blob: 27dfd3b82b0f39cfcd38ff8b2a02c55151651f30 [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 * inode->i_alloc_sem (vmtruncate_range)
25 * mm->mmap_sem
26 * page->flags PG_locked (lock_page)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -070027 * mapping->i_mmap_mutex
Peter Zijlstra2b575eb2011-05-24 17:12:11 -070028 * anon_vma->mutex
Nick Piggin82591e62006-10-19 23:29:10 -070029 * mm->page_table_lock or pte_lock
30 * zone->lru_lock (in mark_page_accessed, isolate_lru_page)
31 * swap_lock (in swap_duplicate, swap_info_get)
32 * mmlist_lock (in mmput, drain_mmlist and others)
33 * mapping->private_lock (in __set_page_dirty_buffers)
Dave Chinner250df6e2011-03-22 22:23:36 +110034 * inode->i_lock (in set_page_dirty's __mark_inode_dirty)
Dave Chinnera66979a2011-03-22 22:23:41 +110035 * inode_wb_list_lock (in set_page_dirty's __mark_inode_dirty)
Nick Piggin82591e62006-10-19 23:29:10 -070036 * sb_lock (within inode_lock in fs/fs-writeback.c)
37 * mapping->tree_lock (widely used, in set_page_dirty,
38 * in arch-dependent flush_dcache_mmap_lock,
Dave Chinnera66979a2011-03-22 22:23:41 +110039 * within inode_wb_list_lock in __sync_single_inode)
Andi Kleen6a460792009-09-16 11:50:15 +020040 *
41 * (code doesn't rely on that order so it could be switched around)
42 * ->tasklist_lock
Peter Zijlstra2b575eb2011-05-24 17:12:11 -070043 * anon_vma->mutex (memory_failure, collect_procs_anon)
Andi Kleen6a460792009-09-16 11:50:15 +020044 * pte map lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070045 */
46
47#include <linux/mm.h>
48#include <linux/pagemap.h>
49#include <linux/swap.h>
50#include <linux/swapops.h>
51#include <linux/slab.h>
52#include <linux/init.h>
Hugh Dickins5ad64682009-12-14 17:59:24 -080053#include <linux/ksm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070054#include <linux/rmap.h>
55#include <linux/rcupdate.h>
Christoph Lametera48d07a2006-02-01 03:05:38 -080056#include <linux/module.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080057#include <linux/memcontrol.h>
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070058#include <linux/mmu_notifier.h>
KOSAKI Motohiro64cdd542009-01-06 14:39:16 -080059#include <linux/migrate.h>
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +090060#include <linux/hugetlb.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070061
62#include <asm/tlbflush.h>
63
Nick Pigginb291f002008-10-18 20:26:44 -070064#include "internal.h"
65
Adrian Bunkfdd2e5f2008-10-18 20:28:38 -070066static struct kmem_cache *anon_vma_cachep;
Rik van Riel5beb4932010-03-05 13:42:07 -080067static struct kmem_cache *anon_vma_chain_cachep;
Adrian Bunkfdd2e5f2008-10-18 20:28:38 -070068
69static inline struct anon_vma *anon_vma_alloc(void)
70{
Peter Zijlstra01d8b202011-03-22 16:32:49 -070071 struct anon_vma *anon_vma;
72
73 anon_vma = kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL);
74 if (anon_vma) {
75 atomic_set(&anon_vma->refcount, 1);
76 /*
77 * Initialise the anon_vma root to point to itself. If called
78 * from fork, the root will be reset to the parents anon_vma.
79 */
80 anon_vma->root = anon_vma;
81 }
82
83 return anon_vma;
Adrian Bunkfdd2e5f2008-10-18 20:28:38 -070084}
85
Peter Zijlstra01d8b202011-03-22 16:32:49 -070086static inline void anon_vma_free(struct anon_vma *anon_vma)
Adrian Bunkfdd2e5f2008-10-18 20:28:38 -070087{
Peter Zijlstra01d8b202011-03-22 16:32:49 -070088 VM_BUG_ON(atomic_read(&anon_vma->refcount));
Peter Zijlstra88c22082011-05-24 17:12:13 -070089
90 /*
91 * Synchronize against page_lock_anon_vma() such that
92 * we can safely hold the lock without the anon_vma getting
93 * freed.
94 *
95 * Relies on the full mb implied by the atomic_dec_and_test() from
96 * put_anon_vma() against the acquire barrier implied by
97 * mutex_trylock() from page_lock_anon_vma(). This orders:
98 *
99 * page_lock_anon_vma() VS put_anon_vma()
100 * mutex_trylock() atomic_dec_and_test()
101 * LOCK MB
102 * atomic_read() mutex_is_locked()
103 *
104 * LOCK should suffice since the actual taking of the lock must
105 * happen _before_ what follows.
106 */
107 if (mutex_is_locked(&anon_vma->root->mutex)) {
108 anon_vma_lock(anon_vma);
109 anon_vma_unlock(anon_vma);
110 }
111
Adrian Bunkfdd2e5f2008-10-18 20:28:38 -0700112 kmem_cache_free(anon_vma_cachep, anon_vma);
113}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114
Linus Torvaldsdd347392011-06-17 19:05:36 -0700115static inline struct anon_vma_chain *anon_vma_chain_alloc(gfp_t gfp)
Rik van Riel5beb4932010-03-05 13:42:07 -0800116{
Linus Torvaldsdd347392011-06-17 19:05:36 -0700117 return kmem_cache_alloc(anon_vma_chain_cachep, gfp);
Rik van Riel5beb4932010-03-05 13:42:07 -0800118}
119
Namhyung Kime574b5f2010-10-26 14:22:02 -0700120static void anon_vma_chain_free(struct anon_vma_chain *anon_vma_chain)
Rik van Riel5beb4932010-03-05 13:42:07 -0800121{
122 kmem_cache_free(anon_vma_chain_cachep, anon_vma_chain);
123}
124
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700125/**
126 * anon_vma_prepare - attach an anon_vma to a memory region
127 * @vma: the memory region in question
128 *
129 * This makes sure the memory mapping described by 'vma' has
130 * an 'anon_vma' attached to it, so that we can associate the
131 * anonymous pages mapped into it with that anon_vma.
132 *
133 * The common case will be that we already have one, but if
Figo.zhang23a07902010-12-27 15:14:06 +0100134 * not we either need to find an adjacent mapping that we
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700135 * can re-use the anon_vma from (very common when the only
136 * reason for splitting a vma has been mprotect()), or we
137 * allocate a new one.
138 *
139 * Anon-vma allocations are very subtle, because we may have
140 * optimistically looked up an anon_vma in page_lock_anon_vma()
141 * and that may actually touch the spinlock even in the newly
142 * allocated vma (it depends on RCU to make sure that the
143 * anon_vma isn't actually destroyed).
144 *
145 * As a result, we need to do proper anon_vma locking even
146 * for the new allocation. At the same time, we do not want
147 * to do any locking for the common case of already having
148 * an anon_vma.
149 *
150 * This must be called with the mmap_sem held for reading.
151 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152int anon_vma_prepare(struct vm_area_struct *vma)
153{
154 struct anon_vma *anon_vma = vma->anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -0800155 struct anon_vma_chain *avc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156
157 might_sleep();
158 if (unlikely(!anon_vma)) {
159 struct mm_struct *mm = vma->vm_mm;
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700160 struct anon_vma *allocated;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161
Linus Torvaldsdd347392011-06-17 19:05:36 -0700162 avc = anon_vma_chain_alloc(GFP_KERNEL);
Rik van Riel5beb4932010-03-05 13:42:07 -0800163 if (!avc)
164 goto out_enomem;
165
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 anon_vma = find_mergeable_anon_vma(vma);
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700167 allocated = NULL;
168 if (!anon_vma) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 anon_vma = anon_vma_alloc();
170 if (unlikely(!anon_vma))
Rik van Riel5beb4932010-03-05 13:42:07 -0800171 goto out_enomem_free_avc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 allocated = anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 }
174
Rik van Rielcba48b92010-08-09 17:18:38 -0700175 anon_vma_lock(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 /* page_table_lock to protect against threads */
177 spin_lock(&mm->page_table_lock);
178 if (likely(!vma->anon_vma)) {
179 vma->anon_vma = anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -0800180 avc->anon_vma = anon_vma;
181 avc->vma = vma;
182 list_add(&avc->same_vma, &vma->anon_vma_chain);
Andrea Arcangeli26ba0cb2010-08-09 17:19:08 -0700183 list_add_tail(&avc->same_anon_vma, &anon_vma->head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 allocated = NULL;
Oleg Nesterov31f2b0e2010-04-23 13:18:01 -0400185 avc = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 }
187 spin_unlock(&mm->page_table_lock);
Rik van Rielcba48b92010-08-09 17:18:38 -0700188 anon_vma_unlock(anon_vma);
Oleg Nesterov31f2b0e2010-04-23 13:18:01 -0400189
190 if (unlikely(allocated))
Peter Zijlstra01d8b202011-03-22 16:32:49 -0700191 put_anon_vma(allocated);
Oleg Nesterov31f2b0e2010-04-23 13:18:01 -0400192 if (unlikely(avc))
Rik van Riel5beb4932010-03-05 13:42:07 -0800193 anon_vma_chain_free(avc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 }
195 return 0;
Rik van Riel5beb4932010-03-05 13:42:07 -0800196
197 out_enomem_free_avc:
198 anon_vma_chain_free(avc);
199 out_enomem:
200 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201}
202
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700203/*
204 * This is a useful helper function for locking the anon_vma root as
205 * we traverse the vma->anon_vma_chain, looping over anon_vma's that
206 * have the same vma.
207 *
208 * Such anon_vma's should have the same root, so you'd expect to see
209 * just a single mutex_lock for the whole traversal.
