blob: a4fd3680038be499a47cf747d1ba657646e5d36c [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * mm/rmap.c - physical to virtual reverse mappings
3 *
4 * Copyright 2001, Rik van Riel <riel@conectiva.com.br>
5 * Released under the General Public License (GPL).
6 *
7 * Simple, low overhead reverse mapping scheme.
8 * Please try to keep this thing as modular as possible.
9 *
10 * Provides methods for unmapping each kind of mapped page:
11 * the anon methods track anonymous pages, and
12 * the file methods track pages belonging to an inode.
13 *
14 * Original design by Rik van Riel <riel@conectiva.com.br> 2001
15 * File methods by Dave McCracken <dmccr@us.ibm.com> 2003, 2004
16 * Anonymous methods by Andrea Arcangeli <andrea@suse.de> 2004
Hugh Dickins98f32602009-05-21 20:33:58 +010017 * Contributions by Hugh Dickins 2003, 2004
Linus Torvalds1da177e2005-04-16 15:20:36 -070018 */
19
20/*
21 * Lock ordering in mm:
22 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080023 * inode->i_mutex (while writing or truncating, not reading or faulting)
Nick Piggin82591e62006-10-19 23:29:10 -070024 * mm->mmap_sem
25 * page->flags PG_locked (lock_page)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -070026 * mapping->i_mmap_mutex
Peter Zijlstra2b575eb2011-05-24 17:12:11 -070027 * anon_vma->mutex
Nick Piggin82591e62006-10-19 23:29:10 -070028 * mm->page_table_lock or pte_lock
29 * zone->lru_lock (in mark_page_accessed, isolate_lru_page)
30 * swap_lock (in swap_duplicate, swap_info_get)
31 * mmlist_lock (in mmput, drain_mmlist and others)
32 * mapping->private_lock (in __set_page_dirty_buffers)
Dave Chinner250df6e2011-03-22 22:23:36 +110033 * inode->i_lock (in set_page_dirty's __mark_inode_dirty)
Christoph Hellwigf758eea2011-04-21 18:19:44 -060034 * bdi.wb->list_lock (in set_page_dirty's __mark_inode_dirty)
Nick Piggin82591e62006-10-19 23:29:10 -070035 * sb_lock (within inode_lock in fs/fs-writeback.c)
36 * mapping->tree_lock (widely used, in set_page_dirty,
37 * in arch-dependent flush_dcache_mmap_lock,
Christoph Hellwigf758eea2011-04-21 18:19:44 -060038 * within bdi.wb->list_lock in __sync_single_inode)
Andi Kleen6a460792009-09-16 11:50:15 +020039 *
Peter Zijlstra9b679322011-06-27 16:18:09 -070040 * anon_vma->mutex,mapping->i_mutex (memory_failure, collect_procs_anon)
41 * ->tasklist_lock
Andi Kleen6a460792009-09-16 11:50:15 +020042 * pte map lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070043 */
44
45#include <linux/mm.h>
46#include <linux/pagemap.h>
47#include <linux/swap.h>
48#include <linux/swapops.h>
49#include <linux/slab.h>
50#include <linux/init.h>
Hugh Dickins5ad64682009-12-14 17:59:24 -080051#include <linux/ksm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#include <linux/rmap.h>
53#include <linux/rcupdate.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040054#include <linux/export.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080055#include <linux/memcontrol.h>
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070056#include <linux/mmu_notifier.h>
KOSAKI Motohiro64cdd542009-01-06 14:39:16 -080057#include <linux/migrate.h>
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +090058#include <linux/hugetlb.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070059
60#include <asm/tlbflush.h>
61
Nick Pigginb291f002008-10-18 20:26:44 -070062#include "internal.h"
63
Adrian Bunkfdd2e5f2008-10-18 20:28:38 -070064static struct kmem_cache *anon_vma_cachep;
Rik van Riel5beb4932010-03-05 13:42:07 -080065static struct kmem_cache *anon_vma_chain_cachep;
Adrian Bunkfdd2e5f2008-10-18 20:28:38 -070066
67static inline struct anon_vma *anon_vma_alloc(void)
68{
Peter Zijlstra01d8b202011-03-22 16:32:49 -070069 struct anon_vma *anon_vma;
70
71 anon_vma = kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL);
72 if (anon_vma) {
73 atomic_set(&anon_vma->refcount, 1);
74 /*
75 * Initialise the anon_vma root to point to itself. If called
76 * from fork, the root will be reset to the parents anon_vma.
77 */
78 anon_vma->root = anon_vma;
79 }
80
81 return anon_vma;
Adrian Bunkfdd2e5f2008-10-18 20:28:38 -070082}
83
Peter Zijlstra01d8b202011-03-22 16:32:49 -070084static inline void anon_vma_free(struct anon_vma *anon_vma)
Adrian Bunkfdd2e5f2008-10-18 20:28:38 -070085{
Peter Zijlstra01d8b202011-03-22 16:32:49 -070086 VM_BUG_ON(atomic_read(&anon_vma->refcount));
Peter Zijlstra88c22082011-05-24 17:12:13 -070087
88 /*
89 * Synchronize against page_lock_anon_vma() such that
90 * we can safely hold the lock without the anon_vma getting
91 * freed.
92 *
93 * Relies on the full mb implied by the atomic_dec_and_test() from
94 * put_anon_vma() against the acquire barrier implied by
95 * mutex_trylock() from page_lock_anon_vma(). This orders:
96 *
97 * page_lock_anon_vma() VS put_anon_vma()
98 * mutex_trylock() atomic_dec_and_test()
99 * LOCK MB
100 * atomic_read() mutex_is_locked()
101 *
102 * LOCK should suffice since the actual taking of the lock must
103 * happen _before_ what follows.
104 */
105 if (mutex_is_locked(&anon_vma->root->mutex)) {
106 anon_vma_lock(anon_vma);
107 anon_vma_unlock(anon_vma);
108 }
109
Adrian Bunkfdd2e5f2008-10-18 20:28:38 -0700110 kmem_cache_free(anon_vma_cachep, anon_vma);
111}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112
Linus Torvaldsdd347392011-06-17 19:05:36 -0700113static inline struct anon_vma_chain *anon_vma_chain_alloc(gfp_t gfp)
Rik van Riel5beb4932010-03-05 13:42:07 -0800114{
Linus Torvaldsdd347392011-06-17 19:05:36 -0700115 return kmem_cache_alloc(anon_vma_chain_cachep, gfp);
Rik van Riel5beb4932010-03-05 13:42:07 -0800116}
117
Namhyung Kime574b5f2010-10-26 14:22:02 -0700118static void anon_vma_chain_free(struct anon_vma_chain *anon_vma_chain)
Rik van Riel5beb4932010-03-05 13:42:07 -0800119{
120 kmem_cache_free(anon_vma_chain_cachep, anon_vma_chain);
121}
122
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700123/**
124 * anon_vma_prepare - attach an anon_vma to a memory region
125 * @vma: the memory region in question
126 *
127 * This makes sure the memory mapping described by 'vma' has
128 * an 'anon_vma' attached to it, so that we can associate the
129 * anonymous pages mapped into it with that anon_vma.
130 *
131 * The common case will be that we already have one, but if
Figo.zhang23a07902010-12-27 15:14:06 +0100132 * not we either need to find an adjacent mapping that we
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700133 * can re-use the anon_vma from (very common when the only
134 * reason for splitting a vma has been mprotect()), or we
135 * allocate a new one.
136 *
137 * Anon-vma allocations are very subtle, because we may have
138 * optimistically looked up an anon_vma in page_lock_anon_vma()
139 * and that may actually touch the spinlock even in the newly
140 * allocated vma (it depends on RCU to make sure that the
141 * anon_vma isn't actually destroyed).
142 *
143 * As a result, we need to do proper anon_vma locking even
144 * for the new allocation. At the same time, we do not want
145 * to do any locking for the common case of already having
146 * an anon_vma.
147 *
148 * This must be called with the mmap_sem held for reading.
149 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150int anon_vma_prepare(struct vm_area_struct *vma)
151{
152 struct anon_vma *anon_vma = vma->anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -0800153 struct anon_vma_chain *avc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154
155 might_sleep();
156 if (unlikely(!anon_vma)) {
157 struct mm_struct *mm = vma->vm_mm;
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700158 struct anon_vma *allocated;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159
Linus Torvaldsdd347392011-06-17 19:05:36 -0700160 avc = anon_vma_chain_alloc(GFP_KERNEL);
Rik van Riel5beb4932010-03-05 13:42:07 -0800161 if (!avc)
162 goto out_enomem;
163
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 anon_vma = find_mergeable_anon_vma(vma);
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700165 allocated = NULL;
166 if (!anon_vma) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 anon_vma = anon_vma_alloc();
168 if (unlikely(!anon_vma))
Rik van Riel5beb4932010-03-05 13:42:07 -0800169 goto out_enomem_free_avc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 allocated = anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 }
172
Rik van Rielcba48b92010-08-09 17:18:38 -0700173 anon_vma_lock(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 /* page_table_lock to protect against threads */
175 spin_lock(&mm->page_table_lock);
176 if (likely(!vma->anon_vma)) {
177 vma->anon_vma = anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -0800178 avc->anon_vma = anon_vma;
179 avc->vma = vma;
180 list_add(&avc->same_vma, &vma->anon_vma_chain);
Andrea Arcangeli26ba0cb2010-08-09 17:19:08 -0700181 list_add_tail(&avc->same_anon_vma, &anon_vma->head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 allocated = NULL;
Oleg Nesterov31f2b0e2010-04-23 13:18:01 -0400183 avc = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 }
185 spin_unlock(&mm->page_table_lock);
Rik van Rielcba48b92010-08-09 17:18:38 -0700186 anon_vma_unlock(anon_vma);
Oleg Nesterov31f2b0e2010-04-23 13:18:01 -0400187
188 if (unlikely(allocated))
Peter Zijlstra01d8b202011-03-22 16:32:49 -0700189 put_anon_vma(allocated);
Oleg Nesterov31f2b0e2010-04-23 13:18:01 -0400190 if (unlikely(avc))
Rik van Riel5beb4932010-03-05 13:42:07 -0800191 anon_vma_chain_free(avc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 }
193 return 0;
Rik van Riel5beb4932010-03-05 13:42:07 -0800194
195 out_enomem_free_avc:
196 anon_vma_chain_free(avc);
197 out_enomem:
198 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199}
200
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700201/*
202 * This is a useful helper function for locking the anon_vma root as
203 * we traverse the vma->anon_vma_chain, looping over anon_vma's that
204 * have the same vma.
