blob: 24dd3f9fee27dfe577c7f605bd0c260f0bc143c7 [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)
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080026 * mapping->i_mmap_rwsem
Ingo Molnar5a505082012-12-02 19:56:46 +000027 * anon_vma->rwsem
Nick Piggin82591e62006-10-19 23:29:10 -070028 * mm->page_table_lock or pte_lock
29 * zone->lru_lock (in mark_page_accessed, isolate_lru_page)
30 * swap_lock (in swap_duplicate, swap_info_get)
31 * mmlist_lock (in mmput, drain_mmlist and others)
32 * mapping->private_lock (in __set_page_dirty_buffers)
Dave Chinner250df6e2011-03-22 22:23:36 +110033 * inode->i_lock (in set_page_dirty's __mark_inode_dirty)
Christoph Hellwigf758eea2011-04-21 18:19:44 -060034 * bdi.wb->list_lock (in set_page_dirty's __mark_inode_dirty)
Nick Piggin82591e62006-10-19 23:29:10 -070035 * sb_lock (within inode_lock in fs/fs-writeback.c)
36 * mapping->tree_lock (widely used, in set_page_dirty,
37 * in arch-dependent flush_dcache_mmap_lock,
Christoph Hellwigf758eea2011-04-21 18:19:44 -060038 * within bdi.wb->list_lock in __sync_single_inode)
Andi Kleen6a460792009-09-16 11:50:15 +020039 *
Ingo Molnar5a505082012-12-02 19:56:46 +000040 * anon_vma->rwsem,mapping->i_mutex (memory_failure, collect_procs_anon)
Peter Zijlstra9b679322011-06-27 16:18:09 -070041 * ->tasklist_lock
Andi Kleen6a460792009-09-16 11:50:15 +020042 * pte map lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070043 */
44
45#include <linux/mm.h>
46#include <linux/pagemap.h>
47#include <linux/swap.h>
48#include <linux/swapops.h>
49#include <linux/slab.h>
50#include <linux/init.h>
Hugh Dickins5ad64682009-12-14 17:59:24 -080051#include <linux/ksm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#include <linux/rmap.h>
53#include <linux/rcupdate.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040054#include <linux/export.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080055#include <linux/memcontrol.h>
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070056#include <linux/mmu_notifier.h>
KOSAKI Motohiro64cdd542009-01-06 14:39:16 -080057#include <linux/migrate.h>
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +090058#include <linux/hugetlb.h>
Jan Karaef5d4372012-10-25 13:37:31 -070059#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
61#include <asm/tlbflush.h>
62
Nick Pigginb291f002008-10-18 20:26:44 -070063#include "internal.h"
64
Adrian Bunkfdd2e5f2008-10-18 20:28:38 -070065static struct kmem_cache *anon_vma_cachep;
Rik van Riel5beb4932010-03-05 13:42:07 -080066static struct kmem_cache *anon_vma_chain_cachep;
Adrian Bunkfdd2e5f2008-10-18 20:28:38 -070067
68static inline struct anon_vma *anon_vma_alloc(void)
69{
Peter Zijlstra01d8b202011-03-22 16:32:49 -070070 struct anon_vma *anon_vma;
71
72 anon_vma = kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL);
73 if (anon_vma) {
74 atomic_set(&anon_vma->refcount, 1);
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -080075 anon_vma->degree = 1; /* Reference for first vma */
76 anon_vma->parent = anon_vma;
Peter Zijlstra01d8b202011-03-22 16:32:49 -070077 /*
78 * Initialise the anon_vma root to point to itself. If called
79 * from fork, the root will be reset to the parents anon_vma.
80 */
81 anon_vma->root = anon_vma;
82 }
83
84 return anon_vma;
Adrian Bunkfdd2e5f2008-10-18 20:28:38 -070085}
86
Peter Zijlstra01d8b202011-03-22 16:32:49 -070087static inline void anon_vma_free(struct anon_vma *anon_vma)
Adrian Bunkfdd2e5f2008-10-18 20:28:38 -070088{
Peter Zijlstra01d8b202011-03-22 16:32:49 -070089 VM_BUG_ON(atomic_read(&anon_vma->refcount));
Peter Zijlstra88c22082011-05-24 17:12:13 -070090
91 /*
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +000092 * Synchronize against page_lock_anon_vma_read() such that
Peter Zijlstra88c22082011-05-24 17:12:13 -070093 * we can safely hold the lock without the anon_vma getting
94 * freed.
95 *
96 * Relies on the full mb implied by the atomic_dec_and_test() from
97 * put_anon_vma() against the acquire barrier implied by
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +000098 * down_read_trylock() from page_lock_anon_vma_read(). This orders:
Peter Zijlstra88c22082011-05-24 17:12:13 -070099 *
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000100 * page_lock_anon_vma_read() VS put_anon_vma()
101 * down_read_trylock() atomic_dec_and_test()
Peter Zijlstra88c22082011-05-24 17:12:13 -0700102 * LOCK MB
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000103 * atomic_read() rwsem_is_locked()
Peter Zijlstra88c22082011-05-24 17:12:13 -0700104 *
105 * LOCK should suffice since the actual taking of the lock must
106 * happen _before_ what follows.
107 */
Hugh Dickins7f39dda2014-06-04 16:05:33 -0700108 might_sleep();
Ingo Molnar5a505082012-12-02 19:56:46 +0000109 if (rwsem_is_locked(&anon_vma->root->rwsem)) {
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000110 anon_vma_lock_write(anon_vma);
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800111 anon_vma_unlock_write(anon_vma);
Peter Zijlstra88c22082011-05-24 17:12:13 -0700112 }
113
Adrian Bunkfdd2e5f2008-10-18 20:28:38 -0700114 kmem_cache_free(anon_vma_cachep, anon_vma);
115}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116
Linus Torvaldsdd347392011-06-17 19:05:36 -0700117static inline struct anon_vma_chain *anon_vma_chain_alloc(gfp_t gfp)
Rik van Riel5beb4932010-03-05 13:42:07 -0800118{
Linus Torvaldsdd347392011-06-17 19:05:36 -0700119 return kmem_cache_alloc(anon_vma_chain_cachep, gfp);
Rik van Riel5beb4932010-03-05 13:42:07 -0800120}
121
Namhyung Kime574b5f2010-10-26 14:22:02 -0700122static void anon_vma_chain_free(struct anon_vma_chain *anon_vma_chain)
Rik van Riel5beb4932010-03-05 13:42:07 -0800123{
124 kmem_cache_free(anon_vma_chain_cachep, anon_vma_chain);
125}
126
Kautuk Consul6583a842012-03-21 16:34:01 -0700127static void anon_vma_chain_link(struct vm_area_struct *vma,
128 struct anon_vma_chain *avc,
129 struct anon_vma *anon_vma)
130{
131 avc->vma = vma;
132 avc->anon_vma = anon_vma;
133 list_add(&avc->same_vma, &vma->anon_vma_chain);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700134 anon_vma_interval_tree_insert(avc, &anon_vma->rb_root);
Kautuk Consul6583a842012-03-21 16:34:01 -0700135}
136
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700137/**
138 * anon_vma_prepare - attach an anon_vma to a memory region
139 * @vma: the memory region in question
140 *
141 * This makes sure the memory mapping described by 'vma' has
142 * an 'anon_vma' attached to it, so that we can associate the
143 * anonymous pages mapped into it with that anon_vma.
144 *
145 * The common case will be that we already have one, but if
Figo.zhang23a07902010-12-27 15:14:06 +0100146 * not we either need to find an adjacent mapping that we
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700147 * can re-use the anon_vma from (very common when the only
148 * reason for splitting a vma has been mprotect()), or we
149 * allocate a new one.
150 *
151 * Anon-vma allocations are very subtle, because we may have
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000152 * optimistically looked up an anon_vma in page_lock_anon_vma_read()
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700153 * and that may actually touch the spinlock even in the newly
154 * allocated vma (it depends on RCU to make sure that the
155 * anon_vma isn't actually destroyed).
156 *
157 * As a result, we need to do proper anon_vma locking even
158 * for the new allocation. At the same time, we do not want
159 * to do any locking for the common case of already having
160 * an anon_vma.
161 *
162 * This must be called with the mmap_sem held for reading.
163 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164int anon_vma_prepare(struct vm_area_struct *vma)
165{
166 struct anon_vma *anon_vma = vma->anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -0800167 struct anon_vma_chain *avc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168
169 might_sleep();
170 if (unlikely(!anon_vma)) {
171 struct mm_struct *mm = vma->vm_mm;
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700172 struct anon_vma *allocated;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173
Linus Torvaldsdd347392011-06-17 19:05:36 -0700174 avc = anon_vma_chain_alloc(GFP_KERNEL);
Rik van Riel5beb4932010-03-05 13:42:07 -0800175 if (!avc)
176 goto out_enomem;
177
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 anon_vma = find_mergeable_anon_vma(vma);
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700179 allocated = NULL;
180 if (!anon_vma) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 anon_vma = anon_vma_alloc();
182 if (unlikely(!anon_vma))
Rik van Riel5beb4932010-03-05 13:42:07 -0800183 goto out_enomem_free_avc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 allocated = anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 }
186
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000187 anon_vma_lock_write(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 /* page_table_lock to protect against threads */
189 spin_lock(&mm->page_table_lock);
190 if (likely(!vma->anon_vma)) {
191 vma->anon_vma = anon_vma;
Kautuk Consul6583a842012-03-21 16:34:01 -0700192 anon_vma_chain_link(vma, avc, anon_vma);
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800193 /* vma reference or self-parent link for new root */
194 anon_vma->degree++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 allocated = NULL;
Oleg Nesterov31f2b0e2010-04-23 13:18:01 -0400196 avc = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 }
198 spin_unlock(&mm->page_table_lock);
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800199 anon_vma_unlock_write(anon_vma);
Oleg Nesterov31f2b0e2010-04-23 13:18:01 -0400200
201 if (unlikely(allocated))
Peter Zijlstra01d8b202011-03-22 16:32:49 -0700202 put_anon_vma(allocated);
Oleg Nesterov31f2b0e2010-04-23 13:18:01 -0400203 if (unlikely(avc))
Rik van Riel5beb4932010-03-05 13:42:07 -0800204 anon_vma_chain_free(avc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 }
206 return 0;
Rik van Riel5beb4932010-03-05 13:42:07 -0800207
208 out_enomem_free_avc:
209 anon_vma_chain_free(avc);
210 out_enomem:
211 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212}
213
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700214/*
215 * This is a useful helper function for locking the anon_vma root as
216 * we traverse the vma->anon_vma_chain, looping over anon_vma's that
217 * have the same vma.
