced_arg *pra = arg;
750 struct mem_cgroup *memcg = pra->memcg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751
Joonsoo Kim9f326242014-01-21 15:49:53 -0800752 if (!mm_match_cgroup(vma->vm_mm, memcg))
753 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754
Joonsoo Kim9f326242014-01-21 15:49:53 -0800755 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756}
757
758/**
759 * page_referenced - test if the page was referenced
760 * @page: the page to test
761 * @is_locked: caller holds lock on the page
Johannes Weiner72835c82012-01-12 17:18:32 -0800762 * @memcg: target memory cgroup
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700763 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 *
765 * Quick test_and_clear_referenced for all mappings to a page,
766 * returns the number of ptes which referenced the page.
767 */
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700768int page_referenced(struct page *page,
769 int is_locked,
Johannes Weiner72835c82012-01-12 17:18:32 -0800770 struct mem_cgroup *memcg,
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700771 unsigned long *vm_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772{
Joonsoo Kim9f326242014-01-21 15:49:53 -0800773 int ret;
Hugh Dickins5ad64682009-12-14 17:59:24 -0800774 int we_locked = 0;
Joonsoo Kim9f326242014-01-21 15:49:53 -0800775 struct page_referenced_arg pra = {
776 .mapcount = page_mapcount(page),
777 .memcg = memcg,
778 };
779 struct rmap_walk_control rwc = {
780 .rmap_one = page_referenced_one,
781 .arg = (void *)&pra,
782 .anon_lock = page_lock_anon_vma_read,
783 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700785 *vm_flags = 0;
Joonsoo Kim9f326242014-01-21 15:49:53 -0800786 if (!page_mapped(page))
787 return 0;
788
789 if (!page_rmapping(page))
790 return 0;
791
792 if (!is_locked && (!PageAnon(page) || PageKsm(page))) {
793 we_locked = trylock_page(page);
794 if (!we_locked)
795 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 }
Joonsoo Kim9f326242014-01-21 15:49:53 -0800797
798 /*
799 * If we are reclaiming on behalf of a cgroup, skip
800 * counting on behalf of references from different
801 * cgroups
802 */
803 if (memcg) {
804 rwc.invalid_vma = invalid_page_referenced_vma;
805 }
806
807 ret = rmap_walk(page, &rwc);
808 *vm_flags = pra.vm_flags;
809
810 if (we_locked)
811 unlock_page(page);
812
813 return pra.referenced;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814}
815
Hugh Dickins1cb17292009-12-14 17:59:01 -0800816static int page_mkclean_one(struct page *page, struct vm_area_struct *vma,
Joonsoo Kim9853a402014-01-21 15:49:55 -0800817 unsigned long address, void *arg)
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700818{
819 struct mm_struct *mm = vma->vm_mm;
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100820 pte_t *pte;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700821 spinlock_t *ptl;
822 int ret = 0;
Joonsoo Kim9853a402014-01-21 15:49:55 -0800823 int *cleaned = arg;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700824
Nick Piggin479db0b2008-08-20 14:09:18 -0700825 pte = page_check_address(page, mm, address, &ptl, 1);
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700826 if (!pte)
827 goto out;
828
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100829 if (pte_dirty(*pte) || pte_write(*pte)) {
830 pte_t entry;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700831
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100832 flush_cache_page(vma, address, pte_pfn(*pte));
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700833 entry = ptep_clear_flush(vma, address, pte);
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100834 entry = pte_wrprotect(entry);
835 entry = pte_mkclean(entry);
Al Virod6e88e62006-12-29 16:48:35 -0800836 set_pte_at(mm, address, pte, entry);
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100837 ret = 1;
838 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700839
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700840 pte_unmap_unlock(pte, ptl);
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700841
Joonsoo Kim9853a402014-01-21 15:49:55 -0800842 if (ret) {
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700843 mmu_notifier_invalidate_page(mm, address);
Joonsoo Kim9853a402014-01-21 15:49:55 -0800844 (*cleaned)++;
845 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700846out:
Joonsoo Kim9853a402014-01-21 15:49:55 -0800847 return SWAP_AGAIN;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700848}
849
Joonsoo Kim9853a402014-01-21 15:49:55 -0800850static bool invalid_mkclean_vma(struct vm_area_struct *vma, void *arg)
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700851{
Joonsoo Kim9853a402014-01-21 15:49:55 -0800852 if (vma->vm_flags & VM_SHARED)
Fengguang Wu871beb82014-01-23 15:53:41 -0800853 return false;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700854
Fengguang Wu871beb82014-01-23 15:53:41 -0800855 return true;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700856}
857
858int page_mkclean(struct page *page)
859{
Joonsoo Kim9853a402014-01-21 15:49:55 -0800860 int cleaned = 0;
861 struct address_space *mapping;
862 struct rmap_walk_control rwc = {
863 .arg = (void *)&cleaned,
864 .rmap_one = page_mkclean_one,
865 .invalid_vma = invalid_mkclean_vma,
866 };
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700867
868 BUG_ON(!PageLocked(page));
869
Joonsoo Kim9853a402014-01-21 15:49:55 -0800870 if (!page_mapped(page))
871 return 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700872
Joonsoo Kim9853a402014-01-21 15:49:55 -0800873 mapping = page_mapping(page);
874 if (!mapping)
875 return 0;
876
877 rmap_walk(page, &rwc);
878
879 return cleaned;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700880}
Jaya Kumar60b59be2007-05-08 00:37:37 -0700881EXPORT_SYMBOL_GPL(page_mkclean);
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700882
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883/**
Rik van Rielc44b6742010-03-05 13:42:09 -0800884 * page_move_anon_rmap - move a page to our anon_vma
885 * @page: the page to move to our anon_vma
886 * @vma: the vma the page belongs to
887 * @address: the user virtual address mapped
888 *
889 * When a page belongs exclusively to one process after a COW event,
890 * that page can be moved into the anon_vma that belongs to just that
891 * process, so the rmap code will not search the parent or sibling
892 * processes.
