blob: 8fc049f9a5a6c5d511ac7a5ac0dd41c698ed99a3 [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 */
Ingo Molnar5a505082012-12-02 19:56:46 +0000106 if (rwsem_is_locked(&anon_vma->root->rwsem)) {
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000107 anon_vma_lock_write(anon_vma);
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800108 anon_vma_unlock_write(anon_vma);
Peter Zijlstra88c22082011-05-24 17:12:13 -0700109 }
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
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000149 * optimistically looked up an anon_vma in page_lock_anon_vma_read()
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700150 * 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
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000184 anon_vma_lock_write(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);
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800194 anon_vma_unlock_write(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))
Ingo Molnar5a505082012-12-02 19:56:46 +0000222 up_write(&root->rwsem);
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700223 root = new_root;
Ingo Molnar5a505082012-12-02 19:56:46 +0000224 down_write(&root->rwsem);
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700225 }
226 return root;
227}
228
229static inline void unlock_anon_vma_root(struct anon_vma *root)
230{
231 if (root)
Ingo Molnar5a505082012-12-02 19:56:46 +0000232 up_write(&root->rwsem);
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700233}
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;
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000309 anon_vma_lock_write(anon_vma);
Rik van Riel5c341ee12010-08-09 17:18:39 -0700310 anon_vma_chain_link(vma, avc, anon_vma);
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800311 anon_vma_unlock_write(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()
Ingo Molnar5a505082012-12-02 19:56:46 +0000352 * needing to write-acquire the anon_vma->root->rwsem.
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200353 */
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
Ingo Molnar5a505082012-12-02 19:56:46 +0000368 init_rwsem(&anon_vma->rwsem);
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 */
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000445struct anon_vma *page_lock_anon_vma_read(struct page *page)
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700446{
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);
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000460 if (down_read_trylock(&root_anon_vma->rwsem)) {
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)) {
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000467 up_read(&root_anon_vma->rwsem);
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();
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000487 anon_vma_lock_read(anon_vma);
Peter Zijlstra88c22082011-05-24 17:12:13 -0700488
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
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000493 * we'll deadlock on the anon_vma_lock_write() recursion.
Peter Zijlstra88c22082011-05-24 17:12:13 -0700494 */
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000495 anon_vma_unlock_read(anon_vma);
Peter Zijlstra88c22082011-05-24 17:12:13 -0700496 __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
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000507void page_unlock_anon_vma_read(struct anon_vma *anon_vma)
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800508{
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000509 anon_vma_unlock_read(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))) {
Jianguo Wu98398c32013-12-18 17:08:59 -0800603 /* when pud is not present, pte will be NULL */
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +0900604 pte = huge_pte_offset(mm, address);
Jianguo Wu98398c32013-12-18 17:08:59 -0800605 if (!pte)
606 return NULL;
607
Kirill A. Shutemovcb900f42013-11-14 14:31:02 -0800608 ptl = huge_pte_lockptr(page_hstate(page), mm, pte);
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +0900609 goto check;
610 }
611
Bob Liu62190492012-12-11 16:00:37 -0800612 pmd = mm_find_pmd(mm, address);
613 if (!pmd)
Hugh Dickinsc0718802005-10-29 18:16:31 -0700614 return NULL;
615
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800616 if (pmd_trans_huge(*pmd))
617 return NULL;
Hugh Dickinsc0718802005-10-29 18:16:31 -0700618
619 pte = pte_offset_map(pmd, address);
620 /* Make a quick check before getting the lock */
Nick Piggin479db0b2008-08-20 14:09:18 -0700621 if (!sync && !pte_present(*pte)) {
Hugh Dickinsc0718802005-10-29 18:16:31 -0700622 pte_unmap(pte);
623 return NULL;
Nikita Danilov81b40822005-05-01 08:58:36 -0700624 }
Hugh Dickinsc0718802005-10-29 18:16:31 -0700625
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700626 ptl = pte_lockptr(mm, pmd);
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +0900627check:
Hugh Dickinsc0718802005-10-29 18:16:31 -0700628 spin_lock(ptl);
629 if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) {
630 *ptlp = ptl;
631 return pte;
632 }
633 pte_unmap_unlock(pte, ptl);
634 return NULL;
Nikita Danilov81b40822005-05-01 08:58:36 -0700635}
636
Nick Pigginb291f002008-10-18 20:26:44 -0700637/**
638 * page_mapped_in_vma - check whether a page is really mapped in a VMA
639 * @page: the page to test
640 * @vma: the VMA to test
641 *
642 * Returns 1 if the page is mapped into the page tables of the VMA, 0
643 * if the page is not mapped into the page tables of this VMA. Only
644 * valid for normal file or anonymous VMAs.
645 */
Andi Kleen6a460792009-09-16 11:50:15 +0200646int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma)
Nick Pigginb291f002008-10-18 20:26:44 -0700647{
648 unsigned long address;
649 pte_t *pte;
650 spinlock_t *ptl;
651
Michel Lespinasse86c2ad12012-10-08 16:31:42 -0700652 address = __vma_address(page, vma);
653 if (unlikely(address < vma->vm_start || address >= vma->vm_end))
Nick Pigginb291f002008-10-18 20:26:44 -0700654 return 0;
655 pte = page_check_address(page, vma->vm_mm, address, &ptl, 1);
656 if (!pte) /* the page is not in this mm */
657 return 0;
658 pte_unmap_unlock(pte, ptl);
659
660 return 1;
661}
662
Joonsoo Kim9f326242014-01-21 15:49:53 -0800663struct page_referenced_arg {
664 int mapcount;
665 int referenced;
666 unsigned long vm_flags;
667 struct mem_cgroup *memcg;
668};
Nikita Danilov81b40822005-05-01 08:58:36 -0700669/*
Joonsoo Kim9f326242014-01-21 15:49:53 -0800670 * arg: page_referenced_arg will be passed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 */
Hugh Dickins5ad64682009-12-14 17:59:24 -0800672int page_referenced_one(struct page *page, struct vm_area_struct *vma,
Joonsoo Kim9f326242014-01-21 15:49:53 -0800673 unsigned long address, void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674{
675 struct mm_struct *mm = vma->vm_mm;
Kirill A. Shutemov117b0792013-11-14 14:30:56 -0800676 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 int referenced = 0;
Joonsoo Kim9f326242014-01-21 15:49:53 -0800678 struct page_referenced_arg *pra = arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800680 if (unlikely(PageTransHuge(page))) {
681 pmd_t *pmd;
682
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800683 /*
684 * rmap might return false positives; we must filter
685 * these out using page_check_address_pmd().
686 */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800687 pmd = page_check_address_pmd(page, mm, address,
Kirill A. Shutemov117b0792013-11-14 14:30:56 -0800688 PAGE_CHECK_ADDRESS_PMD_FLAG, &ptl);
689 if (!pmd)
Joonsoo Kim9f326242014-01-21 15:49:53 -0800690 return SWAP_AGAIN;
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800691
692 if (vma->vm_flags & VM_LOCKED) {
Kirill A. Shutemov117b0792013-11-14 14:30:56 -0800693 spin_unlock(ptl);
Joonsoo Kim9f326242014-01-21 15:49:53 -0800694 pra->vm_flags |= VM_LOCKED;
695 return SWAP_FAIL; /* To break the loop */
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800696 }
697
698 /* go ahead even if the pmd is pmd_trans_splitting() */
699 if (pmdp_clear_flush_young_notify(vma, address, pmd))
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800700 referenced++;
Kirill A. Shutemov117b0792013-11-14 14:30:56 -0800701 spin_unlock(ptl);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800702 } else {
703 pte_t *pte;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800704
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800705 /*
706 * rmap might return false positives; we must filter
707 * these out using page_check_address().
