blob: 807c96bf0dc6d08cdc167ce432a68999b1e21692 [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))) {
603 pte = huge_pte_offset(mm, address);
604 ptl = &mm->page_table_lock;
605 goto check;
606 }
607
Bob Liu62190492012-12-11 16:00:37 -0800608 pmd = mm_find_pmd(mm, address);
609 if (!pmd)
Hugh Dickinsc0718802005-10-29 18:16:31 -0700610 return NULL;
611
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800612 if (pmd_trans_huge(*pmd))
613 return NULL;
Hugh Dickinsc0718802005-10-29 18:16:31 -0700614
615 pte = pte_offset_map(pmd, address);
616 /* Make a quick check before getting the lock */
Nick Piggin479db0b2008-08-20 14:09:18 -0700617 if (!sync && !pte_present(*pte)) {
Hugh Dickinsc0718802005-10-29 18:16:31 -0700618 pte_unmap(pte);
619 return NULL;
Nikita Danilov81b40822005-05-01 08:58:36 -0700620 }
Hugh Dickinsc0718802005-10-29 18:16:31 -0700621
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700622 ptl = pte_lockptr(mm, pmd);
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +0900623check:
Hugh Dickinsc0718802005-10-29 18:16:31 -0700624 spin_lock(ptl);
625 if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) {
626 *ptlp = ptl;
627 return pte;
628 }
629 pte_unmap_unlock(pte, ptl);
630 return NULL;
Nikita Danilov81b40822005-05-01 08:58:36 -0700631}
632
Nick Pigginb291f002008-10-18 20:26:44 -0700633/**
634 * page_mapped_in_vma - check whether a page is really mapped in a VMA
635 * @page: the page to test
636 * @vma: the VMA to test
637 *
638 * Returns 1 if the page is mapped into the page tables of the VMA, 0
639 * if the page is not mapped into the page tables of this VMA. Only
640 * valid for normal file or anonymous VMAs.
641 */
Andi Kleen6a460792009-09-16 11:50:15 +0200642int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma)
Nick Pigginb291f002008-10-18 20:26:44 -0700643{
644 unsigned long address;
645 pte_t *pte;
646 spinlock_t *ptl;
647
Michel Lespinasse86c2ad12012-10-08 16:31:42 -0700648 address = __vma_address(page, vma);
649 if (unlikely(address < vma->vm_start || address >= vma->vm_end))
Nick Pigginb291f002008-10-18 20:26:44 -0700650 return 0;
651 pte = page_check_address(page, vma->vm_mm, address, &ptl, 1);
652 if (!pte) /* the page is not in this mm */
653 return 0;
654 pte_unmap_unlock(pte, ptl);
655
656 return 1;
657}
658
Nikita Danilov81b40822005-05-01 08:58:36 -0700659/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 * Subfunctions of page_referenced: page_referenced_one called
661 * repeatedly from either page_referenced_anon or page_referenced_file.
662 */
Hugh Dickins5ad64682009-12-14 17:59:24 -0800663int page_referenced_one(struct page *page, struct vm_area_struct *vma,
664 unsigned long address, unsigned int *mapcount,
665 unsigned long *vm_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666{
667 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 int referenced = 0;
669
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800670 if (unlikely(PageTransHuge(page))) {
671 pmd_t *pmd;
672
673 spin_lock(&mm->page_table_lock);
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800674 /*
675 * rmap might return false positives; we must filter
676 * these out using page_check_address_pmd().
677 */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800678 pmd = page_check_address_pmd(page, mm, address,
679 PAGE_CHECK_ADDRESS_PMD_FLAG);
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800680 if (!pmd) {
681 spin_unlock(&mm->page_table_lock);
682 goto out;
683 }
684
685 if (vma->vm_flags & VM_LOCKED) {
686 spin_unlock(&mm->page_table_lock);
687 *mapcount = 0; /* break early from loop */
688 *vm_flags |= VM_LOCKED;
689 goto out;
690 }
691
692 /* go ahead even if the pmd is pmd_trans_splitting() */
693 if (pmdp_clear_flush_young_notify(vma, address, pmd))
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800694 referenced++;
695 spin_unlock(&mm->page_table_lock);
696 } else {
697 pte_t *pte;
698 spinlock_t *ptl;
699
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800700 /*
701 * rmap might return false positives; we must filter
702 * these out using page_check_address().
703 */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800704 pte = page_check_address(page, mm, address, &ptl, 0);
705 if (!pte)
706 goto out;
707
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800708 if (vma->vm_flags & VM_LOCKED) {
709 pte_unmap_unlock(pte, ptl);
710 *mapcount = 0; /* break early from loop */
711 *vm_flags |= VM_LOCKED;
712 goto out;
713 }
714
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800715 if (ptep_clear_flush_young_notify(vma, address, pte)) {
716 /*
717 * Don't treat a reference through a sequentially read
718 * mapping as such. If the page has been used in
719 * another mapping, we will catch it; if this other
720 * mapping is already gone, the unmap path will have
721 * set PG_referenced or activated the page.
722 */
723 if (likely(!VM_SequentialReadHint(vma)))
724 referenced++;
725 }
726 pte_unmap_unlock(pte, ptl);
727 }
728
Hugh Dickinsc0718802005-10-29 18:16:31 -0700729 (*mapcount)--;
Huang Shijie273f0472009-12-14 17:58:51 -0800730
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700731 if (referenced)
732 *vm_flags |= vma->vm_flags;
Huang Shijie273f0472009-12-14 17:58:51 -0800733out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 return referenced;
735}
736
Balbir Singhbed71612008-02-07 00:14:01 -0800737static int page_referenced_anon(struct page *page,
Johannes Weiner72835c82012-01-12 17:18:32 -0800738 struct mem_cgroup *memcg,
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700739 unsigned long *vm_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740{
741 unsigned int mapcount;
742 struct anon_vma *anon_vma;
Michel Lespinassebf181b92012-10-08 16:31:39 -0700743 pgoff_t pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -0800744 struct anon_vma_chain *avc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 int referenced = 0;
746
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000747 anon_vma = page_lock_anon_vma_read(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 if (!anon_vma)
749 return referenced;
750
751 mapcount = page_mapcount(page);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700752 pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
753 anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) {
Rik van Riel5beb4932010-03-05 13:42:07 -0800754 struct vm_area_struct *vma = avc->vma;
Hugh Dickins1cb17292009-12-14 17:59:01 -0800755 unsigned long address = vma_address(page, vma);
Balbir Singhbed71612008-02-07 00:14:01 -0800756 /*
757 * If we are reclaiming on behalf of a cgroup, skip
758 * counting on behalf of references from different
759 * cgroups
760 */
Johannes Weiner72835c82012-01-12 17:18:32 -0800761 if (memcg && !mm_match_cgroup(vma->vm_mm, memcg))
Balbir Singhbed71612008-02-07 00:14:01 -0800762 continue;
Hugh Dickins1cb17292009-12-14 17:59:01 -0800763 referenced += page_referenced_one(page, vma, address,
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700764 &mapcount, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 if (!mapcount)
766 break;
767 }
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800768
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000769 page_unlock_anon_vma_read(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 return referenced;
771}
772
773/**
774 * page_referenced_file - referenced check for object-based rmap
775 * @page: the page we're checking references on.