210 */
211static inline struct anon_vma *lock_anon_vma_root(struct anon_vma *root, struct anon_vma *anon_vma)
212{
213 struct anon_vma *new_root = anon_vma->root;
214 if (new_root != root) {
215 if (WARN_ON_ONCE(root))
216 mutex_unlock(&root->mutex);
217 root = new_root;
218 mutex_lock(&root->mutex);
219 }
220 return root;
221}
222
223static inline void unlock_anon_vma_root(struct anon_vma *root)
224{
225 if (root)
226 mutex_unlock(&root->mutex);
227}
228
Rik van Riel5beb4932010-03-05 13:42:07 -0800229static void anon_vma_chain_link(struct vm_area_struct *vma,
230 struct anon_vma_chain *avc,
231 struct anon_vma *anon_vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232{
Rik van Riel5beb4932010-03-05 13:42:07 -0800233 avc->vma = vma;
234 avc->anon_vma = anon_vma;
235 list_add(&avc->same_vma, &vma->anon_vma_chain);
236
Andrea Arcangeli05759d32011-01-13 15:46:53 -0800237 /*
238 * It's critical to add new vmas to the tail of the anon_vma,
239 * see comment in huge_memory.c:__split_huge_page().
240 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800241 list_add_tail(&avc->same_anon_vma, &anon_vma->head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242}
243
Rik van Riel5beb4932010-03-05 13:42:07 -0800244/*
245 * Attach the anon_vmas from src to dst.
246 * Returns 0 on success, -ENOMEM on failure.
247 */
248int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249{
Rik van Riel5beb4932010-03-05 13:42:07 -0800250 struct anon_vma_chain *avc, *pavc;
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700251 struct anon_vma *root = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252
Linus Torvalds646d87b2010-04-11 17:15:03 -0700253 list_for_each_entry_reverse(pavc, &src->anon_vma_chain, same_vma) {
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700254 struct anon_vma *anon_vma;
255
Linus Torvaldsdd347392011-06-17 19:05:36 -0700256 avc = anon_vma_chain_alloc(GFP_NOWAIT | __GFP_NOWARN);
257 if (unlikely(!avc)) {
258 unlock_anon_vma_root(root);
259 root = NULL;
260 avc = anon_vma_chain_alloc(GFP_KERNEL);
261 if (!avc)
262 goto enomem_failure;
263 }
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700264 anon_vma = pavc->anon_vma;
265 root = lock_anon_vma_root(root, anon_vma);
266 anon_vma_chain_link(dst, avc, anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 }
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700268 unlock_anon_vma_root(root);
Rik van Riel5beb4932010-03-05 13:42:07 -0800269 return 0;
270
271 enomem_failure:
272 unlink_anon_vmas(dst);
273 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274}
275
Rik van Riel5beb4932010-03-05 13:42:07 -0800276/*
277 * Attach vma to its own anon_vma, as well as to the anon_vmas that
278 * the corresponding VMA in the parent process is attached to.
279 * Returns 0 on success, non-zero on failure.
280 */
281int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282{
Rik van Riel5beb4932010-03-05 13:42:07 -0800283 struct anon_vma_chain *avc;
284 struct anon_vma *anon_vma;
285
286 /* Don't bother if the parent process has no anon_vma here. */
287 if (!pvma->anon_vma)
288 return 0;
289
290 /*
291 * First, attach the new VMA to the parent VMA's anon_vmas,
292 * so rmap can find non-COWed pages in child processes.
293 */
294 if (anon_vma_clone(vma, pvma))
295 return -ENOMEM;
296
297 /* Then add our own anon_vma. */
298 anon_vma = anon_vma_alloc();
299 if (!anon_vma)
300 goto out_error;
Linus Torvaldsdd347392011-06-17 19:05:36 -0700301 avc = anon_vma_chain_alloc(GFP_KERNEL);
Rik van Riel5beb4932010-03-05 13:42:07 -0800302 if (!avc)
303 goto out_error_free_anon_vma;
Rik van Riel5c341ee12010-08-09 17:18:39 -0700304
305 /*
306 * The root anon_vma's spinlock is the lock actually used when we
307 * lock any of the anon_vmas in this anon_vma tree.
308 */
309 anon_vma->root = pvma->anon_vma->root;
Rik van Riel76545062010-08-09 17:18:41 -0700310 /*
Peter Zijlstra01d8b202011-03-22 16:32:49 -0700311 * With refcounts, an anon_vma can stay around longer than the
312 * process it belongs to. The root anon_vma needs to be pinned until
313 * this anon_vma is freed, because the lock lives in the root.
Rik van Riel76545062010-08-09 17:18:41 -0700314 */
315 get_anon_vma(anon_vma->root);
Rik van Riel5beb4932010-03-05 13:42:07 -0800316 /* Mark this anon_vma as the one where our new (COWed) pages go. */
317 vma->anon_vma = anon_vma;
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700318 anon_vma_lock(anon_vma);
Rik van Riel5c341ee12010-08-09 17:18:39 -0700319 anon_vma_chain_link(vma, avc, anon_vma);
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700320 anon_vma_unlock(anon_vma);
Rik van Riel5beb4932010-03-05 13:42:07 -0800321
322 return 0;
323
324 out_error_free_anon_vma:
Peter Zijlstra01d8b202011-03-22 16:32:49 -0700325 put_anon_vma(anon_vma);
Rik van Riel5beb4932010-03-05 13:42:07 -0800326 out_error:
Rik van Riel4946d542010-04-05 12:13:33 -0400327 unlink_anon_vmas(vma);
Rik van Riel5beb4932010-03-05 13:42:07 -0800328 return -ENOMEM;
329}
330
Rik van Riel5beb4932010-03-05 13:42:07 -0800331void unlink_anon_vmas(struct vm_area_struct *vma)
332{
333 struct anon_vma_chain *avc, *next;
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200334 struct anon_vma *root = NULL;
Rik van Riel5beb4932010-03-05 13:42:07 -0800335
Rik van Riel5c341ee12010-08-09 17:18:39 -0700336 /*
337 * Unlink each anon_vma chained to the VMA. This list is ordered
338 * from newest to oldest, ensuring the root anon_vma gets freed last.
339 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800340 list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) {
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200341 struct anon_vma *anon_vma = avc->anon_vma;
342
343 root = lock_anon_vma_root(root, anon_vma);
344 list_del(&avc->same_anon_vma);
345
346 /*
347 * Leave empty anon_vmas on the list - we'll need
348 * to free them outside the lock.
349 */
350 if (list_empty(&anon_vma->head))
351 continue;
352
353 list_del(&avc->same_vma);
354 anon_vma_chain_free(avc);
355 }
356 unlock_anon_vma_root(root);
357
358 /*
359 * Iterate the list once more, it now only contains empty and unlinked
360 * anon_vmas, destroy them. Could not do before due to __put_anon_vma()
361 * needing to acquire the anon_vma->root->mutex.
362 */
363 list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) {
364 struct anon_vma *anon_vma = avc->anon_vma;
365
366 put_anon_vma(anon_vma);
367
Rik van Riel5beb4932010-03-05 13:42:07 -0800368 list_del(&avc->same_vma);
369 anon_vma_chain_free(avc);
370 }
371}
372
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700373static void anon_vma_ctor(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374{
Christoph Lametera35afb82007-05-16 22:10:57 -0700375 struct anon_vma *anon_vma = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376
Peter Zijlstra2b575eb2011-05-24 17:12:11 -0700377 mutex_init(&anon_vma->mutex);
Peter Zijlstra83813262011-03-22 16:32:48 -0700378 atomic_set(&anon_vma->refcount, 0);
Christoph Lametera35afb82007-05-16 22:10:57 -0700379 INIT_LIST_HEAD(&anon_vma->head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380}
381
382void __init anon_vma_init(void)
383{
384 anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma),
Paul Mundt20c2df82007-07-20 10:11:58 +0900385 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor);
Rik van Riel5beb4932010-03-05 13:42:07 -0800386 anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain, SLAB_PANIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387}
388
389/*
Peter Zijlstra6111e4c2011-05-24 17:12:08 -0700390 * Getting a lock on a stable anon_vma from a page off the LRU is tricky!
391 *
392 * Since there is no serialization what so ever against page_remove_rmap()
393 * the best this function can do is return a locked anon_vma that might
394 * have been relevant to this page.
395 *
396 * The page might have been remapped to a different anon_vma or the anon_vma
397 * returned may already be freed (and even reused).
398 *
Peter Zijlstrabc658c92011-05-29 10:33:44 +0200399 * In case it was remapped to a different anon_vma, the new anon_vma will be a
400 * child of the old anon_vma, and the anon_vma lifetime rules will therefore
401 * ensure that any anon_vma obtained from the page will still be valid for as
402 * long as we observe page_mapped() [ hence all those page_mapped() tests ].
403 *
Peter Zijlstra6111e4c2011-05-24 17:12:08 -0700404 * All users of this function must be very careful when walking the anon_vma
405 * chain and verify that the page in question is indeed mapped in it
406 * [ something equivalent to page_mapped_in_vma() ].
407 *
408 * Since anon_vma's slab is DESTROY_BY_RCU and we know from page_remove_rmap()
409 * that the anon_vma pointer from page->mapping is valid if there is a
410 * mapcount, we can dereference the anon_vma after observing those.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 */
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700412struct anon_vma *page_get_anon_vma(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413{
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700414 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 unsigned long anon_mapping;
416
417 rcu_read_lock();
Hugh Dickins80e148222009-12-14 17:59:29 -0800418 anon_mapping = (unsigned long) ACCESS_ONCE(page->mapping);
Hugh Dickins3ca7b3c2009-12-14 17:58:57 -0800419 if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 goto out;
421 if (!page_mapped(page))
422 goto out;
423
424 anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700425 if (!atomic_inc_not_zero(&anon_vma->refcount)) {
426 anon_vma = NULL;
427 goto out;
428 }
Hugh Dickinsf18194272010-08-25 23:12:54 -0700429
430 /*
431 * If this page is still mapped, then its anon_vma cannot have been
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700432 * freed. But if it has been unmapped, we have no security against the
433 * anon_vma structure being freed and reused (for another anon_vma:
434 * SLAB_DESTROY_BY_RCU guarantees that - so the atomic_inc_not_zero()
435 * above cannot corrupt).
Hugh Dickinsf18194272010-08-25 23:12:54 -0700436 */
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700437 if (!page_mapped(page)) {
438 put_anon_vma(anon_vma);
439 anon_vma = NULL;
440 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441out:
442 rcu_read_unlock();
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700443
444 return anon_vma;
445}
446
Peter Zijlstra88c22082011-05-24 17:12:13 -0700447/*
448 * Similar to page_get_anon_vma() except it locks the anon_vma.