708 */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800709 pte = page_check_address(page, mm, address, &ptl, 0);
710 if (!pte)
Joonsoo Kim9f326242014-01-21 15:49:53 -0800711 return SWAP_AGAIN;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800712
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800713 if (vma->vm_flags & VM_LOCKED) {
714 pte_unmap_unlock(pte, ptl);
Joonsoo Kim9f326242014-01-21 15:49:53 -0800715 pra->vm_flags |= VM_LOCKED;
716 return SWAP_FAIL; /* To break the loop */
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800717 }
718
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800719 if (ptep_clear_flush_young_notify(vma, address, pte)) {
720 /*
721 * Don't treat a reference through a sequentially read
722 * mapping as such. If the page has been used in
723 * another mapping, we will catch it; if this other
724 * mapping is already gone, the unmap path will have
725 * set PG_referenced or activated the page.
726 */
Joe Perches64363aa2013-07-08 16:00:18 -0700727 if (likely(!(vma->vm_flags & VM_SEQ_READ)))
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800728 referenced++;
729 }
730 pte_unmap_unlock(pte, ptl);
731 }
732
Joonsoo Kim9f326242014-01-21 15:49:53 -0800733 if (referenced) {
734 pra->referenced++;
735 pra->vm_flags |= vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 }
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800737
Joonsoo Kim9f326242014-01-21 15:49:53 -0800738 pra->mapcount--;
739 if (!pra->mapcount)
740 return SWAP_SUCCESS; /* To break the loop */
741
742 return SWAP_AGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743}
744
Joonsoo Kim9f326242014-01-21 15:49:53 -0800745static bool invalid_page_referenced_vma(struct vm_area_struct *vma, void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746{
Joonsoo Kim9f326242014-01-21 15:49:53 -0800747 struct page_referenced_arg *pra = arg;
748 struct mem_cgroup *memcg = pra->memcg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749
Joonsoo Kim9f326242014-01-21 15:49:53 -0800750 if (!mm_match_cgroup(vma->vm_mm, memcg))
751 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752
Joonsoo Kim9f326242014-01-21 15:49:53 -0800753 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754}
755
756/**
757 * page_referenced - test if the page was referenced
758 * @page: the page to test
759 * @is_locked: caller holds lock on the page
Johannes Weiner72835c82012-01-12 17:18:32 -0800760 * @memcg: target memory cgroup
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700761 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 *
763 * Quick test_and_clear_referenced for all mappings to a page,
764 * returns the number of ptes which referenced the page.
765 */
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700766int page_referenced(struct page *page,
767 int is_locked,
Johannes Weiner72835c82012-01-12 17:18:32 -0800768 struct mem_cgroup *memcg,
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700769 unsigned long *vm_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770{
Joonsoo Kim9f326242014-01-21 15:49:53 -0800771 int ret;
Hugh Dickins5ad64682009-12-14 17:59:24 -0800772 int we_locked = 0;
Joonsoo Kim9f326242014-01-21 15:49:53 -0800773 struct page_referenced_arg pra = {
774 .mapcount = page_mapcount(page),
775 .memcg = memcg,
776 };
777 struct rmap_walk_control rwc = {
778 .rmap_one = page_referenced_one,
779 .arg = (void *)&pra,
780 .anon_lock = page_lock_anon_vma_read,
781 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700783 *vm_flags = 0;
Joonsoo Kim9f326242014-01-21 15:49:53 -0800784 if (!page_mapped(page))
785 return 0;
786
787 if (!page_rmapping(page))
788 return 0;
789
790 if (!is_locked && (!PageAnon(page) || PageKsm(page))) {
791 we_locked = trylock_page(page);
792 if (!we_locked)
793 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 }
Joonsoo Kim9f326242014-01-21 15:49:53 -0800795
796 /*
797 * If we are reclaiming on behalf of a cgroup, skip
798 * counting on behalf of references from different
799 * cgroups
800 */
801 if (memcg) {
802 rwc.invalid_vma = invalid_page_referenced_vma;
803 }
804
805 ret = rmap_walk(page, &rwc);
806 *vm_flags = pra.vm_flags;
807
808 if (we_locked)
809 unlock_page(page);
810
811 return pra.referenced;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812}
813
Hugh Dickins1cb17292009-12-14 17:59:01 -0800814static int page_mkclean_one(struct page *page, struct vm_area_struct *vma,
Joonsoo Kim9853a402014-01-21 15:49:55 -0800815 unsigned long address, void *arg)
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700816{
817 struct mm_struct *mm = vma->vm_mm;
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100818 pte_t *pte;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700819 spinlock_t *ptl;
820 int ret = 0;
Joonsoo Kim9853a402014-01-21 15:49:55 -0800821 int *cleaned = arg;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700822
Nick Piggin479db0b2008-08-20 14:09:18 -0700823 pte = page_check_address(page, mm, address, &ptl, 1);
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700824 if (!pte)
825 goto out;
826
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100827 if (pte_dirty(*pte) || pte_write(*pte)) {
828 pte_t entry;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700829
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100830 flush_cache_page(vma, address, pte_pfn(*pte));
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700831 entry = ptep_clear_flush(vma, address, pte);
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100832 entry = pte_wrprotect(entry);
833 entry = pte_mkclean(entry);
Al Virod6e88e62006-12-29 16:48:35 -0800834 set_pte_at(mm, address, pte, entry);
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100835 ret = 1;
836 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700837
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700838 pte_unmap_unlock(pte, ptl);
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700839
Joonsoo Kim9853a402014-01-21 15:49:55 -0800840 if (ret) {
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700841 mmu_notifier_invalidate_page(mm, address);
Joonsoo Kim9853a402014-01-21 15:49:55 -0800842 (*cleaned)++;
843 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700844out:
Joonsoo Kim9853a402014-01-21 15:49:55 -0800845 return SWAP_AGAIN;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700846}
847
Joonsoo Kim9853a402014-01-21 15:49:55 -0800848static bool invalid_mkclean_vma(struct vm_area_struct *vma, void *arg)
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700849{
Joonsoo Kim9853a402014-01-21 15:49:55 -0800850 if (vma->vm_flags & VM_SHARED)
Fengguang Wu871beb82014-01-23 15:53:41 -0800851 return false;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700852
Fengguang Wu871beb82014-01-23 15:53:41 -0800853 return true;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700854}
855
856int page_mkclean(struct page *page)
857{
Joonsoo Kim9853a402014-01-21 15:49:55 -0800858 int cleaned = 0;
859 struct address_space *mapping;
860 struct rmap_walk_control rwc = {
861 .arg = (void *)&cleaned,
862 .rmap_one = page_mkclean_one,
863 .invalid_vma = invalid_mkclean_vma,
864 };
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700865
866 BUG_ON(!PageLocked(page));
867
Joonsoo Kim9853a402014-01-21 15:49:55 -0800868 if (!page_mapped(page))
869 return 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700870
Joonsoo Kim9853a402014-01-21 15:49:55 -0800871 mapping = page_mapping(page);
872 if (!mapping)
873 return 0;
874
875 rmap_walk(page, &rwc);
876
877 return cleaned;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700878}
Jaya Kumar60b59be2007-05-08 00:37:37 -0700879EXPORT_SYMBOL_GPL(page_mkclean);
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700880
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881/**
Rik van Rielc44b6742010-03-05 13:42:09 -0800882 * page_move_anon_rmap - move a page to our anon_vma
883 * @page: the page to move to our anon_vma
884 * @vma: the vma the page belongs to
885 * @address: the user virtual address mapped
886 *
887 * When a page belongs exclusively to one process after a COW event,
888 * that page can be moved into the anon_vma that belongs to just that
889 * process, so the rmap code will not search the parent or sibling
890 * processes.