Johannes Weiner72835c82012-01-12 17:18:32 -0800776 * @memcg: target memory control group
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700777 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 *
779 * For an object-based mapped page, find all the places it is mapped and
780 * check/clear the referenced flag. This is done by following the page->mapping
781 * pointer, then walking the chain of vmas it holds. It returns the number
782 * of references it found.
783 *
784 * This function is only called from page_referenced for object-based pages.
785 */
Balbir Singhbed71612008-02-07 00:14:01 -0800786static int page_referenced_file(struct page *page,
Johannes Weiner72835c82012-01-12 17:18:32 -0800787 struct mem_cgroup *memcg,
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700788 unsigned long *vm_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789{
790 unsigned int mapcount;
791 struct address_space *mapping = page->mapping;
792 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
793 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 int referenced = 0;
795
796 /*
797 * The caller's checks on page->mapping and !PageAnon have made
798 * sure that this is a file page: the check for page->mapping
799 * excludes the case just before it gets set on an anon page.
800 */
801 BUG_ON(PageAnon(page));
802
803 /*
804 * The page lock not only makes sure that page->mapping cannot
805 * suddenly be NULLified by truncation, it makes sure that the
806 * structure at mapping cannot be freed and reused yet,
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700807 * so we can safely take mapping->i_mmap_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 */
809 BUG_ON(!PageLocked(page));
810
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700811 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812
813 /*
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700814 * i_mmap_mutex does not stabilize mapcount at all, but mapcount
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 * is more likely to be accurate if we note it after spinning.
816 */
817 mapcount = page_mapcount(page);
818
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700819 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
Hugh Dickins1cb17292009-12-14 17:59:01 -0800820 unsigned long address = vma_address(page, vma);
Balbir Singhbed71612008-02-07 00:14:01 -0800821 /*
822 * If we are reclaiming on behalf of a cgroup, skip
823 * counting on behalf of references from different
824 * cgroups
825 */
Johannes Weiner72835c82012-01-12 17:18:32 -0800826 if (memcg && !mm_match_cgroup(vma->vm_mm, memcg))
Balbir Singhbed71612008-02-07 00:14:01 -0800827 continue;
Hugh Dickins1cb17292009-12-14 17:59:01 -0800828 referenced += page_referenced_one(page, vma, address,
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700829 &mapcount, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 if (!mapcount)
831 break;
832 }
833
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700834 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 return referenced;
836}
837
838/**
839 * page_referenced - test if the page was referenced
840 * @page: the page to test
841 * @is_locked: caller holds lock on the page
Johannes Weiner72835c82012-01-12 17:18:32 -0800842 * @memcg: target memory cgroup
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700843 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 *
845 * Quick test_and_clear_referenced for all mappings to a page,
846 * returns the number of ptes which referenced the page.
847 */
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700848int page_referenced(struct page *page,
849 int is_locked,
Johannes Weiner72835c82012-01-12 17:18:32 -0800850 struct mem_cgroup *memcg,
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700851 unsigned long *vm_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852{
853 int referenced = 0;
Hugh Dickins5ad64682009-12-14 17:59:24 -0800854 int we_locked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700856 *vm_flags = 0;
Hugh Dickins3ca7b3c2009-12-14 17:58:57 -0800857 if (page_mapped(page) && page_rmapping(page)) {
Hugh Dickins5ad64682009-12-14 17:59:24 -0800858 if (!is_locked && (!PageAnon(page) || PageKsm(page))) {
859 we_locked = trylock_page(page);
860 if (!we_locked) {
861 referenced++;
862 goto out;
863 }
864 }
865 if (unlikely(PageKsm(page)))
Johannes Weiner72835c82012-01-12 17:18:32 -0800866 referenced += page_referenced_ksm(page, memcg,
Hugh Dickins5ad64682009-12-14 17:59:24 -0800867 vm_flags);
868 else if (PageAnon(page))
Johannes Weiner72835c82012-01-12 17:18:32 -0800869 referenced += page_referenced_anon(page, memcg,
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700870 vm_flags);
Hugh Dickins5ad64682009-12-14 17:59:24 -0800871 else if (page->mapping)
Johannes Weiner72835c82012-01-12 17:18:32 -0800872 referenced += page_referenced_file(page, memcg,
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700873 vm_flags);
Hugh Dickins5ad64682009-12-14 17:59:24 -0800874 if (we_locked)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 unlock_page(page);
Martin Schwidefsky50a15982011-07-24 10:47:59 +0200876
877 if (page_test_and_clear_young(page_to_pfn(page)))
878 referenced++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 }
Hugh Dickins5ad64682009-12-14 17:59:24 -0800880out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 return referenced;
882}
883
Hugh Dickins1cb17292009-12-14 17:59:01 -0800884static int page_mkclean_one(struct page *page, struct vm_area_struct *vma,
885 unsigned long address)
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700886{
887 struct mm_struct *mm = vma->vm_mm;
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100888 pte_t *pte;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700889 spinlock_t *ptl;
890 int ret = 0;
891
Nick Piggin479db0b2008-08-20 14:09:18 -0700892 pte = page_check_address(page, mm, address, &ptl, 1);
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700893 if (!pte)
894 goto out;
895
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100896 if (pte_dirty(*pte) || pte_write(*pte)) {
897 pte_t entry;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700898
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100899 flush_cache_page(vma, address, pte_pfn(*pte));
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700900 entry = ptep_clear_flush(vma, address, pte);
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100901 entry = pte_wrprotect(entry);
902 entry = pte_mkclean(entry);
Al Virod6e88e62006-12-29 16:48:35 -0800903 set_pte_at(mm, address, pte, entry);
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100904 ret = 1;
905 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700906
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700907 pte_unmap_unlock(pte, ptl);
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700908
909 if (ret)
910 mmu_notifier_invalidate_page(mm, address);
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700911out:
912 return ret;
913}
914
915static int page_mkclean_file(struct address_space *mapping, struct page *page)
916{
917 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
918 struct vm_area_struct *vma;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700919 int ret = 0;
920
921 BUG_ON(PageAnon(page));
922
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700923 mutex_lock(&mapping->i_mmap_mutex);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700924 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
Hugh Dickins1cb17292009-12-14 17:59:01 -0800925 if (vma->vm_flags & VM_SHARED) {
926 unsigned long address = vma_address(page, vma);
Hugh Dickins1cb17292009-12-14 17:59:01 -0800927 ret += page_mkclean_one(page, vma, address);
928 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700929 }
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700930 mutex_unlock(&mapping->i_mmap_mutex);
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700931 return ret;
932}
933
934int page_mkclean(struct page *page)
935{
936 int ret = 0;
937
938 BUG_ON(!PageLocked(page));
939
940 if (page_mapped(page)) {
941 struct address_space *mapping = page_mapping(page);
Jan Karaef5d4372012-10-25 13:37:31 -0700942 if (mapping)
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700943 ret = page_mkclean_file(mapping, page);
944 }
945
946 return ret;
947}
Jaya Kumar60b59be2007-05-08 00:37:37 -0700948EXPORT_SYMBOL_GPL(page_mkclean);
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700949
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950/**
Rik van Rielc44b6742010-03-05 13:42:09 -0800951 * page_move_anon_rmap - move a page to our anon_vma
952 * @page: the page to move to our anon_vma
953 * @vma: the vma the page belongs to
954 * @address: the user virtual address mapped
955 *
956 * When a page belongs exclusively to one process after a COW event,
957 * that page can be moved into the anon_vma that belongs to just that
958 * process, so the rmap code will not search the parent or sibling
959 * processes.