893 */
894void page_move_anon_rmap(struct page *page,
895 struct vm_area_struct *vma, unsigned long address)
896{
897 struct anon_vma *anon_vma = vma->anon_vma;
898
Sasha Levin309381fea2014-01-23 15:52:54 -0800899 VM_BUG_ON_PAGE(!PageLocked(page), page);
Rik van Rielc44b6742010-03-05 13:42:09 -0800900 VM_BUG_ON(!anon_vma);
Sasha Levin309381fea2014-01-23 15:52:54 -0800901 VM_BUG_ON_PAGE(page->index != linear_page_index(vma, address), page);
Rik van Rielc44b6742010-03-05 13:42:09 -0800902
903 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
904 page->mapping = (struct address_space *) anon_vma;
905}
906
907/**
Andi Kleen4e1c1972010-09-22 12:43:56 +0200908 * __page_set_anon_rmap - set up new anonymous rmap
909 * @page: Page to add to rmap
910 * @vma: VM area to add page to.
911 * @address: User virtual address of the mapping
Rik van Riele8a03fe2010-04-14 17:59:28 -0400912 * @exclusive: the page is exclusively owned by the current process
Nick Piggin9617d952006-01-06 00:11:12 -0800913 */
914static void __page_set_anon_rmap(struct page *page,
Rik van Riele8a03fe2010-04-14 17:59:28 -0400915 struct vm_area_struct *vma, unsigned long address, int exclusive)
Nick Piggin9617d952006-01-06 00:11:12 -0800916{
Rik van Riele8a03fe2010-04-14 17:59:28 -0400917 struct anon_vma *anon_vma = vma->anon_vma;
Nick Piggin9617d952006-01-06 00:11:12 -0800918
Rik van Riele8a03fe2010-04-14 17:59:28 -0400919 BUG_ON(!anon_vma);
Linus Torvaldsea900022010-04-12 12:44:29 -0700920
Andi Kleen4e1c1972010-09-22 12:43:56 +0200921 if (PageAnon(page))
922 return;
923
Linus Torvaldsea900022010-04-12 12:44:29 -0700924 /*
Rik van Riele8a03fe2010-04-14 17:59:28 -0400925 * If the page isn't exclusively mapped into this vma,
926 * we must use the _oldest_ possible anon_vma for the
927 * page mapping!
Linus Torvaldsea900022010-04-12 12:44:29 -0700928 */
Andi Kleen4e1c1972010-09-22 12:43:56 +0200929 if (!exclusive)
Andrea Arcangeli288468c2010-08-09 17:19:09 -0700930 anon_vma = anon_vma->root;
Linus Torvaldsea900022010-04-12 12:44:29 -0700931
Nick Piggin9617d952006-01-06 00:11:12 -0800932 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
933 page->mapping = (struct address_space *) anon_vma;
Nick Piggin9617d952006-01-06 00:11:12 -0800934 page->index = linear_page_index(vma, address);
Nick Piggin9617d952006-01-06 00:11:12 -0800935}
936
937/**
Randy Dunlap43d8eac2008-03-19 17:00:43 -0700938 * __page_check_anon_rmap - sanity check anonymous rmap addition
Nick Pigginc97a9e12007-05-16 22:11:21 -0700939 * @page: the page to add the mapping to
940 * @vma: the vm area in which the mapping is added
941 * @address: the user virtual address mapped
942 */
943static void __page_check_anon_rmap(struct page *page,
944 struct vm_area_struct *vma, unsigned long address)
945{
946#ifdef CONFIG_DEBUG_VM
947 /*
948 * The page's anon-rmap details (mapping and index) are guaranteed to
949 * be set up correctly at this point.
950 *
951 * We have exclusion against page_add_anon_rmap because the caller
952 * always holds the page locked, except if called from page_dup_rmap,
953 * in which case the page is already known to be setup.
954 *
955 * We have exclusion against page_add_new_anon_rmap because those pages
956 * are initially only visible via the pagetables, and the pte is locked
957 * over the call to page_add_new_anon_rmap.
958 */
Andrea Arcangeli44ab57a2010-08-09 17:19:10 -0700959 BUG_ON(page_anon_vma(page)->root != vma->anon_vma->root);
Nick Pigginc97a9e12007-05-16 22:11:21 -0700960 BUG_ON(page->index != linear_page_index(vma, address));
961#endif
962}
963
964/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 * page_add_anon_rmap - add pte mapping to an anonymous page
966 * @page: the page to add the mapping to
967 * @vma: the vm area in which the mapping is added
968 * @address: the user virtual address mapped
969 *
Hugh Dickins5ad64682009-12-14 17:59:24 -0800970 * The caller needs to hold the pte lock, and the page must be locked in
Hugh Dickins80e148222009-12-14 17:59:29 -0800971 * the anon_vma case: to serialize mapping,index checking after setting,
972 * and to ensure that PageAnon is not being upgraded racily to PageKsm
973 * (but PageKsm is never downgraded to PageAnon).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 */
975void page_add_anon_rmap(struct page *page,
976 struct vm_area_struct *vma, unsigned long address)
977{
Rik van Rielad8c2ee2010-08-09 17:19:48 -0700978 do_page_add_anon_rmap(page, vma, address, 0);
979}
980
981/*
982 * Special version of the above for do_swap_page, which often runs
983 * into pages that are exclusively owned by the current process.
984 * Everybody else should continue to use page_add_anon_rmap above.
985 */
986void do_page_add_anon_rmap(struct page *page,
987 struct vm_area_struct *vma, unsigned long address, int exclusive)
988{
Hugh Dickins5ad64682009-12-14 17:59:24 -0800989 int first = atomic_inc_and_test(&page->_mapcount);
Andrea Arcangeli79134172011-01-13 15:46:58 -0800990 if (first) {
Jianyu Zhanbea04b02014-06-04 16:09:51 -0700991 /*
992 * We use the irq-unsafe __{inc|mod}_zone_page_stat because
993 * these counters are not modified in interrupt context, and
994 * pte lock(a spinlock) is held, which implies preemption
995 * disabled.