218 *
219 * Such anon_vma's should have the same root, so you'd expect to see
220 * just a single mutex_lock for the whole traversal.
221 */
222static inline struct anon_vma *lock_anon_vma_root(struct anon_vma *root, struct anon_vma *anon_vma)
223{
224 struct anon_vma *new_root = anon_vma->root;
225 if (new_root != root) {
226 if (WARN_ON_ONCE(root))
Ingo Molnar5a505082012-12-02 19:56:46 +0000227 up_write(&root->rwsem);
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700228 root = new_root;
Ingo Molnar5a505082012-12-02 19:56:46 +0000229 down_write(&root->rwsem);
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700230 }
231 return root;
232}
233
234static inline void unlock_anon_vma_root(struct anon_vma *root)
235{
236 if (root)
Ingo Molnar5a505082012-12-02 19:56:46 +0000237 up_write(&root->rwsem);
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700238}
239
Rik van Riel5beb4932010-03-05 13:42:07 -0800240/*
241 * Attach the anon_vmas from src to dst.
242 * Returns 0 on success, -ENOMEM on failure.
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800243 *
244 * If dst->anon_vma is NULL this function tries to find and reuse existing
245 * anon_vma which has no vmas and only one child anon_vma. This prevents
246 * degradation of anon_vma hierarchy to endless linear chain in case of
247 * constantly forking task. On the other hand, an anon_vma with more than one
248 * child isn't reused even if there was no alive vma, thus rmap walker has a
249 * good chance of avoiding scanning the whole hierarchy when it searches where
250 * page is mapped.
Rik van Riel5beb4932010-03-05 13:42:07 -0800251 */
252int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253{
Rik van Riel5beb4932010-03-05 13:42:07 -0800254 struct anon_vma_chain *avc, *pavc;
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700255 struct anon_vma *root = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256
Linus Torvalds646d87b2010-04-11 17:15:03 -0700257 list_for_each_entry_reverse(pavc, &src->anon_vma_chain, same_vma) {
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700258 struct anon_vma *anon_vma;
259
Linus Torvaldsdd347392011-06-17 19:05:36 -0700260 avc = anon_vma_chain_alloc(GFP_NOWAIT | __GFP_NOWARN);
261 if (unlikely(!avc)) {
262 unlock_anon_vma_root(root);
263 root = NULL;
264 avc = anon_vma_chain_alloc(GFP_KERNEL);
265 if (!avc)
266 goto enomem_failure;
267 }
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700268 anon_vma = pavc->anon_vma;
269 root = lock_anon_vma_root(root, anon_vma);
270 anon_vma_chain_link(dst, avc, anon_vma);
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800271
272 /*
273 * Reuse existing anon_vma if its degree lower than two,
274 * that means it has no vma and only one anon_vma child.
275 *
276 * Do not chose parent anon_vma, otherwise first child
277 * will always reuse it. Root anon_vma is never reused:
278 * it has self-parent reference and at least one child.
279 */
280 if (!dst->anon_vma && anon_vma != src->anon_vma &&
281 anon_vma->degree < 2)
282 dst->anon_vma = anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 }
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800284 if (dst->anon_vma)
285 dst->anon_vma->degree++;
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700286 unlock_anon_vma_root(root);
Rik van Riel5beb4932010-03-05 13:42:07 -0800287 return 0;
288
289 enomem_failure:
Leon Yu3fe89b32015-03-25 15:55:11 -0700290 /*
291 * dst->anon_vma is dropped here otherwise its degree can be incorrectly
292 * decremented in unlink_anon_vmas().
293 * We can safely do this because callers of anon_vma_clone() don't care
294 * about dst->anon_vma if anon_vma_clone() failed.
295 */
296 dst->anon_vma = NULL;
Rik van Riel5beb4932010-03-05 13:42:07 -0800297 unlink_anon_vmas(dst);
298 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299}
300
Rik van Riel5beb4932010-03-05 13:42:07 -0800301/*
302 * Attach vma to its own anon_vma, as well as to the anon_vmas that
303 * the corresponding VMA in the parent process is attached to.
304 * Returns 0 on success, non-zero on failure.
305 */
306int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307{
Rik van Riel5beb4932010-03-05 13:42:07 -0800308 struct anon_vma_chain *avc;
309 struct anon_vma *anon_vma;
Daniel Forrestc4ea95d2014-12-02 15:59:42 -0800310 int error;
Rik van Riel5beb4932010-03-05 13:42:07 -0800311
312 /* Don't bother if the parent process has no anon_vma here. */
313 if (!pvma->anon_vma)
314 return 0;
315
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800316 /* Drop inherited anon_vma, we'll reuse existing or allocate new. */
317 vma->anon_vma = NULL;
318
Rik van Riel5beb4932010-03-05 13:42:07 -0800319 /*
320 * First, attach the new VMA to the parent VMA's anon_vmas,
321 * so rmap can find non-COWed pages in child processes.
322 */
Daniel Forrestc4ea95d2014-12-02 15:59:42 -0800323 error = anon_vma_clone(vma, pvma);
324 if (error)
325 return error;
Rik van Riel5beb4932010-03-05 13:42:07 -0800326
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800327 /* An existing anon_vma has been reused, all done then. */
328 if (vma->anon_vma)
329 return 0;
330
Rik van Riel5beb4932010-03-05 13:42:07 -0800331 /* Then add our own anon_vma. */
332 anon_vma = anon_vma_alloc();
333 if (!anon_vma)
334 goto out_error;
Linus Torvaldsdd347392011-06-17 19:05:36 -0700335 avc = anon_vma_chain_alloc(GFP_KERNEL);
Rik van Riel5beb4932010-03-05 13:42:07 -0800336 if (!avc)
337 goto out_error_free_anon_vma;
Rik van Riel5c341ee12010-08-09 17:18:39 -0700338
339 /*
340 * The root anon_vma's spinlock is the lock actually used when we
341 * lock any of the anon_vmas in this anon_vma tree.
342 */
343 anon_vma->root = pvma->anon_vma->root;
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800344 anon_vma->parent = pvma->anon_vma;
Rik van Riel76545062010-08-09 17:18:41 -0700345 /*
Peter Zijlstra01d8b202011-03-22 16:32:49 -0700346 * With refcounts, an anon_vma can stay around longer than the
347 * process it belongs to. The root anon_vma needs to be pinned until
348 * this anon_vma is freed, because the lock lives in the root.
Rik van Riel76545062010-08-09 17:18:41 -0700349 */
350 get_anon_vma(anon_vma->root);
Rik van Riel5beb4932010-03-05 13:42:07 -0800351 /* Mark this anon_vma as the one where our new (COWed) pages go. */
352 vma->anon_vma = anon_vma;
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000353 anon_vma_lock_write(anon_vma);
Rik van Riel5c341ee12010-08-09 17:18:39 -0700354 anon_vma_chain_link(vma, avc, anon_vma);
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800355 anon_vma->parent->degree++;
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800356 anon_vma_unlock_write(anon_vma);
Rik van Riel5beb4932010-03-05 13:42:07 -0800357
358 return 0;
359
360 out_error_free_anon_vma:
Peter Zijlstra01d8b202011-03-22 16:32:49 -0700361 put_anon_vma(anon_vma);
Rik van Riel5beb4932010-03-05 13:42:07 -0800362 out_error:
Rik van Riel4946d542010-04-05 12:13:33 -0400363 unlink_anon_vmas(vma);
Rik van Riel5beb4932010-03-05 13:42:07 -0800364 return -ENOMEM;
365}
366
Rik van Riel5beb4932010-03-05 13:42:07 -0800367void unlink_anon_vmas(struct vm_area_struct *vma)
368{
369 struct anon_vma_chain *avc, *next;
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200370 struct anon_vma *root = NULL;
Rik van Riel5beb4932010-03-05 13:42:07 -0800371
Rik van Riel5c341ee12010-08-09 17:18:39 -0700372 /*
373 * Unlink each anon_vma chained to the VMA. This list is ordered
374 * from newest to oldest, ensuring the root anon_vma gets freed last.
375 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800376 list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) {
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200377 struct anon_vma *anon_vma = avc->anon_vma;
378
379 root = lock_anon_vma_root(root, anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700380 anon_vma_interval_tree_remove(avc, &anon_vma->rb_root);
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200381
382 /*
383 * Leave empty anon_vmas on the list - we'll need
384 * to free them outside the lock.
385 */
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800386 if (RB_EMPTY_ROOT(&anon_vma->rb_root)) {
387 anon_vma->parent->degree--;
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200388 continue;
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800389 }
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200390
391 list_del(&avc->same_vma);
392 anon_vma_chain_free(avc);
393 }
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800394 if (vma->anon_vma)
395 vma->anon_vma->degree--;
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200396 unlock_anon_vma_root(root);
397
398 /*
399 * Iterate the list once more, it now only contains empty and unlinked
400 * anon_vmas, destroy them. Could not do before due to __put_anon_vma()
Ingo Molnar5a505082012-12-02 19:56:46 +0000401 * needing to write-acquire the anon_vma->root->rwsem.