960 */
961void page_move_anon_rmap(struct page *page,
962 struct vm_area_struct *vma, unsigned long address)
963{
964 struct anon_vma *anon_vma = vma->anon_vma;
965
966 VM_BUG_ON(!PageLocked(page));
967 VM_BUG_ON(!anon_vma);
968 VM_BUG_ON(page->index != linear_page_index(vma, address));
969
970 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
971 page->mapping = (struct address_space *) anon_vma;
972}
973
974/**
Andi Kleen4e1c1972010-09-22 12:43:56 +0200975 * __page_set_anon_rmap - set up new anonymous rmap
976 * @page: Page to add to rmap
977 * @vma: VM area to add page to.
978 * @address: User virtual address of the mapping
Rik van Riele8a03fe2010-04-14 17:59:28 -0400979 * @exclusive: the page is exclusively owned by the current process
Nick Piggin9617d952006-01-06 00:11:12 -0800980 */
981static void __page_set_anon_rmap(struct page *page,
Rik van Riele8a03fe2010-04-14 17:59:28 -0400982 struct vm_area_struct *vma, unsigned long address, int exclusive)
Nick Piggin9617d952006-01-06 00:11:12 -0800983{
Rik van Riele8a03fe2010-04-14 17:59:28 -0400984 struct anon_vma *anon_vma = vma->anon_vma;
Nick Piggin9617d952006-01-06 00:11:12 -0800985
Rik van Riele8a03fe2010-04-14 17:59:28 -0400986 BUG_ON(!anon_vma);
Linus Torvaldsea900022010-04-12 12:44:29 -0700987
Andi Kleen4e1c1972010-09-22 12:43:56 +0200988 if (PageAnon(page))
989 return;
990
Linus Torvaldsea900022010-04-12 12:44:29 -0700991 /*
Rik van Riele8a03fe2010-04-14 17:59:28 -0400992 * If the page isn't exclusively mapped into this vma,
993 * we must use the _oldest_ possible anon_vma for the
994 * page mapping!
Linus Torvaldsea900022010-04-12 12:44:29 -0700995 */
Andi Kleen4e1c1972010-09-22 12:43:56 +0200996 if (!exclusive)
Andrea Arcangeli288468c2010-08-09 17:19:09 -0700997 anon_vma = anon_vma->root;
Linus Torvaldsea900022010-04-12 12:44:29 -0700998
Nick Piggin9617d952006-01-06 00:11:12 -0800999 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
1000 page->mapping = (struct address_space *) anon_vma;
Nick Piggin9617d952006-01-06 00:11:12 -08001001 page->index = linear_page_index(vma, address);
Nick Piggin9617d952006-01-06 00:11:12 -08001002}
1003
1004/**
Randy Dunlap43d8eac2008-03-19 17:00:43 -07001005 * __page_check_anon_rmap - sanity check anonymous rmap addition
Nick Pigginc97a9e12007-05-16 22:11:21 -07001006 * @page: the page to add the mapping to
1007 * @vma: the vm area in which the mapping is added
1008 * @address: the user virtual address mapped
1009 */
1010static void __page_check_anon_rmap(struct page *page,
1011 struct vm_area_struct *vma, unsigned long address)
1012{
1013#ifdef CONFIG_DEBUG_VM
1014 /*
1015 * The page's anon-rmap details (mapping and index) are guaranteed to
1016 * be set up correctly at this point.
1017 *
1018 * We have exclusion against page_add_anon_rmap because the caller
1019 * always holds the page locked, except if called from page_dup_rmap,
1020 * in which case the page is already known to be setup.
1021 *
1022 * We have exclusion against page_add_new_anon_rmap because those pages
1023 * are initially only visible via the pagetables, and the pte is locked
1024 * over the call to page_add_new_anon_rmap.
1025 */
Andrea Arcangeli44ab57a2010-08-09 17:19:10 -07001026 BUG_ON(page_anon_vma(page)->root != vma->anon_vma->root);
Nick Pigginc97a9e12007-05-16 22:11:21 -07001027 BUG_ON(page->index != linear_page_index(vma, address));
1028#endif
1029}
1030
1031/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 * page_add_anon_rmap - add pte mapping to an anonymous page
1033 * @page: the page to add the mapping to
1034 * @vma: the vm area in which the mapping is added
1035 * @address: the user virtual address mapped
1036 *
Hugh Dickins5ad64682009-12-14 17:59:24 -08001037 * The caller needs to hold the pte lock, and the page must be locked in
Hugh Dickins80e148222009-12-14 17:59:29 -08001038 * the anon_vma case: to serialize mapping,index checking after setting,
1039 * and to ensure that PageAnon is not being upgraded racily to PageKsm
1040 * (but PageKsm is never downgraded to PageAnon).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 */
1042void page_add_anon_rmap(struct page *page,
1043 struct vm_area_struct *vma, unsigned long address)
1044{
Rik van Rielad8c2ee2010-08-09 17:19:48 -07001045 do_page_add_anon_rmap(page, vma, address, 0);
1046}
1047
1048/*
1049 * Special version of the above for do_swap_page, which often runs
1050 * into pages that are exclusively owned by the current process.
1051 * Everybody else should continue to use page_add_anon_rmap above.
1052 */
1053void do_page_add_anon_rmap(struct page *page,
1054 struct vm_area_struct *vma, unsigned long address, int exclusive)
1055{
Hugh Dickins5ad64682009-12-14 17:59:24 -08001056 int first = atomic_inc_and_test(&page->_mapcount);
Andrea Arcangeli79134172011-01-13 15:46:58 -08001057 if (first) {
1058 if (!PageTransHuge(page))
1059 __inc_zone_page_state(page, NR_ANON_PAGES);
1060 else
1061 __inc_zone_page_state(page,
1062 NR_ANON_TRANSPARENT_HUGEPAGES);
1063 }
Hugh Dickins5ad64682009-12-14 17:59:24 -08001064 if (unlikely(PageKsm(page)))
1065 return;
1066
Nick Pigginc97a9e12007-05-16 22:11:21 -07001067 VM_BUG_ON(!PageLocked(page));
Hugh Dickins5dbe0af2011-05-28 13:17:04 -07001068 /* address might be in next vma when migration races vma_adjust */
Hugh Dickins5ad64682009-12-14 17:59:24 -08001069 if (first)
Rik van Rielad8c2ee2010-08-09 17:19:48 -07001070 __page_set_anon_rmap(page, vma, address, exclusive);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001071 else
Nick Pigginc97a9e12007-05-16 22:11:21 -07001072 __page_check_anon_rmap(page, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073}
1074
Randy Dunlap43d8eac2008-03-19 17:00:43 -07001075/**
Nick Piggin9617d952006-01-06 00:11:12 -08001076 * page_add_new_anon_rmap - add pte mapping to a new anonymous page
1077 * @page: the page to add the mapping to
1078 * @vma: the vm area in which the mapping is added
1079 * @address: the user virtual address mapped
1080 *
1081 * Same as page_add_anon_rmap but must only be called on *new* pages.