205 *
206 * Such anon_vma's should have the same root, so you'd expect to see
207 * just a single mutex_lock for the whole traversal.
208 */
209static inline struct anon_vma *lock_anon_vma_root(struct anon_vma *root, struct anon_vma *anon_vma)
210{
211 struct anon_vma *new_root = anon_vma->root;
212 if (new_root != root) {
213 if (WARN_ON_ONCE(root))
214 mutex_unlock(&root->mutex);
215 root = new_root;
216 mutex_lock(&root->mutex);
217 }
218 return root;
219}
220
221static inline void unlock_anon_vma_root(struct anon_vma *root)
222{
223 if (root)
224 mutex_unlock(&root->mutex);
225}
226
Rik van Riel5beb4932010-03-05 13:42:07 -0800227static void anon_vma_chain_link(struct vm_area_struct *vma,
228 struct anon_vma_chain *avc,
229 struct anon_vma *anon_vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230{
Rik van Riel5beb4932010-03-05 13:42:07 -0800231 avc->vma = vma;
232 avc->anon_vma = anon_vma;
233 list_add(&avc->same_vma, &vma->anon_vma_chain);
234
Andrea Arcangeli05759d32011-01-13 15:46:53 -0800235 /*
236 * It's critical to add new vmas to the tail of the anon_vma,
237 * see comment in huge_memory.c:__split_huge_page().
238 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800239 list_add_tail(&avc->same_anon_vma, &anon_vma->head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240}
241
Rik van Riel5beb4932010-03-05 13:42:07 -0800242/*
243 * Attach the anon_vmas from src to dst.
244 * Returns 0 on success, -ENOMEM on failure.
245 */
246int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247{
Rik van Riel5beb4932010-03-05 13:42:07 -0800248 struct anon_vma_chain *avc, *pavc;
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700249 struct anon_vma *root = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250
Linus Torvalds646d87b2010-04-11 17:15:03 -0700251 list_for_each_entry_reverse(pavc, &src->anon_vma_chain, same_vma) {
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700252 struct anon_vma *anon_vma;
253
Linus Torvaldsdd347392011-06-17 19:05:36 -0700254 avc = anon_vma_chain_alloc(GFP_NOWAIT | __GFP_NOWARN);
255 if (unlikely(!avc)) {
256 unlock_anon_vma_root(root);
257 root = NULL;
258 avc = anon_vma_chain_alloc(GFP_KERNEL);
259 if (!avc)
260 goto enomem_failure;
261 }
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700262 anon_vma = pavc->anon_vma;
263 root = lock_anon_vma_root(root, anon_vma);
264 anon_vma_chain_link(dst, avc, anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 }
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700266 unlock_anon_vma_root(root);
Rik van Riel5beb4932010-03-05 13:42:07 -0800267 return 0;
268
269 enomem_failure:
270 unlink_anon_vmas(dst);
271 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272}
273
Rik van Riel5beb4932010-03-05 13:42:07 -0800274/*
275 * Attach vma to its own anon_vma, as well as to the anon_vmas that
276 * the corresponding VMA in the parent process is attached to.
277 * Returns 0 on success, non-zero on failure.
278 */
279int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280{
Rik van Riel5beb4932010-03-05 13:42:07 -0800281 struct anon_vma_chain *avc;
282 struct anon_vma *anon_vma;
283
284 /* Don't bother if the parent process has no anon_vma here. */
285 if (!pvma->anon_vma)
286 return 0;
287
288 /*
289 * First, attach the new VMA to the parent VMA's anon_vmas,
290 * so rmap can find non-COWed pages in child processes.
291 */
292 if (anon_vma_clone(vma, pvma))
293 return -ENOMEM;
294
295 /* Then add our own anon_vma. */
296 anon_vma = anon_vma_alloc();
297 if (!anon_vma)
298 goto out_error;
Linus Torvaldsdd347392011-06-17 19:05:36 -0700299 avc = anon_vma_chain_alloc(GFP_KERNEL);
Rik van Riel5beb4932010-03-05 13:42:07 -0800300 if (!avc)
301 goto out_error_free_anon_vma;
Rik van Riel5c341ee12010-08-09 17:18:39 -0700302
303 /*
304 * The root anon_vma's spinlock is the lock actually used when we
305 * lock any of the anon_vmas in this anon_vma tree.
306 */
307 anon_vma->root = pvma->anon_vma->root;
Rik van Riel76545062010-08-09 17:18:41 -0700308 /*
Peter Zijlstra01d8b202011-03-22 16:32:49 -0700309 * With refcounts, an anon_vma can stay around longer than the
310 * process it belongs to. The root anon_vma needs to be pinned until
311 * this anon_vma is freed, because the lock lives in the root.
Rik van Riel76545062010-08-09 17:18:41 -0700312 */
313 get_anon_vma(anon_vma->root);
Rik van Riel5beb4932010-03-05 13:42:07 -0800314 /* Mark this anon_vma as the one where our new (COWed) pages go. */
315 vma->anon_vma = anon_vma;
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700316 anon_vma_lock(anon_vma);
Rik van Riel5c341ee12010-08-09 17:18:39 -0700317 anon_vma_chain_link(vma, avc, anon_vma);
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700318 anon_vma_unlock(anon_vma);
Rik van Riel5beb4932010-03-05 13:42:07 -0800319
320 return 0;
321
322 out_error_free_anon_vma:
Peter Zijlstra01d8b202011-03-22 16:32:49 -0700323 put_anon_vma(anon_vma);
Rik van Riel5beb4932010-03-05 13:42:07 -0800324 out_error:
Rik van Riel4946d542010-04-05 12:13:33 -0400325 unlink_anon_vmas(vma);
Rik van Riel5beb4932010-03-05 13:42:07 -0800326 return -ENOMEM;
327}
328
Rik van Riel5beb4932010-03-05 13:42:07 -0800329void unlink_anon_vmas(struct vm_area_struct *vma)
330{
331 struct anon_vma_chain *avc, *next;
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200332 struct anon_vma *root = NULL;
Rik van Riel5beb4932010-03-05 13:42:07 -0800333
Rik van Riel5c341ee12010-08-09 17:18:39 -0700334 /*
335 * Unlink each anon_vma chained to the VMA. This list is ordered
336 * from newest to oldest, ensuring the root anon_vma gets freed last.
337 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800338 list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) {
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200339 struct anon_vma *anon_vma = avc->anon_vma;
340
341 root = lock_anon_vma_root(root, anon_vma);
342 list_del(&avc->same_anon_vma);
343
344 /*
345 * Leave empty anon_vmas on the list - we'll need
346 * to free them outside the lock.
347 */
348 if (list_empty(&anon_vma->head))
349 continue;
350
351 list_del(&avc->same_vma);
352 anon_vma_chain_free(avc);
353 }
354 unlock_anon_vma_root(root);
355
356 /*
357 * Iterate the list once more, it now only contains empty and unlinked
358 * anon_vmas, destroy them. Could not do before due to __put_anon_vma()
359 * needing to acquire the anon_vma->root->mutex.
360 */
361 list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) {
362 struct anon_vma *anon_vma = avc->anon_vma;
363
364 put_anon_vma(anon_vma);
365
Rik van Riel5beb4932010-03-05 13:42:07 -0800366 list_del(&avc->same_vma);
367 anon_vma_chain_free(avc);
368 }
369}
370
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700371static void anon_vma_ctor(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372{
Christoph Lametera35afb82007-05-16 22:10:57 -0700373 struct anon_vma *anon_vma = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374
Peter Zijlstra2b575eb2011-05-24 17:12:11 -0700375 mutex_init(&anon_vma->mutex);
Peter Zijlstra83813262011-03-22 16:32:48 -0700376 atomic_set(&anon_vma->refcount, 0);
Christoph Lametera35afb82007-05-16 22:10:57 -0700377 INIT_LIST_HEAD(&anon_vma->head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378}
379
380void __init anon_vma_init(void)
381{
382 anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma),
Paul Mundt20c2df82007-07-20 10:11:58 +0900383 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor);
Rik van Riel5beb4932010-03-05 13:42:07 -0800384 anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain, SLAB_PANIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385}
386
387/*
Peter Zijlstra6111e4c2011-05-24 17:12:08 -0700388 * Getting a lock on a stable anon_vma from a page off the LRU is tricky!
389 *
390 * Since there is no serialization what so ever against page_remove_rmap()
391 * the best this function can do is return a locked anon_vma that might
392 * have been relevant to this page.
393 *
394 * The page might have been remapped to a different anon_vma or the anon_vma
395 * returned may already be freed (and even reused).
396 *
Peter Zijlstrabc658c92011-05-29 10:33:44 +0200397 * In case it was remapped to a different anon_vma, the new anon_vma will be a
398 * child of the old anon_vma, and the anon_vma lifetime rules will therefore
399 * ensure that any anon_vma obtained from the page will still be valid for as
400 * long as we observe page_mapped() [ hence all those page_mapped() tests ].
401 *
Peter Zijlstra6111e4c2011-05-24 17:12:08 -0700402 * All users of this function must be very careful when walking the anon_vma
403 * chain and verify that the page in question is indeed mapped in it
404 * [ something equivalent to page_mapped_in_vma() ].
405 *
406 * Since anon_vma's slab is DESTROY_BY_RCU and we know from page_remove_rmap()
407 * that the anon_vma pointer from page->mapping is valid if there is a
408 * mapcount, we can dereference the anon_vma after observing those.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 */
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700410struct anon_vma *page_get_anon_vma(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411{
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700412 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 unsigned long anon_mapping;
414
415 rcu_read_lock();
Hugh Dickins80e148222009-12-14 17:59:29 -0800416 anon_mapping = (unsigned long) ACCESS_ONCE(page->mapping);
Hugh Dickins3ca7b3c2009-12-14 17:58:57 -0800417 if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 goto out;
419 if (!page_mapped(page))
420 goto out;
421
422 anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700423 if (!atomic_inc_not_zero(&anon_vma->refcount)) {
424 anon_vma = NULL;
425 goto out;
426 }
Hugh Dickinsf18194272010-08-25 23:12:54 -0700427
428 /*
429 * If this page is still mapped, then its anon_vma cannot have been
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700430 * freed. But if it has been unmapped, we have no security against the
431 * anon_vma structure being freed and reused (for another anon_vma:
432 * SLAB_DESTROY_BY_RCU guarantees that - so the atomic_inc_not_zero()
433 * above cannot corrupt).