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200402 */
403 list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) {
404 struct anon_vma *anon_vma = avc->anon_vma;
405
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800406 BUG_ON(anon_vma->degree);
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200407 put_anon_vma(anon_vma);
408
Rik van Riel5beb4932010-03-05 13:42:07 -0800409 list_del(&avc->same_vma);
410 anon_vma_chain_free(avc);
411 }
412}
413
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700414static void anon_vma_ctor(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415{
Christoph Lametera35afb82007-05-16 22:10:57 -0700416 struct anon_vma *anon_vma = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417
Ingo Molnar5a505082012-12-02 19:56:46 +0000418 init_rwsem(&anon_vma->rwsem);
Peter Zijlstra83813262011-03-22 16:32:48 -0700419 atomic_set(&anon_vma->refcount, 0);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700420 anon_vma->rb_root = RB_ROOT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421}
422
423void __init anon_vma_init(void)
424{
425 anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma),
Paul Mundt20c2df82007-07-20 10:11:58 +0900426 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor);
Rik van Riel5beb4932010-03-05 13:42:07 -0800427 anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain, SLAB_PANIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428}
429
430/*
Peter Zijlstra6111e4c2011-05-24 17:12:08 -0700431 * Getting a lock on a stable anon_vma from a page off the LRU is tricky!
432 *
433 * Since there is no serialization what so ever against page_remove_rmap()
434 * the best this function can do is return a locked anon_vma that might
435 * have been relevant to this page.
436 *
437 * The page might have been remapped to a different anon_vma or the anon_vma
438 * returned may already be freed (and even reused).
439 *
Peter Zijlstrabc658c92011-05-29 10:33:44 +0200440 * In case it was remapped to a different anon_vma, the new anon_vma will be a
441 * child of the old anon_vma, and the anon_vma lifetime rules will therefore
442 * ensure that any anon_vma obtained from the page will still be valid for as
443 * long as we observe page_mapped() [ hence all those page_mapped() tests ].
444 *
Peter Zijlstra6111e4c2011-05-24 17:12:08 -0700445 * All users of this function must be very careful when walking the anon_vma
446 * chain and verify that the page in question is indeed mapped in it
447 * [ something equivalent to page_mapped_in_vma() ].
448 *
449 * Since anon_vma's slab is DESTROY_BY_RCU and we know from page_remove_rmap()
450 * that the anon_vma pointer from page->mapping is valid if there is a
451 * mapcount, we can dereference the anon_vma after observing those.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 */
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700453struct anon_vma *page_get_anon_vma(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454{
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700455 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 unsigned long anon_mapping;
457
458 rcu_read_lock();
Jason Low4db0c3c2015-04-15 16:14:08 -0700459 anon_mapping = (unsigned long)READ_ONCE(page->mapping);
Hugh Dickins3ca7b3c2009-12-14 17:58:57 -0800460 if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 goto out;
462 if (!page_mapped(page))
463 goto out;
464
465 anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700466 if (!atomic_inc_not_zero(&anon_vma->refcount)) {
467 anon_vma = NULL;
468 goto out;
469 }
Hugh Dickinsf18194272010-08-25 23:12:54 -0700470
471 /*
472 * If this page is still mapped, then its anon_vma cannot have been
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700473 * freed. But if it has been unmapped, we have no security against the
474 * anon_vma structure being freed and reused (for another anon_vma:
475 * SLAB_DESTROY_BY_RCU guarantees that - so the atomic_inc_not_zero()
476 * above cannot corrupt).
Hugh Dickinsf18194272010-08-25 23:12:54 -0700477 */
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700478 if (!page_mapped(page)) {
Hugh Dickins7f39dda2014-06-04 16:05:33 -0700479 rcu_read_unlock();
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700480 put_anon_vma(anon_vma);
Hugh Dickins7f39dda2014-06-04 16:05:33 -0700481 return NULL;
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700482 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483out:
484 rcu_read_unlock();
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700485
486 return anon_vma;
487}
488
Peter Zijlstra88c22082011-05-24 17:12:13 -0700489/*
490 * Similar to page_get_anon_vma() except it locks the anon_vma.
491 *
492 * Its a little more complex as it tries to keep the fast path to a single
493 * atomic op -- the trylock. If we fail the trylock, we fall back to getting a
494 * reference like with page_get_anon_vma() and then block on the mutex.
495 */
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000496struct anon_vma *page_lock_anon_vma_read(struct page *page)
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700497{
Peter Zijlstra88c22082011-05-24 17:12:13 -0700498 struct anon_vma *anon_vma = NULL;
Hugh Dickinseee0f252011-05-28 13:20:21 -0700499 struct anon_vma *root_anon_vma;
Peter Zijlstra88c22082011-05-24 17:12:13 -0700500 unsigned long anon_mapping;
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700501
Peter Zijlstra88c22082011-05-24 17:12:13 -0700502 rcu_read_lock();
Jason Low4db0c3c2015-04-15 16:14:08 -0700503 anon_mapping = (unsigned long)READ_ONCE(page->mapping);
Peter Zijlstra88c22082011-05-24 17:12:13 -0700504 if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
505 goto out;
506 if (!page_mapped(page))
507 goto out;
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700508
Peter Zijlstra88c22082011-05-24 17:12:13 -0700509 anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
Jason Low4db0c3c2015-04-15 16:14:08 -0700510 root_anon_vma = READ_ONCE(anon_vma->root);
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000511 if (down_read_trylock(&root_anon_vma->rwsem)) {
Peter Zijlstra88c22082011-05-24 17:12:13 -0700512 /*
Hugh Dickinseee0f252011-05-28 13:20:21 -0700513 * If the page is still mapped, then this anon_vma is still
514 * its anon_vma, and holding the mutex ensures that it will
Peter Zijlstrabc658c92011-05-29 10:33:44 +0200515 * not go away, see anon_vma_free().
Peter Zijlstra88c22082011-05-24 17:12:13 -0700516 */
Hugh Dickinseee0f252011-05-28 13:20:21 -0700517 if (!page_mapped(page)) {
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000518 up_read(&root_anon_vma->rwsem);
Peter Zijlstra88c22082011-05-24 17:12:13 -0700519 anon_vma = NULL;
520 }
521 goto out;
522 }
523
524 /* trylock failed, we got to sleep */
525 if (!atomic_inc_not_zero(&anon_vma->refcount)) {
526 anon_vma = NULL;
527 goto out;
528 }
529
530 if (!page_mapped(page)) {
Hugh Dickins7f39dda2014-06-04 16:05:33 -0700531 rcu_read_unlock();
Peter Zijlstra88c22082011-05-24 17:12:13 -0700532 put_anon_vma(anon_vma);
Hugh Dickins7f39dda2014-06-04 16:05:33 -0700533 return NULL;
Peter Zijlstra88c22082011-05-24 17:12:13 -0700534 }
535
536 /* we pinned the anon_vma, its safe to sleep */
537 rcu_read_unlock();
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000538 anon_vma_lock_read(anon_vma);
Peter Zijlstra88c22082011-05-24 17:12:13 -0700539
540 if (atomic_dec_and_test(&anon_vma->refcount)) {
541 /*
542 * Oops, we held the last refcount, release the lock
543 * and bail -- can't simply use put_anon_vma() because
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000544 * we'll deadlock on the anon_vma_lock_write() recursion.
Peter Zijlstra88c22082011-05-24 17:12:13 -0700545 */
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000546 anon_vma_unlock_read(anon_vma);
Peter Zijlstra88c22082011-05-24 17:12:13 -0700547 __put_anon_vma(anon_vma);
548 anon_vma = NULL;
549 }
550
551 return anon_vma;
552
553out:
554 rcu_read_unlock();
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700555 return anon_vma;
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800556}
557
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000558void page_unlock_anon_vma_read(struct anon_vma *anon_vma)
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800559{
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000560 anon_vma_unlock_read(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561}
562
563/*
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -0800564 * At what user virtual address is page expected in @vma?
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 */
Michel Lespinasse86c2ad12012-10-08 16:31:42 -0700566static inline unsigned long
567__vma_address(struct page *page, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568{
Naoya Horiguchia0f7a752014-07-23 14:00:01 -0700569 pgoff_t pgoff = page_to_pgoff(page);
Michel Lespinasse86c2ad12012-10-08 16:31:42 -0700570 return vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
571}
572
573inline unsigned long
574vma_address(struct page *page, struct vm_area_struct *vma)
575{
576 unsigned long address = __vma_address(page, vma);
577
578 /* page should be within @vma mapping range */
Sasha Levin81d1b092014-10-09 15:28:10 -0700579 VM_BUG_ON_VMA(address < vma->vm_start || address >= vma->vm_end, vma);
Michel Lespinasse86c2ad12012-10-08 16:31:42 -0700580
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 return address;
582}
583
584/*
Huang Shijiebf89c8c2009-10-01 15:44:04 -0700585 * At what user virtual address is page expected in vma?
Naoya Horiguchiab941e02010-05-11 14:06:55 -0700586 * Caller should check the page is actually part of the vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 */
588unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma)
589{
Michel Lespinasse86c2ad12012-10-08 16:31:42 -0700590 unsigned long address;
Andrea Arcangeli21d0d442010-08-09 17:19:10 -0700591 if (PageAnon(page)) {
Hugh Dickins4829b902010-10-02 17:46:06 -0700592 struct anon_vma *page__anon_vma = page_anon_vma(page);
593 /*
594 * Note: swapoff's unuse_vma() is more efficient with this
595 * check, and needs it to match anon_vma when KSM is active.
596 */
597 if (!vma->anon_vma || !page__anon_vma ||
598 vma->anon_vma->root != page__anon_vma->root)
Andrea Arcangeli21d0d442010-08-09 17:19:10 -0700599 return -EFAULT;
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800600 } else if (page->mapping) {
601 if (!vma->vm_file || vma->vm_file->f_mapping != page->mapping)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 return -EFAULT;
603 } else
604 return -EFAULT;
Michel Lespinasse86c2ad12012-10-08 16:31:42 -0700605 address = __vma_address(page, vma);
606 if (unlikely(address < vma->vm_start || address >= vma->vm_end))
607 return -EFAULT;
608 return address;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609}
610
Bob Liu62190492012-12-11 16:00:37 -0800611pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address)
612{
613 pgd_t *pgd;
614 pud_t *pud;
615 pmd_t *pmd = NULL;
Hugh Dickinsf72e7dc2014-06-23 13:22:05 -0700616 pmd_t pmde;
Bob Liu62190492012-12-11 16:00:37 -0800617
618 pgd = pgd_offset(mm, address);
619 if (!pgd_present(*pgd))
620 goto out;
621
622 pud = pud_offset(pgd, address);
623 if (!pud_present(*pud))
624 goto out;
625
626 pmd = pmd_offset(pud, address);
Hugh Dickinsf72e7dc2014-06-23 13:22:05 -0700627 /*
628 * Some THP functions use the sequence pmdp_clear_flush(), set_pmd_at()
629 * without holding anon_vma lock for write. So when looking for a
630 * genuine pmde (in which to find pte), test present and !THP together.