1082 * This means the inc-and-test can be bypassed.
Nick Pigginc97a9e12007-05-16 22:11:21 -07001083 * Page does not have to be locked.
Nick Piggin9617d952006-01-06 00:11:12 -08001084 */
1085void page_add_new_anon_rmap(struct page *page,
1086 struct vm_area_struct *vma, unsigned long address)
1087{
Hugh Dickinsb5934c52009-01-06 14:39:25 -08001088 VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end);
Hugh Dickinscbf84b72009-01-06 14:39:27 -08001089 SetPageSwapBacked(page);
1090 atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */
Andrea Arcangeli79134172011-01-13 15:46:58 -08001091 if (!PageTransHuge(page))
1092 __inc_zone_page_state(page, NR_ANON_PAGES);
1093 else
1094 __inc_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES);
Rik van Riele8a03fe2010-04-14 17:59:28 -04001095 __page_set_anon_rmap(page, vma, address, 1);
Hugh Dickins39b5f292012-10-08 16:33:18 -07001096 if (!mlocked_vma_newpage(vma, page))
Hugh Dickinscbf84b72009-01-06 14:39:27 -08001097 lru_cache_add_lru(page, LRU_ACTIVE_ANON);
Hugh Dickinsb5934c52009-01-06 14:39:25 -08001098 else
1099 add_page_to_unevictable_list(page);
Nick Piggin9617d952006-01-06 00:11:12 -08001100}
1101
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102/**
1103 * page_add_file_rmap - add pte mapping to a file page
1104 * @page: the page to add the mapping to
1105 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -07001106 * The caller needs to hold the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 */
1108void page_add_file_rmap(struct page *page)
1109{
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001110 bool locked;
1111 unsigned long flags;
1112
1113 mem_cgroup_begin_update_page_stat(page, &locked, &flags);
Balbir Singhd69b0422009-06-17 16:26:34 -07001114 if (atomic_inc_and_test(&page->_mapcount)) {
Christoph Lameter65ba55f2006-06-30 01:55:34 -07001115 __inc_zone_page_state(page, NR_FILE_MAPPED);
Greg Thelen2a7106f2011-01-13 15:47:37 -08001116 mem_cgroup_inc_page_stat(page, MEMCG_NR_FILE_MAPPED);
Balbir Singhd69b0422009-06-17 16:26:34 -07001117 }
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001118 mem_cgroup_end_update_page_stat(page, &locked, &flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119}
1120
1121/**
1122 * page_remove_rmap - take down pte mapping from a page
1123 * @page: page to remove mapping from
1124 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -07001125 * The caller needs to hold the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 */
Hugh Dickinsedc315f2009-01-06 14:40:11 -08001127void page_remove_rmap(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128{
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001129 bool anon = PageAnon(page);
1130 bool locked;
1131 unsigned long flags;
1132
1133 /*
1134 * The anon case has no mem_cgroup page_stat to update; but may
1135 * uncharge_page() below, where the lock ordering can deadlock if
1136 * we hold the lock against page_stat move: so avoid it on anon.
1137 */
1138 if (!anon)
1139 mem_cgroup_begin_update_page_stat(page, &locked, &flags);
1140
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001141 /* page still mapped by someone else? */
1142 if (!atomic_add_negative(-1, &page->_mapcount))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001143 goto out;
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001144
1145 /*
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001146 * Hugepages are not counted in NR_ANON_PAGES nor NR_FILE_MAPPED
1147 * and not charged by memcg for now.
1148 */
1149 if (unlikely(PageHuge(page)))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001150 goto out;
1151 if (anon) {
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001152 mem_cgroup_uncharge_page(page);
Andrea Arcangeli79134172011-01-13 15:46:58 -08001153 if (!PageTransHuge(page))
1154 __dec_zone_page_state(page, NR_ANON_PAGES);
1155 else
1156 __dec_zone_page_state(page,
1157 NR_ANON_TRANSPARENT_HUGEPAGES);
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001158 } else {
1159 __dec_zone_page_state(page, NR_FILE_MAPPED);
Greg Thelen2a7106f2011-01-13 15:47:37 -08001160 mem_cgroup_dec_page_stat(page, MEMCG_NR_FILE_MAPPED);
Hugh Dickinse6c509f2012-10-08 16:33:19 -07001161 mem_cgroup_end_update_page_stat(page, &locked, &flags);
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001162 }
Hugh Dickinse6c509f2012-10-08 16:33:19 -07001163 if (unlikely(PageMlocked(page)))
1164 clear_page_mlock(page);
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001165 /*
1166 * It would be tidy to reset the PageAnon mapping here,
1167 * but that might overwrite a racing page_add_anon_rmap
1168 * which increments mapcount after us but sets mapping
1169 * before us: so leave the reset to free_hot_cold_page,
1170 * and remember that it's only reliable while mapped.
1171 * Leaving it set also helps swapoff to reinstate ptes
1172 * faster for those pages still in swapcache.
1173 */
Hugh Dickinse6c509f2012-10-08 16:33:19 -07001174 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001175out:
1176 if (!anon)
1177 mem_cgroup_end_update_page_stat(page, &locked, &flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178}
1179
1180/*
1181 * Subfunctions of try_to_unmap: try_to_unmap_one called
Wanlong Gao99ef0312011-10-31 17:08:51 -07001182 * repeatedly from try_to_unmap_ksm, try_to_unmap_anon or try_to_unmap_file.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 */
Hugh Dickins5ad64682009-12-14 17:59:24 -08001184int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
1185 unsigned long address, enum ttu_flags flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186{
1187 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 pte_t *pte;
1189 pte_t pteval;
Hugh Dickinsc0718802005-10-29 18:16:31 -07001190 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 int ret = SWAP_AGAIN;
1192
Nick Piggin479db0b2008-08-20 14:09:18 -07001193 pte = page_check_address(page, mm, address, &ptl, 0);
Hugh Dickinsc0718802005-10-29 18:16:31 -07001194 if (!pte)
Nikita Danilov81b40822005-05-01 08:58:36 -07001195 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196
1197 /*
1198 * If the page is mlock()d, we cannot swap it out.
1199 * If it's recently referenced (perhaps page_referenced
1200 * skipped over this mm) then we should reactivate it.