996 */
Kirill A. Shutemov3cd14fc2013-09-12 15:13:58 -0700997 if (PageTransHuge(page))
Andrea Arcangeli79134172011-01-13 15:46:58 -0800998 __inc_zone_page_state(page,
999 NR_ANON_TRANSPARENT_HUGEPAGES);
Kirill A. Shutemov3cd14fc2013-09-12 15:13:58 -07001000 __mod_zone_page_state(page_zone(page), NR_ANON_PAGES,
1001 hpage_nr_pages(page));
Andrea Arcangeli79134172011-01-13 15:46:58 -08001002 }
Hugh Dickins5ad64682009-12-14 17:59:24 -08001003 if (unlikely(PageKsm(page)))
1004 return;
1005
Sasha Levin309381fea2014-01-23 15:52:54 -08001006 VM_BUG_ON_PAGE(!PageLocked(page), page);
Hugh Dickins5dbe0af2011-05-28 13:17:04 -07001007 /* address might be in next vma when migration races vma_adjust */
Hugh Dickins5ad64682009-12-14 17:59:24 -08001008 if (first)
Rik van Rielad8c2ee2010-08-09 17:19:48 -07001009 __page_set_anon_rmap(page, vma, address, exclusive);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001010 else
Nick Pigginc97a9e12007-05-16 22:11:21 -07001011 __page_check_anon_rmap(page, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012}
1013
Randy Dunlap43d8eac2008-03-19 17:00:43 -07001014/**
Nick Piggin9617d952006-01-06 00:11:12 -08001015 * page_add_new_anon_rmap - add pte mapping to a new anonymous page
1016 * @page: the page to add the mapping to
1017 * @vma: the vm area in which the mapping is added
1018 * @address: the user virtual address mapped
1019 *
1020 * Same as page_add_anon_rmap but must only be called on *new* pages.
1021 * This means the inc-and-test can be bypassed.
Nick Pigginc97a9e12007-05-16 22:11:21 -07001022 * Page does not have to be locked.
Nick Piggin9617d952006-01-06 00:11:12 -08001023 */
1024void page_add_new_anon_rmap(struct page *page,
1025 struct vm_area_struct *vma, unsigned long address)
1026{
Hugh Dickinsb5934c52009-01-06 14:39:25 -08001027 VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end);
Hugh Dickinscbf84b72009-01-06 14:39:27 -08001028 SetPageSwapBacked(page);
1029 atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */
Kirill A. Shutemov3cd14fc2013-09-12 15:13:58 -07001030 if (PageTransHuge(page))
Andrea Arcangeli79134172011-01-13 15:46:58 -08001031 __inc_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES);
Kirill A. Shutemov3cd14fc2013-09-12 15:13:58 -07001032 __mod_zone_page_state(page_zone(page), NR_ANON_PAGES,
1033 hpage_nr_pages(page));
Rik van Riele8a03fe2010-04-14 17:59:28 -04001034 __page_set_anon_rmap(page, vma, address, 1);
Jianyu Zhan7ee07a42014-06-04 16:09:52 -07001035
1036 VM_BUG_ON_PAGE(PageLRU(page), page);
1037 if (likely((vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) != VM_LOCKED)) {
Mel Gormanc53954a2013-07-03 15:02:34 -07001038 SetPageActive(page);
1039 lru_cache_add(page);
Jianyu Zhan7ee07a42014-06-04 16:09:52 -07001040 return;
1041 }
1042
1043 if (!TestSetPageMlocked(page)) {
1044 /*
1045 * We use the irq-unsafe __mod_zone_page_stat because this
1046 * counter is not modified from interrupt context, and the pte
1047 * lock is held(spinlock), which implies preemption disabled.
1048 */
1049 __mod_zone_page_state(page_zone(page), NR_MLOCK,
1050 hpage_nr_pages(page));
1051 count_vm_event(UNEVICTABLE_PGMLOCKED);
1052 }
1053 add_page_to_unevictable_list(page);
Nick Piggin9617d952006-01-06 00:11:12 -08001054}
1055
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056/**
1057 * page_add_file_rmap - add pte mapping to a file page
1058 * @page: the page to add the mapping to
1059 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -07001060 * The caller needs to hold the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 */
1062void page_add_file_rmap(struct page *page)
1063{
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001064 bool locked;
1065 unsigned long flags;
1066
1067 mem_cgroup_begin_update_page_stat(page, &locked, &flags);
Balbir Singhd69b0422009-06-17 16:26:34 -07001068 if (atomic_inc_and_test(&page->_mapcount)) {
Christoph Lameter65ba55f2006-06-30 01:55:34 -07001069 __inc_zone_page_state(page, NR_FILE_MAPPED);
Sha Zhengju68b48762013-09-12 15:13:50 -07001070 mem_cgroup_inc_page_stat(page, MEM_CGROUP_STAT_FILE_MAPPED);
Balbir Singhd69b0422009-06-17 16:26:34 -07001071 }
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001072 mem_cgroup_end_update_page_stat(page, &locked, &flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073}
1074
1075/**
1076 * page_remove_rmap - take down pte mapping from a page
1077 * @page: page to remove mapping from
1078 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -07001079 * The caller needs to hold the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 */
Hugh Dickinsedc315f2009-01-06 14:40:11 -08001081void page_remove_rmap(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082{
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001083 bool anon = PageAnon(page);
1084 bool locked;
1085 unsigned long flags;
1086
1087 /*
1088 * The anon case has no mem_cgroup page_stat to update; but may
1089 * uncharge_page() below, where the lock ordering can deadlock if
1090 * we hold the lock against page_stat move: so avoid it on anon.
1091 */
1092 if (!anon)
1093 mem_cgroup_begin_update_page_stat(page, &locked, &flags);
1094
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001095 /* page still mapped by someone else? */
1096 if (!atomic_add_negative(-1, &page->_mapcount))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001097 goto out;
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001098
1099 /*
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001100 * Hugepages are not counted in NR_ANON_PAGES nor NR_FILE_MAPPED
1101 * and not charged by memcg for now.
Jianyu Zhanbea04b02014-06-04 16:09:51 -07001102 *
1103 * We use the irq-unsafe __{inc|mod}_zone_page_stat because
1104 * these counters are not modified in interrupt context, and
1105 * these counters are not modified in interrupt context, and
1106 * pte lock(a spinlock) is held, which implies preemption disabled.
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001107 */
1108 if (unlikely(PageHuge(page)))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001109 goto out;
1110 if (anon) {
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001111 mem_cgroup_uncharge_page(page);
Kirill A. Shutemov3cd14fc2013-09-12 15:13:58 -07001112 if (PageTransHuge(page))
Andrea Arcangeli79134172011-01-13 15:46:58 -08001113 __dec_zone_page_state(page,
1114 NR_ANON_TRANSPARENT_HUGEPAGES);
Kirill A. Shutemov3cd14fc2013-09-12 15:13:58 -07001115 __mod_zone_page_state(page_zone(page), NR_ANON_PAGES,
1116 -hpage_nr_pages(page));
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001117 } else {
1118 __dec_zone_page_state(page, NR_FILE_MAPPED);
Sha Zhengju68b48762013-09-12 15:13:50 -07001119 mem_cgroup_dec_page_stat(page, MEM_CGROUP_STAT_FILE_MAPPED);
Hugh Dickinse6c509f2012-10-08 16:33:19 -07001120 mem_cgroup_end_update_page_stat(page, &locked, &flags);
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001121 }
Hugh Dickinse6c509f2012-10-08 16:33:19 -07001122 if (unlikely(PageMlocked(page)))
1123 clear_page_mlock(page);
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001124 /*
1125 * It would be tidy to reset the PageAnon mapping here,
1126 * but that might overwrite a racing page_add_anon_rmap
1127 * which increments mapcount after us but sets mapping
1128 * before us: so leave the reset to free_hot_cold_page,
1129 * and remember that it's only reliable while mapped.