891 */
892void page_move_anon_rmap(struct page *page,
893 struct vm_area_struct *vma, unsigned long address)
894{
895 struct anon_vma *anon_vma = vma->anon_vma;
896
Sasha Levin309381fea2014-01-23 15:52:54 -0800897 VM_BUG_ON_PAGE(!PageLocked(page), page);
Rik van Rielc44b6742010-03-05 13:42:09 -0800898 VM_BUG_ON(!anon_vma);
Sasha Levin309381fea2014-01-23 15:52:54 -0800899 VM_BUG_ON_PAGE(page->index != linear_page_index(vma, address), page);
Rik van Rielc44b6742010-03-05 13:42:09 -0800900
901 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
902 page->mapping = (struct address_space *) anon_vma;
903}
904
905/**
Andi Kleen4e1c1972010-09-22 12:43:56 +0200906 * __page_set_anon_rmap - set up new anonymous rmap
907 * @page: Page to add to rmap
908 * @vma: VM area to add page to.
909 * @address: User virtual address of the mapping
Rik van Riele8a03fe2010-04-14 17:59:28 -0400910 * @exclusive: the page is exclusively owned by the current process
Nick Piggin9617d952006-01-06 00:11:12 -0800911 */
912static void __page_set_anon_rmap(struct page *page,
Rik van Riele8a03fe2010-04-14 17:59:28 -0400913 struct vm_area_struct *vma, unsigned long address, int exclusive)
Nick Piggin9617d952006-01-06 00:11:12 -0800914{
Rik van Riele8a03fe2010-04-14 17:59:28 -0400915 struct anon_vma *anon_vma = vma->anon_vma;
Nick Piggin9617d952006-01-06 00:11:12 -0800916
Rik van Riele8a03fe2010-04-14 17:59:28 -0400917 BUG_ON(!anon_vma);
Linus Torvaldsea900022010-04-12 12:44:29 -0700918
Andi Kleen4e1c1972010-09-22 12:43:56 +0200919 if (PageAnon(page))
920 return;
921
Linus Torvaldsea900022010-04-12 12:44:29 -0700922 /*
Rik van Riele8a03fe2010-04-14 17:59:28 -0400923 * If the page isn't exclusively mapped into this vma,
924 * we must use the _oldest_ possible anon_vma for the
925 * page mapping!
Linus Torvaldsea900022010-04-12 12:44:29 -0700926 */
Andi Kleen4e1c1972010-09-22 12:43:56 +0200927 if (!exclusive)
Andrea Arcangeli288468c2010-08-09 17:19:09 -0700928 anon_vma = anon_vma->root;
Linus Torvaldsea900022010-04-12 12:44:29 -0700929
Nick Piggin9617d952006-01-06 00:11:12 -0800930 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
931 page->mapping = (struct address_space *) anon_vma;
Nick Piggin9617d952006-01-06 00:11:12 -0800932 page->index = linear_page_index(vma, address);
Nick Piggin9617d952006-01-06 00:11:12 -0800933}
934
935/**
Randy Dunlap43d8eac2008-03-19 17:00:43 -0700936 * __page_check_anon_rmap - sanity check anonymous rmap addition
Nick Pigginc97a9e12007-05-16 22:11:21 -0700937 * @page: the page to add the mapping to
938 * @vma: the vm area in which the mapping is added
939 * @address: the user virtual address mapped
940 */
941static void __page_check_anon_rmap(struct page *page,
942 struct vm_area_struct *vma, unsigned long address)
943{
944#ifdef CONFIG_DEBUG_VM
945 /*
946 * The page's anon-rmap details (mapping and index) are guaranteed to
947 * be set up correctly at this point.
948 *
949 * We have exclusion against page_add_anon_rmap because the caller
950 * always holds the page locked, except if called from page_dup_rmap,
951 * in which case the page is already known to be setup.
952 *
953 * We have exclusion against page_add_new_anon_rmap because those pages
954 * are initially only visible via the pagetables, and the pte is locked
955 * over the call to page_add_new_anon_rmap.
956 */
Andrea Arcangeli44ab57a2010-08-09 17:19:10 -0700957 BUG_ON(page_anon_vma(page)->root != vma->anon_vma->root);
Nick Pigginc97a9e12007-05-16 22:11:21 -0700958 BUG_ON(page->index != linear_page_index(vma, address));
959#endif
960}
961
962/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 * page_add_anon_rmap - add pte mapping to an anonymous page
964 * @page: the page to add the mapping to
965 * @vma: the vm area in which the mapping is added
966 * @address: the user virtual address mapped
967 *
Hugh Dickins5ad64682009-12-14 17:59:24 -0800968 * The caller needs to hold the pte lock, and the page must be locked in
Hugh Dickins80e148222009-12-14 17:59:29 -0800969 * the anon_vma case: to serialize mapping,index checking after setting,
970 * and to ensure that PageAnon is not being upgraded racily to PageKsm
971 * (but PageKsm is never downgraded to PageAnon).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 */
973void page_add_anon_rmap(struct page *page,
974 struct vm_area_struct *vma, unsigned long address)
975{
Rik van Rielad8c2ee2010-08-09 17:19:48 -0700976 do_page_add_anon_rmap(page, vma, address, 0);
977}
978
979/*
980 * Special version of the above for do_swap_page, which often runs
981 * into pages that are exclusively owned by the current process.
982 * Everybody else should continue to use page_add_anon_rmap above.