1201 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001202 if (!(flags & TTU_IGNORE_MLOCK)) {
KOSAKI Motohirocaed0f482009-12-14 17:59:45 -08001203 if (vma->vm_flags & VM_LOCKED)
1204 goto out_mlock;
1205
Hugh Dickinsaf8e3352009-12-14 17:58:59 -08001206 if (TTU_ACTION(flags) == TTU_MUNLOCK)
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001207 goto out_unmap;
Andi Kleen14fa31b2009-09-16 11:50:10 +02001208 }
1209 if (!(flags & TTU_IGNORE_ACCESS)) {
Nick Pigginb291f002008-10-18 20:26:44 -07001210 if (ptep_clear_flush_young_notify(vma, address, pte)) {
1211 ret = SWAP_FAIL;
1212 goto out_unmap;
1213 }
1214 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 /* Nuke the page table entry. */
1217 flush_cache_page(vma, address, page_to_pfn(page));
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001218 pteval = ptep_clear_flush(vma, address, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219
1220 /* Move the dirty bit to the physical page now the pte is gone. */
1221 if (pte_dirty(pteval))
1222 set_page_dirty(page);
1223
Hugh Dickins365e9c872005-10-29 18:16:18 -07001224 /* Update high watermark before we lower rss */
1225 update_hiwater_rss(mm);
1226
Andi Kleen888b9f72009-09-16 11:50:11 +02001227 if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) {
Naoya Horiguchi5f24ae52012-12-12 13:52:30 -08001228 if (!PageHuge(page)) {
1229 if (PageAnon(page))
1230 dec_mm_counter(mm, MM_ANONPAGES);
1231 else
1232 dec_mm_counter(mm, MM_FILEPAGES);
1233 }
Andi Kleen888b9f72009-09-16 11:50:11 +02001234 set_pte_at(mm, address, pte,
Naoya Horiguchi5f24ae52012-12-12 13:52:30 -08001235 swp_entry_to_pte(make_hwpoison_entry(page)));
Andi Kleen888b9f72009-09-16 11:50:11 +02001236 } else if (PageAnon(page)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001237 swp_entry_t entry = { .val = page_private(page) };
Christoph Lameter06972122006-06-23 02:03:35 -07001238
1239 if (PageSwapCache(page)) {
1240 /*
1241 * Store the swap location in the pte.
1242 * See handle_pte_fault() ...
1243 */
Hugh Dickins570a335b2009-12-14 17:58:46 -08001244 if (swap_duplicate(entry) < 0) {
1245 set_pte_at(mm, address, pte, pteval);
1246 ret = SWAP_FAIL;
1247 goto out_unmap;
1248 }
Christoph Lameter06972122006-06-23 02:03:35 -07001249 if (list_empty(&mm->mmlist)) {
1250 spin_lock(&mmlist_lock);
1251 if (list_empty(&mm->mmlist))
1252 list_add(&mm->mmlist, &init_mm.mmlist);
1253 spin_unlock(&mmlist_lock);
1254 }
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001255 dec_mm_counter(mm, MM_ANONPAGES);
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08001256 inc_mm_counter(mm, MM_SWAPENTS);
Konstantin Khlebnikovce1744f2012-03-21 16:33:59 -07001257 } else if (IS_ENABLED(CONFIG_MIGRATION)) {
Christoph Lameter06972122006-06-23 02:03:35 -07001258 /*
1259 * Store the pfn of the page in a special migration
1260 * pte. do_swap_page() will wait until the migration
1261 * pte is removed and then restart fault handling.
1262 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001263 BUG_ON(TTU_ACTION(flags) != TTU_MIGRATION);
Christoph Lameter06972122006-06-23 02:03:35 -07001264 entry = make_migration_entry(page, pte_write(pteval));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 }
1266 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
1267 BUG_ON(pte_file(*pte));
Konstantin Khlebnikovce1744f2012-03-21 16:33:59 -07001268 } else if (IS_ENABLED(CONFIG_MIGRATION) &&
1269 (TTU_ACTION(flags) == TTU_MIGRATION)) {
Christoph Lameter04e62a22006-06-23 02:03:38 -07001270 /* Establish migration entry for a file page */
1271 swp_entry_t entry;
1272 entry = make_migration_entry(page, pte_write(pteval));
1273 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
1274 } else
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001275 dec_mm_counter(mm, MM_FILEPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276
Hugh Dickinsedc315f2009-01-06 14:40:11 -08001277 page_remove_rmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 page_cache_release(page);
1279
1280out_unmap:
Hugh Dickinsc0718802005-10-29 18:16:31 -07001281 pte_unmap_unlock(pte, ptl);
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001282 if (ret != SWAP_FAIL)
1283 mmu_notifier_invalidate_page(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284out:
1285 return ret;
KOSAKI Motohirocaed0f482009-12-14 17:59:45 -08001286
1287out_mlock:
1288 pte_unmap_unlock(pte, ptl);
1289
1290
1291 /*
1292 * We need mmap_sem locking, Otherwise VM_LOCKED check makes
1293 * unstable result and race. Plus, We can't wait here because
Ingo Molnar5a505082012-12-02 19:56:46 +00001294 * we now hold anon_vma->rwsem or mapping->i_mmap_mutex.
KOSAKI Motohirocaed0f482009-12-14 17:59:45 -08001295 * if trylock failed, the page remain in evictable lru and later
1296 * vmscan could retry to move the page to unevictable lru if the
1297 * page is actually mlocked.
1298 */
1299 if (down_read_trylock(&vma->vm_mm->mmap_sem)) {
1300 if (vma->vm_flags & VM_LOCKED) {
1301 mlock_vma_page(page);
1302 ret = SWAP_MLOCK;
1303 }
1304 up_read(&vma->vm_mm->mmap_sem);
1305 }
1306 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307}
1308
1309/*
1310 * objrmap doesn't work for nonlinear VMAs because the assumption that
1311 * offset-into-file correlates with offset-into-virtual-addresses does not hold.
1312 * Consequently, given a particular page and its ->index, we cannot locate the
1313 * ptes which are mapping that page without an exhaustive linear search.
1314 *
1315 * So what this code does is a mini "virtual scan" of each nonlinear VMA which
1316 * maps the file to which the target page belongs. The ->vm_private_data field
1317 * holds the current cursor into that scan. Successive searches will circulate
1318 * around the vma's virtual address space.
1319 *
1320 * So as more replacement pressure is applied to the pages in a nonlinear VMA,
1321 * more scanning pressure is placed against them as well. Eventually pages
1322 * will become fully unmapped and are eligible for eviction.
1323 *
1324 * For very sparsely populated VMAs this is a little inefficient - chances are
1325 * there there won't be many ptes located within the scan cluster. In this case
1326 * maybe we could scan further - to the end of the pte page, perhaps.