631 */
Christian Borntraegere37c6982014-12-07 21:41:33 +0100632 pmde = *pmd;
633 barrier();
Hugh Dickinsf72e7dc2014-06-23 13:22:05 -0700634 if (!pmd_present(pmde) || pmd_trans_huge(pmde))
Bob Liu62190492012-12-11 16:00:37 -0800635 pmd = NULL;
636out:
637 return pmd;
638}
639
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640/*
Nikita Danilov81b40822005-05-01 08:58:36 -0700641 * Check that @page is mapped at @address into @mm.
642 *
Nick Piggin479db0b2008-08-20 14:09:18 -0700643 * If @sync is false, page_check_address may perform a racy check to avoid
644 * the page table lock when the pte is not present (helpful when reclaiming
645 * highly shared pages).
646 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -0700647 * On success returns with pte mapped and locked.
Nikita Danilov81b40822005-05-01 08:58:36 -0700648 */
Namhyung Kime9a81a82010-10-26 14:22:01 -0700649pte_t *__page_check_address(struct page *page, struct mm_struct *mm,
Nick Piggin479db0b2008-08-20 14:09:18 -0700650 unsigned long address, spinlock_t **ptlp, int sync)
Nikita Danilov81b40822005-05-01 08:58:36 -0700651{
Nikita Danilov81b40822005-05-01 08:58:36 -0700652 pmd_t *pmd;
653 pte_t *pte;
Hugh Dickinsc0718802005-10-29 18:16:31 -0700654 spinlock_t *ptl;
Nikita Danilov81b40822005-05-01 08:58:36 -0700655
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +0900656 if (unlikely(PageHuge(page))) {
Jianguo Wu98398c32013-12-18 17:08:59 -0800657 /* when pud is not present, pte will be NULL */
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +0900658 pte = huge_pte_offset(mm, address);
Jianguo Wu98398c32013-12-18 17:08:59 -0800659 if (!pte)
660 return NULL;
661
Kirill A. Shutemovcb900f42013-11-14 14:31:02 -0800662 ptl = huge_pte_lockptr(page_hstate(page), mm, pte);
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +0900663 goto check;
664 }
665
Bob Liu62190492012-12-11 16:00:37 -0800666 pmd = mm_find_pmd(mm, address);
667 if (!pmd)
Hugh Dickinsc0718802005-10-29 18:16:31 -0700668 return NULL;
669
Hugh Dickinsc0718802005-10-29 18:16:31 -0700670 pte = pte_offset_map(pmd, address);
671 /* Make a quick check before getting the lock */
Nick Piggin479db0b2008-08-20 14:09:18 -0700672 if (!sync && !pte_present(*pte)) {
Hugh Dickinsc0718802005-10-29 18:16:31 -0700673 pte_unmap(pte);
674 return NULL;
Nikita Danilov81b40822005-05-01 08:58:36 -0700675 }
Hugh Dickinsc0718802005-10-29 18:16:31 -0700676
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700677 ptl = pte_lockptr(mm, pmd);
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +0900678check:
Hugh Dickinsc0718802005-10-29 18:16:31 -0700679 spin_lock(ptl);
680 if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) {
681 *ptlp = ptl;
682 return pte;
683 }
684 pte_unmap_unlock(pte, ptl);
685 return NULL;
Nikita Danilov81b40822005-05-01 08:58:36 -0700686}
687
Nick Pigginb291f002008-10-18 20:26:44 -0700688/**
689 * page_mapped_in_vma - check whether a page is really mapped in a VMA
690 * @page: the page to test
691 * @vma: the VMA to test
692 *
693 * Returns 1 if the page is mapped into the page tables of the VMA, 0
694 * if the page is not mapped into the page tables of this VMA. Only
695 * valid for normal file or anonymous VMAs.
696 */
Andi Kleen6a460792009-09-16 11:50:15 +0200697int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma)
Nick Pigginb291f002008-10-18 20:26:44 -0700698{
699 unsigned long address;
700 pte_t *pte;
701 spinlock_t *ptl;
702
Michel Lespinasse86c2ad12012-10-08 16:31:42 -0700703 address = __vma_address(page, vma);
704 if (unlikely(address < vma->vm_start || address >= vma->vm_end))
Nick Pigginb291f002008-10-18 20:26:44 -0700705 return 0;
706 pte = page_check_address(page, vma->vm_mm, address, &ptl, 1);
707 if (!pte) /* the page is not in this mm */
708 return 0;
709 pte_unmap_unlock(pte, ptl);
710
711 return 1;
712}
713
Joonsoo Kim9f326242014-01-21 15:49:53 -0800714struct page_referenced_arg {
715 int mapcount;
716 int referenced;
717 unsigned long vm_flags;
718 struct mem_cgroup *memcg;
719};
Nikita Danilov81b40822005-05-01 08:58:36 -0700720/*
Joonsoo Kim9f326242014-01-21 15:49:53 -0800721 * arg: page_referenced_arg will be passed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 */
Kirill A. Shutemovac769502014-06-04 16:08:17 -0700723static int page_referenced_one(struct page *page, struct vm_area_struct *vma,
Joonsoo Kim9f326242014-01-21 15:49:53 -0800724 unsigned long address, void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725{
726 struct mm_struct *mm = vma->vm_mm;
Kirill A. Shutemov117b0792013-11-14 14:30:56 -0800727 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 int referenced = 0;
Joonsoo Kim9f326242014-01-21 15:49:53 -0800729 struct page_referenced_arg *pra = arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800731 if (unlikely(PageTransHuge(page))) {
732 pmd_t *pmd;
733
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800734 /*
735 * rmap might return false positives; we must filter
736 * these out using page_check_address_pmd().
737 */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800738 pmd = page_check_address_pmd(page, mm, address,
Kirill A. Shutemov117b0792013-11-14 14:30:56 -0800739 PAGE_CHECK_ADDRESS_PMD_FLAG, &ptl);
740 if (!pmd)
Joonsoo Kim9f326242014-01-21 15:49:53 -0800741 return SWAP_AGAIN;
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800742
743 if (vma->vm_flags & VM_LOCKED) {
Kirill A. Shutemov117b0792013-11-14 14:30:56 -0800744 spin_unlock(ptl);
Joonsoo Kim9f326242014-01-21 15:49:53 -0800745 pra->vm_flags |= VM_LOCKED;
746 return SWAP_FAIL; /* To break the loop */
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800747 }
748
749 /* go ahead even if the pmd is pmd_trans_splitting() */
750 if (pmdp_clear_flush_young_notify(vma, address, pmd))
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800751 referenced++;
Kirill A. Shutemov117b0792013-11-14 14:30:56 -0800752 spin_unlock(ptl);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800753 } else {
754 pte_t *pte;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800755
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800756 /*
757 * rmap might return false positives; we must filter
758 * these out using page_check_address().
759 */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800760 pte = page_check_address(page, mm, address, &ptl, 0);
761 if (!pte)
Joonsoo Kim9f326242014-01-21 15:49:53 -0800762 return SWAP_AGAIN;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800763
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800764 if (vma->vm_flags & VM_LOCKED) {
765 pte_unmap_unlock(pte, ptl);
Joonsoo Kim9f326242014-01-21 15:49:53 -0800766 pra->vm_flags |= VM_LOCKED;
767 return SWAP_FAIL; /* To break the loop */
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800768 }
769
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800770 if (ptep_clear_flush_young_notify(vma, address, pte)) {
771 /*
772 * Don't treat a reference through a sequentially read
773 * mapping as such. If the page has been used in
774 * another mapping, we will catch it; if this other
775 * mapping is already gone, the unmap path will have
776 * set PG_referenced or activated the page.
777 */
Joe Perches64363aa2013-07-08 16:00:18 -0700778 if (likely(!(vma->vm_flags & VM_SEQ_READ)))
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800779 referenced++;
780 }
781 pte_unmap_unlock(pte, ptl);
782 }
783
Joonsoo Kim9f326242014-01-21 15:49:53 -0800784 if (referenced) {
785 pra->referenced++;
786 pra->vm_flags |= vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 }
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800788
Joonsoo Kim9f326242014-01-21 15:49:53 -0800789 pra->mapcount--;
790 if (!pra->mapcount)
791 return SWAP_SUCCESS; /* To break the loop */
792
793 return SWAP_AGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794}
795
Joonsoo Kim9f326242014-01-21 15:49:53 -0800796static bool invalid_page_referenced_vma(struct vm_area_struct *vma, void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797{
Joonsoo Kim9f326242014-01-21 15:49:53 -0800798 struct page_referenced_arg *pra = arg;
799 struct mem_cgroup *memcg = pra->memcg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800
Joonsoo Kim9f326242014-01-21 15:49:53 -0800801 if (!mm_match_cgroup(vma->vm_mm, memcg))
802 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803
Joonsoo Kim9f326242014-01-21 15:49:53 -0800804 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805}
806
807/**
808 * page_referenced - test if the page was referenced
809 * @page: the page to test
810 * @is_locked: caller holds lock on the page
Johannes Weiner72835c82012-01-12 17:18:32 -0800811 * @memcg: target memory cgroup
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700812 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 *
814 * Quick test_and_clear_referenced for all mappings to a page,
815 * returns the number of ptes which referenced the page.