449 *
450 * Its a little more complex as it tries to keep the fast path to a single
451 * atomic op -- the trylock. If we fail the trylock, we fall back to getting a
452 * reference like with page_get_anon_vma() and then block on the mutex.
453 */
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700454struct anon_vma *page_lock_anon_vma(struct page *page)
455{
Peter Zijlstra88c22082011-05-24 17:12:13 -0700456 struct anon_vma *anon_vma = NULL;
Hugh Dickinseee0f252011-05-28 13:20:21 -0700457 struct anon_vma *root_anon_vma;
Peter Zijlstra88c22082011-05-24 17:12:13 -0700458 unsigned long anon_mapping;
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700459
Peter Zijlstra88c22082011-05-24 17:12:13 -0700460 rcu_read_lock();
461 anon_mapping = (unsigned long) ACCESS_ONCE(page->mapping);
462 if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
463 goto out;
464 if (!page_mapped(page))
465 goto out;
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700466
Peter Zijlstra88c22082011-05-24 17:12:13 -0700467 anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
Hugh Dickinseee0f252011-05-28 13:20:21 -0700468 root_anon_vma = ACCESS_ONCE(anon_vma->root);
469 if (mutex_trylock(&root_anon_vma->mutex)) {
Peter Zijlstra88c22082011-05-24 17:12:13 -0700470 /*
Hugh Dickinseee0f252011-05-28 13:20:21 -0700471 * If the page is still mapped, then this anon_vma is still
472 * its anon_vma, and holding the mutex ensures that it will
Peter Zijlstrabc658c92011-05-29 10:33:44 +0200473 * not go away, see anon_vma_free().
Peter Zijlstra88c22082011-05-24 17:12:13 -0700474 */
Hugh Dickinseee0f252011-05-28 13:20:21 -0700475 if (!page_mapped(page)) {
476 mutex_unlock(&root_anon_vma->mutex);
Peter Zijlstra88c22082011-05-24 17:12:13 -0700477 anon_vma = NULL;
478 }
479 goto out;
480 }
481
482 /* trylock failed, we got to sleep */
483 if (!atomic_inc_not_zero(&anon_vma->refcount)) {
484 anon_vma = NULL;
485 goto out;
486 }
487
488 if (!page_mapped(page)) {
489 put_anon_vma(anon_vma);
490 anon_vma = NULL;
491 goto out;
492 }
493
494 /* we pinned the anon_vma, its safe to sleep */
495 rcu_read_unlock();
496 anon_vma_lock(anon_vma);
497
498 if (atomic_dec_and_test(&anon_vma->refcount)) {
499 /*
500 * Oops, we held the last refcount, release the lock
501 * and bail -- can't simply use put_anon_vma() because
502 * we'll deadlock on the anon_vma_lock() recursion.
503 */
504 anon_vma_unlock(anon_vma);
505 __put_anon_vma(anon_vma);
506 anon_vma = NULL;
507 }
508
509 return anon_vma;
510
511out:
512 rcu_read_unlock();
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700513 return anon_vma;
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800514}
515
Andi Kleen10be22d2009-09-16 11:50:04 +0200516void page_unlock_anon_vma(struct anon_vma *anon_vma)
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800517{
Rik van Rielcba48b92010-08-09 17:18:38 -0700518 anon_vma_unlock(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519}
520
521/*
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -0800522 * At what user virtual address is page expected in @vma?
523 * Returns virtual address or -EFAULT if page's index/offset is not
524 * within the range mapped the @vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800526inline unsigned long
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527vma_address(struct page *page, struct vm_area_struct *vma)
528{
529 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
530 unsigned long address;
531
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +0900532 if (unlikely(is_vm_hugetlb_page(vma)))
533 pgoff = page->index << huge_page_order(page_hstate(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
535 if (unlikely(address < vma->vm_start || address >= vma->vm_end)) {
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -0800536 /* page should be within @vma mapping range */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 return -EFAULT;
538 }
539 return address;
540}
541
542/*
Huang Shijiebf89c8c2009-10-01 15:44:04 -0700543 * At what user virtual address is page expected in vma?
Naoya Horiguchiab941e02010-05-11 14:06:55 -0700544 * Caller should check the page is actually part of the vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 */
546unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma)
547{
Andrea Arcangeli21d0d442010-08-09 17:19:10 -0700548 if (PageAnon(page)) {
Hugh Dickins4829b902010-10-02 17:46:06 -0700549 struct anon_vma *page__anon_vma = page_anon_vma(page);
550 /*
551 * Note: swapoff's unuse_vma() is more efficient with this
552 * check, and needs it to match anon_vma when KSM is active.
553 */
554 if (!vma->anon_vma || !page__anon_vma ||
555 vma->anon_vma->root != page__anon_vma->root)
Andrea Arcangeli21d0d442010-08-09 17:19:10 -0700556 return -EFAULT;
557 } else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) {
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800558 if (!vma->vm_file ||
559 vma->vm_file->f_mapping != page->mapping)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 return -EFAULT;
561 } else
562 return -EFAULT;
563 return vma_address(page, vma);
564}
565
566/*
Nikita Danilov81b40822005-05-01 08:58:36 -0700567 * Check that @page is mapped at @address into @mm.
568 *
Nick Piggin479db0b2008-08-20 14:09:18 -0700569 * If @sync is false, page_check_address may perform a racy check to avoid
570 * the page table lock when the pte is not present (helpful when reclaiming
571 * highly shared pages).
572 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -0700573 * On success returns with pte mapped and locked.
Nikita Danilov81b40822005-05-01 08:58:36 -0700574 */
Namhyung Kime9a81a82010-10-26 14:22:01 -0700575pte_t *__page_check_address(struct page *page, struct mm_struct *mm,
Nick Piggin479db0b2008-08-20 14:09:18 -0700576 unsigned long address, spinlock_t **ptlp, int sync)
Nikita Danilov81b40822005-05-01 08:58:36 -0700577{
578 pgd_t *pgd;
579 pud_t *pud;
580 pmd_t *pmd;
581 pte_t *pte;
Hugh Dickinsc0718802005-10-29 18:16:31 -0700582 spinlock_t *ptl;
Nikita Danilov81b40822005-05-01 08:58:36 -0700583
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +0900584 if (unlikely(PageHuge(page))) {
585 pte = huge_pte_offset(mm, address);
586 ptl = &mm->page_table_lock;
587 goto check;
588 }
589
Nikita Danilov81b40822005-05-01 08:58:36 -0700590 pgd = pgd_offset(mm, address);
Hugh Dickinsc0718802005-10-29 18:16:31 -0700591 if (!pgd_present(*pgd))
592 return NULL;
593
594 pud = pud_offset(pgd, address);
595 if (!pud_present(*pud))
596 return NULL;
597
598 pmd = pmd_offset(pud, address);
599 if (!pmd_present(*pmd))
600 return NULL;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800601 if (pmd_trans_huge(*pmd))
602 return NULL;
Hugh Dickinsc0718802005-10-29 18:16:31 -0700603
604 pte = pte_offset_map(pmd, address);
605 /* Make a quick check before getting the lock */
Nick Piggin479db0b2008-08-20 14:09:18 -0700606 if (!sync && !pte_present(*pte)) {
Hugh Dickinsc0718802005-10-29 18:16:31 -0700607 pte_unmap(pte);
608 return NULL;
Nikita Danilov81b40822005-05-01 08:58:36 -0700609 }
Hugh Dickinsc0718802005-10-29 18:16:31 -0700610
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700611 ptl = pte_lockptr(mm, pmd);
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +0900612check:
Hugh Dickinsc0718802005-10-29 18:16:31 -0700613 spin_lock(ptl);
614 if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) {
615 *ptlp = ptl;
616 return pte;
617 }
618 pte_unmap_unlock(pte, ptl);
619 return NULL;
Nikita Danilov81b40822005-05-01 08:58:36 -0700620}
621
Nick Pigginb291f002008-10-18 20:26:44 -0700622/**
623 * page_mapped_in_vma - check whether a page is really mapped in a VMA
624 * @page: the page to test
625 * @vma: the VMA to test
626 *
627 * Returns 1 if the page is mapped into the page tables of the VMA, 0
628 * if the page is not mapped into the page tables of this VMA. Only
629 * valid for normal file or anonymous VMAs.
630 */
Andi Kleen6a460792009-09-16 11:50:15 +0200631int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma)
Nick Pigginb291f002008-10-18 20:26:44 -0700632{
633 unsigned long address;
634 pte_t *pte;
635 spinlock_t *ptl;
636
637 address = vma_address(page, vma);
638 if (address == -EFAULT) /* out of vma range */
639 return 0;
640 pte = page_check_address(page, vma->vm_mm, address, &ptl, 1);
641 if (!pte) /* the page is not in this mm */
642 return 0;
643 pte_unmap_unlock(pte, ptl);
644
645 return 1;
646}
647
Nikita Danilov81b40822005-05-01 08:58:36 -0700648/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 * Subfunctions of page_referenced: page_referenced_one called
650 * repeatedly from either page_referenced_anon or page_referenced_file.
651 */
Hugh Dickins5ad64682009-12-14 17:59:24 -0800652int page_referenced_one(struct page *page, struct vm_area_struct *vma,
653 unsigned long address, unsigned int *mapcount,
654 unsigned long *vm_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655{
656 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 int referenced = 0;
658
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800659 if (unlikely(PageTransHuge(page))) {
660 pmd_t *pmd;
661
662 spin_lock(&mm->page_table_lock);
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800663 /*
664 * rmap might return false positives; we must filter
665 * these out using page_check_address_pmd().
666 */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800667 pmd = page_check_address_pmd(page, mm, address,
668 PAGE_CHECK_ADDRESS_PMD_FLAG);
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800669 if (!pmd) {
670 spin_unlock(&mm->page_table_lock);
671 goto out;
672 }
673
674 if (vma->vm_flags & VM_LOCKED) {
675 spin_unlock(&mm->page_table_lock);
676 *mapcount = 0; /* break early from loop */
677 *vm_flags |= VM_LOCKED;
678 goto out;
679 }
680
681 /* go ahead even if the pmd is pmd_trans_splitting() */
682 if (pmdp_clear_flush_young_notify(vma, address, pmd))
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800683 referenced++;
684 spin_unlock(&mm->page_table_lock);
685 } else {
686 pte_t *pte;
687 spinlock_t *ptl;
688
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800689 /*
690 * rmap might return false positives; we must filter
691 * these out using page_check_address().