1130 * Leaving it set also helps swapoff to reinstate ptes
1131 * faster for those pages still in swapcache.
1132 */
Hugh Dickinse6c509f2012-10-08 16:33:19 -07001133 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001134out:
1135 if (!anon)
1136 mem_cgroup_end_update_page_stat(page, &locked, &flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137}
1138
1139/*
Joonsoo Kim52629502014-01-21 15:49:50 -08001140 * @arg: enum ttu_flags will be passed to this argument
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 */
Kirill A. Shutemovac769502014-06-04 16:08:17 -07001142static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
Joonsoo Kim52629502014-01-21 15:49:50 -08001143 unsigned long address, void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144{
1145 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 pte_t *pte;
1147 pte_t pteval;
Hugh Dickinsc0718802005-10-29 18:16:31 -07001148 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 int ret = SWAP_AGAIN;
Joonsoo Kim52629502014-01-21 15:49:50 -08001150 enum ttu_flags flags = (enum ttu_flags)arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151
Nick Piggin479db0b2008-08-20 14:09:18 -07001152 pte = page_check_address(page, mm, address, &ptl, 0);
Hugh Dickinsc0718802005-10-29 18:16:31 -07001153 if (!pte)
Nikita Danilov81b40822005-05-01 08:58:36 -07001154 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155
1156 /*
1157 * If the page is mlock()d, we cannot swap it out.
1158 * If it's recently referenced (perhaps page_referenced
1159 * skipped over this mm) then we should reactivate it.
1160 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001161 if (!(flags & TTU_IGNORE_MLOCK)) {
KOSAKI Motohirocaed0f482009-12-14 17:59:45 -08001162 if (vma->vm_flags & VM_LOCKED)
1163 goto out_mlock;
1164
Hugh Dickinsaf8e3352009-12-14 17:58:59 -08001165 if (TTU_ACTION(flags) == TTU_MUNLOCK)
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001166 goto out_unmap;
Andi Kleen14fa31b2009-09-16 11:50:10 +02001167 }
1168 if (!(flags & TTU_IGNORE_ACCESS)) {
Nick Pigginb291f002008-10-18 20:26:44 -07001169 if (ptep_clear_flush_young_notify(vma, address, pte)) {
1170 ret = SWAP_FAIL;
1171 goto out_unmap;
1172 }
1173 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 /* Nuke the page table entry. */
1176 flush_cache_page(vma, address, page_to_pfn(page));
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001177 pteval = ptep_clear_flush(vma, address, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178
1179 /* Move the dirty bit to the physical page now the pte is gone. */
1180 if (pte_dirty(pteval))
1181 set_page_dirty(page);
1182
Hugh Dickins365e9c872005-10-29 18:16:18 -07001183 /* Update high watermark before we lower rss */
1184 update_hiwater_rss(mm);
1185
Andi Kleen888b9f72009-09-16 11:50:11 +02001186 if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) {
Naoya Horiguchi5f24ae52012-12-12 13:52:30 -08001187 if (!PageHuge(page)) {
1188 if (PageAnon(page))
1189 dec_mm_counter(mm, MM_ANONPAGES);
1190 else
1191 dec_mm_counter(mm, MM_FILEPAGES);
1192 }
Andi Kleen888b9f72009-09-16 11:50:11 +02001193 set_pte_at(mm, address, pte,
Naoya Horiguchi5f24ae52012-12-12 13:52:30 -08001194 swp_entry_to_pte(make_hwpoison_entry(page)));
Konstantin Weitz45961722013-04-17 13:59:32 +02001195 } else if (pte_unused(pteval)) {
1196 /*
1197 * The guest indicated that the page content is of no
1198 * interest anymore. Simply discard the pte, vmscan
1199 * will take care of the rest.
1200 */
1201 if (PageAnon(page))
1202 dec_mm_counter(mm, MM_ANONPAGES);
1203 else
1204 dec_mm_counter(mm, MM_FILEPAGES);
Andi Kleen888b9f72009-09-16 11:50:11 +02001205 } else if (PageAnon(page)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001206 swp_entry_t entry = { .val = page_private(page) };
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001207 pte_t swp_pte;
Christoph Lameter06972122006-06-23 02:03:35 -07001208
1209 if (PageSwapCache(page)) {
1210 /*
1211 * Store the swap location in the pte.
1212 * See handle_pte_fault() ...
1213 */
Hugh Dickins570a335b2009-12-14 17:58:46 -08001214 if (swap_duplicate(entry) < 0) {
1215 set_pte_at(mm, address, pte, pteval);
1216 ret = SWAP_FAIL;
1217 goto out_unmap;
1218 }
Christoph Lameter06972122006-06-23 02:03:35 -07001219 if (list_empty(&mm->mmlist)) {
1220 spin_lock(&mmlist_lock);
1221 if (list_empty(&mm->mmlist))
1222 list_add(&mm->mmlist, &init_mm.mmlist);
1223 spin_unlock(&mmlist_lock);
1224 }
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001225 dec_mm_counter(mm, MM_ANONPAGES);
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08001226 inc_mm_counter(mm, MM_SWAPENTS);
Konstantin Khlebnikovce1744f2012-03-21 16:33:59 -07001227 } else if (IS_ENABLED(CONFIG_MIGRATION)) {
Christoph Lameter06972122006-06-23 02:03:35 -07001228 /*
1229 * Store the pfn of the page in a special migration
1230 * pte. do_swap_page() will wait until the migration
1231 * pte is removed and then restart fault handling.