Nick Pigginb291f002008-10-18 20:26:44 -07001327 *
1328 * Mlocked pages: check VM_LOCKED under mmap_sem held for read, if we can
1329 * acquire it without blocking. If vma locked, mlock the pages in the cluster,
1330 * rather than unmapping them. If we encounter the "check_page" that vmscan is
1331 * trying to unmap, return SWAP_MLOCK, else default SWAP_AGAIN.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 */
1333#define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE)
1334#define CLUSTER_MASK (~(CLUSTER_SIZE - 1))
1335
Nick Pigginb291f002008-10-18 20:26:44 -07001336static int try_to_unmap_cluster(unsigned long cursor, unsigned int *mapcount,
1337 struct vm_area_struct *vma, struct page *check_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338{
1339 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 pmd_t *pmd;
Hugh Dickinsc0718802005-10-29 18:16:31 -07001341 pte_t *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 pte_t pteval;
Hugh Dickinsc0718802005-10-29 18:16:31 -07001343 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 struct page *page;
1345 unsigned long address;
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001346 unsigned long mmun_start; /* For mmu_notifiers */
1347 unsigned long mmun_end; /* For mmu_notifiers */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 unsigned long end;
Nick Pigginb291f002008-10-18 20:26:44 -07001349 int ret = SWAP_AGAIN;
1350 int locked_vma = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 address = (vma->vm_start + cursor) & CLUSTER_MASK;
1353 end = address + CLUSTER_SIZE;
1354 if (address < vma->vm_start)
1355 address = vma->vm_start;
1356 if (end > vma->vm_end)
1357 end = vma->vm_end;
1358
Bob Liu62190492012-12-11 16:00:37 -08001359 pmd = mm_find_pmd(mm, address);
1360 if (!pmd)
Nick Pigginb291f002008-10-18 20:26:44 -07001361 return ret;
1362
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001363 mmun_start = address;
1364 mmun_end = end;
1365 mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
1366
Nick Pigginb291f002008-10-18 20:26:44 -07001367 /*
Hugh Dickinsaf8e3352009-12-14 17:58:59 -08001368 * If we can acquire the mmap_sem for read, and vma is VM_LOCKED,
Nick Pigginb291f002008-10-18 20:26:44 -07001369 * keep the sem while scanning the cluster for mlocking pages.
1370 */
Hugh Dickinsaf8e3352009-12-14 17:58:59 -08001371 if (down_read_trylock(&vma->vm_mm->mmap_sem)) {
Nick Pigginb291f002008-10-18 20:26:44 -07001372 locked_vma = (vma->vm_flags & VM_LOCKED);
1373 if (!locked_vma)
1374 up_read(&vma->vm_mm->mmap_sem); /* don't need it */
1375 }
Hugh Dickinsc0718802005-10-29 18:16:31 -07001376
1377 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378
Hugh Dickins365e9c872005-10-29 18:16:18 -07001379 /* Update high watermark before we lower rss */
1380 update_hiwater_rss(mm);
1381
Hugh Dickinsc0718802005-10-29 18:16:31 -07001382 for (; address < end; pte++, address += PAGE_SIZE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 if (!pte_present(*pte))
1384 continue;
Linus Torvalds6aab3412005-11-28 14:34:23 -08001385 page = vm_normal_page(vma, address, *pte);
1386 BUG_ON(!page || PageAnon(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387
Nick Pigginb291f002008-10-18 20:26:44 -07001388 if (locked_vma) {
1389 mlock_vma_page(page); /* no-op if already mlocked */
1390 if (page == check_page)
1391 ret = SWAP_MLOCK;
1392 continue; /* don't unmap */
1393 }
1394
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001395 if (ptep_clear_flush_young_notify(vma, address, pte))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 continue;
1397
1398 /* Nuke the page table entry. */
Ben Collinseca35132005-11-29 11:45:26 -08001399 flush_cache_page(vma, address, pte_pfn(*pte));
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001400 pteval = ptep_clear_flush(vma, address, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401
1402 /* If nonlinear, store the file page offset in the pte. */
1403 if (page->index != linear_page_index(vma, address))
1404 set_pte_at(mm, address, pte, pgoff_to_pte(page->index));
1405
1406 /* Move the dirty bit to the physical page now the pte is gone. */
1407 if (pte_dirty(pteval))
1408 set_page_dirty(page);
1409
Hugh Dickinsedc315f2009-01-06 14:40:11 -08001410 page_remove_rmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 page_cache_release(page);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001412 dec_mm_counter(mm, MM_FILEPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 (*mapcount)--;
1414 }
Hugh Dickinsc0718802005-10-29 18:16:31 -07001415 pte_unmap_unlock(pte - 1, ptl);
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001416 mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
Nick Pigginb291f002008-10-18 20:26:44 -07001417 if (locked_vma)
1418 up_read(&vma->vm_mm->mmap_sem);
1419 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420}
1421
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001422bool is_vma_temporary_stack(struct vm_area_struct *vma)
Mel Gormana8bef8ff2010-05-24 14:32:24 -07001423{
1424 int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP);
1425
1426 if (!maybe_stack)
1427 return false;
1428
1429 if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) ==
1430 VM_STACK_INCOMPLETE_SETUP)
1431 return true;
1432
1433 return false;
1434}
1435
Nick Pigginb291f002008-10-18 20:26:44 -07001436/**
1437 * try_to_unmap_anon - unmap or unlock anonymous page using the object-based
1438 * rmap method
1439 * @page: the page to unmap/unlock
Huang Shijie8051be52009-12-14 17:58:50 -08001440 * @flags: action and flags
Nick Pigginb291f002008-10-18 20:26:44 -07001441 *
1442 * Find all the mappings of a page using the mapping pointer and the vma chains
1443 * contained in the anon_vma struct it points to.
1444 *
1445 * This function is only called from try_to_unmap/try_to_munlock for
1446 * anonymous pages.
1447 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1448 * where the page was found will be held for write. So, we won't recheck
1449 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1450 * 'LOCKED.
1451 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001452static int try_to_unmap_anon(struct page *page, enum ttu_flags flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453{
1454 struct anon_vma *anon_vma;
Michel Lespinassebf181b92012-10-08 16:31:39 -07001455 pgoff_t pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001456 struct anon_vma_chain *avc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 int ret = SWAP_AGAIN;
Nick Pigginb291f002008-10-18 20:26:44 -07001458
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +00001459 anon_vma = page_lock_anon_vma_read(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 if (!anon_vma)
1461 return ret;
1462
Michel Lespinassebf181b92012-10-08 16:31:39 -07001463 pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
1464 anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) {
Rik van Riel5beb4932010-03-05 13:42:07 -08001465 struct vm_area_struct *vma = avc->vma;
Mel Gormana8bef8ff2010-05-24 14:32:24 -07001466 unsigned long address;
1467
1468 /*
1469 * During exec, a temporary VMA is setup and later moved.
1470 * The VMA is moved under the anon_vma lock but not the
1471 * page tables leading to a race where migration cannot
1472 * find the migration ptes. Rather than increasing the
1473 * locking requirements of exec(), migration skips
1474 * temporary VMAs until after exec() completes.