983 */
984void do_page_add_anon_rmap(struct page *page,
985 struct vm_area_struct *vma, unsigned long address, int exclusive)
986{
Hugh Dickins5ad64682009-12-14 17:59:24 -0800987 int first = atomic_inc_and_test(&page->_mapcount);
Andrea Arcangeli79134172011-01-13 15:46:58 -0800988 if (first) {
Kirill A. Shutemov3cd14fc2013-09-12 15:13:58 -0700989 if (PageTransHuge(page))
Andrea Arcangeli79134172011-01-13 15:46:58 -0800990 __inc_zone_page_state(page,
991 NR_ANON_TRANSPARENT_HUGEPAGES);
Kirill A. Shutemov3cd14fc2013-09-12 15:13:58 -0700992 __mod_zone_page_state(page_zone(page), NR_ANON_PAGES,
993 hpage_nr_pages(page));
Andrea Arcangeli79134172011-01-13 15:46:58 -0800994 }
Hugh Dickins5ad64682009-12-14 17:59:24 -0800995 if (unlikely(PageKsm(page)))
996 return;
997
Sasha Levin309381fea2014-01-23 15:52:54 -0800998 VM_BUG_ON_PAGE(!PageLocked(page), page);
Hugh Dickins5dbe0af2011-05-28 13:17:04 -0700999 /* address might be in next vma when migration races vma_adjust */
Hugh Dickins5ad64682009-12-14 17:59:24 -08001000 if (first)
Rik van Rielad8c2ee2010-08-09 17:19:48 -07001001 __page_set_anon_rmap(page, vma, address, exclusive);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001002 else
Nick Pigginc97a9e12007-05-16 22:11:21 -07001003 __page_check_anon_rmap(page, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004}
1005
Randy Dunlap43d8eac2008-03-19 17:00:43 -07001006/**
Nick Piggin9617d952006-01-06 00:11:12 -08001007 * page_add_new_anon_rmap - add pte mapping to a new anonymous page
1008 * @page: the page to add the mapping to
1009 * @vma: the vm area in which the mapping is added
1010 * @address: the user virtual address mapped
1011 *
1012 * Same as page_add_anon_rmap but must only be called on *new* pages.
1013 * This means the inc-and-test can be bypassed.
Nick Pigginc97a9e12007-05-16 22:11:21 -07001014 * Page does not have to be locked.
Nick Piggin9617d952006-01-06 00:11:12 -08001015 */
1016void page_add_new_anon_rmap(struct page *page,
1017 struct vm_area_struct *vma, unsigned long address)
1018{
Hugh Dickinsb5934c52009-01-06 14:39:25 -08001019 VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end);
Hugh Dickinscbf84b72009-01-06 14:39:27 -08001020 SetPageSwapBacked(page);
1021 atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */
Kirill A. Shutemov3cd14fc2013-09-12 15:13:58 -07001022 if (PageTransHuge(page))
Andrea Arcangeli79134172011-01-13 15:46:58 -08001023 __inc_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES);
Kirill A. Shutemov3cd14fc2013-09-12 15:13:58 -07001024 __mod_zone_page_state(page_zone(page), NR_ANON_PAGES,
1025 hpage_nr_pages(page));
Rik van Riele8a03fe2010-04-14 17:59:28 -04001026 __page_set_anon_rmap(page, vma, address, 1);
Mel Gormanc53954a2013-07-03 15:02:34 -07001027 if (!mlocked_vma_newpage(vma, page)) {
1028 SetPageActive(page);
1029 lru_cache_add(page);
1030 } else
Hugh Dickinsb5934c52009-01-06 14:39:25 -08001031 add_page_to_unevictable_list(page);
Nick Piggin9617d952006-01-06 00:11:12 -08001032}
1033
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034/**
1035 * page_add_file_rmap - add pte mapping to a file page
1036 * @page: the page to add the mapping to
1037 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -07001038 * The caller needs to hold the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 */
1040void page_add_file_rmap(struct page *page)
1041{
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001042 bool locked;
1043 unsigned long flags;
1044
1045 mem_cgroup_begin_update_page_stat(page, &locked, &flags);
Balbir Singhd69b0422009-06-17 16:26:34 -07001046 if (atomic_inc_and_test(&page->_mapcount)) {
Christoph Lameter65ba55f2006-06-30 01:55:34 -07001047 __inc_zone_page_state(page, NR_FILE_MAPPED);
Sha Zhengju68b48762013-09-12 15:13:50 -07001048 mem_cgroup_inc_page_stat(page, MEM_CGROUP_STAT_FILE_MAPPED);
Balbir Singhd69b0422009-06-17 16:26:34 -07001049 }
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001050 mem_cgroup_end_update_page_stat(page, &locked, &flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051}
1052
1053/**
1054 * page_remove_rmap - take down pte mapping from a page
1055 * @page: page to remove mapping from
1056 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -07001057 * The caller needs to hold the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 */
Hugh Dickinsedc315f2009-01-06 14:40:11 -08001059void page_remove_rmap(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060{
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001061 bool anon = PageAnon(page);
1062 bool locked;
1063 unsigned long flags;
1064
1065 /*
1066 * The anon case has no mem_cgroup page_stat to update; but may
1067 * uncharge_page() below, where the lock ordering can deadlock if
1068 * we hold the lock against page_stat move: so avoid it on anon.
1069 */
1070 if (!anon)
1071 mem_cgroup_begin_update_page_stat(page, &locked, &flags);
1072
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001073 /* page still mapped by someone else? */
1074 if (!atomic_add_negative(-1, &page->_mapcount))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001075 goto out;
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001076
1077 /*
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001078 * Hugepages are not counted in NR_ANON_PAGES nor NR_FILE_MAPPED
1079 * and not charged by memcg for now.
1080 */
1081 if (unlikely(PageHuge(page)))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001082 goto out;
1083 if (anon) {
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001084 mem_cgroup_uncharge_page(page);
Kirill A. Shutemov3cd14fc2013-09-12 15:13:58 -07001085 if (PageTransHuge(page))
Andrea Arcangeli79134172011-01-13 15:46:58 -08001086 __dec_zone_page_state(page,
1087 NR_ANON_TRANSPARENT_HUGEPAGES);
Kirill A. Shutemov3cd14fc2013-09-12 15:13:58 -07001088 __mod_zone_page_state(page_zone(page), NR_ANON_PAGES,
1089 -hpage_nr_pages(page));
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001090 } else {
1091 __dec_zone_page_state(page, NR_FILE_MAPPED);
Sha Zhengju68b48762013-09-12 15:13:50 -07001092 mem_cgroup_dec_page_stat(page, MEM_CGROUP_STAT_FILE_MAPPED);
Hugh Dickinse6c509f2012-10-08 16:33:19 -07001093 mem_cgroup_end_update_page_stat(page, &locked, &flags);
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001094 }
Hugh Dickinse6c509f2012-10-08 16:33:19 -07001095 if (unlikely(PageMlocked(page)))
1096 clear_page_mlock(page);
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001097 /*
1098 * It would be tidy to reset the PageAnon mapping here,
1099 * but that might overwrite a racing page_add_anon_rmap
1100 * which increments mapcount after us but sets mapping
1101 * before us: so leave the reset to free_hot_cold_page,
1102 * and remember that it's only reliable while mapped.
1103 * Leaving it set also helps swapoff to reinstate ptes
1104 * faster for those pages still in swapcache.
1105 */
Hugh Dickinse6c509f2012-10-08 16:33:19 -07001106 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001107out:
1108 if (!anon)
1109 mem_cgroup_end_update_page_stat(page, &locked, &flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110}
1111
1112/*
Joonsoo Kim52629502014-01-21 15:49:50 -08001113 * @arg: enum ttu_flags will be passed to this argument
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 */
Hugh Dickins5ad64682009-12-14 17:59:24 -08001115int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
Joonsoo Kim52629502014-01-21 15:49:50 -08001116 unsigned long address, void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117{
1118 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 pte_t *pte;
1120 pte_t pteval;
Hugh Dickinsc0718802005-10-29 18:16:31 -07001121 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 int ret = SWAP_AGAIN;
Joonsoo Kim52629502014-01-21 15:49:50 -08001123 enum ttu_flags flags = (enum ttu_flags)arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124
Nick Piggin479db0b2008-08-20 14:09:18 -07001125 pte = page_check_address(page, mm, address, &ptl, 0);
Hugh Dickinsc0718802005-10-29 18:16:31 -07001126 if (!pte)
Nikita Danilov81b40822005-05-01 08:58:36 -07001127 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128
1129 /*
1130 * If the page is mlock()d, we cannot swap it out.