1232 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001233 BUG_ON(TTU_ACTION(flags) != TTU_MIGRATION);
Christoph Lameter06972122006-06-23 02:03:35 -07001234 entry = make_migration_entry(page, pte_write(pteval));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 }
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001236 swp_pte = swp_entry_to_pte(entry);
1237 if (pte_soft_dirty(pteval))
1238 swp_pte = pte_swp_mksoft_dirty(swp_pte);
1239 set_pte_at(mm, address, pte, swp_pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 BUG_ON(pte_file(*pte));
Konstantin Khlebnikovce1744f2012-03-21 16:33:59 -07001241 } else if (IS_ENABLED(CONFIG_MIGRATION) &&
1242 (TTU_ACTION(flags) == TTU_MIGRATION)) {
Christoph Lameter04e62a22006-06-23 02:03:38 -07001243 /* Establish migration entry for a file page */
1244 swp_entry_t entry;
1245 entry = make_migration_entry(page, pte_write(pteval));
1246 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
1247 } else
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001248 dec_mm_counter(mm, MM_FILEPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249
Hugh Dickinsedc315f2009-01-06 14:40:11 -08001250 page_remove_rmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 page_cache_release(page);
1252
1253out_unmap:
Hugh Dickinsc0718802005-10-29 18:16:31 -07001254 pte_unmap_unlock(pte, ptl);
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001255 if (ret != SWAP_FAIL)
1256 mmu_notifier_invalidate_page(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257out:
1258 return ret;
KOSAKI Motohirocaed0f482009-12-14 17:59:45 -08001259
1260out_mlock:
1261 pte_unmap_unlock(pte, ptl);
1262
1263
1264 /*
1265 * We need mmap_sem locking, Otherwise VM_LOCKED check makes
1266 * unstable result and race. Plus, We can't wait here because
Ingo Molnar5a505082012-12-02 19:56:46 +00001267 * we now hold anon_vma->rwsem or mapping->i_mmap_mutex.
KOSAKI Motohirocaed0f482009-12-14 17:59:45 -08001268 * if trylock failed, the page remain in evictable lru and later
1269 * vmscan could retry to move the page to unevictable lru if the
1270 * page is actually mlocked.
1271 */
1272 if (down_read_trylock(&vma->vm_mm->mmap_sem)) {
1273 if (vma->vm_flags & VM_LOCKED) {
1274 mlock_vma_page(page);
1275 ret = SWAP_MLOCK;
1276 }
1277 up_read(&vma->vm_mm->mmap_sem);
1278 }
1279 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280}
1281
1282/*
1283 * objrmap doesn't work for nonlinear VMAs because the assumption that
1284 * offset-into-file correlates with offset-into-virtual-addresses does not hold.
1285 * Consequently, given a particular page and its ->index, we cannot locate the
1286 * ptes which are mapping that page without an exhaustive linear search.
1287 *
1288 * So what this code does is a mini "virtual scan" of each nonlinear VMA which
1289 * maps the file to which the target page belongs. The ->vm_private_data field
1290 * holds the current cursor into that scan. Successive searches will circulate
1291 * around the vma's virtual address space.
1292 *
1293 * So as more replacement pressure is applied to the pages in a nonlinear VMA,
1294 * more scanning pressure is placed against them as well. Eventually pages
1295 * will become fully unmapped and are eligible for eviction.
1296 *
1297 * For very sparsely populated VMAs this is a little inefficient - chances are
1298 * there there won't be many ptes located within the scan cluster. In this case
1299 * maybe we could scan further - to the end of the pte page, perhaps.
Nick Pigginb291f002008-10-18 20:26:44 -07001300 *
1301 * Mlocked pages: check VM_LOCKED under mmap_sem held for read, if we can
1302 * acquire it without blocking. If vma locked, mlock the pages in the cluster,
1303 * rather than unmapping them. If we encounter the "check_page" that vmscan is
1304 * trying to unmap, return SWAP_MLOCK, else default SWAP_AGAIN.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 */
1306#define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE)
1307#define CLUSTER_MASK (~(CLUSTER_SIZE - 1))
1308
Nick Pigginb291f002008-10-18 20:26:44 -07001309static int try_to_unmap_cluster(unsigned long cursor, unsigned int *mapcount,
1310 struct vm_area_struct *vma, struct page *check_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311{
1312 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 pmd_t *pmd;
Hugh Dickinsc0718802005-10-29 18:16:31 -07001314 pte_t *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 pte_t pteval;
Hugh Dickinsc0718802005-10-29 18:16:31 -07001316 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317 struct page *page;
1318 unsigned long address;
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001319 unsigned long mmun_start; /* For mmu_notifiers */
1320 unsigned long mmun_end; /* For mmu_notifiers */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 unsigned long end;
Nick Pigginb291f002008-10-18 20:26:44 -07001322 int ret = SWAP_AGAIN;
1323 int locked_vma = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 address = (vma->vm_start + cursor) & CLUSTER_MASK;
1326 end = address + CLUSTER_SIZE;
1327 if (address < vma->vm_start)
1328 address = vma->vm_start;
1329 if (end > vma->vm_end)
1330 end = vma->vm_end;
1331
Bob Liu62190492012-12-11 16:00:37 -08001332 pmd = mm_find_pmd(mm, address);
1333 if (!pmd)
Nick Pigginb291f002008-10-18 20:26:44 -07001334 return ret;
1335
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001336 mmun_start = address;
1337 mmun_end = end;
1338 mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
1339
Nick Pigginb291f002008-10-18 20:26:44 -07001340 /*
Hugh Dickinsaf8e3352009-12-14 17:58:59 -08001341 * If we can acquire the mmap_sem for read, and vma is VM_LOCKED,
Nick Pigginb291f002008-10-18 20:26:44 -07001342 * keep the sem while scanning the cluster for mlocking pages.
1343 */
Hugh Dickinsaf8e3352009-12-14 17:58:59 -08001344 if (down_read_trylock(&vma->vm_mm->mmap_sem)) {
Nick Pigginb291f002008-10-18 20:26:44 -07001345 locked_vma = (vma->vm_flags & VM_LOCKED);
1346 if (!locked_vma)
1347 up_read(&vma->vm_mm->mmap_sem); /* don't need it */
1348 }
Hugh Dickinsc0718802005-10-29 18:16:31 -07001349
1350 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351
Hugh Dickins365e9c872005-10-29 18:16:18 -07001352 /* Update high watermark before we lower rss */
1353 update_hiwater_rss(mm);
1354
Hugh Dickinsc0718802005-10-29 18:16:31 -07001355 for (; address < end; pte++, address += PAGE_SIZE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 if (!pte_present(*pte))
1357 continue;
Linus Torvalds6aab3412005-11-28 14:34:23 -08001358 page = vm_normal_page(vma, address, *pte);
1359 BUG_ON(!page || PageAnon(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360
Nick Pigginb291f002008-10-18 20:26:44 -07001361 if (locked_vma) {
Vlastimil Babka57e68e92014-04-07 15:37:50 -07001362 if (page == check_page) {
1363 /* we know we have check_page locked */
1364 mlock_vma_page(page);
Nick Pigginb291f002008-10-18 20:26:44 -07001365 ret = SWAP_MLOCK;
Vlastimil Babka57e68e92014-04-07 15:37:50 -07001366 } else if (trylock_page(page)) {
1367 /*
1368 * If we can lock the page, perform mlock.