Hugh Dickinsf18194272010-08-25 23:12:54 -0700434 */
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700435 if (!page_mapped(page)) {
436 put_anon_vma(anon_vma);
437 anon_vma = NULL;
438 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439out:
440 rcu_read_unlock();
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700441
442 return anon_vma;
443}
444
Peter Zijlstra88c22082011-05-24 17:12:13 -0700445/*
446 * Similar to page_get_anon_vma() except it locks the anon_vma.
447 *
448 * Its a little more complex as it tries to keep the fast path to a single
449 * atomic op -- the trylock. If we fail the trylock, we fall back to getting a
450 * reference like with page_get_anon_vma() and then block on the mutex.
451 */
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700452struct anon_vma *page_lock_anon_vma(struct page *page)
453{
Peter Zijlstra88c22082011-05-24 17:12:13 -0700454 struct anon_vma *anon_vma = NULL;
Hugh Dickinseee0f252011-05-28 13:20:21 -0700455 struct anon_vma *root_anon_vma;
Peter Zijlstra88c22082011-05-24 17:12:13 -0700456 unsigned long anon_mapping;
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700457
Peter Zijlstra88c22082011-05-24 17:12:13 -0700458 rcu_read_lock();
459 anon_mapping = (unsigned long) ACCESS_ONCE(page->mapping);
460 if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
461 goto out;
462 if (!page_mapped(page))
463 goto out;
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700464
Peter Zijlstra88c22082011-05-24 17:12:13 -0700465 anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
Hugh Dickinseee0f252011-05-28 13:20:21 -0700466 root_anon_vma = ACCESS_ONCE(anon_vma->root);
467 if (mutex_trylock(&root_anon_vma->mutex)) {
Peter Zijlstra88c22082011-05-24 17:12:13 -0700468 /*
Hugh Dickinseee0f252011-05-28 13:20:21 -0700469 * If the page is still mapped, then this anon_vma is still
470 * its anon_vma, and holding the mutex ensures that it will
Peter Zijlstrabc658c92011-05-29 10:33:44 +0200471 * not go away, see anon_vma_free().
Peter Zijlstra88c22082011-05-24 17:12:13 -0700472 */
Hugh Dickinseee0f252011-05-28 13:20:21 -0700473 if (!page_mapped(page)) {
474 mutex_unlock(&root_anon_vma->mutex);
Peter Zijlstra88c22082011-05-24 17:12:13 -0700475 anon_vma = NULL;
476 }
477 goto out;
478 }
479
480 /* trylock failed, we got to sleep */
481 if (!atomic_inc_not_zero(&anon_vma->refcount)) {
482 anon_vma = NULL;
483 goto out;
484 }
485
486 if (!page_mapped(page)) {
487 put_anon_vma(anon_vma);
488 anon_vma = NULL;
489 goto out;
490 }
491
492 /* we pinned the anon_vma, its safe to sleep */
493 rcu_read_unlock();
494 anon_vma_lock(anon_vma);
495
496 if (atomic_dec_and_test(&anon_vma->refcount)) {
497 /*
498 * Oops, we held the last refcount, release the lock
499 * and bail -- can't simply use put_anon_vma() because
500 * we'll deadlock on the anon_vma_lock() recursion.
501 */
502 anon_vma_unlock(anon_vma);
503 __put_anon_vma(anon_vma);
504 anon_vma = NULL;
505 }
506
507 return anon_vma;
508
509out:
510 rcu_read_unlock();
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700511 return anon_vma;
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800512}
513
Andi Kleen10be22d2009-09-16 11:50:04 +0200514void page_unlock_anon_vma(struct anon_vma *anon_vma)
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800515{
Rik van Rielcba48b92010-08-09 17:18:38 -0700516 anon_vma_unlock(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517}
518
519/*
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -0800520 * At what user virtual address is page expected in @vma?
521 * Returns virtual address or -EFAULT if page's index/offset is not
522 * within the range mapped the @vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800524inline unsigned long
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525vma_address(struct page *page, struct vm_area_struct *vma)
526{
527 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
528 unsigned long address;
529
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +0900530 if (unlikely(is_vm_hugetlb_page(vma)))
531 pgoff = page->index << huge_page_order(page_hstate(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
533 if (unlikely(address < vma->vm_start || address >= vma->vm_end)) {
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -0800534 /* page should be within @vma mapping range */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 return -EFAULT;
536 }
537 return address;
538}
539
540/*
Huang Shijiebf89c8c2009-10-01 15:44:04 -0700541 * At what user virtual address is page expected in vma?
Naoya Horiguchiab941e02010-05-11 14:06:55 -0700542 * Caller should check the page is actually part of the vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 */
544unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma)
545{
Andrea Arcangeli21d0d442010-08-09 17:19:10 -0700546 if (PageAnon(page)) {
Hugh Dickins4829b902010-10-02 17:46:06 -0700547 struct anon_vma *page__anon_vma = page_anon_vma(page);
548 /*
549 * Note: swapoff's unuse_vma() is more efficient with this
550 * check, and needs it to match anon_vma when KSM is active.
551 */
552 if (!vma->anon_vma || !page__anon_vma ||
553 vma->anon_vma->root != page__anon_vma->root)
Andrea Arcangeli21d0d442010-08-09 17:19:10 -0700554 return -EFAULT;
555 } else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) {
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800556 if (!vma->vm_file ||
557 vma->vm_file->f_mapping != page->mapping)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 return -EFAULT;
559 } else
560 return -EFAULT;
561 return vma_address(page, vma);
562}
563
564/*
Nikita Danilov81b40822005-05-01 08:58:36 -0700565 * Check that @page is mapped at @address into @mm.
566 *
Nick Piggin479db0b2008-08-20 14:09:18 -0700567 * If @sync is false, page_check_address may perform a racy check to avoid
568 * the page table lock when the pte is not present (helpful when reclaiming
569 * highly shared pages).
570 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -0700571 * On success returns with pte mapped and locked.
Nikita Danilov81b40822005-05-01 08:58:36 -0700572 */
Namhyung Kime9a81a82010-10-26 14:22:01 -0700573pte_t *__page_check_address(struct page *page, struct mm_struct *mm,
Nick Piggin479db0b2008-08-20 14:09:18 -0700574 unsigned long address, spinlock_t **ptlp, int sync)
Nikita Danilov81b40822005-05-01 08:58:36 -0700575{
576 pgd_t *pgd;
577 pud_t *pud;
578 pmd_t *pmd;
579 pte_t *pte;
Hugh Dickinsc0718802005-10-29 18:16:31 -0700580 spinlock_t *ptl;
Nikita Danilov81b40822005-05-01 08:58:36 -0700581
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +0900582 if (unlikely(PageHuge(page))) {
583 pte = huge_pte_offset(mm, address);
584 ptl = &mm->page_table_lock;
585 goto check;
586 }
587
Nikita Danilov81b40822005-05-01 08:58:36 -0700588 pgd = pgd_offset(mm, address);
Hugh Dickinsc0718802005-10-29 18:16:31 -0700589 if (!pgd_present(*pgd))
590 return NULL;
591
592 pud = pud_offset(pgd, address);
593 if (!pud_present(*pud))
594 return NULL;
595
596 pmd = pmd_offset(pud, address);
597 if (!pmd_present(*pmd))
598 return NULL;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800599 if (pmd_trans_huge(*pmd))
600 return NULL;
Hugh Dickinsc0718802005-10-29 18:16:31 -0700601
602 pte = pte_offset_map(pmd, address);
603 /* Make a quick check before getting the lock */
Nick Piggin479db0b2008-08-20 14:09:18 -0700604 if (!sync && !pte_present(*pte)) {
Hugh Dickinsc0718802005-10-29 18:16:31 -0700605 pte_unmap(pte);
606 return NULL;
Nikita Danilov81b40822005-05-01 08:58:36 -0700607 }
Hugh Dickinsc0718802005-10-29 18:16:31 -0700608
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700609 ptl = pte_lockptr(mm, pmd);
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +0900610check:
Hugh Dickinsc0718802005-10-29 18:16:31 -0700611 spin_lock(ptl);
612 if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) {
613 *ptlp = ptl;
614 return pte;
615 }
616 pte_unmap_unlock(pte, ptl);
617 return NULL;
Nikita Danilov81b40822005-05-01 08:58:36 -0700618}
619
Nick Pigginb291f002008-10-18 20:26:44 -0700620/**
621 * page_mapped_in_vma - check whether a page is really mapped in a VMA
622 * @page: the page to test
623 * @vma: the VMA to test
624 *
625 * Returns 1 if the page is mapped into the page tables of the VMA, 0
626 * if the page is not mapped into the page tables of this VMA. Only
627 * valid for normal file or anonymous VMAs.
628 */
Andi Kleen6a460792009-09-16 11:50:15 +0200629int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma)
Nick Pigginb291f002008-10-18 20:26:44 -0700630{
631 unsigned long address;
632 pte_t *pte;
633 spinlock_t *ptl;
634
635 address = vma_address(page, vma);
636 if (address == -EFAULT) /* out of vma range */
637 return 0;
638 pte = page_check_address(page, vma->vm_mm, address, &ptl, 1);
639 if (!pte) /* the page is not in this mm */
640 return 0;
641 pte_unmap_unlock(pte, ptl);
642
643 return 1;
644}
645
Nikita Danilov81b40822005-05-01 08:58:36 -0700646/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 * Subfunctions of page_referenced: page_referenced_one called
648 * repeatedly from either page_referenced_anon or page_referenced_file.
649 */
Hugh Dickins5ad64682009-12-14 17:59:24 -0800650int page_referenced_one(struct page *page, struct vm_area_struct *vma,
651 unsigned long address, unsigned int *mapcount,
652 unsigned long *vm_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653{
654 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 int referenced = 0;
656
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800657 if (unlikely(PageTransHuge(page))) {
658 pmd_t *pmd;
659
660 spin_lock(&mm->page_table_lock);
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800661 /*
662 * rmap might return false positives; we must filter
663 * these out using page_check_address_pmd().