1131 * If it's recently referenced (perhaps page_referenced
1132 * skipped over this mm) then we should reactivate it.
1133 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001134 if (!(flags & TTU_IGNORE_MLOCK)) {
KOSAKI Motohirocaed0f482009-12-14 17:59:45 -08001135 if (vma->vm_flags & VM_LOCKED)
1136 goto out_mlock;
1137
Hugh Dickinsaf8e3352009-12-14 17:58:59 -08001138 if (TTU_ACTION(flags) == TTU_MUNLOCK)
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001139 goto out_unmap;
Andi Kleen14fa31b2009-09-16 11:50:10 +02001140 }
1141 if (!(flags & TTU_IGNORE_ACCESS)) {
Nick Pigginb291f002008-10-18 20:26:44 -07001142 if (ptep_clear_flush_young_notify(vma, address, pte)) {
1143 ret = SWAP_FAIL;
1144 goto out_unmap;
1145 }
1146 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 /* Nuke the page table entry. */
1149 flush_cache_page(vma, address, page_to_pfn(page));
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001150 pteval = ptep_clear_flush(vma, address, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151
1152 /* Move the dirty bit to the physical page now the pte is gone. */
1153 if (pte_dirty(pteval))
1154 set_page_dirty(page);
1155
Hugh Dickins365e9c872005-10-29 18:16:18 -07001156 /* Update high watermark before we lower rss */
1157 update_hiwater_rss(mm);
1158
Andi Kleen888b9f72009-09-16 11:50:11 +02001159 if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) {
Naoya Horiguchi5f24ae52012-12-12 13:52:30 -08001160 if (!PageHuge(page)) {
1161 if (PageAnon(page))
1162 dec_mm_counter(mm, MM_ANONPAGES);
1163 else
1164 dec_mm_counter(mm, MM_FILEPAGES);
1165 }
Andi Kleen888b9f72009-09-16 11:50:11 +02001166 set_pte_at(mm, address, pte,
Naoya Horiguchi5f24ae52012-12-12 13:52:30 -08001167 swp_entry_to_pte(make_hwpoison_entry(page)));
Andi Kleen888b9f72009-09-16 11:50:11 +02001168 } else if (PageAnon(page)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001169 swp_entry_t entry = { .val = page_private(page) };
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001170 pte_t swp_pte;
Christoph Lameter06972122006-06-23 02:03:35 -07001171
1172 if (PageSwapCache(page)) {
1173 /*
1174 * Store the swap location in the pte.
1175 * See handle_pte_fault() ...
1176 */
Hugh Dickins570a335b2009-12-14 17:58:46 -08001177 if (swap_duplicate(entry) < 0) {
1178 set_pte_at(mm, address, pte, pteval);
1179 ret = SWAP_FAIL;
1180 goto out_unmap;
1181 }
Christoph Lameter06972122006-06-23 02:03:35 -07001182 if (list_empty(&mm->mmlist)) {
1183 spin_lock(&mmlist_lock);
1184 if (list_empty(&mm->mmlist))
1185 list_add(&mm->mmlist, &init_mm.mmlist);
1186 spin_unlock(&mmlist_lock);
1187 }
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001188 dec_mm_counter(mm, MM_ANONPAGES);
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08001189 inc_mm_counter(mm, MM_SWAPENTS);
Konstantin Khlebnikovce1744f2012-03-21 16:33:59 -07001190 } else if (IS_ENABLED(CONFIG_MIGRATION)) {
Christoph Lameter06972122006-06-23 02:03:35 -07001191 /*
1192 * Store the pfn of the page in a special migration
1193 * pte. do_swap_page() will wait until the migration
1194 * pte is removed and then restart fault handling.
1195 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001196 BUG_ON(TTU_ACTION(flags) != TTU_MIGRATION);
Christoph Lameter06972122006-06-23 02:03:35 -07001197 entry = make_migration_entry(page, pte_write(pteval));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 }
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001199 swp_pte = swp_entry_to_pte(entry);
1200 if (pte_soft_dirty(pteval))
1201 swp_pte = pte_swp_mksoft_dirty(swp_pte);
1202 set_pte_at(mm, address, pte, swp_pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 BUG_ON(pte_file(*pte));
Konstantin Khlebnikovce1744f2012-03-21 16:33:59 -07001204 } else if (IS_ENABLED(CONFIG_MIGRATION) &&
1205 (TTU_ACTION(flags) == TTU_MIGRATION)) {
Christoph Lameter04e62a22006-06-23 02:03:38 -07001206 /* Establish migration entry for a file page */
1207 swp_entry_t entry;
1208 entry = make_migration_entry(page, pte_write(pteval));
1209 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
1210 } else
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001211 dec_mm_counter(mm, MM_FILEPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212
Hugh Dickinsedc315f2009-01-06 14:40:11 -08001213 page_remove_rmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 page_cache_release(page);
1215
1216out_unmap:
Hugh Dickinsc0718802005-10-29 18:16:31 -07001217 pte_unmap_unlock(pte, ptl);
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001218 if (ret != SWAP_FAIL)
1219 mmu_notifier_invalidate_page(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220out:
1221 return ret;
KOSAKI Motohirocaed0f482009-12-14 17:59:45 -08001222
1223out_mlock:
1224 pte_unmap_unlock(pte, ptl);
1225
1226
1227 /*
1228 * We need mmap_sem locking, Otherwise VM_LOCKED check makes
1229 * unstable result and race. Plus, We can't wait here because
Ingo Molnar5a505082012-12-02 19:56:46 +00001230 * we now hold anon_vma->rwsem or mapping->i_mmap_mutex.
KOSAKI Motohirocaed0f482009-12-14 17:59:45 -08001231 * if trylock failed, the page remain in evictable lru and later
1232 * vmscan could retry to move the page to unevictable lru if the
1233 * page is actually mlocked.
1234 */
1235 if (down_read_trylock(&vma->vm_mm->mmap_sem)) {
1236 if (vma->vm_flags & VM_LOCKED) {
1237 mlock_vma_page(page);
1238 ret = SWAP_MLOCK;
1239 }
1240 up_read(&vma->vm_mm->mmap_sem);
1241 }
1242 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243}
1244
1245/*
1246 * objrmap doesn't work for nonlinear VMAs because the assumption that
1247 * offset-into-file correlates with offset-into-virtual-addresses does not hold.
1248 * Consequently, given a particular page and its ->index, we cannot locate the
1249 * ptes which are mapping that page without an exhaustive linear search.
1250 *
1251 * So what this code does is a mini "virtual scan" of each nonlinear VMA which
1252 * maps the file to which the target page belongs. The ->vm_private_data field
1253 * holds the current cursor into that scan. Successive searches will circulate
1254 * around the vma's virtual address space.
1255 *
1256 * So as more replacement pressure is applied to the pages in a nonlinear VMA,
1257 * more scanning pressure is placed against them as well. Eventually pages
1258 * will become fully unmapped and are eligible for eviction.