1369 * Otherwise leave the page alone, it will be
1370 * eventually encountered again later.
1371 */
1372 mlock_vma_page(page);
1373 unlock_page(page);
1374 }
Nick Pigginb291f002008-10-18 20:26:44 -07001375 continue; /* don't unmap */
1376 }
1377
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001378 if (ptep_clear_flush_young_notify(vma, address, pte))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 continue;
1380
1381 /* Nuke the page table entry. */
Ben Collinseca35132005-11-29 11:45:26 -08001382 flush_cache_page(vma, address, pte_pfn(*pte));
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001383 pteval = ptep_clear_flush(vma, address, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384
1385 /* If nonlinear, store the file page offset in the pte. */
Cyrill Gorcunov41bb3472013-08-13 16:00:51 -07001386 if (page->index != linear_page_index(vma, address)) {
1387 pte_t ptfile = pgoff_to_pte(page->index);
1388 if (pte_soft_dirty(pteval))
Cyrill Gorcunovb43790e2014-06-04 16:06:42 -07001389 ptfile = pte_file_mksoft_dirty(ptfile);
Cyrill Gorcunov41bb3472013-08-13 16:00:51 -07001390 set_pte_at(mm, address, pte, ptfile);
1391 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392
1393 /* Move the dirty bit to the physical page now the pte is gone. */
1394 if (pte_dirty(pteval))
1395 set_page_dirty(page);
1396
Hugh Dickinsedc315f2009-01-06 14:40:11 -08001397 page_remove_rmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 page_cache_release(page);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001399 dec_mm_counter(mm, MM_FILEPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 (*mapcount)--;
1401 }
Hugh Dickinsc0718802005-10-29 18:16:31 -07001402 pte_unmap_unlock(pte - 1, ptl);
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001403 mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
Nick Pigginb291f002008-10-18 20:26:44 -07001404 if (locked_vma)
1405 up_read(&vma->vm_mm->mmap_sem);
1406 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407}
1408
Joonsoo Kim0f843c62014-01-21 15:49:45 -08001409static int try_to_unmap_nonlinear(struct page *page,
Hugh Dickins7e09e732014-03-20 21:52:17 -07001410 struct address_space *mapping, void *arg)
Joonsoo Kim0f843c62014-01-21 15:49:45 -08001411{
Hugh Dickins7e09e732014-03-20 21:52:17 -07001412 struct vm_area_struct *vma;
Joonsoo Kim0f843c62014-01-21 15:49:45 -08001413 int ret = SWAP_AGAIN;
1414 unsigned long cursor;
1415 unsigned long max_nl_cursor = 0;
1416 unsigned long max_nl_size = 0;
1417 unsigned int mapcount;
1418
1419 list_for_each_entry(vma,
1420 &mapping->i_mmap_nonlinear, shared.nonlinear) {
1421
1422 cursor = (unsigned long) vma->vm_private_data;
1423 if (cursor > max_nl_cursor)
1424 max_nl_cursor = cursor;
1425 cursor = vma->vm_end - vma->vm_start;
1426 if (cursor > max_nl_size)
1427 max_nl_size = cursor;
1428 }
1429
1430 if (max_nl_size == 0) { /* all nonlinears locked or reserved ? */
1431 return SWAP_FAIL;
1432 }
1433
1434 /*
1435 * We don't try to search for this page in the nonlinear vmas,
1436 * and page_referenced wouldn't have found it anyway. Instead
1437 * just walk the nonlinear vmas trying to age and unmap some.
1438 * The mapcount of the page we came in with is irrelevant,
1439 * but even so use it as a guide to how hard we should try?
1440 */
1441 mapcount = page_mapcount(page);
1442 if (!mapcount)
1443 return ret;
1444
1445 cond_resched();
1446
1447 max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK;
1448 if (max_nl_cursor == 0)
1449 max_nl_cursor = CLUSTER_SIZE;
1450
1451 do {
1452 list_for_each_entry(vma,
1453 &mapping->i_mmap_nonlinear, shared.nonlinear) {
1454
1455 cursor = (unsigned long) vma->vm_private_data;
1456 while (cursor < max_nl_cursor &&
1457 cursor < vma->vm_end - vma->vm_start) {
1458 if (try_to_unmap_cluster(cursor, &mapcount,
1459 vma, page) == SWAP_MLOCK)
1460 ret = SWAP_MLOCK;
1461 cursor += CLUSTER_SIZE;
1462 vma->vm_private_data = (void *) cursor;
1463 if ((int)mapcount <= 0)
1464 return ret;
1465 }
1466 vma->vm_private_data = (void *) max_nl_cursor;
1467 }
1468 cond_resched();
1469 max_nl_cursor += CLUSTER_SIZE;
1470 } while (max_nl_cursor <= max_nl_size);
1471
1472 /*
1473 * Don't loop forever (perhaps all the remaining pages are
1474 * in locked vmas). Reset cursor on all unreserved nonlinear
1475 * vmas, now forgetting on which ones it had fallen behind.
1476 */
1477 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.nonlinear)
1478 vma->vm_private_data = NULL;
1479
1480 return ret;
1481}
1482
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001483bool is_vma_temporary_stack(struct vm_area_struct *vma)
Mel Gormana8bef8ff2010-05-24 14:32:24 -07001484{
1485 int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP);
1486
1487 if (!maybe_stack)
1488 return false;
1489
1490 if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) ==
1491 VM_STACK_INCOMPLETE_SETUP)
1492 return true;
1493
1494 return false;
1495}
1496
Joonsoo Kim52629502014-01-21 15:49:50 -08001497static bool invalid_migration_vma(struct vm_area_struct *vma, void *arg)
1498{
1499 return is_vma_temporary_stack(vma);
1500}
1501
Joonsoo Kim52629502014-01-21 15:49:50 -08001502static int page_not_mapped(struct page *page)
1503{
1504 return !page_mapped(page);
1505};
1506
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507/**
1508 * try_to_unmap - try to remove all page table mappings to a page
1509 * @page: the page to get unmapped
Andi Kleen14fa31b2009-09-16 11:50:10 +02001510 * @flags: action and flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 *
1512 * Tries to remove all the page table entries which are mapping this
1513 * page, used in the pageout path. Caller must hold the page lock.