664 */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800665 pmd = page_check_address_pmd(page, mm, address,
666 PAGE_CHECK_ADDRESS_PMD_FLAG);
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800667 if (!pmd) {
668 spin_unlock(&mm->page_table_lock);
669 goto out;
670 }
671
672 if (vma->vm_flags & VM_LOCKED) {
673 spin_unlock(&mm->page_table_lock);
674 *mapcount = 0; /* break early from loop */
675 *vm_flags |= VM_LOCKED;
676 goto out;
677 }
678
679 /* go ahead even if the pmd is pmd_trans_splitting() */
680 if (pmdp_clear_flush_young_notify(vma, address, pmd))
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800681 referenced++;
682 spin_unlock(&mm->page_table_lock);
683 } else {
684 pte_t *pte;
685 spinlock_t *ptl;
686
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800687 /*
688 * rmap might return false positives; we must filter
689 * these out using page_check_address().
690 */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800691 pte = page_check_address(page, mm, address, &ptl, 0);
692 if (!pte)
693 goto out;
694
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800695 if (vma->vm_flags & VM_LOCKED) {
696 pte_unmap_unlock(pte, ptl);
697 *mapcount = 0; /* break early from loop */
698 *vm_flags |= VM_LOCKED;
699 goto out;
700 }
701
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800702 if (ptep_clear_flush_young_notify(vma, address, pte)) {
703 /*
704 * Don't treat a reference through a sequentially read
705 * mapping as such. If the page has been used in
706 * another mapping, we will catch it; if this other
707 * mapping is already gone, the unmap path will have
708 * set PG_referenced or activated the page.
709 */
710 if (likely(!VM_SequentialReadHint(vma)))
711 referenced++;
712 }
713 pte_unmap_unlock(pte, ptl);
714 }
715
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800716 /* Pretend the page is referenced if the task has the
717 swap token and is in the middle of a page fault. */
718 if (mm != current->mm && has_swap_token(mm) &&
719 rwsem_is_locked(&mm->mmap_sem))
720 referenced++;
721
Hugh Dickinsc0718802005-10-29 18:16:31 -0700722 (*mapcount)--;
Huang Shijie273f0472009-12-14 17:58:51 -0800723
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700724 if (referenced)
725 *vm_flags |= vma->vm_flags;
Huang Shijie273f0472009-12-14 17:58:51 -0800726out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 return referenced;
728}
729
Balbir Singhbed71612008-02-07 00:14:01 -0800730static int page_referenced_anon(struct page *page,
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700731 struct mem_cgroup *mem_cont,
732 unsigned long *vm_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733{
734 unsigned int mapcount;
735 struct anon_vma *anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -0800736 struct anon_vma_chain *avc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 int referenced = 0;
738
739 anon_vma = page_lock_anon_vma(page);
740 if (!anon_vma)
741 return referenced;
742
743 mapcount = page_mapcount(page);
Rik van Riel5beb4932010-03-05 13:42:07 -0800744 list_for_each_entry(avc, &anon_vma->head, same_anon_vma) {
745 struct vm_area_struct *vma = avc->vma;
Hugh Dickins1cb17292009-12-14 17:59:01 -0800746 unsigned long address = vma_address(page, vma);
747 if (address == -EFAULT)
748 continue;
Balbir Singhbed71612008-02-07 00:14:01 -0800749 /*
750 * If we are reclaiming on behalf of a cgroup, skip
751 * counting on behalf of references from different
752 * cgroups
753 */
Hugh Dickinsbd845e32008-03-04 14:29:01 -0800754 if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont))
Balbir Singhbed71612008-02-07 00:14:01 -0800755 continue;
Hugh Dickins1cb17292009-12-14 17:59:01 -0800756 referenced += page_referenced_one(page, vma, address,
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700757 &mapcount, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 if (!mapcount)
759 break;
760 }
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800761
762 page_unlock_anon_vma(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 return referenced;
764}
765
766/**
767 * page_referenced_file - referenced check for object-based rmap
768 * @page: the page we're checking references on.
Randy Dunlap43d8eac2008-03-19 17:00:43 -0700769 * @mem_cont: target memory controller
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700770 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771 *
772 * For an object-based mapped page, find all the places it is mapped and
773 * check/clear the referenced flag. This is done by following the page->mapping
774 * pointer, then walking the chain of vmas it holds. It returns the number
775 * of references it found.
776 *
777 * This function is only called from page_referenced for object-based pages.
778 */
Balbir Singhbed71612008-02-07 00:14:01 -0800779static int page_referenced_file(struct page *page,
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700780 struct mem_cgroup *mem_cont,
781 unsigned long *vm_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782{
783 unsigned int mapcount;
784 struct address_space *mapping = page->mapping;
785 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
786 struct vm_area_struct *vma;
787 struct prio_tree_iter iter;
788 int referenced = 0;
789
790 /*
791 * The caller's checks on page->mapping and !PageAnon have made
792 * sure that this is a file page: the check for page->mapping
793 * excludes the case just before it gets set on an anon page.
794 */
795 BUG_ON(PageAnon(page));
796
797 /*
798 * The page lock not only makes sure that page->mapping cannot
799 * suddenly be NULLified by truncation, it makes sure that the
800 * structure at mapping cannot be freed and reused yet,
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700801 * so we can safely take mapping->i_mmap_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 */
803 BUG_ON(!PageLocked(page));
804
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700805 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806
807 /*
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700808 * i_mmap_mutex does not stabilize mapcount at all, but mapcount
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 * is more likely to be accurate if we note it after spinning.
810 */
811 mapcount = page_mapcount(page);
812
813 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
Hugh Dickins1cb17292009-12-14 17:59:01 -0800814 unsigned long address = vma_address(page, vma);
815 if (address == -EFAULT)
816 continue;
Balbir Singhbed71612008-02-07 00:14:01 -0800817 /*
818 * If we are reclaiming on behalf of a cgroup, skip
819 * counting on behalf of references from different
820 * cgroups
821 */
Hugh Dickinsbd845e32008-03-04 14:29:01 -0800822 if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont))
Balbir Singhbed71612008-02-07 00:14:01 -0800823 continue;
Hugh Dickins1cb17292009-12-14 17:59:01 -0800824 referenced += page_referenced_one(page, vma, address,
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700825 &mapcount, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 if (!mapcount)
827 break;
828 }
829
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700830 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 return referenced;
832}
833
834/**
835 * page_referenced - test if the page was referenced
836 * @page: the page to test
837 * @is_locked: caller holds lock on the page
Randy Dunlap43d8eac2008-03-19 17:00:43 -0700838 * @mem_cont: target memory controller
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700839 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 *
841 * Quick test_and_clear_referenced for all mappings to a page,
842 * returns the number of ptes which referenced the page.
843 */
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700844int page_referenced(struct page *page,
845 int is_locked,
846 struct mem_cgroup *mem_cont,
847 unsigned long *vm_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848{
849 int referenced = 0;
Hugh Dickins5ad64682009-12-14 17:59:24 -0800850 int we_locked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700852 *vm_flags = 0;
Hugh Dickins3ca7b3c2009-12-14 17:58:57 -0800853 if (page_mapped(page) && page_rmapping(page)) {
Hugh Dickins5ad64682009-12-14 17:59:24 -0800854 if (!is_locked && (!PageAnon(page) || PageKsm(page))) {
855 we_locked = trylock_page(page);
856 if (!we_locked) {
857 referenced++;
858 goto out;
859 }
860 }
861 if (unlikely(PageKsm(page)))
862 referenced += page_referenced_ksm(page, mem_cont,
863 vm_flags);
864 else if (PageAnon(page))
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700865 referenced += page_referenced_anon(page, mem_cont,
866 vm_flags);
Hugh Dickins5ad64682009-12-14 17:59:24 -0800867 else if (page->mapping)
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700868 referenced += page_referenced_file(page, mem_cont,
869 vm_flags);
Hugh Dickins5ad64682009-12-14 17:59:24 -0800870 if (we_locked)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 unlock_page(page);
Martin Schwidefsky50a15982011-07-24 10:47:59 +0200872
873 if (page_test_and_clear_young(page_to_pfn(page)))
874 referenced++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 }
Hugh Dickins5ad64682009-12-14 17:59:24 -0800876out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 return referenced;
878}
879
Hugh Dickins1cb17292009-12-14 17:59:01 -0800880static int page_mkclean_one(struct page *page, struct vm_area_struct *vma,
881 unsigned long address)
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700882{
883 struct mm_struct *mm = vma->vm_mm;
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100884 pte_t *pte;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700885 spinlock_t *ptl;
886 int ret = 0;
887
Nick Piggin479db0b2008-08-20 14:09:18 -0700888 pte = page_check_address(page, mm, address, &ptl, 1);
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700889 if (!pte)
890 goto out;
891
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100892 if (pte_dirty(*pte) || pte_write(*pte)) {
893 pte_t entry;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700894
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100895 flush_cache_page(vma, address, pte_pfn(*pte));
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700896 entry = ptep_clear_flush_notify(vma, address, pte);
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100897 entry = pte_wrprotect(entry);
898 entry = pte_mkclean(entry);
Al Virod6e88e62006-12-29 16:48:35 -0800899 set_pte_at(mm, address, pte, entry);
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100900 ret = 1;
901 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700902
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700903 pte_unmap_unlock(pte, ptl);
904out:
905 return ret;
906}
907
908static int page_mkclean_file(struct address_space *mapping, struct page *page)
909{
910 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
911 struct vm_area_struct *vma;
912 struct prio_tree_iter iter;
913 int ret = 0;
914
915 BUG_ON(PageAnon(page));
916
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700917 mutex_lock(&mapping->i_mmap_mutex);
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700918 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
Hugh Dickins1cb17292009-12-14 17:59:01 -0800919 if (vma->vm_flags & VM_SHARED) {
920 unsigned long address = vma_address(page, vma);
921 if (address == -EFAULT)
922 continue;
923 ret += page_mkclean_one(page, vma, address);
924 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700925 }
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700926 mutex_unlock(&mapping->i_mmap_mutex);
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700927 return ret;
928}
929
930int page_mkclean(struct page *page)
931{
932 int ret = 0;
933
934 BUG_ON(!PageLocked(page));
935
936 if (page_mapped(page)) {
937 struct address_space *mapping = page_mapping(page);
Christian Borntraegerce7e9fa2007-11-20 11:13:36 +0100938 if (mapping) {
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700939 ret = page_mkclean_file(mapping, page);
Martin Schwidefsky2d425522011-05-23 10:24:39 +0200940 if (page_test_and_clear_dirty(page_to_pfn(page), 1))
Christian Borntraegerce7e9fa2007-11-20 11:13:36 +0100941 ret = 1;
Martin Schwidefsky6c210482007-04-27 16:01:57 +0200942 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700943 }
944
945 return ret;
946}
Jaya Kumar60b59be2007-05-08 00:37:37 -0700947EXPORT_SYMBOL_GPL(page_mkclean);
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700948
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949/**
Rik van Rielc44b6742010-03-05 13:42:09 -0800950 * page_move_anon_rmap - move a page to our anon_vma
951 * @page: the page to move to our anon_vma
952 * @vma: the vma the page belongs to
953 * @address: the user virtual address mapped
954 *
955 * When a page belongs exclusively to one process after a COW event,
956 * that page can be moved into the anon_vma that belongs to just that
957 * process, so the rmap code will not search the parent or sibling
958 * processes.