1475 */
Konstantin Khlebnikovce1744f2012-03-21 16:33:59 -07001476 if (IS_ENABLED(CONFIG_MIGRATION) && (flags & TTU_MIGRATION) &&
Mel Gormana8bef8ff2010-05-24 14:32:24 -07001477 is_vma_temporary_stack(vma))
1478 continue;
1479
1480 address = vma_address(page, vma);
Hugh Dickins1cb17292009-12-14 17:59:01 -08001481 ret = try_to_unmap_one(page, vma, address, flags);
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001482 if (ret != SWAP_AGAIN || !page_mapped(page))
1483 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 }
Oleg Nesterov34bbd702007-02-28 20:13:49 -08001485
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +00001486 page_unlock_anon_vma_read(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 return ret;
1488}
1489
1490/**
Nick Pigginb291f002008-10-18 20:26:44 -07001491 * try_to_unmap_file - unmap/unlock file page using the object-based rmap method
1492 * @page: the page to unmap/unlock
Andi Kleen14fa31b2009-09-16 11:50:10 +02001493 * @flags: action and flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 *
1495 * Find all the mappings of a page using the mapping pointer and the vma chains
1496 * contained in the address_space struct it points to.
1497 *
Nick Pigginb291f002008-10-18 20:26:44 -07001498 * This function is only called from try_to_unmap/try_to_munlock for
1499 * object-based pages.
1500 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1501 * where the page was found will be held for write. So, we won't recheck
1502 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1503 * 'LOCKED.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001505static int try_to_unmap_file(struct page *page, enum ttu_flags flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506{
1507 struct address_space *mapping = page->mapping;
1508 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
1509 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 int ret = SWAP_AGAIN;
1511 unsigned long cursor;
1512 unsigned long max_nl_cursor = 0;
1513 unsigned long max_nl_size = 0;
1514 unsigned int mapcount;
1515
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07001516 mutex_lock(&mapping->i_mmap_mutex);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07001517 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
Hugh Dickins1cb17292009-12-14 17:59:01 -08001518 unsigned long address = vma_address(page, vma);
Hugh Dickins1cb17292009-12-14 17:59:01 -08001519 ret = try_to_unmap_one(page, vma, address, flags);
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001520 if (ret != SWAP_AGAIN || !page_mapped(page))
1521 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 }
1523
1524 if (list_empty(&mapping->i_mmap_nonlinear))
1525 goto out;
1526
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001527 /*
1528 * We don't bother to try to find the munlocked page in nonlinears.
1529 * It's costly. Instead, later, page reclaim logic may call
1530 * try_to_unmap(TTU_MUNLOCK) and recover PG_mlocked lazily.
1531 */
1532 if (TTU_ACTION(flags) == TTU_MUNLOCK)
1533 goto out;
1534
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07001536 shared.nonlinear) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537 cursor = (unsigned long) vma->vm_private_data;
1538 if (cursor > max_nl_cursor)
1539 max_nl_cursor = cursor;
1540 cursor = vma->vm_end - vma->vm_start;
1541 if (cursor > max_nl_size)
1542 max_nl_size = cursor;
1543 }
1544
Nick Pigginb291f002008-10-18 20:26:44 -07001545 if (max_nl_size == 0) { /* all nonlinears locked or reserved ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 ret = SWAP_FAIL;
1547 goto out;
1548 }
1549
1550 /*
1551 * We don't try to search for this page in the nonlinear vmas,
1552 * and page_referenced wouldn't have found it anyway. Instead
1553 * just walk the nonlinear vmas trying to age and unmap some.
1554 * The mapcount of the page we came in with is irrelevant,
1555 * but even so use it as a guide to how hard we should try?
1556 */
1557 mapcount = page_mapcount(page);
1558 if (!mapcount)
1559 goto out;
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07001560 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561
1562 max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK;
1563 if (max_nl_cursor == 0)
1564 max_nl_cursor = CLUSTER_SIZE;
1565
1566 do {
1567 list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07001568 shared.nonlinear) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 cursor = (unsigned long) vma->vm_private_data;
Hugh Dickins839b9682005-09-03 15:54:43 -07001570 while ( cursor < max_nl_cursor &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 cursor < vma->vm_end - vma->vm_start) {
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001572 if (try_to_unmap_cluster(cursor, &mapcount,
1573 vma, page) == SWAP_MLOCK)
1574 ret = SWAP_MLOCK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 cursor += CLUSTER_SIZE;
1576 vma->vm_private_data = (void *) cursor;
1577 if ((int)mapcount <= 0)
1578 goto out;
1579 }
1580 vma->vm_private_data = (void *) max_nl_cursor;
1581 }
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07001582 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 max_nl_cursor += CLUSTER_SIZE;
1584 } while (max_nl_cursor <= max_nl_size);
1585
1586 /*
1587 * Don't loop forever (perhaps all the remaining pages are
1588 * in locked vmas). Reset cursor on all unreserved nonlinear
1589 * vmas, now forgetting on which ones it had fallen behind.
1590 */
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07001591 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.nonlinear)
Hugh Dickins101d2be2005-11-21 21:32:16 -08001592 vma->vm_private_data = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593out:
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07001594 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595 return ret;
1596}
1597
1598/**
1599 * try_to_unmap - try to remove all page table mappings to a page
1600 * @page: the page to get unmapped
Andi Kleen14fa31b2009-09-16 11:50:10 +02001601 * @flags: action and flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602 *
1603 * Tries to remove all the page table entries which are mapping this
1604 * page, used in the pageout path. Caller must hold the page lock.
1605 * Return values are:
1606 *
1607 * SWAP_SUCCESS - we succeeded in removing all mappings
1608 * SWAP_AGAIN - we missed a mapping, try again later
1609 * SWAP_FAIL - the page is unswappable
Nick Pigginb291f002008-10-18 20:26:44 -07001610 * SWAP_MLOCK - page is mlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001612int try_to_unmap(struct page *page, enum ttu_flags flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613{
1614 int ret;
1615
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616 BUG_ON(!PageLocked(page));
Andrea Arcangeli91600e92011-01-13 15:47:16 -08001617 VM_BUG_ON(!PageHuge(page) && PageTransHuge(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618
Hugh Dickins5ad64682009-12-14 17:59:24 -08001619 if (unlikely(PageKsm(page)))
1620 ret = try_to_unmap_ksm(page, flags);
1621 else if (PageAnon(page))
Andi Kleen14fa31b2009-09-16 11:50:10 +02001622 ret = try_to_unmap_anon(page, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 else
Andi Kleen14fa31b2009-09-16 11:50:10 +02001624 ret = try_to_unmap_file(page, flags);
Nick Pigginb291f002008-10-18 20:26:44 -07001625 if (ret != SWAP_MLOCK && !page_mapped(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 ret = SWAP_SUCCESS;
1627 return ret;
1628}
Nikita Danilov81b40822005-05-01 08:58:36 -07001629
Nick Pigginb291f002008-10-18 20:26:44 -07001630/**
1631 * try_to_munlock - try to munlock a page
1632 * @page: the page to be munlocked
1633 *
1634 * Called from munlock code. Checks all of the VMAs mapping the page
1635 * to make sure nobody else has this page mlocked. The page will be
1636 * returned with PG_mlocked cleared if no other vmas have it mlocked.
1637 *
1638 * Return values are:
1639 *
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001640 * SWAP_AGAIN - no vma is holding page mlocked, or,
Nick Pigginb291f002008-10-18 20:26:44 -07001641 * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem
Hugh Dickins5ad64682009-12-14 17:59:24 -08001642 * SWAP_FAIL - page cannot be located at present
Nick Pigginb291f002008-10-18 20:26:44 -07001643 * SWAP_MLOCK - page is now mlocked.