816 */
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700817int page_referenced(struct page *page,
818 int is_locked,
Johannes Weiner72835c82012-01-12 17:18:32 -0800819 struct mem_cgroup *memcg,
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700820 unsigned long *vm_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821{
Joonsoo Kim9f326242014-01-21 15:49:53 -0800822 int ret;
Hugh Dickins5ad64682009-12-14 17:59:24 -0800823 int we_locked = 0;
Joonsoo Kim9f326242014-01-21 15:49:53 -0800824 struct page_referenced_arg pra = {
825 .mapcount = page_mapcount(page),
826 .memcg = memcg,
827 };
828 struct rmap_walk_control rwc = {
829 .rmap_one = page_referenced_one,
830 .arg = (void *)&pra,
831 .anon_lock = page_lock_anon_vma_read,
832 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700834 *vm_flags = 0;
Joonsoo Kim9f326242014-01-21 15:49:53 -0800835 if (!page_mapped(page))
836 return 0;
837
838 if (!page_rmapping(page))
839 return 0;
840
841 if (!is_locked && (!PageAnon(page) || PageKsm(page))) {
842 we_locked = trylock_page(page);
843 if (!we_locked)
844 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 }
Joonsoo Kim9f326242014-01-21 15:49:53 -0800846
847 /*
848 * If we are reclaiming on behalf of a cgroup, skip
849 * counting on behalf of references from different
850 * cgroups
851 */
852 if (memcg) {
853 rwc.invalid_vma = invalid_page_referenced_vma;
854 }
855
856 ret = rmap_walk(page, &rwc);
857 *vm_flags = pra.vm_flags;
858
859 if (we_locked)
860 unlock_page(page);
861
862 return pra.referenced;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863}
864
Hugh Dickins1cb17292009-12-14 17:59:01 -0800865static int page_mkclean_one(struct page *page, struct vm_area_struct *vma,
Joonsoo Kim9853a402014-01-21 15:49:55 -0800866 unsigned long address, void *arg)
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700867{
868 struct mm_struct *mm = vma->vm_mm;
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100869 pte_t *pte;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700870 spinlock_t *ptl;
871 int ret = 0;
Joonsoo Kim9853a402014-01-21 15:49:55 -0800872 int *cleaned = arg;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700873
Nick Piggin479db0b2008-08-20 14:09:18 -0700874 pte = page_check_address(page, mm, address, &ptl, 1);
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700875 if (!pte)
876 goto out;
877
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100878 if (pte_dirty(*pte) || pte_write(*pte)) {
879 pte_t entry;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700880
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100881 flush_cache_page(vma, address, pte_pfn(*pte));
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700882 entry = ptep_clear_flush(vma, address, pte);
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100883 entry = pte_wrprotect(entry);
884 entry = pte_mkclean(entry);
Al Virod6e88e62006-12-29 16:48:35 -0800885 set_pte_at(mm, address, pte, entry);
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100886 ret = 1;
887 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700888
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700889 pte_unmap_unlock(pte, ptl);
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700890
Joonsoo Kim9853a402014-01-21 15:49:55 -0800891 if (ret) {
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700892 mmu_notifier_invalidate_page(mm, address);
Joonsoo Kim9853a402014-01-21 15:49:55 -0800893 (*cleaned)++;
894 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700895out:
Joonsoo Kim9853a402014-01-21 15:49:55 -0800896 return SWAP_AGAIN;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700897}
898
Joonsoo Kim9853a402014-01-21 15:49:55 -0800899static bool invalid_mkclean_vma(struct vm_area_struct *vma, void *arg)
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700900{
Joonsoo Kim9853a402014-01-21 15:49:55 -0800901 if (vma->vm_flags & VM_SHARED)
Fengguang Wu871beb82014-01-23 15:53:41 -0800902 return false;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700903
Fengguang Wu871beb82014-01-23 15:53:41 -0800904 return true;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700905}
906
907int page_mkclean(struct page *page)
908{
Joonsoo Kim9853a402014-01-21 15:49:55 -0800909 int cleaned = 0;
910 struct address_space *mapping;
911 struct rmap_walk_control rwc = {
912 .arg = (void *)&cleaned,
913 .rmap_one = page_mkclean_one,
914 .invalid_vma = invalid_mkclean_vma,
915 };
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700916
917 BUG_ON(!PageLocked(page));
918
Joonsoo Kim9853a402014-01-21 15:49:55 -0800919 if (!page_mapped(page))
920 return 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700921
Joonsoo Kim9853a402014-01-21 15:49:55 -0800922 mapping = page_mapping(page);
923 if (!mapping)
924 return 0;
925
926 rmap_walk(page, &rwc);
927
928 return cleaned;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700929}
Jaya Kumar60b59be2007-05-08 00:37:37 -0700930EXPORT_SYMBOL_GPL(page_mkclean);
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700931
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932/**
Rik van Rielc44b6742010-03-05 13:42:09 -0800933 * page_move_anon_rmap - move a page to our anon_vma
934 * @page: the page to move to our anon_vma
935 * @vma: the vma the page belongs to
936 * @address: the user virtual address mapped
937 *
938 * When a page belongs exclusively to one process after a COW event,
939 * that page can be moved into the anon_vma that belongs to just that
940 * process, so the rmap code will not search the parent or sibling
941 * processes.
942 */
943void page_move_anon_rmap(struct page *page,
944 struct vm_area_struct *vma, unsigned long address)
945{
946 struct anon_vma *anon_vma = vma->anon_vma;
947
Sasha Levin309381fea2014-01-23 15:52:54 -0800948 VM_BUG_ON_PAGE(!PageLocked(page), page);
Sasha Levin81d1b092014-10-09 15:28:10 -0700949 VM_BUG_ON_VMA(!anon_vma, vma);
Sasha Levin309381fea2014-01-23 15:52:54 -0800950 VM_BUG_ON_PAGE(page->index != linear_page_index(vma, address), page);
Rik van Rielc44b6742010-03-05 13:42:09 -0800951
952 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
953 page->mapping = (struct address_space *) anon_vma;
954}
955
956/**
Andi Kleen4e1c1972010-09-22 12:43:56 +0200957 * __page_set_anon_rmap - set up new anonymous rmap
958 * @page: Page to add to rmap
959 * @vma: VM area to add page to.
960 * @address: User virtual address of the mapping
Rik van Riele8a03fe2010-04-14 17:59:28 -0400961 * @exclusive: the page is exclusively owned by the current process
Nick Piggin9617d952006-01-06 00:11:12 -0800962 */
963static void __page_set_anon_rmap(struct page *page,
Rik van Riele8a03fe2010-04-14 17:59:28 -0400964 struct vm_area_struct *vma, unsigned long address, int exclusive)
Nick Piggin9617d952006-01-06 00:11:12 -0800965{
Rik van Riele8a03fe2010-04-14 17:59:28 -0400966 struct anon_vma *anon_vma = vma->anon_vma;
Nick Piggin9617d952006-01-06 00:11:12 -0800967
Rik van Riele8a03fe2010-04-14 17:59:28 -0400968 BUG_ON(!anon_vma);
Linus Torvaldsea900022010-04-12 12:44:29 -0700969
Andi Kleen4e1c1972010-09-22 12:43:56 +0200970 if (PageAnon(page))
971 return;
972
Linus Torvaldsea900022010-04-12 12:44:29 -0700973 /*
Rik van Riele8a03fe2010-04-14 17:59:28 -0400974 * If the page isn't exclusively mapped into this vma,
975 * we must use the _oldest_ possible anon_vma for the
976 * page mapping!
Linus Torvaldsea900022010-04-12 12:44:29 -0700977 */
Andi Kleen4e1c1972010-09-22 12:43:56 +0200978 if (!exclusive)
Andrea Arcangeli288468c2010-08-09 17:19:09 -0700979 anon_vma = anon_vma->root;
Linus Torvaldsea900022010-04-12 12:44:29 -0700980
Nick Piggin9617d952006-01-06 00:11:12 -0800981 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
982 page->mapping = (struct address_space *) anon_vma;
Nick Piggin9617d952006-01-06 00:11:12 -0800983 page->index = linear_page_index(vma, address);
Nick Piggin9617d952006-01-06 00:11:12 -0800984}
985
986/**
Randy Dunlap43d8eac2008-03-19 17:00:43 -0700987 * __page_check_anon_rmap - sanity check anonymous rmap addition
Nick Pigginc97a9e12007-05-16 22:11:21 -0700988 * @page: the page to add the mapping to
989 * @vma: the vm area in which the mapping is added
990 * @address: the user virtual address mapped
991 */
992static void __page_check_anon_rmap(struct page *page,
993 struct vm_area_struct *vma, unsigned long address)
994{
995#ifdef CONFIG_DEBUG_VM
996 /*
997 * The page's anon-rmap details (mapping and index) are guaranteed to
998 * be set up correctly at this point.
999 *
1000 * We have exclusion against page_add_anon_rmap because the caller
1001 * always holds the page locked, except if called from page_dup_rmap,
1002 * in which case the page is already known to be setup.
1003 *
1004 * We have exclusion against page_add_new_anon_rmap because those pages
1005 * are initially only visible via the pagetables, and the pte is locked
1006 * over the call to page_add_new_anon_rmap.
1007 */
Andrea Arcangeli44ab57a2010-08-09 17:19:10 -07001008 BUG_ON(page_anon_vma(page)->root != vma->anon_vma->root);
Nick Pigginc97a9e12007-05-16 22:11:21 -07001009 BUG_ON(page->index != linear_page_index(vma, address));
1010#endif
1011}
1012
1013/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 * page_add_anon_rmap - add pte mapping to an anonymous page
1015 * @page: the page to add the mapping to
1016 * @vma: the vm area in which the mapping is added
1017 * @address: the user virtual address mapped
1018 *
Hugh Dickins5ad64682009-12-14 17:59:24 -08001019 * The caller needs to hold the pte lock, and the page must be locked in
Hugh Dickins80e148222009-12-14 17:59:29 -08001020 * the anon_vma case: to serialize mapping,index checking after setting,
1021 * and to ensure that PageAnon is not being upgraded racily to PageKsm
1022 * (but PageKsm is never downgraded to PageAnon).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 */
1024void page_add_anon_rmap(struct page *page,
1025 struct vm_area_struct *vma, unsigned long address)
1026{
Rik van Rielad8c2ee2010-08-09 17:19:48 -07001027 do_page_add_anon_rmap(page, vma, address, 0);
1028}
1029
1030/*
1031 * Special version of the above for do_swap_page, which often runs
1032 * into pages that are exclusively owned by the current process.