959 */
960void page_move_anon_rmap(struct page *page,
961 struct vm_area_struct *vma, unsigned long address)
962{
963 struct anon_vma *anon_vma = vma->anon_vma;
964
965 VM_BUG_ON(!PageLocked(page));
966 VM_BUG_ON(!anon_vma);
967 VM_BUG_ON(page->index != linear_page_index(vma, address));
968
969 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
970 page->mapping = (struct address_space *) anon_vma;
971}
972
973/**
Andi Kleen4e1c1972010-09-22 12:43:56 +0200974 * __page_set_anon_rmap - set up new anonymous rmap
975 * @page: Page to add to rmap
976 * @vma: VM area to add page to.
977 * @address: User virtual address of the mapping
Rik van Riele8a03fe2010-04-14 17:59:28 -0400978 * @exclusive: the page is exclusively owned by the current process
Nick Piggin9617d952006-01-06 00:11:12 -0800979 */
980static void __page_set_anon_rmap(struct page *page,
Rik van Riele8a03fe2010-04-14 17:59:28 -0400981 struct vm_area_struct *vma, unsigned long address, int exclusive)
Nick Piggin9617d952006-01-06 00:11:12 -0800982{
Rik van Riele8a03fe2010-04-14 17:59:28 -0400983 struct anon_vma *anon_vma = vma->anon_vma;
Nick Piggin9617d952006-01-06 00:11:12 -0800984
Rik van Riele8a03fe2010-04-14 17:59:28 -0400985 BUG_ON(!anon_vma);
Linus Torvaldsea900022010-04-12 12:44:29 -0700986
Andi Kleen4e1c1972010-09-22 12:43:56 +0200987 if (PageAnon(page))
988 return;
989
Linus Torvaldsea900022010-04-12 12:44:29 -0700990 /*
Rik van Riele8a03fe2010-04-14 17:59:28 -0400991 * If the page isn't exclusively mapped into this vma,
992 * we must use the _oldest_ possible anon_vma for the
993 * page mapping!
Linus Torvaldsea900022010-04-12 12:44:29 -0700994 */
Andi Kleen4e1c1972010-09-22 12:43:56 +0200995 if (!exclusive)
Andrea Arcangeli288468c2010-08-09 17:19:09 -0700996 anon_vma = anon_vma->root;
Linus Torvaldsea900022010-04-12 12:44:29 -0700997
Nick Piggin9617d952006-01-06 00:11:12 -0800998 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
999 page->mapping = (struct address_space *) anon_vma;
Nick Piggin9617d952006-01-06 00:11:12 -08001000 page->index = linear_page_index(vma, address);
Nick Piggin9617d952006-01-06 00:11:12 -08001001}
1002
1003/**
Randy Dunlap43d8eac2008-03-19 17:00:43 -07001004 * __page_check_anon_rmap - sanity check anonymous rmap addition
Nick Pigginc97a9e12007-05-16 22:11:21 -07001005 * @page: the page to add the mapping to
1006 * @vma: the vm area in which the mapping is added
1007 * @address: the user virtual address mapped
1008 */
1009static void __page_check_anon_rmap(struct page *page,
1010 struct vm_area_struct *vma, unsigned long address)
1011{
1012#ifdef CONFIG_DEBUG_VM
1013 /*
1014 * The page's anon-rmap details (mapping and index) are guaranteed to
1015 * be set up correctly at this point.
1016 *
1017 * We have exclusion against page_add_anon_rmap because the caller
1018 * always holds the page locked, except if called from page_dup_rmap,
1019 * in which case the page is already known to be setup.
1020 *
1021 * We have exclusion against page_add_new_anon_rmap because those pages
1022 * are initially only visible via the pagetables, and the pte is locked
1023 * over the call to page_add_new_anon_rmap.
1024 */
Andrea Arcangeli44ab57a2010-08-09 17:19:10 -07001025 BUG_ON(page_anon_vma(page)->root != vma->anon_vma->root);
Nick Pigginc97a9e12007-05-16 22:11:21 -07001026 BUG_ON(page->index != linear_page_index(vma, address));
1027#endif
1028}
1029
1030/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 * page_add_anon_rmap - add pte mapping to an anonymous page
1032 * @page: the page to add the mapping to
1033 * @vma: the vm area in which the mapping is added
1034 * @address: the user virtual address mapped
1035 *
Hugh Dickins5ad64682009-12-14 17:59:24 -08001036 * The caller needs to hold the pte lock, and the page must be locked in
Hugh Dickins80e148222009-12-14 17:59:29 -08001037 * the anon_vma case: to serialize mapping,index checking after setting,
1038 * and to ensure that PageAnon is not being upgraded racily to PageKsm
1039 * (but PageKsm is never downgraded to PageAnon).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 */
1041void page_add_anon_rmap(struct page *page,
1042 struct vm_area_struct *vma, unsigned long address)
1043{
Rik van Rielad8c2ee2010-08-09 17:19:48 -07001044 do_page_add_anon_rmap(page, vma, address, 0);
1045}
1046
1047/*
1048 * Special version of the above for do_swap_page, which often runs
1049 * into pages that are exclusively owned by the current process.
1050 * Everybody else should continue to use page_add_anon_rmap above.
1051 */
1052void do_page_add_anon_rmap(struct page *page,
1053 struct vm_area_struct *vma, unsigned long address, int exclusive)
1054{
Hugh Dickins5ad64682009-12-14 17:59:24 -08001055 int first = atomic_inc_and_test(&page->_mapcount);
Andrea Arcangeli79134172011-01-13 15:46:58 -08001056 if (first) {
1057 if (!PageTransHuge(page))
1058 __inc_zone_page_state(page, NR_ANON_PAGES);
1059 else
1060 __inc_zone_page_state(page,
1061 NR_ANON_TRANSPARENT_HUGEPAGES);
1062 }
Hugh Dickins5ad64682009-12-14 17:59:24 -08001063 if (unlikely(PageKsm(page)))
1064 return;
1065
Nick Pigginc97a9e12007-05-16 22:11:21 -07001066 VM_BUG_ON(!PageLocked(page));
Hugh Dickins5dbe0af2011-05-28 13:17:04 -07001067 /* address might be in next vma when migration races vma_adjust */
Hugh Dickins5ad64682009-12-14 17:59:24 -08001068 if (first)
Rik van Rielad8c2ee2010-08-09 17:19:48 -07001069 __page_set_anon_rmap(page, vma, address, exclusive);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001070 else
Nick Pigginc97a9e12007-05-16 22:11:21 -07001071 __page_check_anon_rmap(page, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072}
1073
Randy Dunlap43d8eac2008-03-19 17:00:43 -07001074/**
Nick Piggin9617d952006-01-06 00:11:12 -08001075 * page_add_new_anon_rmap - add pte mapping to a new anonymous page
1076 * @page: the page to add the mapping to
1077 * @vma: the vm area in which the mapping is added
1078 * @address: the user virtual address mapped
1079 *
1080 * Same as page_add_anon_rmap but must only be called on *new* pages.
1081 * This means the inc-and-test can be bypassed.
Nick Pigginc97a9e12007-05-16 22:11:21 -07001082 * Page does not have to be locked.
Nick Piggin9617d952006-01-06 00:11:12 -08001083 */
1084void page_add_new_anon_rmap(struct page *page,
1085 struct vm_area_struct *vma, unsigned long address)
1086{
Hugh Dickinsb5934c52009-01-06 14:39:25 -08001087 VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end);
Hugh Dickinscbf84b72009-01-06 14:39:27 -08001088 SetPageSwapBacked(page);
1089 atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */
Andrea Arcangeli79134172011-01-13 15:46:58 -08001090 if (!PageTransHuge(page))
1091 __inc_zone_page_state(page, NR_ANON_PAGES);
1092 else
1093 __inc_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES);
Rik van Riele8a03fe2010-04-14 17:59:28 -04001094 __page_set_anon_rmap(page, vma, address, 1);
Hugh Dickinsb5934c52009-01-06 14:39:25 -08001095 if (page_evictable(page, vma))
Hugh Dickinscbf84b72009-01-06 14:39:27 -08001096 lru_cache_add_lru(page, LRU_ACTIVE_ANON);
Hugh Dickinsb5934c52009-01-06 14:39:25 -08001097 else
1098 add_page_to_unevictable_list(page);
Nick Piggin9617d952006-01-06 00:11:12 -08001099}
1100
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101/**
1102 * page_add_file_rmap - add pte mapping to a file page
1103 * @page: the page to add the mapping to
1104 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -07001105 * The caller needs to hold the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 */
1107void page_add_file_rmap(struct page *page)
1108{
Balbir Singhd69b0422009-06-17 16:26:34 -07001109 if (atomic_inc_and_test(&page->_mapcount)) {
Christoph Lameter65ba55f2006-06-30 01:55:34 -07001110 __inc_zone_page_state(page, NR_FILE_MAPPED);
Greg Thelen2a7106f2011-01-13 15:47:37 -08001111 mem_cgroup_inc_page_stat(page, MEMCG_NR_FILE_MAPPED);
Balbir Singhd69b0422009-06-17 16:26:34 -07001112 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113}
1114
1115/**
1116 * page_remove_rmap - take down pte mapping from a page
1117 * @page: page to remove mapping from
1118 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -07001119 * The caller needs to hold the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 */
Hugh Dickinsedc315f2009-01-06 14:40:11 -08001121void page_remove_rmap(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122{
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001123 /* page still mapped by someone else? */
1124 if (!atomic_add_negative(-1, &page->_mapcount))
1125 return;
1126
1127 /*
1128 * Now that the last pte has gone, s390 must transfer dirty
1129 * flag from storage key to struct page. We can usually skip
1130 * this if the page is anon, so about to be freed; but perhaps
1131 * not if it's in swapcache - there might be another pte slot
1132 * containing the swap entry, but page not yet written to swap.