1514 * Return values are:
1515 *
1516 * SWAP_SUCCESS - we succeeded in removing all mappings
1517 * SWAP_AGAIN - we missed a mapping, try again later
1518 * SWAP_FAIL - the page is unswappable
Nick Pigginb291f002008-10-18 20:26:44 -07001519 * SWAP_MLOCK - page is mlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001521int try_to_unmap(struct page *page, enum ttu_flags flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522{
1523 int ret;
Joonsoo Kim52629502014-01-21 15:49:50 -08001524 struct rmap_walk_control rwc = {
1525 .rmap_one = try_to_unmap_one,
1526 .arg = (void *)flags,
1527 .done = page_not_mapped,
1528 .file_nonlinear = try_to_unmap_nonlinear,
1529 .anon_lock = page_lock_anon_vma_read,
1530 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531
Sasha Levin309381fea2014-01-23 15:52:54 -08001532 VM_BUG_ON_PAGE(!PageHuge(page) && PageTransHuge(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533
Joonsoo Kim52629502014-01-21 15:49:50 -08001534 /*
1535 * During exec, a temporary VMA is setup and later moved.
1536 * The VMA is moved under the anon_vma lock but not the
1537 * page tables leading to a race where migration cannot
1538 * find the migration ptes. Rather than increasing the
1539 * locking requirements of exec(), migration skips
1540 * temporary VMAs until after exec() completes.
1541 */
1542 if (flags & TTU_MIGRATION && !PageKsm(page) && PageAnon(page))
1543 rwc.invalid_vma = invalid_migration_vma;
1544
1545 ret = rmap_walk(page, &rwc);
1546
Nick Pigginb291f002008-10-18 20:26:44 -07001547 if (ret != SWAP_MLOCK && !page_mapped(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 ret = SWAP_SUCCESS;
1549 return ret;
1550}
Nikita Danilov81b40822005-05-01 08:58:36 -07001551
Nick Pigginb291f002008-10-18 20:26:44 -07001552/**
1553 * try_to_munlock - try to munlock a page
1554 * @page: the page to be munlocked
1555 *
1556 * Called from munlock code. Checks all of the VMAs mapping the page
1557 * to make sure nobody else has this page mlocked. The page will be
1558 * returned with PG_mlocked cleared if no other vmas have it mlocked.
1559 *
1560 * Return values are:
1561 *
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001562 * SWAP_AGAIN - no vma is holding page mlocked, or,
Nick Pigginb291f002008-10-18 20:26:44 -07001563 * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem
Hugh Dickins5ad64682009-12-14 17:59:24 -08001564 * SWAP_FAIL - page cannot be located at present
Nick Pigginb291f002008-10-18 20:26:44 -07001565 * SWAP_MLOCK - page is now mlocked.
1566 */
1567int try_to_munlock(struct page *page)
1568{
Joonsoo Kime8351ac2014-01-21 15:49:52 -08001569 int ret;
1570 struct rmap_walk_control rwc = {
1571 .rmap_one = try_to_unmap_one,
1572 .arg = (void *)TTU_MUNLOCK,
1573 .done = page_not_mapped,
1574 /*
1575 * We don't bother to try to find the munlocked page in
1576 * nonlinears. It's costly. Instead, later, page reclaim logic
1577 * may call try_to_unmap() and recover PG_mlocked lazily.
1578 */
1579 .file_nonlinear = NULL,
1580 .anon_lock = page_lock_anon_vma_read,
1581
1582 };
1583
Sasha Levin309381fea2014-01-23 15:52:54 -08001584 VM_BUG_ON_PAGE(!PageLocked(page) || PageLRU(page), page);
Nick Pigginb291f002008-10-18 20:26:44 -07001585
Joonsoo Kime8351ac2014-01-21 15:49:52 -08001586 ret = rmap_walk(page, &rwc);
1587 return ret;
Nick Pigginb291f002008-10-18 20:26:44 -07001588}
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001589
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001590void __put_anon_vma(struct anon_vma *anon_vma)
Rik van Riel76545062010-08-09 17:18:41 -07001591{
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001592 struct anon_vma *root = anon_vma->root;
Rik van Riel76545062010-08-09 17:18:41 -07001593
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001594 if (root != anon_vma && atomic_dec_and_test(&root->refcount))
1595 anon_vma_free(root);
Rik van Riel76545062010-08-09 17:18:41 -07001596
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001597 anon_vma_free(anon_vma);
Rik van Riel76545062010-08-09 17:18:41 -07001598}
Rik van Riel76545062010-08-09 17:18:41 -07001599
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001600static struct anon_vma *rmap_walk_anon_lock(struct page *page,
1601 struct rmap_walk_control *rwc)
Joonsoo Kimfaecd8d2014-01-21 15:49:46 -08001602{
1603 struct anon_vma *anon_vma;
1604
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001605 if (rwc->anon_lock)
1606 return rwc->anon_lock(page);
1607
Joonsoo Kimfaecd8d2014-01-21 15:49:46 -08001608 /*
1609 * Note: remove_migration_ptes() cannot use page_lock_anon_vma_read()
1610 * because that depends on page_mapped(); but not all its usages
1611 * are holding mmap_sem. Users without mmap_sem are required to
1612 * take a reference count to prevent the anon_vma disappearing
1613 */
1614 anon_vma = page_anon_vma(page);
1615 if (!anon_vma)
1616 return NULL;
1617
1618 anon_vma_lock_read(anon_vma);
1619 return anon_vma;
1620}
1621
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001622/*
Joonsoo Kime8351ac2014-01-21 15:49:52 -08001623 * rmap_walk_anon - do something to anonymous page using the object-based
1624 * rmap method
1625 * @page: the page to be handled
1626 * @rwc: control variable according to each walk type
1627 *
1628 * Find all the mappings of a page using the mapping pointer and the vma chains
1629 * contained in the anon_vma struct it points to.
1630 *
1631 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1632 * where the page was found will be held for write. So, we won't recheck
1633 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1634 * LOCKED.
