blob: 171b68768df1478355bcddd5e30c2edd616ba05b [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)
Greg Thelenc4843a72015-05-22 17:13:16 -040033 * mem_cgroup_{begin,end}_page_stat (memcg->move_lock)
34 * mapping->tree_lock (widely used)
Dave Chinner250df6e2011-03-22 22:23:36 +110035 * inode->i_lock (in set_page_dirty's __mark_inode_dirty)
Christoph Hellwigf758eea2011-04-21 18:19:44 -060036 * bdi.wb->list_lock (in set_page_dirty's __mark_inode_dirty)
Nick Piggin82591e62006-10-19 23:29:10 -070037 * sb_lock (within inode_lock in fs/fs-writeback.c)
38 * mapping->tree_lock (widely used, in set_page_dirty,
39 * in arch-dependent flush_dcache_mmap_lock,
Christoph Hellwigf758eea2011-04-21 18:19:44 -060040 * within bdi.wb->list_lock in __sync_single_inode)
Andi Kleen6a460792009-09-16 11:50:15 +020041 *
Ingo Molnar5a505082012-12-02 19:56:46 +000042 * anon_vma->rwsem,mapping->i_mutex (memory_failure, collect_procs_anon)
Peter Zijlstra9b679322011-06-27 16:18:09 -070043 * ->tasklist_lock
Andi Kleen6a460792009-09-16 11:50:15 +020044 * pte map lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070045 */
46
47#include <linux/mm.h>
48#include <linux/pagemap.h>
49#include <linux/swap.h>
50#include <linux/swapops.h>
51#include <linux/slab.h>
52#include <linux/init.h>
Hugh Dickins5ad64682009-12-14 17:59:24 -080053#include <linux/ksm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070054#include <linux/rmap.h>
55#include <linux/rcupdate.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040056#include <linux/export.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080057#include <linux/memcontrol.h>
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070058#include <linux/mmu_notifier.h>
KOSAKI Motohiro64cdd542009-01-06 14:39:16 -080059#include <linux/migrate.h>
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +090060#include <linux/hugetlb.h>
Jan Karaef5d4372012-10-25 13:37:31 -070061#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070062
63#include <asm/tlbflush.h>
64
Nick Pigginb291f002008-10-18 20:26:44 -070065#include "internal.h"
66
Adrian Bunkfdd2e5f2008-10-18 20:28:38 -070067static struct kmem_cache *anon_vma_cachep;
Rik van Riel5beb4932010-03-05 13:42:07 -080068static struct kmem_cache *anon_vma_chain_cachep;
Adrian Bunkfdd2e5f2008-10-18 20:28:38 -070069
70static inline struct anon_vma *anon_vma_alloc(void)
71{
Peter Zijlstra01d8b202011-03-22 16:32:49 -070072 struct anon_vma *anon_vma;
73
74 anon_vma = kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL);
75 if (anon_vma) {
76 atomic_set(&anon_vma->refcount, 1);
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -080077 anon_vma->degree = 1; /* Reference for first vma */
78 anon_vma->parent = anon_vma;
Peter Zijlstra01d8b202011-03-22 16:32:49 -070079 /*
80 * Initialise the anon_vma root to point to itself. If called
81 * from fork, the root will be reset to the parents anon_vma.
82 */
83 anon_vma->root = anon_vma;
84 }
85
86 return anon_vma;
Adrian Bunkfdd2e5f2008-10-18 20:28:38 -070087}
88
Peter Zijlstra01d8b202011-03-22 16:32:49 -070089static inline void anon_vma_free(struct anon_vma *anon_vma)
Adrian Bunkfdd2e5f2008-10-18 20:28:38 -070090{
Peter Zijlstra01d8b202011-03-22 16:32:49 -070091 VM_BUG_ON(atomic_read(&anon_vma->refcount));
Peter Zijlstra88c22082011-05-24 17:12:13 -070092
93 /*
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +000094 * Synchronize against page_lock_anon_vma_read() such that
Peter Zijlstra88c22082011-05-24 17:12:13 -070095 * we can safely hold the lock without the anon_vma getting
96 * freed.
97 *
98 * Relies on the full mb implied by the atomic_dec_and_test() from
99 * put_anon_vma() against the acquire barrier implied by
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000100 * down_read_trylock() from page_lock_anon_vma_read(). This orders:
Peter Zijlstra88c22082011-05-24 17:12:13 -0700101 *
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000102 * page_lock_anon_vma_read() VS put_anon_vma()
103 * down_read_trylock() atomic_dec_and_test()
Peter Zijlstra88c22082011-05-24 17:12:13 -0700104 * LOCK MB
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000105 * atomic_read() rwsem_is_locked()
Peter Zijlstra88c22082011-05-24 17:12:13 -0700106 *
107 * LOCK should suffice since the actual taking of the lock must
108 * happen _before_ what follows.