1133 */
Martin Schwidefsky2d425522011-05-23 10:24:39 +02001134 if ((!PageAnon(page) || PageSwapCache(page)) &&
1135 page_test_and_clear_dirty(page_to_pfn(page), 1))
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001136 set_page_dirty(page);
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001137 /*
1138 * Hugepages are not counted in NR_ANON_PAGES nor NR_FILE_MAPPED
1139 * and not charged by memcg for now.
1140 */
1141 if (unlikely(PageHuge(page)))
1142 return;
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001143 if (PageAnon(page)) {
1144 mem_cgroup_uncharge_page(page);
Andrea Arcangeli79134172011-01-13 15:46:58 -08001145 if (!PageTransHuge(page))
1146 __dec_zone_page_state(page, NR_ANON_PAGES);
1147 else
1148 __dec_zone_page_state(page,
1149 NR_ANON_TRANSPARENT_HUGEPAGES);
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001150 } else {
1151 __dec_zone_page_state(page, NR_FILE_MAPPED);
Greg Thelen2a7106f2011-01-13 15:47:37 -08001152 mem_cgroup_dec_page_stat(page, MEMCG_NR_FILE_MAPPED);
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001153 }
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001154 /*
1155 * It would be tidy to reset the PageAnon mapping here,
1156 * but that might overwrite a racing page_add_anon_rmap
1157 * which increments mapcount after us but sets mapping
1158 * before us: so leave the reset to free_hot_cold_page,
1159 * and remember that it's only reliable while mapped.
1160 * Leaving it set also helps swapoff to reinstate ptes
1161 * faster for those pages still in swapcache.
1162 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163}
1164
1165/*
1166 * Subfunctions of try_to_unmap: try_to_unmap_one called
Wanlong Gao99ef0312011-10-31 17:08:51 -07001167 * repeatedly from try_to_unmap_ksm, try_to_unmap_anon or try_to_unmap_file.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 */
Hugh Dickins5ad64682009-12-14 17:59:24 -08001169int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
1170 unsigned long address, enum ttu_flags flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171{
1172 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 pte_t *pte;
1174 pte_t pteval;
Hugh Dickinsc0718802005-10-29 18:16:31 -07001175 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 int ret = SWAP_AGAIN;
1177
Nick Piggin479db0b2008-08-20 14:09:18 -07001178 pte = page_check_address(page, mm, address, &ptl, 0);
Hugh Dickinsc0718802005-10-29 18:16:31 -07001179 if (!pte)
Nikita Danilov81b40822005-05-01 08:58:36 -07001180 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181
1182 /*
1183 * If the page is mlock()d, we cannot swap it out.
1184 * If it's recently referenced (perhaps page_referenced
1185 * skipped over this mm) then we should reactivate it.
1186 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001187 if (!(flags & TTU_IGNORE_MLOCK)) {
KOSAKI Motohirocaed0f482009-12-14 17:59:45 -08001188 if (vma->vm_flags & VM_LOCKED)
1189 goto out_mlock;
1190
Hugh Dickinsaf8e3352009-12-14 17:58:59 -08001191 if (TTU_ACTION(flags) == TTU_MUNLOCK)
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001192 goto out_unmap;
Andi Kleen14fa31b2009-09-16 11:50:10 +02001193 }
1194 if (!(flags & TTU_IGNORE_ACCESS)) {
Nick Pigginb291f002008-10-18 20:26:44 -07001195 if (ptep_clear_flush_young_notify(vma, address, pte)) {
1196 ret = SWAP_FAIL;
1197 goto out_unmap;
1198 }
1199 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 /* Nuke the page table entry. */
1202 flush_cache_page(vma, address, page_to_pfn(page));
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001203 pteval = ptep_clear_flush_notify(vma, address, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204
1205 /* Move the dirty bit to the physical page now the pte is gone. */
1206 if (pte_dirty(pteval))
1207 set_page_dirty(page);
1208
Hugh Dickins365e9c872005-10-29 18:16:18 -07001209 /* Update high watermark before we lower rss */
1210 update_hiwater_rss(mm);
1211
Andi Kleen888b9f72009-09-16 11:50:11 +02001212 if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) {
1213 if (PageAnon(page))
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001214 dec_mm_counter(mm, MM_ANONPAGES);
Andi Kleen888b9f72009-09-16 11:50:11 +02001215 else
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001216 dec_mm_counter(mm, MM_FILEPAGES);
Andi Kleen888b9f72009-09-16 11:50:11 +02001217 set_pte_at(mm, address, pte,
1218 swp_entry_to_pte(make_hwpoison_entry(page)));
1219 } else if (PageAnon(page)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001220 swp_entry_t entry = { .val = page_private(page) };
Christoph Lameter06972122006-06-23 02:03:35 -07001221
1222 if (PageSwapCache(page)) {
1223 /*
1224 * Store the swap location in the pte.
1225 * See handle_pte_fault() ...
1226 */
Hugh Dickins570a335b2009-12-14 17:58:46 -08001227 if (swap_duplicate(entry) < 0) {
1228 set_pte_at(mm, address, pte, pteval);
1229 ret = SWAP_FAIL;
1230 goto out_unmap;
1231 }
Christoph Lameter06972122006-06-23 02:03:35 -07001232 if (list_empty(&mm->mmlist)) {
1233 spin_lock(&mmlist_lock);
1234 if (list_empty(&mm->mmlist))
1235 list_add(&mm->mmlist, &init_mm.mmlist);
1236 spin_unlock(&mmlist_lock);
1237 }
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001238 dec_mm_counter(mm, MM_ANONPAGES);
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08001239 inc_mm_counter(mm, MM_SWAPENTS);
KOSAKI Motohiro64cdd542009-01-06 14:39:16 -08001240 } else if (PAGE_MIGRATION) {
Christoph Lameter06972122006-06-23 02:03:35 -07001241 /*
1242 * Store the pfn of the page in a special migration
1243 * pte. do_swap_page() will wait until the migration
1244 * pte is removed and then restart fault handling.
1245 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001246 BUG_ON(TTU_ACTION(flags) != TTU_MIGRATION);
Christoph Lameter06972122006-06-23 02:03:35 -07001247 entry = make_migration_entry(page, pte_write(pteval));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 }
1249 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
1250 BUG_ON(pte_file(*pte));
Andi Kleen14fa31b2009-09-16 11:50:10 +02001251 } else if (PAGE_MIGRATION && (TTU_ACTION(flags) == TTU_MIGRATION)) {
Christoph Lameter04e62a22006-06-23 02:03:38 -07001252 /* Establish migration entry for a file page */
1253 swp_entry_t entry;
1254 entry = make_migration_entry(page, pte_write(pteval));
1255 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
1256 } else
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001257 dec_mm_counter(mm, MM_FILEPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258
Hugh Dickinsedc315f2009-01-06 14:40:11 -08001259 page_remove_rmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 page_cache_release(page);
1261
1262out_unmap:
Hugh Dickinsc0718802005-10-29 18:16:31 -07001263 pte_unmap_unlock(pte, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264out:
1265 return ret;
KOSAKI Motohirocaed0f482009-12-14 17:59:45 -08001266
1267out_mlock:
1268 pte_unmap_unlock(pte, ptl);
1269
1270
1271 /*
1272 * We need mmap_sem locking, Otherwise VM_LOCKED check makes
1273 * unstable result and race. Plus, We can't wait here because
Peter Zijlstra2b575eb2011-05-24 17:12:11 -07001274 * we now hold anon_vma->mutex or mapping->i_mmap_mutex.
KOSAKI Motohirocaed0f482009-12-14 17:59:45 -08001275 * if trylock failed, the page remain in evictable lru and later
1276 * vmscan could retry to move the page to unevictable lru if the
1277 * page is actually mlocked.
1278 */
1279 if (down_read_trylock(&vma->vm_mm->mmap_sem)) {
1280 if (vma->vm_flags & VM_LOCKED) {
1281 mlock_vma_page(page);
1282 ret = SWAP_MLOCK;
1283 }
1284 up_read(&vma->vm_mm->mmap_sem);
1285 }
1286 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287}
1288
1289/*
1290 * objrmap doesn't work for nonlinear VMAs because the assumption that
1291 * offset-into-file correlates with offset-into-virtual-addresses does not hold.
1292 * Consequently, given a particular page and its ->index, we cannot locate the
1293 * ptes which are mapping that page without an exhaustive linear search.
1294 *
1295 * So what this code does is a mini "virtual scan" of each nonlinear VMA which
1296 * maps the file to which the target page belongs. The ->vm_private_data field
1297 * holds the current cursor into that scan. Successive searches will circulate
1298 * around the vma's virtual address space.
1299 *
1300 * So as more replacement pressure is applied to the pages in a nonlinear VMA,
1301 * more scanning pressure is placed against them as well. Eventually pages
1302 * will become fully unmapped and are eligible for eviction.
1303 *
1304 * For very sparsely populated VMAs this is a little inefficient - chances are
1305 * there there won't be many ptes located within the scan cluster. In this case
1306 * maybe we could scan further - to the end of the pte page, perhaps.