692 */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800693 pte = page_check_address(page, mm, address, &ptl, 0);
694 if (!pte)
695 goto out;
696
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800697 if (vma->vm_flags & VM_LOCKED) {
698 pte_unmap_unlock(pte, ptl);
699 *mapcount = 0; /* break early from loop */
700 *vm_flags |= VM_LOCKED;
701 goto out;
702 }
703
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800704 if (ptep_clear_flush_young_notify(vma, address, pte)) {
705 /*
706 * Don't treat a reference through a sequentially read
707 * mapping as such. If the page has been used in
708 * another mapping, we will catch it; if this other
709 * mapping is already gone, the unmap path will have
710 * set PG_referenced or activated the page.
711 */
712 if (likely(!VM_SequentialReadHint(vma)))
713 referenced++;
714 }
715 pte_unmap_unlock(pte, ptl);
716 }
717
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800718 /* Pretend the page is referenced if the task has the
719 swap token and is in the middle of a page fault. */
720 if (mm != current->mm && has_swap_token(mm) &&
721 rwsem_is_locked(&mm->mmap_sem))
722 referenced++;
723
Hugh Dickinsc0718802005-10-29 18:16:31 -0700724 (*mapcount)--;
Huang Shijie273f0472009-12-14 17:58:51 -0800725
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700726 if (referenced)
727 *vm_flags |= vma->vm_flags;
Huang Shijie273f0472009-12-14 17:58:51 -0800728out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 return referenced;
730}
731
Balbir Singhbed71612008-02-07 00:14:01 -0800732static int page_referenced_anon(struct page *page,
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700733 struct mem_cgroup *mem_cont,
734 unsigned long *vm_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735{
736 unsigned int mapcount;
737 struct anon_vma *anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -0800738 struct anon_vma_chain *avc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 int referenced = 0;
740
741 anon_vma = page_lock_anon_vma(page);
742 if (!anon_vma)
743 return referenced;
744
745 mapcount = page_mapcount(page);
Rik van Riel5beb4932010-03-05 13:42:07 -0800746 list_for_each_entry(avc, &anon_vma->head, same_anon_vma) {
747 struct vm_area_struct *vma = avc->vma;
Hugh Dickins1cb17292009-12-14 17:59:01 -0800748 unsigned long address = vma_address(page, vma);
749 if (address == -EFAULT)
750 continue;
Balbir Singhbed71612008-02-07 00:14:01 -0800751 /*
752 * If we are reclaiming on behalf of a cgroup, skip
753 * counting on behalf of references from different
754 * cgroups
755 */
Hugh Dickinsbd845e32008-03-04 14:29:01 -0800756 if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont))
Balbir Singhbed71612008-02-07 00:14:01 -0800757 continue;
Hugh Dickins1cb17292009-12-14 17:59:01 -0800758 referenced += page_referenced_one(page, vma, address,
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700759 &mapcount, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 if (!mapcount)
761 break;
762 }
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800763
764 page_unlock_anon_vma(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 return referenced;
766}
767
768/**
769 * page_referenced_file - referenced check for object-based rmap
770 * @page: the page we're checking references on.
Randy Dunlap43d8eac2008-03-19 17:00:43 -0700771 * @mem_cont: target memory controller
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700772 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 *
774 * For an object-based mapped page, find all the places it is mapped and
775 * check/clear the referenced flag. This is done by following the page->mapping
776 * pointer, then walking the chain of vmas it holds. It returns the number
777 * of references it found.
778 *
779 * This function is only called from page_referenced for object-based pages.
780 */
Balbir Singhbed71612008-02-07 00:14:01 -0800781static int page_referenced_file(struct page *page,
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700782 struct mem_cgroup *mem_cont,
783 unsigned long *vm_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784{
785 unsigned int mapcount;
786 struct address_space *mapping = page->mapping;
787 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
788 struct vm_area_struct *vma;
789 struct prio_tree_iter iter;
790 int referenced = 0;
791
792 /*
793 * The caller's checks on page->mapping and !PageAnon have made
794 * sure that this is a file page: the check for page->mapping
795 * excludes the case just before it gets set on an anon page.
796 */
797 BUG_ON(PageAnon(page));
798
799 /*
800 * The page lock not only makes sure that page->mapping cannot
801 * suddenly be NULLified by truncation, it makes sure that the
802 * structure at mapping cannot be freed and reused yet,
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700803 * so we can safely take mapping->i_mmap_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 */
805 BUG_ON(!PageLocked(page));
806
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700807 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808
809 /*
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700810 * i_mmap_mutex does not stabilize mapcount at all, but mapcount
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 * is more likely to be accurate if we note it after spinning.
812 */
813 mapcount = page_mapcount(page);
814
815 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
Hugh Dickins1cb17292009-12-14 17:59:01 -0800816 unsigned long address = vma_address(page, vma);
817 if (address == -EFAULT)
818 continue;
Balbir Singhbed71612008-02-07 00:14:01 -0800819 /*
820 * If we are reclaiming on behalf of a cgroup, skip
821 * counting on behalf of references from different
822 * cgroups
823 */
Hugh Dickinsbd845e32008-03-04 14:29:01 -0800824 if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont))
Balbir Singhbed71612008-02-07 00:14:01 -0800825 continue;
Hugh Dickins1cb17292009-12-14 17:59:01 -0800826 referenced += page_referenced_one(page, vma, address,
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700827 &mapcount, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 if (!mapcount)
829 break;
830 }
831
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700832 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 return referenced;
834}
835
836/**
837 * page_referenced - test if the page was referenced
838 * @page: the page to test
839 * @is_locked: caller holds lock on the page
Randy Dunlap43d8eac2008-03-19 17:00:43 -0700840 * @mem_cont: target memory controller
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700841 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 *
843 * Quick test_and_clear_referenced for all mappings to a page,
844 * returns the number of ptes which referenced the page.
845 */
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700846int page_referenced(struct page *page,
847 int is_locked,
848 struct mem_cgroup *mem_cont,
849 unsigned long *vm_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850{
851 int referenced = 0;
Hugh Dickins5ad64682009-12-14 17:59:24 -0800852 int we_locked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700854 *vm_flags = 0;
Hugh Dickins3ca7b3c2009-12-14 17:58:57 -0800855 if (page_mapped(page) && page_rmapping(page)) {
Hugh Dickins5ad64682009-12-14 17:59:24 -0800856 if (!is_locked && (!PageAnon(page) || PageKsm(page))) {
857 we_locked = trylock_page(page);
858 if (!we_locked) {
859 referenced++;
860 goto out;
861 }
862 }
863 if (unlikely(PageKsm(page)))
864 referenced += page_referenced_ksm(page, mem_cont,
865 vm_flags);
866 else if (PageAnon(page))
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700867 referenced += page_referenced_anon(page, mem_cont,
868 vm_flags);
Hugh Dickins5ad64682009-12-14 17:59:24 -0800869 else if (page->mapping)
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700870 referenced += page_referenced_file(page, mem_cont,
871 vm_flags);
Hugh Dickins5ad64682009-12-14 17:59:24 -0800872 if (we_locked)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 }
Hugh Dickins5ad64682009-12-14 17:59:24 -0800875out:
Martin Schwidefsky2d425522011-05-23 10:24:39 +0200876 if (page_test_and_clear_young(page_to_pfn(page)))
Christian Borntraeger5b7baf02008-03-25 18:47:12 +0100877 referenced++;
878
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 return referenced;
880}
881
Hugh Dickins1cb17292009-12-14 17:59:01 -0800882static int page_mkclean_one(struct page *page, struct vm_area_struct *vma,
883 unsigned long address)
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700884{
885 struct mm_struct *mm = vma->vm_mm;
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100886 pte_t *pte;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700887 spinlock_t *ptl;
888 int ret = 0;
889
Nick Piggin479db0b2008-08-20 14:09:18 -0700890 pte = page_check_address(page, mm, address, &ptl, 1);
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700891 if (!pte)
892 goto out;
893
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100894 if (pte_dirty(*pte) || pte_write(*pte)) {
895 pte_t entry;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700896
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100897 flush_cache_page(vma, address, pte_pfn(*pte));
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700898 entry = ptep_clear_flush_notify(vma, address, pte);
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100899 entry = pte_wrprotect(entry);
900 entry = pte_mkclean(entry);
Al Virod6e88e62006-12-29 16:48:35 -0800901 set_pte_at(mm, address, pte, entry);
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100902 ret = 1;
903 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700904
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700905 pte_unmap_unlock(pte, ptl);
906out:
907 return ret;
908}
909
910static int page_mkclean_file(struct address_space *mapping, struct page *page)
911{
912 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
913 struct vm_area_struct *vma;
914 struct prio_tree_iter iter;
915 int ret = 0;
916
917 BUG_ON(PageAnon(page));
918
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700919 mutex_lock(&mapping->i_mmap_mutex);
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700920 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
Hugh Dickins1cb17292009-12-14 17:59:01 -0800921 if (vma->vm_flags & VM_SHARED) {
922 unsigned long address = vma_address(page, vma);
923 if (address == -EFAULT)
924 continue;
925 ret += page_mkclean_one(page, vma, address);
926 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700927 }
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700928 mutex_unlock(&mapping->i_mmap_mutex);
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700929 return ret;
930}
931
932int page_mkclean(struct page *page)
933{
934 int ret = 0;
935
936 BUG_ON(!PageLocked(page));
937
938 if (page_mapped(page)) {
939 struct address_space *mapping = page_mapping(page);
Christian Borntraegerce7e9fa2007-11-20 11:13:36 +0100940 if (mapping) {
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700941 ret = page_mkclean_file(mapping, page);
Martin Schwidefsky2d425522011-05-23 10:24:39 +0200942 if (page_test_and_clear_dirty(page_to_pfn(page), 1))
Christian Borntraegerce7e9fa2007-11-20 11:13:36 +0100943 ret = 1;
Martin Schwidefsky6c210482007-04-27 16:01:57 +0200944 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700945 }
946
947 return ret;
948}
Jaya Kumar60b59be2007-05-08 00:37:37 -0700949EXPORT_SYMBOL_GPL(page_mkclean);
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700950
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951/**
Rik van Rielc44b6742010-03-05 13:42:09 -0800952 * page_move_anon_rmap - move a page to our anon_vma
953 * @page: the page to move to our anon_vma
954 * @vma: the vma the page belongs to
955 * @address: the user virtual address mapped
956 *
957 * When a page belongs exclusively to one process after a COW event,
958 * that page can be moved into the anon_vma that belongs to just that
959 * process, so the rmap code will not search the parent or sibling
960 * processes.