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001635 */
Joonsoo Kim051ac832014-01-21 15:49:48 -08001636static int rmap_walk_anon(struct page *page, struct rmap_walk_control *rwc)
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001637{
1638 struct anon_vma *anon_vma;
Michel Lespinassebf181b92012-10-08 16:31:39 -07001639 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
Rik van Riel5beb4932010-03-05 13:42:07 -08001640 struct anon_vma_chain *avc;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001641 int ret = SWAP_AGAIN;
1642
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001643 anon_vma = rmap_walk_anon_lock(page, rwc);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001644 if (!anon_vma)
1645 return ret;
Joonsoo Kimfaecd8d2014-01-21 15:49:46 -08001646
Michel Lespinassebf181b92012-10-08 16:31:39 -07001647 anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) {
Rik van Riel5beb4932010-03-05 13:42:07 -08001648 struct vm_area_struct *vma = avc->vma;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001649 unsigned long address = vma_address(page, vma);
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001650
1651 if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg))
1652 continue;
1653
Joonsoo Kim051ac832014-01-21 15:49:48 -08001654 ret = rwc->rmap_one(page, vma, address, rwc->arg);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001655 if (ret != SWAP_AGAIN)
1656 break;
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001657 if (rwc->done && rwc->done(page))
1658 break;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001659 }
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +00001660 anon_vma_unlock_read(anon_vma);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001661 return ret;
1662}
1663
Joonsoo Kime8351ac2014-01-21 15:49:52 -08001664/*
1665 * rmap_walk_file - do something to file page using the object-based rmap method
1666 * @page: the page to be handled
1667 * @rwc: control variable according to each walk type
1668 *
1669 * Find all the mappings of a page using the mapping pointer and the vma chains
1670 * contained in the address_space struct it points to.
1671 *
1672 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1673 * where the page was found will be held for write. So, we won't recheck
1674 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1675 * LOCKED.
1676 */
Joonsoo Kim051ac832014-01-21 15:49:48 -08001677static int rmap_walk_file(struct page *page, struct rmap_walk_control *rwc)
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001678{
1679 struct address_space *mapping = page->mapping;
Joonsoo Kimb854f712014-01-21 15:49:43 -08001680 pgoff_t pgoff = page->index << compound_order(page);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001681 struct vm_area_struct *vma;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001682 int ret = SWAP_AGAIN;
1683
Joonsoo Kim9f326242014-01-21 15:49:53 -08001684 /*
1685 * The page lock not only makes sure that page->mapping cannot
1686 * suddenly be NULLified by truncation, it makes sure that the
1687 * structure at mapping cannot be freed and reused yet,
1688 * so we can safely take mapping->i_mmap_mutex.
1689 */
1690 VM_BUG_ON(!PageLocked(page));
1691
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001692 if (!mapping)
1693 return ret;
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07001694 mutex_lock(&mapping->i_mmap_mutex);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07001695 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001696 unsigned long address = vma_address(page, vma);
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001697
1698 if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg))
1699 continue;
1700
Joonsoo Kim051ac832014-01-21 15:49:48 -08001701 ret = rwc->rmap_one(page, vma, address, rwc->arg);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001702 if (ret != SWAP_AGAIN)
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001703 goto done;
1704 if (rwc->done && rwc->done(page))
1705 goto done;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001706 }
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001707
1708 if (!rwc->file_nonlinear)
1709 goto done;
1710
1711 if (list_empty(&mapping->i_mmap_nonlinear))
1712 goto done;
1713
Hugh Dickins7e09e732014-03-20 21:52:17 -07001714 ret = rwc->file_nonlinear(page, mapping, rwc->arg);
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001715
1716done:
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07001717 mutex_unlock(&mapping->i_mmap_mutex);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001718 return ret;
1719}
1720
Joonsoo Kim051ac832014-01-21 15:49:48 -08001721int rmap_walk(struct page *page, struct rmap_walk_control *rwc)
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001722{
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001723 if (unlikely(PageKsm(page)))
Joonsoo Kim051ac832014-01-21 15:49:48 -08001724 return rmap_walk_ksm(page, rwc);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001725 else if (PageAnon(page))
Joonsoo Kim051ac832014-01-21 15:49:48 -08001726 return rmap_walk_anon(page, rwc);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001727 else
Joonsoo Kim051ac832014-01-21 15:49:48 -08001728 return rmap_walk_file(page, rwc);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001729}
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001730
Naoya Horiguchie3390f62010-06-15 13:18:13 +09001731#ifdef CONFIG_HUGETLB_PAGE
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001732/*
1733 * The following three functions are for anonymous (private mapped) hugepages.
1734 * Unlike common anonymous pages, anonymous hugepages have no accounting code
1735 * and no lru code, because we handle hugepages differently from common pages.
1736 */
1737static void __hugepage_set_anon_rmap(struct page *page,
1738 struct vm_area_struct *vma, unsigned long address, int exclusive)
1739{
1740 struct anon_vma *anon_vma = vma->anon_vma;
Naoya Horiguchi433abed2010-09-10 13:23:03 +09001741
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001742 BUG_ON(!anon_vma);
Naoya Horiguchi433abed2010-09-10 13:23:03 +09001743
1744 if (PageAnon(page))
1745 return;
1746 if (!exclusive)
1747 anon_vma = anon_vma->root;
1748
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001749 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
1750 page->mapping = (struct address_space *) anon_vma;
1751 page->index = linear_page_index(vma, address);
1752}
1753
1754void hugepage_add_anon_rmap(struct page *page,
1755 struct vm_area_struct *vma, unsigned long address)
1756{
1757 struct anon_vma *anon_vma = vma->anon_vma;
1758 int first;
Naoya Horiguchia850ea32010-09-10 13:23:06 +09001759
1760 BUG_ON(!PageLocked(page));
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001761 BUG_ON(!anon_vma);
Hugh Dickins5dbe0af2011-05-28 13:17:04 -07001762 /* address might be in next vma when migration races vma_adjust */
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001763 first = atomic_inc_and_test(&page->_mapcount);
1764 if (first)
1765 __hugepage_set_anon_rmap(page, vma, address, 0);
1766}
1767
1768void hugepage_add_new_anon_rmap(struct page *page,
1769 struct vm_area_struct *vma, unsigned long address)
1770{
1771 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
1772 atomic_set(&page->_mapcount, 0);
1773 __hugepage_set_anon_rmap(page, vma, address, 1);
1774}
Naoya Horiguchie3390f62010-06-15 13:18:13 +09001775#endif /* CONFIG_HUGETLB_PAGE */