1259 *
1260 * For very sparsely populated VMAs this is a little inefficient - chances are
1261 * there there won't be many ptes located within the scan cluster. In this case
1262 * maybe we could scan further - to the end of the pte page, perhaps.
Nick Pigginb291f002008-10-18 20:26:44 -07001263 *
1264 * Mlocked pages: check VM_LOCKED under mmap_sem held for read, if we can
1265 * acquire it without blocking. If vma locked, mlock the pages in the cluster,
1266 * rather than unmapping them. If we encounter the "check_page" that vmscan is
1267 * trying to unmap, return SWAP_MLOCK, else default SWAP_AGAIN.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 */
1269#define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE)
1270#define CLUSTER_MASK (~(CLUSTER_SIZE - 1))
1271
Nick Pigginb291f002008-10-18 20:26:44 -07001272static int try_to_unmap_cluster(unsigned long cursor, unsigned int *mapcount,
1273 struct vm_area_struct *vma, struct page *check_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274{
1275 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 pmd_t *pmd;
Hugh Dickinsc0718802005-10-29 18:16:31 -07001277 pte_t *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 pte_t pteval;
Hugh Dickinsc0718802005-10-29 18:16:31 -07001279 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 struct page *page;
1281 unsigned long address;
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001282 unsigned long mmun_start; /* For mmu_notifiers */
1283 unsigned long mmun_end; /* For mmu_notifiers */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 unsigned long end;
Nick Pigginb291f002008-10-18 20:26:44 -07001285 int ret = SWAP_AGAIN;
1286 int locked_vma = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 address = (vma->vm_start + cursor) & CLUSTER_MASK;
1289 end = address + CLUSTER_SIZE;
1290 if (address < vma->vm_start)
1291 address = vma->vm_start;
1292 if (end > vma->vm_end)
1293 end = vma->vm_end;
1294
Bob Liu62190492012-12-11 16:00:37 -08001295 pmd = mm_find_pmd(mm, address);
1296 if (!pmd)
Nick Pigginb291f002008-10-18 20:26:44 -07001297 return ret;
1298
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001299 mmun_start = address;
1300 mmun_end = end;
1301 mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
1302
Nick Pigginb291f002008-10-18 20:26:44 -07001303 /*
Hugh Dickinsaf8e3352009-12-14 17:58:59 -08001304 * If we can acquire the mmap_sem for read, and vma is VM_LOCKED,
Nick Pigginb291f002008-10-18 20:26:44 -07001305 * keep the sem while scanning the cluster for mlocking pages.
1306 */
Hugh Dickinsaf8e3352009-12-14 17:58:59 -08001307 if (down_read_trylock(&vma->vm_mm->mmap_sem)) {
Nick Pigginb291f002008-10-18 20:26:44 -07001308 locked_vma = (vma->vm_flags & VM_LOCKED);
1309 if (!locked_vma)
1310 up_read(&vma->vm_mm->mmap_sem); /* don't need it */
1311 }
Hugh Dickinsc0718802005-10-29 18:16:31 -07001312
1313 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314
Hugh Dickins365e9c872005-10-29 18:16:18 -07001315 /* Update high watermark before we lower rss */
1316 update_hiwater_rss(mm);
1317
Hugh Dickinsc0718802005-10-29 18:16:31 -07001318 for (; address < end; pte++, address += PAGE_SIZE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 if (!pte_present(*pte))
1320 continue;
Linus Torvalds6aab3412005-11-28 14:34:23 -08001321 page = vm_normal_page(vma, address, *pte);
1322 BUG_ON(!page || PageAnon(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323
Nick Pigginb291f002008-10-18 20:26:44 -07001324 if (locked_vma) {
1325 mlock_vma_page(page); /* no-op if already mlocked */
1326 if (page == check_page)
1327 ret = SWAP_MLOCK;
1328 continue; /* don't unmap */
1329 }
1330
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001331 if (ptep_clear_flush_young_notify(vma, address, pte))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 continue;
1333
1334 /* Nuke the page table entry. */
Ben Collinseca35132005-11-29 11:45:26 -08001335 flush_cache_page(vma, address, pte_pfn(*pte));
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001336 pteval = ptep_clear_flush(vma, address, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337
1338 /* If nonlinear, store the file page offset in the pte. */
Cyrill Gorcunov41bb3472013-08-13 16:00:51 -07001339 if (page->index != linear_page_index(vma, address)) {
1340 pte_t ptfile = pgoff_to_pte(page->index);
1341 if (pte_soft_dirty(pteval))
1342 pte_file_mksoft_dirty(ptfile);
1343 set_pte_at(mm, address, pte, ptfile);
1344 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345
1346 /* Move the dirty bit to the physical page now the pte is gone. */
1347 if (pte_dirty(pteval))
1348 set_page_dirty(page);
1349
Hugh Dickinsedc315f2009-01-06 14:40:11 -08001350 page_remove_rmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 page_cache_release(page);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001352 dec_mm_counter(mm, MM_FILEPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 (*mapcount)--;
1354 }
Hugh Dickinsc0718802005-10-29 18:16:31 -07001355 pte_unmap_unlock(pte - 1, ptl);
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001356 mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
Nick Pigginb291f002008-10-18 20:26:44 -07001357 if (locked_vma)
1358 up_read(&vma->vm_mm->mmap_sem);
1359 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360}
1361
Joonsoo Kim0f843c62014-01-21 15:49:45 -08001362static int try_to_unmap_nonlinear(struct page *page,
Hugh Dickins7e09e732014-03-20 21:52:17 -07001363 struct address_space *mapping, void *arg)
Joonsoo Kim0f843c62014-01-21 15:49:45 -08001364{
Hugh Dickins7e09e732014-03-20 21:52:17 -07001365 struct vm_area_struct *vma;
Joonsoo Kim0f843c62014-01-21 15:49:45 -08001366 int ret = SWAP_AGAIN;
1367 unsigned long cursor;
1368 unsigned long max_nl_cursor = 0;
1369 unsigned long max_nl_size = 0;
1370 unsigned int mapcount;
1371
1372 list_for_each_entry(vma,
1373 &mapping->i_mmap_nonlinear, shared.nonlinear) {
1374
1375 cursor = (unsigned long) vma->vm_private_data;
1376 if (cursor > max_nl_cursor)
1377 max_nl_cursor = cursor;
1378 cursor = vma->vm_end - vma->vm_start;
1379 if (cursor > max_nl_size)
1380 max_nl_size = cursor;
1381 }
1382
1383 if (max_nl_size == 0) { /* all nonlinears locked or reserved ? */
1384 return SWAP_FAIL;
1385 }
1386
1387 /*
1388 * We don't try to search for this page in the nonlinear vmas,
1389 * and page_referenced wouldn't have found it anyway. Instead
1390 * just walk the nonlinear vmas trying to age and unmap some.
1391 * The mapcount of the page we came in with is irrelevant,
1392 * but even so use it as a guide to how hard we should try?