109 */
Hugh Dickins7f39dda2014-06-04 16:05:33 -0700110 might_sleep();
Ingo Molnar5a505082012-12-02 19:56:46 +0000111 if (rwsem_is_locked(&anon_vma->root->rwsem)) {
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000112 anon_vma_lock_write(anon_vma);
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800113 anon_vma_unlock_write(anon_vma);
Peter Zijlstra88c22082011-05-24 17:12:13 -0700114 }
115
Adrian Bunkfdd2e5f2008-10-18 20:28:38 -0700116 kmem_cache_free(anon_vma_cachep, anon_vma);
117}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118
Linus Torvaldsdd347392011-06-17 19:05:36 -0700119static inline struct anon_vma_chain *anon_vma_chain_alloc(gfp_t gfp)
Rik van Riel5beb4932010-03-05 13:42:07 -0800120{
Linus Torvaldsdd347392011-06-17 19:05:36 -0700121 return kmem_cache_alloc(anon_vma_chain_cachep, gfp);
Rik van Riel5beb4932010-03-05 13:42:07 -0800122}
123
Namhyung Kime574b5f2010-10-26 14:22:02 -0700124static void anon_vma_chain_free(struct anon_vma_chain *anon_vma_chain)
Rik van Riel5beb4932010-03-05 13:42:07 -0800125{
126 kmem_cache_free(anon_vma_chain_cachep, anon_vma_chain);
127}
128
Kautuk Consul6583a842012-03-21 16:34:01 -0700129static void anon_vma_chain_link(struct vm_area_struct *vma,
130 struct anon_vma_chain *avc,
131 struct anon_vma *anon_vma)
132{
133 avc->vma = vma;
134 avc->anon_vma = anon_vma;
135 list_add(&avc->same_vma, &vma->anon_vma_chain);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700136 anon_vma_interval_tree_insert(avc, &anon_vma->rb_root);
Kautuk Consul6583a842012-03-21 16:34:01 -0700137}
138
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700139/**
140 * anon_vma_prepare - attach an anon_vma to a memory region
141 * @vma: the memory region in question
142 *
143 * This makes sure the memory mapping described by 'vma' has
144 * an 'anon_vma' attached to it, so that we can associate the
145 * anonymous pages mapped into it with that anon_vma.
146 *
147 * The common case will be that we already have one, but if
Figo.zhang23a07902010-12-27 15:14:06 +0100148 * not we either need to find an adjacent mapping that we
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700149 * can re-use the anon_vma from (very common when the only
150 * reason for splitting a vma has been mprotect()), or we
151 * allocate a new one.
152 *
153 * Anon-vma allocations are very subtle, because we may have
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000154 * optimistically looked up an anon_vma in page_lock_anon_vma_read()
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700155 * and that may actually touch the spinlock even in the newly
156 * allocated vma (it depends on RCU to make sure that the
157 * anon_vma isn't actually destroyed).
158 *
159 * As a result, we need to do proper anon_vma locking even
160 * for the new allocation. At the same time, we do not want
161 * to do any locking for the common case of already having
162 * an anon_vma.
163 *
164 * This must be called with the mmap_sem held for reading.
165 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166int anon_vma_prepare(struct vm_area_struct *vma)
167{
168 struct anon_vma *anon_vma = vma->anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -0800169 struct anon_vma_chain *avc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170
171 might_sleep();
172 if (unlikely(!anon_vma)) {
173 struct mm_struct *mm = vma->vm_mm;
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700174 struct anon_vma *allocated;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175
Linus Torvaldsdd347392011-06-17 19:05:36 -0700176 avc = anon_vma_chain_alloc(GFP_KERNEL);
Rik van Riel5beb4932010-03-05 13:42:07 -0800177 if (!avc)
178 goto out_enomem;
179
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 anon_vma = find_mergeable_anon_vma(vma);
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700181 allocated = NULL;
182 if (!anon_vma) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 anon_vma = anon_vma_alloc();
184 if (unlikely(!anon_vma))
Rik van Riel5beb4932010-03-05 13:42:07 -0800185 goto out_enomem_free_avc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 allocated = anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 }
188
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000189 anon_vma_lock_write(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 /* page_table_lock to protect against threads */
191 spin_lock(&mm->page_table_lock);
192 if (likely(!vma->anon_vma)) {
193 vma->anon_vma = anon_vma;
Kautuk Consul6583a842012-03-21 16:34:01 -0700194 anon_vma_chain_link(vma, avc, anon_vma);
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800195 /* vma reference or self-parent link for new root */
196 anon_vma->degree++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 allocated = NULL;
Oleg Nesterov31f2b0e2010-04-23 13:18:01 -0400198 avc = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 }
200 spin_unlock(&mm->page_table_lock);
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800201 anon_vma_unlock_write(anon_vma);
Oleg Nesterov31f2b0e2010-04-23 13:18:01 -0400202
203 if (unlikely(allocated))
Peter Zijlstra01d8b202011-03-22 16:32:49 -0700204 put_anon_vma(allocated);
Oleg Nesterov31f2b0e2010-04-23 13:18:01 -0400205 if (unlikely(avc))
Rik van Riel5beb4932010-03-05 13:42:07 -0800206 anon_vma_chain_free(avc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 }
208 return 0;
Rik van Riel5beb4932010-03-05 13:42:07 -0800209
210 out_enomem_free_avc:
211 anon_vma_chain_free(avc);
212 out_enomem:
213 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214}
215
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700216/*
217 * This is a useful helper function for locking the anon_vma root as
218 * we traverse the vma->anon_vma_chain, looping over anon_vma's that
219 * have the same vma.