961 */
962void page_move_anon_rmap(struct page *page,
963 struct vm_area_struct *vma, unsigned long address)
964{
965 struct anon_vma *anon_vma = vma->anon_vma;
966
967 VM_BUG_ON(!PageLocked(page));
968 VM_BUG_ON(!anon_vma);
969 VM_BUG_ON(page->index != linear_page_index(vma, address));
970
971 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
972 page->mapping = (struct address_space *) anon_vma;
973}
974
975/**
Andi Kleen4e1c1972010-09-22 12:43:56 +0200976 * __page_set_anon_rmap - set up new anonymous rmap
977 * @page: Page to add to rmap
978 * @vma: VM area to add page to.
979 * @address: User virtual address of the mapping
Rik van Riele8a03fe2010-04-14 17:59:28 -0400980 * @exclusive: the page is exclusively owned by the current process
Nick Piggin9617d952006-01-06 00:11:12 -0800981 */
982static void __page_set_anon_rmap(struct page *page,
Rik van Riele8a03fe2010-04-14 17:59:28 -0400983 struct vm_area_struct *vma, unsigned long address, int exclusive)
Nick Piggin9617d952006-01-06 00:11:12 -0800984{
Rik van Riele8a03fe2010-04-14 17:59:28 -0400985 struct anon_vma *anon_vma = vma->anon_vma;
Nick Piggin9617d952006-01-06 00:11:12 -0800986
Rik van Riele8a03fe2010-04-14 17:59:28 -0400987 BUG_ON(!anon_vma);
Linus Torvaldsea900022010-04-12 12:44:29 -0700988
Andi Kleen4e1c1972010-09-22 12:43:56 +0200989 if (PageAnon(page))
990 return;
991
Linus Torvaldsea900022010-04-12 12:44:29 -0700992 /*
Rik van Riele8a03fe2010-04-14 17:59:28 -0400993 * If the page isn't exclusively mapped into this vma,
994 * we must use the _oldest_ possible anon_vma for the
995 * page mapping!
Linus Torvaldsea900022010-04-12 12:44:29 -0700996 */
Andi Kleen4e1c1972010-09-22 12:43:56 +0200997 if (!exclusive)
Andrea Arcangeli288468c2010-08-09 17:19:09 -0700998 anon_vma = anon_vma->root;
Linus Torvaldsea900022010-04-12 12:44:29 -0700999
Nick Piggin9617d952006-01-06 00:11:12 -08001000 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
1001 page->mapping = (struct address_space *) anon_vma;
Nick Piggin9617d952006-01-06 00:11:12 -08001002 page->index = linear_page_index(vma, address);
Nick Piggin9617d952006-01-06 00:11:12 -08001003}
1004
1005/**
Randy Dunlap43d8eac2008-03-19 17:00:43 -07001006 * __page_check_anon_rmap - sanity check anonymous rmap addition
Nick Pigginc97a9e12007-05-16 22:11:21 -07001007 * @page: the page to add the mapping to
1008 * @vma: the vm area in which the mapping is added
1009 * @address: the user virtual address mapped
1010 */
1011static void __page_check_anon_rmap(struct page *page,
1012 struct vm_area_struct *vma, unsigned long address)
1013{
1014#ifdef CONFIG_DEBUG_VM
1015 /*
1016 * The page's anon-rmap details (mapping and index) are guaranteed to
1017 * be set up correctly at this point.
1018 *
1019 * We have exclusion against page_add_anon_rmap because the caller
1020 * always holds the page locked, except if called from page_dup_rmap,
1021 * in which case the page is already known to be setup.
1022 *
1023 * We have exclusion against page_add_new_anon_rmap because those pages
1024 * are initially only visible via the pagetables, and the pte is locked
1025 * over the call to page_add_new_anon_rmap.
1026 */
Andrea Arcangeli44ab57a2010-08-09 17:19:10 -07001027 BUG_ON(page_anon_vma(page)->root != vma->anon_vma->root);
Nick Pigginc97a9e12007-05-16 22:11:21 -07001028 BUG_ON(page->index != linear_page_index(vma, address));
1029#endif
1030}
1031
1032/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 * page_add_anon_rmap - add pte mapping to an anonymous page
1034 * @page: the page to add the mapping to
1035 * @vma: the vm area in which the mapping is added
1036 * @address: the user virtual address mapped
1037 *
Hugh Dickins5ad64682009-12-14 17:59:24 -08001038 * The caller needs to hold the pte lock, and the page must be locked in
Hugh Dickins80e148222009-12-14 17:59:29 -08001039 * the anon_vma case: to serialize mapping,index checking after setting,
1040 * and to ensure that PageAnon is not being upgraded racily to PageKsm
1041 * (but PageKsm is never downgraded to PageAnon).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 */
1043void page_add_anon_rmap(struct page *page,
1044 struct vm_area_struct *vma, unsigned long address)
1045{
Rik van Rielad8c2ee2010-08-09 17:19:48 -07001046 do_page_add_anon_rmap(page, vma, address, 0);
1047}
1048
1049/*
1050 * Special version of the above for do_swap_page, which often runs
1051 * into pages that are exclusively owned by the current process.
1052 * Everybody else should continue to use page_add_anon_rmap above.
1053 */
1054void do_page_add_anon_rmap(struct page *page,
1055 struct vm_area_struct *vma, unsigned long address, int exclusive)
1056{
Hugh Dickins5ad64682009-12-14 17:59:24 -08001057 int first = atomic_inc_and_test(&page->_mapcount);
Andrea Arcangeli79134172011-01-13 15:46:58 -08001058 if (first) {
1059 if (!PageTransHuge(page))
1060 __inc_zone_page_state(page, NR_ANON_PAGES);
1061 else
1062 __inc_zone_page_state(page,
1063 NR_ANON_TRANSPARENT_HUGEPAGES);
1064 }
Hugh Dickins5ad64682009-12-14 17:59:24 -08001065 if (unlikely(PageKsm(page)))
1066 return;
1067
Nick Pigginc97a9e12007-05-16 22:11:21 -07001068 VM_BUG_ON(!PageLocked(page));
Hugh Dickins5dbe0af2011-05-28 13:17:04 -07001069 /* address might be in next vma when migration races vma_adjust */
Hugh Dickins5ad64682009-12-14 17:59:24 -08001070 if (first)
Rik van Rielad8c2ee2010-08-09 17:19:48 -07001071 __page_set_anon_rmap(page, vma, address, exclusive);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001072 else
Nick Pigginc97a9e12007-05-16 22:11:21 -07001073 __page_check_anon_rmap(page, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074}
1075
Randy Dunlap43d8eac2008-03-19 17:00:43 -07001076/**
Nick Piggin9617d952006-01-06 00:11:12 -08001077 * page_add_new_anon_rmap - add pte mapping to a new anonymous page
1078 * @page: the page to add the mapping to
1079 * @vma: the vm area in which the mapping is added
1080 * @address: the user virtual address mapped
1081 *
1082 * Same as page_add_anon_rmap but must only be called on *new* pages.
1083 * This means the inc-and-test can be bypassed.
Nick Pigginc97a9e12007-05-16 22:11:21 -07001084 * Page does not have to be locked.
Nick Piggin9617d952006-01-06 00:11:12 -08001085 */
1086void page_add_new_anon_rmap(struct page *page,
1087 struct vm_area_struct *vma, unsigned long address)
1088{
Hugh Dickinsb5934c52009-01-06 14:39:25 -08001089 VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end);
Hugh Dickinscbf84b72009-01-06 14:39:27 -08001090 SetPageSwapBacked(page);
1091 atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */
Andrea Arcangeli79134172011-01-13 15:46:58 -08001092 if (!PageTransHuge(page))
1093 __inc_zone_page_state(page, NR_ANON_PAGES);
1094 else
1095 __inc_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES);
Rik van Riele8a03fe2010-04-14 17:59:28 -04001096 __page_set_anon_rmap(page, vma, address, 1);
Hugh Dickinsb5934c52009-01-06 14:39:25 -08001097 if (page_evictable(page, vma))
Hugh Dickinscbf84b72009-01-06 14:39:27 -08001098 lru_cache_add_lru(page, LRU_ACTIVE_ANON);
Hugh Dickinsb5934c52009-01-06 14:39:25 -08001099 else
1100 add_page_to_unevictable_list(page);
Nick Piggin9617d952006-01-06 00:11:12 -08001101}
1102
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103/**
1104 * page_add_file_rmap - add pte mapping to a file page
1105 * @page: the page to add the mapping to
1106 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -07001107 * The caller needs to hold the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 */
1109void page_add_file_rmap(struct page *page)
1110{
Balbir Singhd69b0422009-06-17 16:26:34 -07001111 if (atomic_inc_and_test(&page->_mapcount)) {
Christoph Lameter65ba55f2006-06-30 01:55:34 -07001112 __inc_zone_page_state(page, NR_FILE_MAPPED);
Greg Thelen2a7106f2011-01-13 15:47:37 -08001113 mem_cgroup_inc_page_stat(page, MEMCG_NR_FILE_MAPPED);
Balbir Singhd69b0422009-06-17 16:26:34 -07001114 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115}
1116
1117/**
1118 * page_remove_rmap - take down pte mapping from a page
1119 * @page: page to remove mapping from
1120 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -07001121 * The caller needs to hold the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 */
Hugh Dickinsedc315f2009-01-06 14:40:11 -08001123void page_remove_rmap(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124{
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001125 /* page still mapped by someone else? */
1126 if (!atomic_add_negative(-1, &page->_mapcount))
1127 return;
1128
1129 /*
1130 * Now that the last pte has gone, s390 must transfer dirty
1131 * flag from storage key to struct page. We can usually skip
1132 * this if the page is anon, so about to be freed; but perhaps
1133 * not if it's in swapcache - there might be another pte slot
1134 * containing the swap entry, but page not yet written to swap.