1644 */
1645int try_to_munlock(struct page *page)
1646{
1647 VM_BUG_ON(!PageLocked(page) || PageLRU(page));
1648
Hugh Dickins5ad64682009-12-14 17:59:24 -08001649 if (unlikely(PageKsm(page)))
1650 return try_to_unmap_ksm(page, TTU_MUNLOCK);
1651 else if (PageAnon(page))
Andi Kleen14fa31b2009-09-16 11:50:10 +02001652 return try_to_unmap_anon(page, TTU_MUNLOCK);
Nick Pigginb291f002008-10-18 20:26:44 -07001653 else
Andi Kleen14fa31b2009-09-16 11:50:10 +02001654 return try_to_unmap_file(page, TTU_MUNLOCK);
Nick Pigginb291f002008-10-18 20:26:44 -07001655}
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001656
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001657void __put_anon_vma(struct anon_vma *anon_vma)
Rik van Riel76545062010-08-09 17:18:41 -07001658{
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001659 struct anon_vma *root = anon_vma->root;
Rik van Riel76545062010-08-09 17:18:41 -07001660
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001661 if (root != anon_vma && atomic_dec_and_test(&root->refcount))
1662 anon_vma_free(root);
Rik van Riel76545062010-08-09 17:18:41 -07001663
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001664 anon_vma_free(anon_vma);
Rik van Riel76545062010-08-09 17:18:41 -07001665}
Rik van Riel76545062010-08-09 17:18:41 -07001666
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001667#ifdef CONFIG_MIGRATION
1668/*
1669 * rmap_walk() and its helpers rmap_walk_anon() and rmap_walk_file():
1670 * Called by migrate.c to remove migration ptes, but might be used more later.
1671 */
1672static int rmap_walk_anon(struct page *page, int (*rmap_one)(struct page *,
1673 struct vm_area_struct *, unsigned long, void *), void *arg)
1674{
1675 struct anon_vma *anon_vma;
Michel Lespinassebf181b92012-10-08 16:31:39 -07001676 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
Rik van Riel5beb4932010-03-05 13:42:07 -08001677 struct anon_vma_chain *avc;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001678 int ret = SWAP_AGAIN;
1679
1680 /*
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +00001681 * Note: remove_migration_ptes() cannot use page_lock_anon_vma_read()
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001682 * because that depends on page_mapped(); but not all its usages
Mel Gorman3f6c8272010-05-24 14:32:17 -07001683 * are holding mmap_sem. Users without mmap_sem are required to
1684 * take a reference count to prevent the anon_vma disappearing
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001685 */
1686 anon_vma = page_anon_vma(page);
1687 if (!anon_vma)
1688 return ret;
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +00001689 anon_vma_lock_read(anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -07001690 anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) {
Rik van Riel5beb4932010-03-05 13:42:07 -08001691 struct vm_area_struct *vma = avc->vma;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001692 unsigned long address = vma_address(page, vma);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001693 ret = rmap_one(page, vma, address, arg);
1694 if (ret != SWAP_AGAIN)
1695 break;
1696 }
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +00001697 anon_vma_unlock_read(anon_vma);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001698 return ret;
1699}
1700
1701static int rmap_walk_file(struct page *page, int (*rmap_one)(struct page *,
1702 struct vm_area_struct *, unsigned long, void *), void *arg)
1703{
1704 struct address_space *mapping = page->mapping;
1705 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
1706 struct vm_area_struct *vma;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001707 int ret = SWAP_AGAIN;
1708
1709 if (!mapping)
1710 return ret;
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07001711 mutex_lock(&mapping->i_mmap_mutex);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07001712 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001713 unsigned long address = vma_address(page, vma);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001714 ret = rmap_one(page, vma, address, arg);
1715 if (ret != SWAP_AGAIN)
1716 break;
1717 }
1718 /*
1719 * No nonlinear handling: being always shared, nonlinear vmas
1720 * never contain migration ptes. Decide what to do about this
1721 * limitation to linear when we need rmap_walk() on nonlinear.
1722 */
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07001723 mutex_unlock(&mapping->i_mmap_mutex);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001724 return ret;
1725}
1726
1727int rmap_walk(struct page *page, int (*rmap_one)(struct page *,
1728 struct vm_area_struct *, unsigned long, void *), void *arg)
1729{
1730 VM_BUG_ON(!PageLocked(page));
1731
1732 if (unlikely(PageKsm(page)))
1733 return rmap_walk_ksm(page, rmap_one, arg);
1734 else if (PageAnon(page))
1735 return rmap_walk_anon(page, rmap_one, arg);
1736 else
1737 return rmap_walk_file(page, rmap_one, arg);
1738}
1739#endif /* CONFIG_MIGRATION */
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001740
Naoya Horiguchie3390f62010-06-15 13:18:13 +09001741#ifdef CONFIG_HUGETLB_PAGE
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001742/*
1743 * The following three functions are for anonymous (private mapped) hugepages.
1744 * Unlike common anonymous pages, anonymous hugepages have no accounting code
1745 * and no lru code, because we handle hugepages differently from common pages.
1746 */
1747static void __hugepage_set_anon_rmap(struct page *page,
1748 struct vm_area_struct *vma, unsigned long address, int exclusive)
1749{
1750 struct anon_vma *anon_vma = vma->anon_vma;
Naoya Horiguchi433abed2010-09-10 13:23:03 +09001751
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001752 BUG_ON(!anon_vma);
Naoya Horiguchi433abed2010-09-10 13:23:03 +09001753
1754 if (PageAnon(page))
1755 return;
1756 if (!exclusive)
1757 anon_vma = anon_vma->root;
1758
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001759 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
1760 page->mapping = (struct address_space *) anon_vma;
1761 page->index = linear_page_index(vma, address);
1762}
1763
1764void hugepage_add_anon_rmap(struct page *page,
1765 struct vm_area_struct *vma, unsigned long address)
1766{
1767 struct anon_vma *anon_vma = vma->anon_vma;
1768 int first;
Naoya Horiguchia850ea32010-09-10 13:23:06 +09001769
1770 BUG_ON(!PageLocked(page));
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001771 BUG_ON(!anon_vma);
Hugh Dickins5dbe0af2011-05-28 13:17:04 -07001772 /* address might be in next vma when migration races vma_adjust */
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001773 first = atomic_inc_and_test(&page->_mapcount);
1774 if (first)
1775 __hugepage_set_anon_rmap(page, vma, address, 0);
1776}
1777
1778void hugepage_add_new_anon_rmap(struct page *page,
1779 struct vm_area_struct *vma, unsigned long address)
1780{
1781 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
1782 atomic_set(&page->_mapcount, 0);
1783 __hugepage_set_anon_rmap(page, vma, address, 1);
1784}
Naoya Horiguchie3390f62010-06-15 13:18:13 +09001785#endif /* CONFIG_HUGETLB_PAGE */