Nick Pigginb291f002008-10-18 20:26:44 -07001307 *
1308 * Mlocked pages: check VM_LOCKED under mmap_sem held for read, if we can
1309 * acquire it without blocking. If vma locked, mlock the pages in the cluster,
1310 * rather than unmapping them. If we encounter the "check_page" that vmscan is
1311 * trying to unmap, return SWAP_MLOCK, else default SWAP_AGAIN.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 */
1313#define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE)
1314#define CLUSTER_MASK (~(CLUSTER_SIZE - 1))
1315
Nick Pigginb291f002008-10-18 20:26:44 -07001316static int try_to_unmap_cluster(unsigned long cursor, unsigned int *mapcount,
1317 struct vm_area_struct *vma, struct page *check_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318{
1319 struct mm_struct *mm = vma->vm_mm;
1320 pgd_t *pgd;
1321 pud_t *pud;
1322 pmd_t *pmd;
Hugh Dickinsc0718802005-10-29 18:16:31 -07001323 pte_t *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 pte_t pteval;
Hugh Dickinsc0718802005-10-29 18:16:31 -07001325 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 struct page *page;
1327 unsigned long address;
1328 unsigned long end;
Nick Pigginb291f002008-10-18 20:26:44 -07001329 int ret = SWAP_AGAIN;
1330 int locked_vma = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 address = (vma->vm_start + cursor) & CLUSTER_MASK;
1333 end = address + CLUSTER_SIZE;
1334 if (address < vma->vm_start)
1335 address = vma->vm_start;
1336 if (end > vma->vm_end)
1337 end = vma->vm_end;
1338
1339 pgd = pgd_offset(mm, address);
1340 if (!pgd_present(*pgd))
Nick Pigginb291f002008-10-18 20:26:44 -07001341 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342
1343 pud = pud_offset(pgd, address);
1344 if (!pud_present(*pud))
Nick Pigginb291f002008-10-18 20:26:44 -07001345 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346
1347 pmd = pmd_offset(pud, address);
1348 if (!pmd_present(*pmd))
Nick Pigginb291f002008-10-18 20:26:44 -07001349 return ret;
1350
1351 /*
Hugh Dickinsaf8e3352009-12-14 17:58:59 -08001352 * If we can acquire the mmap_sem for read, and vma is VM_LOCKED,
Nick Pigginb291f002008-10-18 20:26:44 -07001353 * keep the sem while scanning the cluster for mlocking pages.
1354 */
Hugh Dickinsaf8e3352009-12-14 17:58:59 -08001355 if (down_read_trylock(&vma->vm_mm->mmap_sem)) {
Nick Pigginb291f002008-10-18 20:26:44 -07001356 locked_vma = (vma->vm_flags & VM_LOCKED);
1357 if (!locked_vma)
1358 up_read(&vma->vm_mm->mmap_sem); /* don't need it */
1359 }
Hugh Dickinsc0718802005-10-29 18:16:31 -07001360
1361 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362
Hugh Dickins365e9c872005-10-29 18:16:18 -07001363 /* Update high watermark before we lower rss */
1364 update_hiwater_rss(mm);
1365
Hugh Dickinsc0718802005-10-29 18:16:31 -07001366 for (; address < end; pte++, address += PAGE_SIZE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 if (!pte_present(*pte))
1368 continue;
Linus Torvalds6aab3412005-11-28 14:34:23 -08001369 page = vm_normal_page(vma, address, *pte);
1370 BUG_ON(!page || PageAnon(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371
Nick Pigginb291f002008-10-18 20:26:44 -07001372 if (locked_vma) {
1373 mlock_vma_page(page); /* no-op if already mlocked */
1374 if (page == check_page)
1375 ret = SWAP_MLOCK;
1376 continue; /* don't unmap */
1377 }
1378
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001379 if (ptep_clear_flush_young_notify(vma, address, pte))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 continue;
1381
1382 /* Nuke the page table entry. */
Ben Collinseca35132005-11-29 11:45:26 -08001383 flush_cache_page(vma, address, pte_pfn(*pte));
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001384 pteval = ptep_clear_flush_notify(vma, address, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385
1386 /* If nonlinear, store the file page offset in the pte. */
1387 if (page->index != linear_page_index(vma, address))
1388 set_pte_at(mm, address, pte, pgoff_to_pte(page->index));
1389
1390 /* Move the dirty bit to the physical page now the pte is gone. */
1391 if (pte_dirty(pteval))
1392 set_page_dirty(page);
1393
Hugh Dickinsedc315f2009-01-06 14:40:11 -08001394 page_remove_rmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 page_cache_release(page);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001396 dec_mm_counter(mm, MM_FILEPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 (*mapcount)--;
1398 }
Hugh Dickinsc0718802005-10-29 18:16:31 -07001399 pte_unmap_unlock(pte - 1, ptl);
Nick Pigginb291f002008-10-18 20:26:44 -07001400 if (locked_vma)
1401 up_read(&vma->vm_mm->mmap_sem);
1402 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403}
1404
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001405bool is_vma_temporary_stack(struct vm_area_struct *vma)
Mel Gormana8bef8ff2010-05-24 14:32:24 -07001406{
1407 int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP);
1408
1409 if (!maybe_stack)
1410 return false;
1411
1412 if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) ==
1413 VM_STACK_INCOMPLETE_SETUP)
1414 return true;
1415
1416 return false;
1417}
1418
Nick Pigginb291f002008-10-18 20:26:44 -07001419/**
1420 * try_to_unmap_anon - unmap or unlock anonymous page using the object-based
1421 * rmap method
1422 * @page: the page to unmap/unlock
Huang Shijie8051be52009-12-14 17:58:50 -08001423 * @flags: action and flags
Nick Pigginb291f002008-10-18 20:26:44 -07001424 *
1425 * Find all the mappings of a page using the mapping pointer and the vma chains
1426 * contained in the anon_vma struct it points to.
1427 *
1428 * This function is only called from try_to_unmap/try_to_munlock for
1429 * anonymous pages.
1430 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1431 * where the page was found will be held for write. So, we won't recheck
1432 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1433 * 'LOCKED.
1434 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001435static int try_to_unmap_anon(struct page *page, enum ttu_flags flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436{
1437 struct anon_vma *anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08001438 struct anon_vma_chain *avc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 int ret = SWAP_AGAIN;
Nick Pigginb291f002008-10-18 20:26:44 -07001440
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 anon_vma = page_lock_anon_vma(page);
1442 if (!anon_vma)
1443 return ret;
1444
Rik van Riel5beb4932010-03-05 13:42:07 -08001445 list_for_each_entry(avc, &anon_vma->head, same_anon_vma) {
1446 struct vm_area_struct *vma = avc->vma;
Mel Gormana8bef8ff2010-05-24 14:32:24 -07001447 unsigned long address;
1448
1449 /*
1450 * During exec, a temporary VMA is setup and later moved.
1451 * The VMA is moved under the anon_vma lock but not the
1452 * page tables leading to a race where migration cannot
1453 * find the migration ptes. Rather than increasing the
1454 * locking requirements of exec(), migration skips
1455 * temporary VMAs until after exec() completes.
1456 */
1457 if (PAGE_MIGRATION && (flags & TTU_MIGRATION) &&
1458 is_vma_temporary_stack(vma))
1459 continue;
1460
1461 address = vma_address(page, vma);
Hugh Dickins1cb17292009-12-14 17:59:01 -08001462 if (address == -EFAULT)
1463 continue;
1464 ret = try_to_unmap_one(page, vma, address, flags);
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001465 if (ret != SWAP_AGAIN || !page_mapped(page))
1466 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 }
Oleg Nesterov34bbd702007-02-28 20:13:49 -08001468
1469 page_unlock_anon_vma(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 return ret;
1471}
1472
1473/**
Nick Pigginb291f002008-10-18 20:26:44 -07001474 * try_to_unmap_file - unmap/unlock file page using the object-based rmap method
1475 * @page: the page to unmap/unlock
Andi Kleen14fa31b2009-09-16 11:50:10 +02001476 * @flags: action and flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 *
1478 * Find all the mappings of a page using the mapping pointer and the vma chains
1479 * contained in the address_space struct it points to.
1480 *
Nick Pigginb291f002008-10-18 20:26:44 -07001481 * This function is only called from try_to_unmap/try_to_munlock for
1482 * object-based pages.
1483 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1484 * where the page was found will be held for write. So, we won't recheck
1485 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1486 * 'LOCKED.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001488static int try_to_unmap_file(struct page *page, enum ttu_flags flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489{
1490 struct address_space *mapping = page->mapping;
1491 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
1492 struct vm_area_struct *vma;
1493 struct prio_tree_iter iter;
1494 int ret = SWAP_AGAIN;
1495 unsigned long cursor;
1496 unsigned long max_nl_cursor = 0;
1497 unsigned long max_nl_size = 0;
1498 unsigned int mapcount;
1499
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07001500 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
Hugh Dickins1cb17292009-12-14 17:59:01 -08001502 unsigned long address = vma_address(page, vma);
1503 if (address == -EFAULT)
1504 continue;
1505 ret = try_to_unmap_one(page, vma, address, flags);
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001506 if (ret != SWAP_AGAIN || !page_mapped(page))
1507 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 }
1509
1510 if (list_empty(&mapping->i_mmap_nonlinear))
1511 goto out;
1512
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001513 /*
1514 * We don't bother to try to find the munlocked page in nonlinears.
1515 * It's costly. Instead, later, page reclaim logic may call
1516 * try_to_unmap(TTU_MUNLOCK) and recover PG_mlocked lazily.
1517 */
1518 if (TTU_ACTION(flags) == TTU_MUNLOCK)
1519 goto out;
1520
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
1522 shared.vm_set.list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 cursor = (unsigned long) vma->vm_private_data;
1524 if (cursor > max_nl_cursor)
1525 max_nl_cursor = cursor;
1526 cursor = vma->vm_end - vma->vm_start;
1527 if (cursor > max_nl_size)
1528 max_nl_size = cursor;
1529 }
1530
Nick Pigginb291f002008-10-18 20:26:44 -07001531 if (max_nl_size == 0) { /* all nonlinears locked or reserved ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 ret = SWAP_FAIL;
1533 goto out;
1534 }
1535
1536 /*
1537 * We don't try to search for this page in the nonlinear vmas,
1538 * and page_referenced wouldn't have found it anyway. Instead
1539 * just walk the nonlinear vmas trying to age and unmap some.
1540 * The mapcount of the page we came in with is irrelevant,
1541 * but even so use it as a guide to how hard we should try?