1033 * Everybody else should continue to use page_add_anon_rmap above.
1034 */
1035void do_page_add_anon_rmap(struct page *page,
1036 struct vm_area_struct *vma, unsigned long address, int exclusive)
1037{
Hugh Dickins5ad64682009-12-14 17:59:24 -08001038 int first = atomic_inc_and_test(&page->_mapcount);
Andrea Arcangeli79134172011-01-13 15:46:58 -08001039 if (first) {
Jianyu Zhanbea04b02014-06-04 16:09:51 -07001040 /*
1041 * We use the irq-unsafe __{inc|mod}_zone_page_stat because
1042 * these counters are not modified in interrupt context, and
1043 * pte lock(a spinlock) is held, which implies preemption
1044 * disabled.
1045 */
Kirill A. Shutemov3cd14fc2013-09-12 15:13:58 -07001046 if (PageTransHuge(page))
Andrea Arcangeli79134172011-01-13 15:46:58 -08001047 __inc_zone_page_state(page,
1048 NR_ANON_TRANSPARENT_HUGEPAGES);
Kirill A. Shutemov3cd14fc2013-09-12 15:13:58 -07001049 __mod_zone_page_state(page_zone(page), NR_ANON_PAGES,
1050 hpage_nr_pages(page));
Andrea Arcangeli79134172011-01-13 15:46:58 -08001051 }
Hugh Dickins5ad64682009-12-14 17:59:24 -08001052 if (unlikely(PageKsm(page)))
1053 return;
1054
Sasha Levin309381fea2014-01-23 15:52:54 -08001055 VM_BUG_ON_PAGE(!PageLocked(page), page);
Hugh Dickins5dbe0af2011-05-28 13:17:04 -07001056 /* address might be in next vma when migration races vma_adjust */
Hugh Dickins5ad64682009-12-14 17:59:24 -08001057 if (first)
Rik van Rielad8c2ee2010-08-09 17:19:48 -07001058 __page_set_anon_rmap(page, vma, address, exclusive);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001059 else
Nick Pigginc97a9e12007-05-16 22:11:21 -07001060 __page_check_anon_rmap(page, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061}
1062
Randy Dunlap43d8eac2008-03-19 17:00:43 -07001063/**
Nick Piggin9617d952006-01-06 00:11:12 -08001064 * page_add_new_anon_rmap - add pte mapping to a new anonymous page
1065 * @page: the page to add the mapping to
1066 * @vma: the vm area in which the mapping is added
1067 * @address: the user virtual address mapped
1068 *
1069 * Same as page_add_anon_rmap but must only be called on *new* pages.
1070 * This means the inc-and-test can be bypassed.
Nick Pigginc97a9e12007-05-16 22:11:21 -07001071 * Page does not have to be locked.
Nick Piggin9617d952006-01-06 00:11:12 -08001072 */
1073void page_add_new_anon_rmap(struct page *page,
1074 struct vm_area_struct *vma, unsigned long address)
1075{
Sasha Levin81d1b092014-10-09 15:28:10 -07001076 VM_BUG_ON_VMA(address < vma->vm_start || address >= vma->vm_end, vma);
Hugh Dickinscbf84b72009-01-06 14:39:27 -08001077 SetPageSwapBacked(page);
1078 atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */
Kirill A. Shutemov3cd14fc2013-09-12 15:13:58 -07001079 if (PageTransHuge(page))
Andrea Arcangeli79134172011-01-13 15:46:58 -08001080 __inc_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES);
Kirill A. Shutemov3cd14fc2013-09-12 15:13:58 -07001081 __mod_zone_page_state(page_zone(page), NR_ANON_PAGES,
1082 hpage_nr_pages(page));
Rik van Riele8a03fe2010-04-14 17:59:28 -04001083 __page_set_anon_rmap(page, vma, address, 1);
Nick Piggin9617d952006-01-06 00:11:12 -08001084}
1085
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086/**
1087 * page_add_file_rmap - add pte mapping to a file page
1088 * @page: the page to add the mapping to
1089 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -07001090 * The caller needs to hold the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 */
1092void page_add_file_rmap(struct page *page)
1093{
Johannes Weinerd7365e72014-10-29 14:50:48 -07001094 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001095
Johannes Weiner6de22612015-02-11 15:25:01 -08001096 memcg = mem_cgroup_begin_page_stat(page);
Balbir Singhd69b0422009-06-17 16:26:34 -07001097 if (atomic_inc_and_test(&page->_mapcount)) {
Christoph Lameter65ba55f2006-06-30 01:55:34 -07001098 __inc_zone_page_state(page, NR_FILE_MAPPED);
Johannes Weinerd7365e72014-10-29 14:50:48 -07001099 mem_cgroup_inc_page_stat(memcg, MEM_CGROUP_STAT_FILE_MAPPED);
Balbir Singhd69b0422009-06-17 16:26:34 -07001100 }
Johannes Weiner6de22612015-02-11 15:25:01 -08001101 mem_cgroup_end_page_stat(memcg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102}
1103
Johannes Weiner8186eb62014-10-29 14:50:51 -07001104static void page_remove_file_rmap(struct page *page)
1105{
1106 struct mem_cgroup *memcg;
Johannes Weiner8186eb62014-10-29 14:50:51 -07001107
Johannes Weiner6de22612015-02-11 15:25:01 -08001108 memcg = mem_cgroup_begin_page_stat(page);
Johannes Weiner8186eb62014-10-29 14:50:51 -07001109
1110 /* page still mapped by someone else? */
1111 if (!atomic_add_negative(-1, &page->_mapcount))
1112 goto out;
1113
1114 /* Hugepages are not counted in NR_FILE_MAPPED for now. */
1115 if (unlikely(PageHuge(page)))
1116 goto out;
1117
1118 /*
1119 * We use the irq-unsafe __{inc|mod}_zone_page_stat because
1120 * these counters are not modified in interrupt context, and
1121 * pte lock(a spinlock) is held, which implies preemption disabled.
1122 */
1123 __dec_zone_page_state(page, NR_FILE_MAPPED);
1124 mem_cgroup_dec_page_stat(memcg, MEM_CGROUP_STAT_FILE_MAPPED);
1125
1126 if (unlikely(PageMlocked(page)))
1127 clear_page_mlock(page);
1128out:
Johannes Weiner6de22612015-02-11 15:25:01 -08001129 mem_cgroup_end_page_stat(memcg);
Johannes Weiner8186eb62014-10-29 14:50:51 -07001130}
1131
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132/**
1133 * page_remove_rmap - take down pte mapping from a page
1134 * @page: page to remove mapping from
1135 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -07001136 * The caller needs to hold the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 */
Hugh Dickinsedc315f2009-01-06 14:40:11 -08001138void page_remove_rmap(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139{
Johannes Weiner8186eb62014-10-29 14:50:51 -07001140 if (!PageAnon(page)) {
1141 page_remove_file_rmap(page);
1142 return;
1143 }
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001144
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001145 /* page still mapped by someone else? */
1146 if (!atomic_add_negative(-1, &page->_mapcount))
Johannes Weiner8186eb62014-10-29 14:50:51 -07001147 return;
1148
1149 /* Hugepages are not counted in NR_ANON_PAGES for now. */
1150 if (unlikely(PageHuge(page)))
1151 return;
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001152
1153 /*
Jianyu Zhanbea04b02014-06-04 16:09:51 -07001154 * We use the irq-unsafe __{inc|mod}_zone_page_stat because
1155 * these counters are not modified in interrupt context, and
Jianyu Zhanbea04b02014-06-04 16:09:51 -07001156 * pte lock(a spinlock) is held, which implies preemption disabled.
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001157 */
Johannes Weiner8186eb62014-10-29 14:50:51 -07001158 if (PageTransHuge(page))
1159 __dec_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES);
1160
1161 __mod_zone_page_state(page_zone(page), NR_ANON_PAGES,
1162 -hpage_nr_pages(page));
1163
Hugh Dickinse6c509f2012-10-08 16:33:19 -07001164 if (unlikely(PageMlocked(page)))
1165 clear_page_mlock(page);
Johannes Weiner8186eb62014-10-29 14:50:51 -07001166
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001167 /*
1168 * It would be tidy to reset the PageAnon mapping here,
1169 * but that might overwrite a racing page_add_anon_rmap
1170 * which increments mapcount after us but sets mapping
1171 * before us: so leave the reset to free_hot_cold_page,
1172 * and remember that it's only reliable while mapped.
1173 * Leaving it set also helps swapoff to reinstate ptes
1174 * faster for those pages still in swapcache.
1175 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176}
1177
1178/*
Joonsoo Kim52629502014-01-21 15:49:50 -08001179 * @arg: enum ttu_flags will be passed to this argument
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 */
Kirill A. Shutemovac769502014-06-04 16:08:17 -07001181static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
Joonsoo Kim52629502014-01-21 15:49:50 -08001182 unsigned long address, void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183{
1184 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 pte_t *pte;
1186 pte_t pteval;
Hugh Dickinsc0718802005-10-29 18:16:31 -07001187 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 int ret = SWAP_AGAIN;
Joonsoo Kim52629502014-01-21 15:49:50 -08001189 enum ttu_flags flags = (enum ttu_flags)arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190
Nick Piggin479db0b2008-08-20 14:09:18 -07001191 pte = page_check_address(page, mm, address, &ptl, 0);
Hugh Dickinsc0718802005-10-29 18:16:31 -07001192 if (!pte)
Nikita Danilov81b40822005-05-01 08:58:36 -07001193 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194
1195 /*
1196 * If the page is mlock()d, we cannot swap it out.
1197 * If it's recently referenced (perhaps page_referenced
1198 * skipped over this mm) then we should reactivate it.