1393 */
1394 mapcount = page_mapcount(page);
1395 if (!mapcount)
1396 return ret;
1397
1398 cond_resched();
1399
1400 max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK;
1401 if (max_nl_cursor == 0)
1402 max_nl_cursor = CLUSTER_SIZE;
1403
1404 do {
1405 list_for_each_entry(vma,
1406 &mapping->i_mmap_nonlinear, shared.nonlinear) {
1407
1408 cursor = (unsigned long) vma->vm_private_data;
1409 while (cursor < max_nl_cursor &&
1410 cursor < vma->vm_end - vma->vm_start) {
1411 if (try_to_unmap_cluster(cursor, &mapcount,
1412 vma, page) == SWAP_MLOCK)
1413 ret = SWAP_MLOCK;
1414 cursor += CLUSTER_SIZE;
1415 vma->vm_private_data = (void *) cursor;
1416 if ((int)mapcount <= 0)
1417 return ret;
1418 }
1419 vma->vm_private_data = (void *) max_nl_cursor;
1420 }
1421 cond_resched();
1422 max_nl_cursor += CLUSTER_SIZE;
1423 } while (max_nl_cursor <= max_nl_size);
1424
1425 /*
1426 * Don't loop forever (perhaps all the remaining pages are
1427 * in locked vmas). Reset cursor on all unreserved nonlinear
1428 * vmas, now forgetting on which ones it had fallen behind.
1429 */
1430 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.nonlinear)
1431 vma->vm_private_data = NULL;
1432
1433 return ret;
1434}
1435
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001436bool is_vma_temporary_stack(struct vm_area_struct *vma)
Mel Gormana8bef8ff2010-05-24 14:32:24 -07001437{
1438 int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP);
1439
1440 if (!maybe_stack)
1441 return false;
1442
1443 if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) ==
1444 VM_STACK_INCOMPLETE_SETUP)
1445 return true;
1446
1447 return false;
1448}
1449
Joonsoo Kim52629502014-01-21 15:49:50 -08001450static bool invalid_migration_vma(struct vm_area_struct *vma, void *arg)
1451{
1452 return is_vma_temporary_stack(vma);
1453}
1454
Joonsoo Kim52629502014-01-21 15:49:50 -08001455static int page_not_mapped(struct page *page)
1456{
1457 return !page_mapped(page);
1458};
1459
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460/**
1461 * try_to_unmap - try to remove all page table mappings to a page
1462 * @page: the page to get unmapped
Andi Kleen14fa31b2009-09-16 11:50:10 +02001463 * @flags: action and flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 *
1465 * Tries to remove all the page table entries which are mapping this
1466 * page, used in the pageout path. Caller must hold the page lock.
1467 * Return values are:
1468 *
1469 * SWAP_SUCCESS - we succeeded in removing all mappings
1470 * SWAP_AGAIN - we missed a mapping, try again later
1471 * SWAP_FAIL - the page is unswappable
Nick Pigginb291f002008-10-18 20:26:44 -07001472 * SWAP_MLOCK - page is mlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001474int try_to_unmap(struct page *page, enum ttu_flags flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475{
1476 int ret;
Joonsoo Kim52629502014-01-21 15:49:50 -08001477 struct rmap_walk_control rwc = {
1478 .rmap_one = try_to_unmap_one,
1479 .arg = (void *)flags,
1480 .done = page_not_mapped,
1481 .file_nonlinear = try_to_unmap_nonlinear,
1482 .anon_lock = page_lock_anon_vma_read,
1483 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484
Sasha Levin309381fea2014-01-23 15:52:54 -08001485 VM_BUG_ON_PAGE(!PageHuge(page) && PageTransHuge(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486
Joonsoo Kim52629502014-01-21 15:49:50 -08001487 /*
1488 * During exec, a temporary VMA is setup and later moved.
1489 * The VMA is moved under the anon_vma lock but not the
1490 * page tables leading to a race where migration cannot
1491 * find the migration ptes. Rather than increasing the
1492 * locking requirements of exec(), migration skips
1493 * temporary VMAs until after exec() completes.
1494 */
1495 if (flags & TTU_MIGRATION && !PageKsm(page) && PageAnon(page))
1496 rwc.invalid_vma = invalid_migration_vma;
1497
1498 ret = rmap_walk(page, &rwc);
1499
Nick Pigginb291f002008-10-18 20:26:44 -07001500 if (ret != SWAP_MLOCK && !page_mapped(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 ret = SWAP_SUCCESS;
1502 return ret;
1503}
Nikita Danilov81b40822005-05-01 08:58:36 -07001504
Nick Pigginb291f002008-10-18 20:26:44 -07001505/**
1506 * try_to_munlock - try to munlock a page
1507 * @page: the page to be munlocked
1508 *
1509 * Called from munlock code. Checks all of the VMAs mapping the page
1510 * to make sure nobody else has this page mlocked. The page will be
1511 * returned with PG_mlocked cleared if no other vmas have it mlocked.
1512 *
1513 * Return values are:
1514 *
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001515 * SWAP_AGAIN - no vma is holding page mlocked, or,
Nick Pigginb291f002008-10-18 20:26:44 -07001516 * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem
Hugh Dickins5ad64682009-12-14 17:59:24 -08001517 * SWAP_FAIL - page cannot be located at present
Nick Pigginb291f002008-10-18 20:26:44 -07001518 * SWAP_MLOCK - page is now mlocked.
1519 */
1520int try_to_munlock(struct page *page)
1521{
Joonsoo Kime8351ac2014-01-21 15:49:52 -08001522 int ret;
1523 struct rmap_walk_control rwc = {
1524 .rmap_one = try_to_unmap_one,
1525 .arg = (void *)TTU_MUNLOCK,
1526 .done = page_not_mapped,
1527 /*
1528 * We don't bother to try to find the munlocked page in
1529 * nonlinears. It's costly. Instead, later, page reclaim logic
1530 * may call try_to_unmap() and recover PG_mlocked lazily.
1531 */
1532 .file_nonlinear = NULL,
1533 .anon_lock = page_lock_anon_vma_read,
1534
1535 };
1536
Sasha Levin309381fea2014-01-23 15:52:54 -08001537 VM_BUG_ON_PAGE(!PageLocked(page) || PageLRU(page), page);
Nick Pigginb291f002008-10-18 20:26:44 -07001538
Joonsoo Kime8351ac2014-01-21 15:49:52 -08001539 ret = rmap_walk(page, &rwc);
1540 return ret;
Nick Pigginb291f002008-10-18 20:26:44 -07001541}
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001542
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001543void __put_anon_vma(struct anon_vma *anon_vma)
Rik van Riel76545062010-08-09 17:18:41 -07001544{
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001545 struct anon_vma *root = anon_vma->root;
Rik van Riel76545062010-08-09 17:18:41 -07001546
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001547 if (root != anon_vma && atomic_dec_and_test(&root->refcount))
1548 anon_vma_free(root);
Rik van Riel76545062010-08-09 17:18:41 -07001549
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001550 anon_vma_free(anon_vma);
Rik van Riel76545062010-08-09 17:18:41 -07001551}
Rik van Riel76545062010-08-09 17:18:41 -07001552
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001553static struct anon_vma *rmap_walk_anon_lock(struct page *page,
1554 struct rmap_walk_control *rwc)
Joonsoo Kimfaecd8d2014-01-21 15:49:46 -08001555{
1556 struct anon_vma *anon_vma;
1557
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001558 if (rwc->anon_lock)
1559 return rwc->anon_lock(page);
1560
Joonsoo Kimfaecd8d2014-01-21 15:49:46 -08001561 /*
1562 * Note: remove_migration_ptes() cannot use page_lock_anon_vma_read()
1563 * because that depends on page_mapped(); but not all its usages
1564 * are holding mmap_sem. Users without mmap_sem are required to
1565 * take a reference count to prevent the anon_vma disappearing
1566 */
1567 anon_vma = page_anon_vma(page);
1568 if (!anon_vma)
1569 return NULL;
1570
1571 anon_vma_lock_read(anon_vma);
1572 return anon_vma;
1573}
1574
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001575/*
Joonsoo Kime8351ac2014-01-21 15:49:52 -08001576 * rmap_walk_anon - do something to anonymous page using the object-based
1577 * rmap method
1578 * @page: the page to be handled
1579 * @rwc: control variable according to each walk type
1580 *
1581 * Find all the mappings of a page using the mapping pointer and the vma chains
1582 * contained in the anon_vma struct it points to.