1135 */
Martin Schwidefsky2d425522011-05-23 10:24:39 +02001136 if ((!PageAnon(page) || PageSwapCache(page)) &&
1137 page_test_and_clear_dirty(page_to_pfn(page), 1))
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001138 set_page_dirty(page);
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001139 /*
1140 * Hugepages are not counted in NR_ANON_PAGES nor NR_FILE_MAPPED
1141 * and not charged by memcg for now.
1142 */
1143 if (unlikely(PageHuge(page)))
1144 return;
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001145 if (PageAnon(page)) {
1146 mem_cgroup_uncharge_page(page);
Andrea Arcangeli79134172011-01-13 15:46:58 -08001147 if (!PageTransHuge(page))
1148 __dec_zone_page_state(page, NR_ANON_PAGES);
1149 else
1150 __dec_zone_page_state(page,
1151 NR_ANON_TRANSPARENT_HUGEPAGES);
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001152 } else {
1153 __dec_zone_page_state(page, NR_FILE_MAPPED);
Greg Thelen2a7106f2011-01-13 15:47:37 -08001154 mem_cgroup_dec_page_stat(page, MEMCG_NR_FILE_MAPPED);
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001155 }
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001156 /*
1157 * It would be tidy to reset the PageAnon mapping here,
1158 * but that might overwrite a racing page_add_anon_rmap
1159 * which increments mapcount after us but sets mapping
1160 * before us: so leave the reset to free_hot_cold_page,
1161 * and remember that it's only reliable while mapped.
1162 * Leaving it set also helps swapoff to reinstate ptes
1163 * faster for those pages still in swapcache.
1164 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165}
1166
1167/*
1168 * Subfunctions of try_to_unmap: try_to_unmap_one called
1169 * repeatedly from either try_to_unmap_anon or try_to_unmap_file.
1170 */
Hugh Dickins5ad64682009-12-14 17:59:24 -08001171int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
1172 unsigned long address, enum ttu_flags flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173{
1174 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 pte_t *pte;
1176 pte_t pteval;
Hugh Dickinsc0718802005-10-29 18:16:31 -07001177 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 int ret = SWAP_AGAIN;
1179
Nick Piggin479db0b2008-08-20 14:09:18 -07001180 pte = page_check_address(page, mm, address, &ptl, 0);
Hugh Dickinsc0718802005-10-29 18:16:31 -07001181 if (!pte)
Nikita Danilov81b40822005-05-01 08:58:36 -07001182 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183
1184 /*
1185 * If the page is mlock()d, we cannot swap it out.
1186 * If it's recently referenced (perhaps page_referenced
1187 * skipped over this mm) then we should reactivate it.
1188 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001189 if (!(flags & TTU_IGNORE_MLOCK)) {
KOSAKI Motohirocaed0f482009-12-14 17:59:45 -08001190 if (vma->vm_flags & VM_LOCKED)
1191 goto out_mlock;
1192
Hugh Dickinsaf8e3352009-12-14 17:58:59 -08001193 if (TTU_ACTION(flags) == TTU_MUNLOCK)
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001194 goto out_unmap;
Andi Kleen14fa31b2009-09-16 11:50:10 +02001195 }
1196 if (!(flags & TTU_IGNORE_ACCESS)) {
Nick Pigginb291f002008-10-18 20:26:44 -07001197 if (ptep_clear_flush_young_notify(vma, address, pte)) {
1198 ret = SWAP_FAIL;
1199 goto out_unmap;
1200 }
1201 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 /* Nuke the page table entry. */
1204 flush_cache_page(vma, address, page_to_pfn(page));
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001205 pteval = ptep_clear_flush_notify(vma, address, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206
1207 /* Move the dirty bit to the physical page now the pte is gone. */
1208 if (pte_dirty(pteval))
1209 set_page_dirty(page);
1210
Hugh Dickins365e9c872005-10-29 18:16:18 -07001211 /* Update high watermark before we lower rss */
1212 update_hiwater_rss(mm);
1213
Andi Kleen888b9f72009-09-16 11:50:11 +02001214 if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) {
1215 if (PageAnon(page))
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001216 dec_mm_counter(mm, MM_ANONPAGES);
Andi Kleen888b9f72009-09-16 11:50:11 +02001217 else
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001218 dec_mm_counter(mm, MM_FILEPAGES);
Andi Kleen888b9f72009-09-16 11:50:11 +02001219 set_pte_at(mm, address, pte,
1220 swp_entry_to_pte(make_hwpoison_entry(page)));
1221 } else if (PageAnon(page)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001222 swp_entry_t entry = { .val = page_private(page) };
Christoph Lameter06972122006-06-23 02:03:35 -07001223
1224 if (PageSwapCache(page)) {
1225 /*
1226 * Store the swap location in the pte.
1227 * See handle_pte_fault() ...
1228 */
Hugh Dickins570a335b2009-12-14 17:58:46 -08001229 if (swap_duplicate(entry) < 0) {
1230 set_pte_at(mm, address, pte, pteval);
1231 ret = SWAP_FAIL;
1232 goto out_unmap;
1233 }
Christoph Lameter06972122006-06-23 02:03:35 -07001234 if (list_empty(&mm->mmlist)) {
1235 spin_lock(&mmlist_lock);
1236 if (list_empty(&mm->mmlist))
1237 list_add(&mm->mmlist, &init_mm.mmlist);
1238 spin_unlock(&mmlist_lock);
1239 }
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001240 dec_mm_counter(mm, MM_ANONPAGES);
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08001241 inc_mm_counter(mm, MM_SWAPENTS);
KOSAKI Motohiro64cdd542009-01-06 14:39:16 -08001242 } else if (PAGE_MIGRATION) {
Christoph Lameter06972122006-06-23 02:03:35 -07001243 /*
1244 * Store the pfn of the page in a special migration
1245 * pte. do_swap_page() will wait until the migration
1246 * pte is removed and then restart fault handling.
1247 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001248 BUG_ON(TTU_ACTION(flags) != TTU_MIGRATION);
Christoph Lameter06972122006-06-23 02:03:35 -07001249 entry = make_migration_entry(page, pte_write(pteval));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 }
1251 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
1252 BUG_ON(pte_file(*pte));
Andi Kleen14fa31b2009-09-16 11:50:10 +02001253 } else if (PAGE_MIGRATION && (TTU_ACTION(flags) == TTU_MIGRATION)) {
Christoph Lameter04e62a22006-06-23 02:03:38 -07001254 /* Establish migration entry for a file page */
1255 swp_entry_t entry;
1256 entry = make_migration_entry(page, pte_write(pteval));
1257 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
1258 } else
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001259 dec_mm_counter(mm, MM_FILEPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260
Hugh Dickinsedc315f2009-01-06 14:40:11 -08001261 page_remove_rmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 page_cache_release(page);
1263
1264out_unmap:
Hugh Dickinsc0718802005-10-29 18:16:31 -07001265 pte_unmap_unlock(pte, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266out:
1267 return ret;
KOSAKI Motohirocaed0f482009-12-14 17:59:45 -08001268
1269out_mlock:
1270 pte_unmap_unlock(pte, ptl);
1271
1272
1273 /*
1274 * We need mmap_sem locking, Otherwise VM_LOCKED check makes
1275 * unstable result and race. Plus, We can't wait here because
Peter Zijlstra2b575eb2011-05-24 17:12:11 -07001276 * we now hold anon_vma->mutex or mapping->i_mmap_mutex.
KOSAKI Motohirocaed0f482009-12-14 17:59:45 -08001277 * if trylock failed, the page remain in evictable lru and later
1278 * vmscan could retry to move the page to unevictable lru if the
1279 * page is actually mlocked.
1280 */
1281 if (down_read_trylock(&vma->vm_mm->mmap_sem)) {
1282 if (vma->vm_flags & VM_LOCKED) {
1283 mlock_vma_page(page);
1284 ret = SWAP_MLOCK;
1285 }
1286 up_read(&vma->vm_mm->mmap_sem);
1287 }
1288 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289}
1290
1291/*
1292 * objrmap doesn't work for nonlinear VMAs because the assumption that
1293 * offset-into-file correlates with offset-into-virtual-addresses does not hold.
1294 * Consequently, given a particular page and its ->index, we cannot locate the
1295 * ptes which are mapping that page without an exhaustive linear search.
1296 *
1297 * So what this code does is a mini "virtual scan" of each nonlinear VMA which
1298 * maps the file to which the target page belongs. The ->vm_private_data field
1299 * holds the current cursor into that scan. Successive searches will circulate
1300 * around the vma's virtual address space.
1301 *
1302 * So as more replacement pressure is applied to the pages in a nonlinear VMA,
1303 * more scanning pressure is placed against them as well. Eventually pages
1304 * will become fully unmapped and are eligible for eviction.
1305 *
1306 * For very sparsely populated VMAs this is a little inefficient - chances are
1307 * there there won't be many ptes located within the scan cluster. In this case
1308 * maybe we could scan further - to the end of the pte page, perhaps.