1199 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001200 if (!(flags & TTU_IGNORE_MLOCK)) {
KOSAKI Motohirocaed0f482009-12-14 17:59:45 -08001201 if (vma->vm_flags & VM_LOCKED)
1202 goto out_mlock;
1203
Konstantin Khlebnikovdaa5ba72014-06-04 16:10:52 -07001204 if (flags & TTU_MUNLOCK)
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001205 goto out_unmap;
Andi Kleen14fa31b2009-09-16 11:50:10 +02001206 }
1207 if (!(flags & TTU_IGNORE_ACCESS)) {
Nick Pigginb291f002008-10-18 20:26:44 -07001208 if (ptep_clear_flush_young_notify(vma, address, pte)) {
1209 ret = SWAP_FAIL;
1210 goto out_unmap;
1211 }
1212 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 /* Nuke the page table entry. */
1215 flush_cache_page(vma, address, page_to_pfn(page));
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001216 pteval = ptep_clear_flush(vma, address, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217
1218 /* Move the dirty bit to the physical page now the pte is gone. */
1219 if (pte_dirty(pteval))
1220 set_page_dirty(page);
1221
Hugh Dickins365e9c872005-10-29 18:16:18 -07001222 /* Update high watermark before we lower rss */
1223 update_hiwater_rss(mm);
1224
Andi Kleen888b9f72009-09-16 11:50:11 +02001225 if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) {
Naoya Horiguchi5f24ae52012-12-12 13:52:30 -08001226 if (!PageHuge(page)) {
1227 if (PageAnon(page))
1228 dec_mm_counter(mm, MM_ANONPAGES);
1229 else
1230 dec_mm_counter(mm, MM_FILEPAGES);
1231 }
Andi Kleen888b9f72009-09-16 11:50:11 +02001232 set_pte_at(mm, address, pte,
Naoya Horiguchi5f24ae52012-12-12 13:52:30 -08001233 swp_entry_to_pte(make_hwpoison_entry(page)));
Konstantin Weitz45961722013-04-17 13:59:32 +02001234 } else if (pte_unused(pteval)) {
1235 /*
1236 * The guest indicated that the page content is of no
1237 * interest anymore. Simply discard the pte, vmscan
1238 * will take care of the rest.
1239 */
1240 if (PageAnon(page))
1241 dec_mm_counter(mm, MM_ANONPAGES);
1242 else
1243 dec_mm_counter(mm, MM_FILEPAGES);
Andi Kleen888b9f72009-09-16 11:50:11 +02001244 } else if (PageAnon(page)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001245 swp_entry_t entry = { .val = page_private(page) };
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001246 pte_t swp_pte;
Christoph Lameter06972122006-06-23 02:03:35 -07001247
1248 if (PageSwapCache(page)) {
1249 /*
1250 * Store the swap location in the pte.
1251 * See handle_pte_fault() ...
1252 */
Hugh Dickins570a335b2009-12-14 17:58:46 -08001253 if (swap_duplicate(entry) < 0) {
1254 set_pte_at(mm, address, pte, pteval);
1255 ret = SWAP_FAIL;
1256 goto out_unmap;
1257 }
Christoph Lameter06972122006-06-23 02:03:35 -07001258 if (list_empty(&mm->mmlist)) {
1259 spin_lock(&mmlist_lock);
1260 if (list_empty(&mm->mmlist))
1261 list_add(&mm->mmlist, &init_mm.mmlist);
1262 spin_unlock(&mmlist_lock);
1263 }
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001264 dec_mm_counter(mm, MM_ANONPAGES);
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08001265 inc_mm_counter(mm, MM_SWAPENTS);
Konstantin Khlebnikovce1744f2012-03-21 16:33:59 -07001266 } else if (IS_ENABLED(CONFIG_MIGRATION)) {
Christoph Lameter06972122006-06-23 02:03:35 -07001267 /*
1268 * Store the pfn of the page in a special migration
1269 * pte. do_swap_page() will wait until the migration
1270 * pte is removed and then restart fault handling.
1271 */
Konstantin Khlebnikovdaa5ba72014-06-04 16:10:52 -07001272 BUG_ON(!(flags & TTU_MIGRATION));
Christoph Lameter06972122006-06-23 02:03:35 -07001273 entry = make_migration_entry(page, pte_write(pteval));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 }
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001275 swp_pte = swp_entry_to_pte(entry);
1276 if (pte_soft_dirty(pteval))
1277 swp_pte = pte_swp_mksoft_dirty(swp_pte);
1278 set_pte_at(mm, address, pte, swp_pte);
Konstantin Khlebnikovce1744f2012-03-21 16:33:59 -07001279 } else if (IS_ENABLED(CONFIG_MIGRATION) &&
Konstantin Khlebnikovdaa5ba72014-06-04 16:10:52 -07001280 (flags & TTU_MIGRATION)) {
Christoph Lameter04e62a22006-06-23 02:03:38 -07001281 /* Establish migration entry for a file page */
1282 swp_entry_t entry;
1283 entry = make_migration_entry(page, pte_write(pteval));
1284 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
1285 } else
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001286 dec_mm_counter(mm, MM_FILEPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287
Hugh Dickinsedc315f2009-01-06 14:40:11 -08001288 page_remove_rmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 page_cache_release(page);
1290
1291out_unmap:
Hugh Dickinsc0718802005-10-29 18:16:31 -07001292 pte_unmap_unlock(pte, ptl);
Konstantin Khlebnikovdaa5ba72014-06-04 16:10:52 -07001293 if (ret != SWAP_FAIL && !(flags & TTU_MUNLOCK))
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001294 mmu_notifier_invalidate_page(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295out:
1296 return ret;
KOSAKI Motohirocaed0f482009-12-14 17:59:45 -08001297
1298out_mlock:
1299 pte_unmap_unlock(pte, ptl);
1300
1301
1302 /*
1303 * We need mmap_sem locking, Otherwise VM_LOCKED check makes
1304 * unstable result and race. Plus, We can't wait here because
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08001305 * we now hold anon_vma->rwsem or mapping->i_mmap_rwsem.
KOSAKI Motohirocaed0f482009-12-14 17:59:45 -08001306 * if trylock failed, the page remain in evictable lru and later
1307 * vmscan could retry to move the page to unevictable lru if the
1308 * page is actually mlocked.
1309 */
1310 if (down_read_trylock(&vma->vm_mm->mmap_sem)) {
1311 if (vma->vm_flags & VM_LOCKED) {
1312 mlock_vma_page(page);
1313 ret = SWAP_MLOCK;
1314 }
1315 up_read(&vma->vm_mm->mmap_sem);
1316 }
1317 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318}
1319
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001320bool is_vma_temporary_stack(struct vm_area_struct *vma)
Mel Gormana8bef8ff2010-05-24 14:32:24 -07001321{
1322 int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP);
1323
1324 if (!maybe_stack)
1325 return false;
1326
1327 if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) ==
1328 VM_STACK_INCOMPLETE_SETUP)
1329 return true;
1330
1331 return false;
1332}
1333
Joonsoo Kim52629502014-01-21 15:49:50 -08001334static bool invalid_migration_vma(struct vm_area_struct *vma, void *arg)
1335{
1336 return is_vma_temporary_stack(vma);
1337}
1338
Joonsoo Kim52629502014-01-21 15:49:50 -08001339static int page_not_mapped(struct page *page)
1340{
1341 return !page_mapped(page);
1342};
1343
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344/**
1345 * try_to_unmap - try to remove all page table mappings to a page
1346 * @page: the page to get unmapped
Andi Kleen14fa31b2009-09-16 11:50:10 +02001347 * @flags: action and flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 *
1349 * Tries to remove all the page table entries which are mapping this
1350 * page, used in the pageout path. Caller must hold the page lock.
1351 * Return values are:
1352 *
1353 * SWAP_SUCCESS - we succeeded in removing all mappings
1354 * SWAP_AGAIN - we missed a mapping, try again later
1355 * SWAP_FAIL - the page is unswappable
Nick Pigginb291f002008-10-18 20:26:44 -07001356 * SWAP_MLOCK - page is mlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001358int try_to_unmap(struct page *page, enum ttu_flags flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359{
1360 int ret;
Joonsoo Kim52629502014-01-21 15:49:50 -08001361 struct rmap_walk_control rwc = {
1362 .rmap_one = try_to_unmap_one,
1363 .arg = (void *)flags,
1364 .done = page_not_mapped,
Joonsoo Kim52629502014-01-21 15:49:50 -08001365 .anon_lock = page_lock_anon_vma_read,
1366 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367
Sasha Levin309381fea2014-01-23 15:52:54 -08001368 VM_BUG_ON_PAGE(!PageHuge(page) && PageTransHuge(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369
Joonsoo Kim52629502014-01-21 15:49:50 -08001370 /*
1371 * During exec, a temporary VMA is setup and later moved.
1372 * The VMA is moved under the anon_vma lock but not the
1373 * page tables leading to a race where migration cannot
1374 * find the migration ptes. Rather than increasing the
1375 * locking requirements of exec(), migration skips
1376 * temporary VMAs until after exec() completes.
1377 */
Konstantin Khlebnikovdaa5ba72014-06-04 16:10:52 -07001378 if ((flags & TTU_MIGRATION) && !PageKsm(page) && PageAnon(page))
Joonsoo Kim52629502014-01-21 15:49:50 -08001379 rwc.invalid_vma = invalid_migration_vma;
1380
1381 ret = rmap_walk(page, &rwc);
1382
Nick Pigginb291f002008-10-18 20:26:44 -07001383 if (ret != SWAP_MLOCK && !page_mapped(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 ret = SWAP_SUCCESS;
1385 return ret;
1386}
Nikita Danilov81b40822005-05-01 08:58:36 -07001387
Nick Pigginb291f002008-10-18 20:26:44 -07001388/**
1389 * try_to_munlock - try to munlock a page
1390 * @page: the page to be munlocked
1391 *
1392 * Called from munlock code. Checks all of the VMAs mapping the page
1393 * to make sure nobody else has this page mlocked. The page will be
1394 * returned with PG_mlocked cleared if no other vmas have it mlocked.