1583 *
1584 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1585 * where the page was found will be held for write. So, we won't recheck
1586 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1587 * LOCKED.
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001588 */
Joonsoo Kim051ac832014-01-21 15:49:48 -08001589static int rmap_walk_anon(struct page *page, struct rmap_walk_control *rwc)
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001590{
1591 struct anon_vma *anon_vma;
Michel Lespinassebf181b92012-10-08 16:31:39 -07001592 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
Rik van Riel5beb4932010-03-05 13:42:07 -08001593 struct anon_vma_chain *avc;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001594 int ret = SWAP_AGAIN;
1595
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001596 anon_vma = rmap_walk_anon_lock(page, rwc);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001597 if (!anon_vma)
1598 return ret;
Joonsoo Kimfaecd8d2014-01-21 15:49:46 -08001599
Michel Lespinassebf181b92012-10-08 16:31:39 -07001600 anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) {
Rik van Riel5beb4932010-03-05 13:42:07 -08001601 struct vm_area_struct *vma = avc->vma;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001602 unsigned long address = vma_address(page, vma);
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001603
1604 if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg))
1605 continue;
1606
Joonsoo Kim051ac832014-01-21 15:49:48 -08001607 ret = rwc->rmap_one(page, vma, address, rwc->arg);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001608 if (ret != SWAP_AGAIN)
1609 break;
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001610 if (rwc->done && rwc->done(page))
1611 break;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001612 }
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +00001613 anon_vma_unlock_read(anon_vma);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001614 return ret;
1615}
1616
Joonsoo Kime8351ac2014-01-21 15:49:52 -08001617/*
1618 * rmap_walk_file - do something to file page using the object-based rmap method
1619 * @page: the page to be handled
1620 * @rwc: control variable according to each walk type
1621 *
1622 * Find all the mappings of a page using the mapping pointer and the vma chains
1623 * contained in the address_space struct it points to.
1624 *
1625 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1626 * where the page was found will be held for write. So, we won't recheck
1627 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1628 * LOCKED.
1629 */
Joonsoo Kim051ac832014-01-21 15:49:48 -08001630static int rmap_walk_file(struct page *page, struct rmap_walk_control *rwc)
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001631{
1632 struct address_space *mapping = page->mapping;
Joonsoo Kimb854f712014-01-21 15:49:43 -08001633 pgoff_t pgoff = page->index << compound_order(page);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001634 struct vm_area_struct *vma;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001635 int ret = SWAP_AGAIN;
1636
Joonsoo Kim9f326242014-01-21 15:49:53 -08001637 /*
1638 * The page lock not only makes sure that page->mapping cannot
1639 * suddenly be NULLified by truncation, it makes sure that the
1640 * structure at mapping cannot be freed and reused yet,
1641 * so we can safely take mapping->i_mmap_mutex.
1642 */
1643 VM_BUG_ON(!PageLocked(page));
1644
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001645 if (!mapping)
1646 return ret;
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07001647 mutex_lock(&mapping->i_mmap_mutex);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07001648 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
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)
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001656 goto done;
1657 if (rwc->done && rwc->done(page))
1658 goto done;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001659 }
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001660
1661 if (!rwc->file_nonlinear)
1662 goto done;
1663
1664 if (list_empty(&mapping->i_mmap_nonlinear))
1665 goto done;
1666
Hugh Dickins7e09e732014-03-20 21:52:17 -07001667 ret = rwc->file_nonlinear(page, mapping, rwc->arg);
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001668
1669done:
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07001670 mutex_unlock(&mapping->i_mmap_mutex);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001671 return ret;
1672}
1673
Joonsoo Kim051ac832014-01-21 15:49:48 -08001674int rmap_walk(struct page *page, struct rmap_walk_control *rwc)
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001675{
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001676 if (unlikely(PageKsm(page)))
Joonsoo Kim051ac832014-01-21 15:49:48 -08001677 return rmap_walk_ksm(page, rwc);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001678 else if (PageAnon(page))
Joonsoo Kim051ac832014-01-21 15:49:48 -08001679 return rmap_walk_anon(page, rwc);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001680 else
Joonsoo Kim051ac832014-01-21 15:49:48 -08001681 return rmap_walk_file(page, rwc);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001682}
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001683
Naoya Horiguchie3390f62010-06-15 13:18:13 +09001684#ifdef CONFIG_HUGETLB_PAGE
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001685/*
1686 * The following three functions are for anonymous (private mapped) hugepages.
1687 * Unlike common anonymous pages, anonymous hugepages have no accounting code
1688 * and no lru code, because we handle hugepages differently from common pages.
1689 */
1690static void __hugepage_set_anon_rmap(struct page *page,
1691 struct vm_area_struct *vma, unsigned long address, int exclusive)
1692{
1693 struct anon_vma *anon_vma = vma->anon_vma;
Naoya Horiguchi433abed2010-09-10 13:23:03 +09001694
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001695 BUG_ON(!anon_vma);
Naoya Horiguchi433abed2010-09-10 13:23:03 +09001696
1697 if (PageAnon(page))
1698 return;
1699 if (!exclusive)
1700 anon_vma = anon_vma->root;
1701
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001702 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
1703 page->mapping = (struct address_space *) anon_vma;
1704 page->index = linear_page_index(vma, address);
1705}
1706
1707void hugepage_add_anon_rmap(struct page *page,
1708 struct vm_area_struct *vma, unsigned long address)
1709{
1710 struct anon_vma *anon_vma = vma->anon_vma;
1711 int first;
Naoya Horiguchia850ea32010-09-10 13:23:06 +09001712
1713 BUG_ON(!PageLocked(page));
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001714 BUG_ON(!anon_vma);
Hugh Dickins5dbe0af2011-05-28 13:17:04 -07001715 /* address might be in next vma when migration races vma_adjust */
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001716 first = atomic_inc_and_test(&page->_mapcount);
1717 if (first)
1718 __hugepage_set_anon_rmap(page, vma, address, 0);
1719}
1720
1721void hugepage_add_new_anon_rmap(struct page *page,
1722 struct vm_area_struct *vma, unsigned long address)
1723{
1724 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
1725 atomic_set(&page->_mapcount, 0);
1726 __hugepage_set_anon_rmap(page, vma, address, 1);
1727}
Naoya Horiguchie3390f62010-06-15 13:18:13 +09001728#endif /* CONFIG_HUGETLB_PAGE */