Nick Pigginb291f002008-10-18 20:26:44 -07001309 *
1310 * Mlocked pages: check VM_LOCKED under mmap_sem held for read, if we can
1311 * acquire it without blocking. If vma locked, mlock the pages in the cluster,
1312 * rather than unmapping them. If we encounter the "check_page" that vmscan is
1313 * trying to unmap, return SWAP_MLOCK, else default SWAP_AGAIN.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 */
1315#define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE)
1316#define CLUSTER_MASK (~(CLUSTER_SIZE - 1))
1317
Nick Pigginb291f002008-10-18 20:26:44 -07001318static int try_to_unmap_cluster(unsigned long cursor, unsigned int *mapcount,
1319 struct vm_area_struct *vma, struct page *check_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320{
1321 struct mm_struct *mm = vma->vm_mm;
1322 pgd_t *pgd;
1323 pud_t *pud;
1324 pmd_t *pmd;
Hugh Dickinsc0718802005-10-29 18:16:31 -07001325 pte_t *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 pte_t pteval;
Hugh Dickinsc0718802005-10-29 18:16:31 -07001327 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 struct page *page;
1329 unsigned long address;
1330 unsigned long end;
Nick Pigginb291f002008-10-18 20:26:44 -07001331 int ret = SWAP_AGAIN;
1332 int locked_vma = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 address = (vma->vm_start + cursor) & CLUSTER_MASK;
1335 end = address + CLUSTER_SIZE;
1336 if (address < vma->vm_start)
1337 address = vma->vm_start;
1338 if (end > vma->vm_end)
1339 end = vma->vm_end;
1340
1341 pgd = pgd_offset(mm, address);
1342 if (!pgd_present(*pgd))
Nick Pigginb291f002008-10-18 20:26:44 -07001343 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344
1345 pud = pud_offset(pgd, address);
1346 if (!pud_present(*pud))
Nick Pigginb291f002008-10-18 20:26:44 -07001347 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348
1349 pmd = pmd_offset(pud, address);
1350 if (!pmd_present(*pmd))
Nick Pigginb291f002008-10-18 20:26:44 -07001351 return ret;
1352
1353 /*
Hugh Dickinsaf8e3352009-12-14 17:58:59 -08001354 * If we can acquire the mmap_sem for read, and vma is VM_LOCKED,
Nick Pigginb291f002008-10-18 20:26:44 -07001355 * keep the sem while scanning the cluster for mlocking pages.
1356 */
Hugh Dickinsaf8e3352009-12-14 17:58:59 -08001357 if (down_read_trylock(&vma->vm_mm->mmap_sem)) {
Nick Pigginb291f002008-10-18 20:26:44 -07001358 locked_vma = (vma->vm_flags & VM_LOCKED);
1359 if (!locked_vma)
1360 up_read(&vma->vm_mm->mmap_sem); /* don't need it */
1361 }
Hugh Dickinsc0718802005-10-29 18:16:31 -07001362
1363 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364
Hugh Dickins365e9c872005-10-29 18:16:18 -07001365 /* Update high watermark before we lower rss */
1366 update_hiwater_rss(mm);
1367
Hugh Dickinsc0718802005-10-29 18:16:31 -07001368 for (; address < end; pte++, address += PAGE_SIZE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 if (!pte_present(*pte))
1370 continue;
Linus Torvalds6aab3412005-11-28 14:34:23 -08001371 page = vm_normal_page(vma, address, *pte);
1372 BUG_ON(!page || PageAnon(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373
Nick Pigginb291f002008-10-18 20:26:44 -07001374 if (locked_vma) {
1375 mlock_vma_page(page); /* no-op if already mlocked */
1376 if (page == check_page)
1377 ret = SWAP_MLOCK;
1378 continue; /* don't unmap */
1379 }
1380
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001381 if (ptep_clear_flush_young_notify(vma, address, pte))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 continue;
1383
1384 /* Nuke the page table entry. */
Ben Collinseca35132005-11-29 11:45:26 -08001385 flush_cache_page(vma, address, pte_pfn(*pte));
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001386 pteval = ptep_clear_flush_notify(vma, address, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387
1388 /* If nonlinear, store the file page offset in the pte. */
1389 if (page->index != linear_page_index(vma, address))
1390 set_pte_at(mm, address, pte, pgoff_to_pte(page->index));
1391
1392 /* Move the dirty bit to the physical page now the pte is gone. */
1393 if (pte_dirty(pteval))
1394 set_page_dirty(page);
1395
Hugh Dickinsedc315f2009-01-06 14:40:11 -08001396 page_remove_rmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 page_cache_release(page);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001398 dec_mm_counter(mm, MM_FILEPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 (*mapcount)--;
1400 }
Hugh Dickinsc0718802005-10-29 18:16:31 -07001401 pte_unmap_unlock(pte - 1, ptl);
Nick Pigginb291f002008-10-18 20:26:44 -07001402 if (locked_vma)
1403 up_read(&vma->vm_mm->mmap_sem);
1404 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405}
1406
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001407bool is_vma_temporary_stack(struct vm_area_struct *vma)
Mel Gormana8bef8ff2010-05-24 14:32:24 -07001408{
1409 int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP);
1410
1411 if (!maybe_stack)
1412 return false;
1413
1414 if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) ==
1415 VM_STACK_INCOMPLETE_SETUP)
1416 return true;
1417
1418 return false;
1419}
1420
Nick Pigginb291f002008-10-18 20:26:44 -07001421/**
1422 * try_to_unmap_anon - unmap or unlock anonymous page using the object-based
1423 * rmap method
1424 * @page: the page to unmap/unlock
Huang Shijie8051be52009-12-14 17:58:50 -08001425 * @flags: action and flags
Nick Pigginb291f002008-10-18 20:26:44 -07001426 *
1427 * Find all the mappings of a page using the mapping pointer and the vma chains
1428 * contained in the anon_vma struct it points to.
1429 *
1430 * This function is only called from try_to_unmap/try_to_munlock for
1431 * anonymous pages.
1432 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1433 * where the page was found will be held for write. So, we won't recheck
1434 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1435 * 'LOCKED.
1436 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001437static int try_to_unmap_anon(struct page *page, enum ttu_flags flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438{
1439 struct anon_vma *anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08001440 struct anon_vma_chain *avc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 int ret = SWAP_AGAIN;
Nick Pigginb291f002008-10-18 20:26:44 -07001442
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 anon_vma = page_lock_anon_vma(page);
1444 if (!anon_vma)
1445 return ret;
1446
Rik van Riel5beb4932010-03-05 13:42:07 -08001447 list_for_each_entry(avc, &anon_vma->head, same_anon_vma) {
1448 struct vm_area_struct *vma = avc->vma;
Mel Gormana8bef8ff2010-05-24 14:32:24 -07001449 unsigned long address;
1450
1451 /*
1452 * During exec, a temporary VMA is setup and later moved.
1453 * The VMA is moved under the anon_vma lock but not the
1454 * page tables leading to a race where migration cannot
1455 * find the migration ptes. Rather than increasing the
1456 * locking requirements of exec(), migration skips
1457 * temporary VMAs until after exec() completes.
1458 */
1459 if (PAGE_MIGRATION && (flags & TTU_MIGRATION) &&
1460 is_vma_temporary_stack(vma))
1461 continue;
1462
1463 address = vma_address(page, vma);
Hugh Dickins1cb17292009-12-14 17:59:01 -08001464 if (address == -EFAULT)
1465 continue;
1466 ret = try_to_unmap_one(page, vma, address, flags);
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001467 if (ret != SWAP_AGAIN || !page_mapped(page))
1468 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 }
Oleg Nesterov34bbd702007-02-28 20:13:49 -08001470
1471 page_unlock_anon_vma(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 return ret;
1473}
1474
1475/**
Nick Pigginb291f002008-10-18 20:26:44 -07001476 * try_to_unmap_file - unmap/unlock file page using the object-based rmap method
1477 * @page: the page to unmap/unlock
Andi Kleen14fa31b2009-09-16 11:50:10 +02001478 * @flags: action and flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 *
1480 * Find all the mappings of a page using the mapping pointer and the vma chains
1481 * contained in the address_space struct it points to.
1482 *
Nick Pigginb291f002008-10-18 20:26:44 -07001483 * This function is only called from try_to_unmap/try_to_munlock for
1484 * object-based pages.
1485 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1486 * where the page was found will be held for write. So, we won't recheck
1487 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1488 * 'LOCKED.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001490static int try_to_unmap_file(struct page *page, enum ttu_flags flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491{
1492 struct address_space *mapping = page->mapping;
1493 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
1494 struct vm_area_struct *vma;
1495 struct prio_tree_iter iter;
1496 int ret = SWAP_AGAIN;
1497 unsigned long cursor;
1498 unsigned long max_nl_cursor = 0;
1499 unsigned long max_nl_size = 0;
1500 unsigned int mapcount;
1501
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07001502 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
Hugh Dickins1cb17292009-12-14 17:59:01 -08001504 unsigned long address = vma_address(page, vma);
1505 if (address == -EFAULT)
1506 continue;
1507 ret = try_to_unmap_one(page, vma, address, flags);
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001508 if (ret != SWAP_AGAIN || !page_mapped(page))
1509 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 }
1511
1512 if (list_empty(&mapping->i_mmap_nonlinear))
1513 goto out;
1514
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001515 /*
1516 * We don't bother to try to find the munlocked page in nonlinears.
1517 * It's costly. Instead, later, page reclaim logic may call
1518 * try_to_unmap(TTU_MUNLOCK) and recover PG_mlocked lazily.
1519 */
1520 if (TTU_ACTION(flags) == TTU_MUNLOCK)
1521 goto out;
1522
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
1524 shared.vm_set.list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 cursor = (unsigned long) vma->vm_private_data;
1526 if (cursor > max_nl_cursor)
1527 max_nl_cursor = cursor;
1528 cursor = vma->vm_end - vma->vm_start;
1529 if (cursor > max_nl_size)
1530 max_nl_size = cursor;
1531 }
1532
Nick Pigginb291f002008-10-18 20:26:44 -07001533 if (max_nl_size == 0) { /* all nonlinears locked or reserved ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 ret = SWAP_FAIL;
1535 goto out;
1536 }
1537
1538 /*
1539 * We don't try to search for this page in the nonlinear vmas,
1540 * and page_referenced wouldn't have found it anyway. Instead
1541 * just walk the nonlinear vmas trying to age and unmap some.
1542 * The mapcount of the page we came in with is irrelevant,
1543 * but even so use it as a guide to how hard we should try?
1544 */
1545 mapcount = page_mapcount(page);
1546 if (!mapcount)
1547 goto out;
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07001548 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549
1550 max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK;
1551 if (max_nl_cursor == 0)
1552 max_nl_cursor = CLUSTER_SIZE;
1553
1554 do {
1555 list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
1556 shared.vm_set.list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 cursor = (unsigned long) vma->vm_private_data;
Hugh Dickins839b9682005-09-03 15:54:43 -07001558 while ( cursor < max_nl_cursor &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 cursor < vma->vm_end - vma->vm_start) {
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001560 if (try_to_unmap_cluster(cursor, &mapcount,
1561 vma, page) == SWAP_MLOCK)
1562 ret = SWAP_MLOCK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 cursor += CLUSTER_SIZE;
1564 vma->vm_private_data = (void *) cursor;
1565 if ((int)mapcount <= 0)
1566 goto out;
1567 }
1568 vma->vm_private_data = (void *) max_nl_cursor;
1569 }
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07001570 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 max_nl_cursor += CLUSTER_SIZE;
1572 } while (max_nl_cursor <= max_nl_size);
1573
1574 /*
1575 * Don't loop forever (perhaps all the remaining pages are
1576 * in locked vmas). Reset cursor on all unreserved nonlinear
1577 * vmas, now forgetting on which ones it had fallen behind.