1395 *
1396 * Return values are:
1397 *
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001398 * SWAP_AGAIN - no vma is holding page mlocked, or,
Nick Pigginb291f002008-10-18 20:26:44 -07001399 * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem
Hugh Dickins5ad64682009-12-14 17:59:24 -08001400 * SWAP_FAIL - page cannot be located at present
Nick Pigginb291f002008-10-18 20:26:44 -07001401 * SWAP_MLOCK - page is now mlocked.
1402 */
1403int try_to_munlock(struct page *page)
1404{
Joonsoo Kime8351ac2014-01-21 15:49:52 -08001405 int ret;
1406 struct rmap_walk_control rwc = {
1407 .rmap_one = try_to_unmap_one,
1408 .arg = (void *)TTU_MUNLOCK,
1409 .done = page_not_mapped,
Joonsoo Kime8351ac2014-01-21 15:49:52 -08001410 .anon_lock = page_lock_anon_vma_read,
1411
1412 };
1413
Sasha Levin309381fea2014-01-23 15:52:54 -08001414 VM_BUG_ON_PAGE(!PageLocked(page) || PageLRU(page), page);
Nick Pigginb291f002008-10-18 20:26:44 -07001415
Joonsoo Kime8351ac2014-01-21 15:49:52 -08001416 ret = rmap_walk(page, &rwc);
1417 return ret;
Nick Pigginb291f002008-10-18 20:26:44 -07001418}
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001419
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001420void __put_anon_vma(struct anon_vma *anon_vma)
Rik van Riel76545062010-08-09 17:18:41 -07001421{
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001422 struct anon_vma *root = anon_vma->root;
Rik van Riel76545062010-08-09 17:18:41 -07001423
Andrey Ryabinin624483f2014-06-06 19:09:30 +04001424 anon_vma_free(anon_vma);
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001425 if (root != anon_vma && atomic_dec_and_test(&root->refcount))
1426 anon_vma_free(root);
Rik van Riel76545062010-08-09 17:18:41 -07001427}
Rik van Riel76545062010-08-09 17:18:41 -07001428
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001429static struct anon_vma *rmap_walk_anon_lock(struct page *page,
1430 struct rmap_walk_control *rwc)
Joonsoo Kimfaecd8d2014-01-21 15:49:46 -08001431{
1432 struct anon_vma *anon_vma;
1433
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001434 if (rwc->anon_lock)
1435 return rwc->anon_lock(page);
1436
Joonsoo Kimfaecd8d2014-01-21 15:49:46 -08001437 /*
1438 * Note: remove_migration_ptes() cannot use page_lock_anon_vma_read()
1439 * because that depends on page_mapped(); but not all its usages
1440 * are holding mmap_sem. Users without mmap_sem are required to
1441 * take a reference count to prevent the anon_vma disappearing
1442 */
1443 anon_vma = page_anon_vma(page);
1444 if (!anon_vma)
1445 return NULL;
1446
1447 anon_vma_lock_read(anon_vma);
1448 return anon_vma;
1449}
1450
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001451/*
Joonsoo Kime8351ac2014-01-21 15:49:52 -08001452 * rmap_walk_anon - do something to anonymous page using the object-based
1453 * rmap method
1454 * @page: the page to be handled
1455 * @rwc: control variable according to each walk type
1456 *
1457 * Find all the mappings of a page using the mapping pointer and the vma chains
1458 * contained in the anon_vma struct it points to.
1459 *
1460 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1461 * where the page was found will be held for write. So, we won't recheck
1462 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1463 * LOCKED.
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001464 */
Joonsoo Kim051ac832014-01-21 15:49:48 -08001465static int rmap_walk_anon(struct page *page, struct rmap_walk_control *rwc)
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001466{
1467 struct anon_vma *anon_vma;
Davidlohr Buesob258d862014-12-12 16:55:04 -08001468 pgoff_t pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001469 struct anon_vma_chain *avc;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001470 int ret = SWAP_AGAIN;
1471
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001472 anon_vma = rmap_walk_anon_lock(page, rwc);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001473 if (!anon_vma)
1474 return ret;
Joonsoo Kimfaecd8d2014-01-21 15:49:46 -08001475
Davidlohr Buesob258d862014-12-12 16:55:04 -08001476 pgoff = page_to_pgoff(page);
Michel Lespinassebf181b92012-10-08 16:31:39 -07001477 anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) {
Rik van Riel5beb4932010-03-05 13:42:07 -08001478 struct vm_area_struct *vma = avc->vma;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001479 unsigned long address = vma_address(page, vma);
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001480
1481 if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg))
1482 continue;
1483
Joonsoo Kim051ac832014-01-21 15:49:48 -08001484 ret = rwc->rmap_one(page, vma, address, rwc->arg);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001485 if (ret != SWAP_AGAIN)
1486 break;
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001487 if (rwc->done && rwc->done(page))
1488 break;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001489 }
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +00001490 anon_vma_unlock_read(anon_vma);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001491 return ret;
1492}
1493
Joonsoo Kime8351ac2014-01-21 15:49:52 -08001494/*
1495 * rmap_walk_file - do something to file page using the object-based rmap method
1496 * @page: the page to be handled
1497 * @rwc: control variable according to each walk type
1498 *
1499 * Find all the mappings of a page using the mapping pointer and the vma chains
1500 * contained in the address_space struct it points to.
1501 *
1502 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1503 * where the page was found will be held for write. So, we won't recheck
1504 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1505 * LOCKED.
1506 */
Joonsoo Kim051ac832014-01-21 15:49:48 -08001507static int rmap_walk_file(struct page *page, struct rmap_walk_control *rwc)
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001508{
1509 struct address_space *mapping = page->mapping;
Davidlohr Buesob258d862014-12-12 16:55:04 -08001510 pgoff_t pgoff;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001511 struct vm_area_struct *vma;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001512 int ret = SWAP_AGAIN;
1513
Joonsoo Kim9f326242014-01-21 15:49:53 -08001514 /*
1515 * The page lock not only makes sure that page->mapping cannot
1516 * suddenly be NULLified by truncation, it makes sure that the
1517 * structure at mapping cannot be freed and reused yet,
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08001518 * so we can safely take mapping->i_mmap_rwsem.
Joonsoo Kim9f326242014-01-21 15:49:53 -08001519 */
Sasha Levin81d1b092014-10-09 15:28:10 -07001520 VM_BUG_ON_PAGE(!PageLocked(page), page);
Joonsoo Kim9f326242014-01-21 15:49:53 -08001521
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001522 if (!mapping)
1523 return ret;
Davidlohr Bueso3dec0ba2014-12-12 16:54:27 -08001524
Davidlohr Buesob258d862014-12-12 16:55:04 -08001525 pgoff = page_to_pgoff(page);
Davidlohr Bueso3dec0ba2014-12-12 16:54:27 -08001526 i_mmap_lock_read(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07001527 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001528 unsigned long address = vma_address(page, vma);
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001529
1530 if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg))
1531 continue;
1532
Joonsoo Kim051ac832014-01-21 15:49:48 -08001533 ret = rwc->rmap_one(page, vma, address, rwc->arg);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001534 if (ret != SWAP_AGAIN)
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001535 goto done;
1536 if (rwc->done && rwc->done(page))
1537 goto done;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001538 }
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001539
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001540done:
Davidlohr Bueso3dec0ba2014-12-12 16:54:27 -08001541 i_mmap_unlock_read(mapping);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001542 return ret;
1543}
1544
Joonsoo Kim051ac832014-01-21 15:49:48 -08001545int rmap_walk(struct page *page, struct rmap_walk_control *rwc)
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001546{
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001547 if (unlikely(PageKsm(page)))
Joonsoo Kim051ac832014-01-21 15:49:48 -08001548 return rmap_walk_ksm(page, rwc);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001549 else if (PageAnon(page))
Joonsoo Kim051ac832014-01-21 15:49:48 -08001550 return rmap_walk_anon(page, rwc);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001551 else
Joonsoo Kim051ac832014-01-21 15:49:48 -08001552 return rmap_walk_file(page, rwc);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001553}
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001554
Naoya Horiguchie3390f62010-06-15 13:18:13 +09001555#ifdef CONFIG_HUGETLB_PAGE
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001556/*
1557 * The following three functions are for anonymous (private mapped) hugepages.
1558 * Unlike common anonymous pages, anonymous hugepages have no accounting code
1559 * and no lru code, because we handle hugepages differently from common pages.
1560 */
1561static void __hugepage_set_anon_rmap(struct page *page,
1562 struct vm_area_struct *vma, unsigned long address, int exclusive)
1563{
1564 struct anon_vma *anon_vma = vma->anon_vma;
Naoya Horiguchi433abed2010-09-10 13:23:03 +09001565
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001566 BUG_ON(!anon_vma);
Naoya Horiguchi433abed2010-09-10 13:23:03 +09001567
1568 if (PageAnon(page))
1569 return;
1570 if (!exclusive)
1571 anon_vma = anon_vma->root;
1572
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001573 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
1574 page->mapping = (struct address_space *) anon_vma;
1575 page->index = linear_page_index(vma, address);
1576}
1577
1578void hugepage_add_anon_rmap(struct page *page,
1579 struct vm_area_struct *vma, unsigned long address)
1580{
1581 struct anon_vma *anon_vma = vma->anon_vma;
1582 int first;
Naoya Horiguchia850ea32010-09-10 13:23:06 +09001583
1584 BUG_ON(!PageLocked(page));
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001585 BUG_ON(!anon_vma);
Hugh Dickins5dbe0af2011-05-28 13:17:04 -07001586 /* address might be in next vma when migration races vma_adjust */
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001587 first = atomic_inc_and_test(&page->_mapcount);
1588 if (first)
1589 __hugepage_set_anon_rmap(page, vma, address, 0);
1590}
1591
1592void hugepage_add_new_anon_rmap(struct page *page,
1593 struct vm_area_struct *vma, unsigned long address)
1594{
1595 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
1596 atomic_set(&page->_mapcount, 0);
1597 __hugepage_set_anon_rmap(page, vma, address, 1);
1598}
Naoya Horiguchie3390f62010-06-15 13:18:13 +09001599#endif /* CONFIG_HUGETLB_PAGE */