1542 */
1543 mapcount = page_mapcount(page);
1544 if (!mapcount)
1545 goto out;
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07001546 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547
1548 max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK;
1549 if (max_nl_cursor == 0)
1550 max_nl_cursor = CLUSTER_SIZE;
1551
1552 do {
1553 list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
1554 shared.vm_set.list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 cursor = (unsigned long) vma->vm_private_data;
Hugh Dickins839b9682005-09-03 15:54:43 -07001556 while ( cursor < max_nl_cursor &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 cursor < vma->vm_end - vma->vm_start) {
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001558 if (try_to_unmap_cluster(cursor, &mapcount,
1559 vma, page) == SWAP_MLOCK)
1560 ret = SWAP_MLOCK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 cursor += CLUSTER_SIZE;
1562 vma->vm_private_data = (void *) cursor;
1563 if ((int)mapcount <= 0)
1564 goto out;
1565 }
1566 vma->vm_private_data = (void *) max_nl_cursor;
1567 }
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07001568 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 max_nl_cursor += CLUSTER_SIZE;
1570 } while (max_nl_cursor <= max_nl_size);
1571
1572 /*
1573 * Don't loop forever (perhaps all the remaining pages are
1574 * in locked vmas). Reset cursor on all unreserved nonlinear
1575 * vmas, now forgetting on which ones it had fallen behind.
1576 */
Hugh Dickins101d2be2005-11-21 21:32:16 -08001577 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
1578 vma->vm_private_data = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579out:
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07001580 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 return ret;
1582}
1583
1584/**
1585 * try_to_unmap - try to remove all page table mappings to a page
1586 * @page: the page to get unmapped
Andi Kleen14fa31b2009-09-16 11:50:10 +02001587 * @flags: action and flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 *
1589 * Tries to remove all the page table entries which are mapping this
1590 * page, used in the pageout path. Caller must hold the page lock.
1591 * Return values are:
1592 *
1593 * SWAP_SUCCESS - we succeeded in removing all mappings
1594 * SWAP_AGAIN - we missed a mapping, try again later
1595 * SWAP_FAIL - the page is unswappable
Nick Pigginb291f002008-10-18 20:26:44 -07001596 * SWAP_MLOCK - page is mlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001598int try_to_unmap(struct page *page, enum ttu_flags flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599{
1600 int ret;
1601
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602 BUG_ON(!PageLocked(page));
Andrea Arcangeli91600e92011-01-13 15:47:16 -08001603 VM_BUG_ON(!PageHuge(page) && PageTransHuge(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604
Hugh Dickins5ad64682009-12-14 17:59:24 -08001605 if (unlikely(PageKsm(page)))
1606 ret = try_to_unmap_ksm(page, flags);
1607 else if (PageAnon(page))
Andi Kleen14fa31b2009-09-16 11:50:10 +02001608 ret = try_to_unmap_anon(page, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 else
Andi Kleen14fa31b2009-09-16 11:50:10 +02001610 ret = try_to_unmap_file(page, flags);
Nick Pigginb291f002008-10-18 20:26:44 -07001611 if (ret != SWAP_MLOCK && !page_mapped(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 ret = SWAP_SUCCESS;
1613 return ret;
1614}
Nikita Danilov81b40822005-05-01 08:58:36 -07001615
Nick Pigginb291f002008-10-18 20:26:44 -07001616/**
1617 * try_to_munlock - try to munlock a page
1618 * @page: the page to be munlocked
1619 *
1620 * Called from munlock code. Checks all of the VMAs mapping the page
1621 * to make sure nobody else has this page mlocked. The page will be
1622 * returned with PG_mlocked cleared if no other vmas have it mlocked.
1623 *
1624 * Return values are:
1625 *
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001626 * SWAP_AGAIN - no vma is holding page mlocked, or,
Nick Pigginb291f002008-10-18 20:26:44 -07001627 * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem
Hugh Dickins5ad64682009-12-14 17:59:24 -08001628 * SWAP_FAIL - page cannot be located at present
Nick Pigginb291f002008-10-18 20:26:44 -07001629 * SWAP_MLOCK - page is now mlocked.
1630 */
1631int try_to_munlock(struct page *page)
1632{
1633 VM_BUG_ON(!PageLocked(page) || PageLRU(page));
1634
Hugh Dickins5ad64682009-12-14 17:59:24 -08001635 if (unlikely(PageKsm(page)))
1636 return try_to_unmap_ksm(page, TTU_MUNLOCK);
1637 else if (PageAnon(page))
Andi Kleen14fa31b2009-09-16 11:50:10 +02001638 return try_to_unmap_anon(page, TTU_MUNLOCK);
Nick Pigginb291f002008-10-18 20:26:44 -07001639 else
Andi Kleen14fa31b2009-09-16 11:50:10 +02001640 return try_to_unmap_file(page, TTU_MUNLOCK);
Nick Pigginb291f002008-10-18 20:26:44 -07001641}
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001642
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001643void __put_anon_vma(struct anon_vma *anon_vma)
Rik van Riel76545062010-08-09 17:18:41 -07001644{
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001645 struct anon_vma *root = anon_vma->root;
Rik van Riel76545062010-08-09 17:18:41 -07001646
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001647 if (root != anon_vma && atomic_dec_and_test(&root->refcount))
1648 anon_vma_free(root);
Rik van Riel76545062010-08-09 17:18:41 -07001649
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001650 anon_vma_free(anon_vma);
Rik van Riel76545062010-08-09 17:18:41 -07001651}
Rik van Riel76545062010-08-09 17:18:41 -07001652
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001653#ifdef CONFIG_MIGRATION
1654/*
1655 * rmap_walk() and its helpers rmap_walk_anon() and rmap_walk_file():
1656 * Called by migrate.c to remove migration ptes, but might be used more later.
1657 */
1658static int rmap_walk_anon(struct page *page, int (*rmap_one)(struct page *,
1659 struct vm_area_struct *, unsigned long, void *), void *arg)
1660{
1661 struct anon_vma *anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08001662 struct anon_vma_chain *avc;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001663 int ret = SWAP_AGAIN;
1664
1665 /*
1666 * Note: remove_migration_ptes() cannot use page_lock_anon_vma()
1667 * because that depends on page_mapped(); but not all its usages
Mel Gorman3f6c8272010-05-24 14:32:17 -07001668 * are holding mmap_sem. Users without mmap_sem are required to
1669 * take a reference count to prevent the anon_vma disappearing
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001670 */
1671 anon_vma = page_anon_vma(page);
1672 if (!anon_vma)
1673 return ret;
Rik van Rielcba48b92010-08-09 17:18:38 -07001674 anon_vma_lock(anon_vma);
Rik van Riel5beb4932010-03-05 13:42:07 -08001675 list_for_each_entry(avc, &anon_vma->head, same_anon_vma) {
1676 struct vm_area_struct *vma = avc->vma;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001677 unsigned long address = vma_address(page, vma);
1678 if (address == -EFAULT)
1679 continue;
1680 ret = rmap_one(page, vma, address, arg);
1681 if (ret != SWAP_AGAIN)
1682 break;
1683 }
Rik van Rielcba48b92010-08-09 17:18:38 -07001684 anon_vma_unlock(anon_vma);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001685 return ret;
1686}
1687
1688static int rmap_walk_file(struct page *page, int (*rmap_one)(struct page *,
1689 struct vm_area_struct *, unsigned long, void *), void *arg)
1690{
1691 struct address_space *mapping = page->mapping;
1692 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
1693 struct vm_area_struct *vma;
1694 struct prio_tree_iter iter;
1695 int ret = SWAP_AGAIN;
1696
1697 if (!mapping)
1698 return ret;
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07001699 mutex_lock(&mapping->i_mmap_mutex);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001700 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
1701 unsigned long address = vma_address(page, vma);
1702 if (address == -EFAULT)
1703 continue;
1704 ret = rmap_one(page, vma, address, arg);
1705 if (ret != SWAP_AGAIN)
1706 break;
1707 }
1708 /*
1709 * No nonlinear handling: being always shared, nonlinear vmas
1710 * never contain migration ptes. Decide what to do about this
1711 * limitation to linear when we need rmap_walk() on nonlinear.
1712 */
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07001713 mutex_unlock(&mapping->i_mmap_mutex);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001714 return ret;
1715}
1716
1717int rmap_walk(struct page *page, int (*rmap_one)(struct page *,
1718 struct vm_area_struct *, unsigned long, void *), void *arg)
1719{
1720 VM_BUG_ON(!PageLocked(page));
1721
1722 if (unlikely(PageKsm(page)))
1723 return rmap_walk_ksm(page, rmap_one, arg);
1724 else if (PageAnon(page))
1725 return rmap_walk_anon(page, rmap_one, arg);
1726 else
1727 return rmap_walk_file(page, rmap_one, arg);
1728}
1729#endif /* CONFIG_MIGRATION */
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001730
Naoya Horiguchie3390f62010-06-15 13:18:13 +09001731#ifdef CONFIG_HUGETLB_PAGE
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001732/*
1733 * The following three functions are for anonymous (private mapped) hugepages.
1734 * Unlike common anonymous pages, anonymous hugepages have no accounting code
1735 * and no lru code, because we handle hugepages differently from common pages.
1736 */
1737static void __hugepage_set_anon_rmap(struct page *page,
1738 struct vm_area_struct *vma, unsigned long address, int exclusive)
1739{
1740 struct anon_vma *anon_vma = vma->anon_vma;
Naoya Horiguchi433abed2010-09-10 13:23:03 +09001741
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001742 BUG_ON(!anon_vma);
Naoya Horiguchi433abed2010-09-10 13:23:03 +09001743
1744 if (PageAnon(page))
1745 return;
1746 if (!exclusive)
1747 anon_vma = anon_vma->root;
1748
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001749 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
1750 page->mapping = (struct address_space *) anon_vma;
1751 page->index = linear_page_index(vma, address);
1752}
1753
1754void hugepage_add_anon_rmap(struct page *page,
1755 struct vm_area_struct *vma, unsigned long address)
1756{
1757 struct anon_vma *anon_vma = vma->anon_vma;
1758 int first;
Naoya Horiguchia850ea32010-09-10 13:23:06 +09001759
1760 BUG_ON(!PageLocked(page));
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001761 BUG_ON(!anon_vma);
Hugh Dickins5dbe0af2011-05-28 13:17:04 -07001762 /* address might be in next vma when migration races vma_adjust */
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001763 first = atomic_inc_and_test(&page->_mapcount);
1764 if (first)
1765 __hugepage_set_anon_rmap(page, vma, address, 0);
1766}
1767
1768void hugepage_add_new_anon_rmap(struct page *page,
1769 struct vm_area_struct *vma, unsigned long address)
1770{
1771 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
1772 atomic_set(&page->_mapcount, 0);
1773 __hugepage_set_anon_rmap(page, vma, address, 1);
1774}
Naoya Horiguchie3390f62010-06-15 13:18:13 +09001775#endif /* CONFIG_HUGETLB_PAGE */