1578 */
Hugh Dickins101d2be2005-11-21 21:32:16 -08001579 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
1580 vma->vm_private_data = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581out:
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07001582 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 return ret;
1584}
1585
1586/**
1587 * try_to_unmap - try to remove all page table mappings to a page
1588 * @page: the page to get unmapped
Andi Kleen14fa31b2009-09-16 11:50:10 +02001589 * @flags: action and flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 *
1591 * Tries to remove all the page table entries which are mapping this
1592 * page, used in the pageout path. Caller must hold the page lock.
1593 * Return values are:
1594 *
1595 * SWAP_SUCCESS - we succeeded in removing all mappings
1596 * SWAP_AGAIN - we missed a mapping, try again later
1597 * SWAP_FAIL - the page is unswappable
Nick Pigginb291f002008-10-18 20:26:44 -07001598 * SWAP_MLOCK - page is mlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001600int try_to_unmap(struct page *page, enum ttu_flags flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601{
1602 int ret;
1603
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 BUG_ON(!PageLocked(page));
Andrea Arcangeli91600e92011-01-13 15:47:16 -08001605 VM_BUG_ON(!PageHuge(page) && PageTransHuge(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606
Hugh Dickins5ad64682009-12-14 17:59:24 -08001607 if (unlikely(PageKsm(page)))
1608 ret = try_to_unmap_ksm(page, flags);
1609 else if (PageAnon(page))
Andi Kleen14fa31b2009-09-16 11:50:10 +02001610 ret = try_to_unmap_anon(page, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 else
Andi Kleen14fa31b2009-09-16 11:50:10 +02001612 ret = try_to_unmap_file(page, flags);
Nick Pigginb291f002008-10-18 20:26:44 -07001613 if (ret != SWAP_MLOCK && !page_mapped(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 ret = SWAP_SUCCESS;
1615 return ret;
1616}
Nikita Danilov81b40822005-05-01 08:58:36 -07001617
Nick Pigginb291f002008-10-18 20:26:44 -07001618/**
1619 * try_to_munlock - try to munlock a page
1620 * @page: the page to be munlocked
1621 *
1622 * Called from munlock code. Checks all of the VMAs mapping the page
1623 * to make sure nobody else has this page mlocked. The page will be
1624 * returned with PG_mlocked cleared if no other vmas have it mlocked.
1625 *
1626 * Return values are:
1627 *
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001628 * SWAP_AGAIN - no vma is holding page mlocked, or,
Nick Pigginb291f002008-10-18 20:26:44 -07001629 * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem
Hugh Dickins5ad64682009-12-14 17:59:24 -08001630 * SWAP_FAIL - page cannot be located at present
Nick Pigginb291f002008-10-18 20:26:44 -07001631 * SWAP_MLOCK - page is now mlocked.
1632 */
1633int try_to_munlock(struct page *page)
1634{
1635 VM_BUG_ON(!PageLocked(page) || PageLRU(page));
1636
Hugh Dickins5ad64682009-12-14 17:59:24 -08001637 if (unlikely(PageKsm(page)))
1638 return try_to_unmap_ksm(page, TTU_MUNLOCK);
1639 else if (PageAnon(page))
Andi Kleen14fa31b2009-09-16 11:50:10 +02001640 return try_to_unmap_anon(page, TTU_MUNLOCK);
Nick Pigginb291f002008-10-18 20:26:44 -07001641 else
Andi Kleen14fa31b2009-09-16 11:50:10 +02001642 return try_to_unmap_file(page, TTU_MUNLOCK);
Nick Pigginb291f002008-10-18 20:26:44 -07001643}
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001644
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001645void __put_anon_vma(struct anon_vma *anon_vma)
Rik van Riel76545062010-08-09 17:18:41 -07001646{
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001647 struct anon_vma *root = anon_vma->root;
Rik van Riel76545062010-08-09 17:18:41 -07001648
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001649 if (root != anon_vma && atomic_dec_and_test(&root->refcount))
1650 anon_vma_free(root);
Rik van Riel76545062010-08-09 17:18:41 -07001651
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001652 anon_vma_free(anon_vma);
Rik van Riel76545062010-08-09 17:18:41 -07001653}
Rik van Riel76545062010-08-09 17:18:41 -07001654
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001655#ifdef CONFIG_MIGRATION
1656/*
1657 * rmap_walk() and its helpers rmap_walk_anon() and rmap_walk_file():
1658 * Called by migrate.c to remove migration ptes, but might be used more later.
1659 */
1660static int rmap_walk_anon(struct page *page, int (*rmap_one)(struct page *,
1661 struct vm_area_struct *, unsigned long, void *), void *arg)
1662{
1663 struct anon_vma *anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08001664 struct anon_vma_chain *avc;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001665 int ret = SWAP_AGAIN;
1666
1667 /*
1668 * Note: remove_migration_ptes() cannot use page_lock_anon_vma()
1669 * because that depends on page_mapped(); but not all its usages
Mel Gorman3f6c8272010-05-24 14:32:17 -07001670 * are holding mmap_sem. Users without mmap_sem are required to
1671 * take a reference count to prevent the anon_vma disappearing
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001672 */
1673 anon_vma = page_anon_vma(page);
1674 if (!anon_vma)
1675 return ret;
Rik van Rielcba48b92010-08-09 17:18:38 -07001676 anon_vma_lock(anon_vma);
Rik van Riel5beb4932010-03-05 13:42:07 -08001677 list_for_each_entry(avc, &anon_vma->head, same_anon_vma) {
1678 struct vm_area_struct *vma = avc->vma;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001679 unsigned long address = vma_address(page, vma);
1680 if (address == -EFAULT)
1681 continue;
1682 ret = rmap_one(page, vma, address, arg);
1683 if (ret != SWAP_AGAIN)
1684 break;
1685 }
Rik van Rielcba48b92010-08-09 17:18:38 -07001686 anon_vma_unlock(anon_vma);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001687 return ret;
1688}
1689
1690static int rmap_walk_file(struct page *page, int (*rmap_one)(struct page *,
1691 struct vm_area_struct *, unsigned long, void *), void *arg)
1692{
1693 struct address_space *mapping = page->mapping;
1694 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
1695 struct vm_area_struct *vma;
1696 struct prio_tree_iter iter;
1697 int ret = SWAP_AGAIN;
1698
1699 if (!mapping)
1700 return ret;
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07001701 mutex_lock(&mapping->i_mmap_mutex);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001702 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
1703 unsigned long address = vma_address(page, vma);
1704 if (address == -EFAULT)
1705 continue;
1706 ret = rmap_one(page, vma, address, arg);
1707 if (ret != SWAP_AGAIN)
1708 break;
1709 }
1710 /*
1711 * No nonlinear handling: being always shared, nonlinear vmas
1712 * never contain migration ptes. Decide what to do about this
1713 * limitation to linear when we need rmap_walk() on nonlinear.
1714 */
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07001715 mutex_unlock(&mapping->i_mmap_mutex);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001716 return ret;
1717}
1718
1719int rmap_walk(struct page *page, int (*rmap_one)(struct page *,
1720 struct vm_area_struct *, unsigned long, void *), void *arg)
1721{
1722 VM_BUG_ON(!PageLocked(page));
1723
1724 if (unlikely(PageKsm(page)))
1725 return rmap_walk_ksm(page, rmap_one, arg);
1726 else if (PageAnon(page))
1727 return rmap_walk_anon(page, rmap_one, arg);
1728 else
1729 return rmap_walk_file(page, rmap_one, arg);
1730}
1731#endif /* CONFIG_MIGRATION */
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001732
Naoya Horiguchie3390f62010-06-15 13:18:13 +09001733#ifdef CONFIG_HUGETLB_PAGE
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001734/*
1735 * The following three functions are for anonymous (private mapped) hugepages.
1736 * Unlike common anonymous pages, anonymous hugepages have no accounting code
1737 * and no lru code, because we handle hugepages differently from common pages.
1738 */
1739static void __hugepage_set_anon_rmap(struct page *page,
1740 struct vm_area_struct *vma, unsigned long address, int exclusive)
1741{
1742 struct anon_vma *anon_vma = vma->anon_vma;
Naoya Horiguchi433abed2010-09-10 13:23:03 +09001743
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001744 BUG_ON(!anon_vma);
Naoya Horiguchi433abed2010-09-10 13:23:03 +09001745
1746 if (PageAnon(page))
1747 return;
1748 if (!exclusive)
1749 anon_vma = anon_vma->root;
1750
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001751 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
1752 page->mapping = (struct address_space *) anon_vma;
1753 page->index = linear_page_index(vma, address);
1754}
1755
1756void hugepage_add_anon_rmap(struct page *page,
1757 struct vm_area_struct *vma, unsigned long address)
1758{
1759 struct anon_vma *anon_vma = vma->anon_vma;
1760 int first;
Naoya Horiguchia850ea32010-09-10 13:23:06 +09001761
1762 BUG_ON(!PageLocked(page));
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001763 BUG_ON(!anon_vma);
Hugh Dickins5dbe0af2011-05-28 13:17:04 -07001764 /* address might be in next vma when migration races vma_adjust */
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001765 first = atomic_inc_and_test(&page->_mapcount);
1766 if (first)
1767 __hugepage_set_anon_rmap(page, vma, address, 0);
1768}
1769
1770void hugepage_add_new_anon_rmap(struct page *page,
1771 struct vm_area_struct *vma, unsigned long address)
1772{
1773 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
1774 atomic_set(&page->_mapcount, 0);
1775 __hugepage_set_anon_rmap(page, vma, address, 1);
1776}
Naoya Horiguchie3390f62010-06-15 13:18:13 +09001777#endif /* CONFIG_HUGETLB_PAGE */