220 *
221 * Such anon_vma's should have the same root, so you'd expect to see
222 * just a single mutex_lock for the whole traversal.
223 */
224static inline struct anon_vma *lock_anon_vma_root(struct anon_vma *root, struct anon_vma *anon_vma)
225{
226 struct anon_vma *new_root = anon_vma->root;
227 if (new_root != root) {
228 if (WARN_ON_ONCE(root))
Ingo Molnar5a505082012-12-02 19:56:46 +0000229 up_write(&root->rwsem);
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700230 root = new_root;
Ingo Molnar5a505082012-12-02 19:56:46 +0000231 down_write(&root->rwsem);
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700232 }
233 return root;
234}
235
236static inline void unlock_anon_vma_root(struct anon_vma *root)
237{
238 if (root)
Ingo Molnar5a505082012-12-02 19:56:46 +0000239 up_write(&root->rwsem);
Linus Torvaldsbb4aa392011-06-16 20:44:51 -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.
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800245 *
246 * If dst->anon_vma is NULL this function tries to find and reuse existing
247 * anon_vma which has no vmas and only one child anon_vma. This prevents
248 * degradation of anon_vma hierarchy to endless linear chain in case of
249 * constantly forking task. On the other hand, an anon_vma with more than one
250 * child isn't reused even if there was no alive vma, thus rmap walker has a
251 * good chance of avoiding scanning the whole hierarchy when it searches where
252 * page is mapped.
Rik van Riel5beb4932010-03-05 13:42:07 -0800253 */
254int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255{
Rik van Riel5beb4932010-03-05 13:42:07 -0800256 struct anon_vma_chain *avc, *pavc;
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700257 struct anon_vma *root = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258
Linus Torvalds646d87b2010-04-11 17:15:03 -0700259 list_for_each_entry_reverse(pavc, &src->anon_vma_chain, same_vma) {
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700260 struct anon_vma *anon_vma;
261
Linus Torvaldsdd347392011-06-17 19:05:36 -0700262 avc = anon_vma_chain_alloc(GFP_NOWAIT | __GFP_NOWARN);
263 if (unlikely(!avc)) {
264 unlock_anon_vma_root(root);
265 root = NULL;
266 avc = anon_vma_chain_alloc(GFP_KERNEL);
267 if (!avc)
268 goto enomem_failure;
269 }
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700270 anon_vma = pavc->anon_vma;
271 root = lock_anon_vma_root(root, anon_vma);
272 anon_vma_chain_link(dst, avc, anon_vma);
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800273
274 /*
275 * Reuse existing anon_vma if its degree lower than two,
276 * that means it has no vma and only one anon_vma child.
277 *
278 * Do not chose parent anon_vma, otherwise first child
279 * will always reuse it. Root anon_vma is never reused:
280 * it has self-parent reference and at least one child.
281 */
282 if (!dst->anon_vma && anon_vma != src->anon_vma &&
283 anon_vma->degree < 2)
284 dst->anon_vma = anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 }
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800286 if (dst->anon_vma)
287 dst->anon_vma->degree++;
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700288 unlock_anon_vma_root(root);
Rik van Riel5beb4932010-03-05 13:42:07 -0800289 return 0;
290
291 enomem_failure:
Leon Yu3fe89b32015-03-25 15:55:11 -0700292 /*
293 * dst->anon_vma is dropped here otherwise its degree can be incorrectly
294 * decremented in unlink_anon_vmas().
295 * We can safely do this because callers of anon_vma_clone() don't care
296 * about dst->anon_vma if anon_vma_clone() failed.
297 */
298 dst->anon_vma = NULL;
Rik van Riel5beb4932010-03-05 13:42:07 -0800299 unlink_anon_vmas(dst);
300 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301}
302
Rik van Riel5beb4932010-03-05 13:42:07 -0800303/*
304 * Attach vma to its own anon_vma, as well as to the anon_vmas that
305 * the corresponding VMA in the parent process is attached to.
306 * Returns 0 on success, non-zero on failure.
307 */
308int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309{
Rik van Riel5beb4932010-03-05 13:42:07 -0800310 struct anon_vma_chain *avc;
311 struct anon_vma *anon_vma;
Daniel Forrestc4ea95d2014-12-02 15:59:42 -0800312 int error;
Rik van Riel5beb4932010-03-05 13:42:07 -0800313
314 /* Don't bother if the parent process has no anon_vma here. */
315 if (!pvma->anon_vma)
316 return 0;
317
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800318 /* Drop inherited anon_vma, we'll reuse existing or allocate new. */
319 vma->anon_vma = NULL;
320
Rik van Riel5beb4932010-03-05 13:42:07 -0800321 /*
322 * First, attach the new VMA to the parent VMA's anon_vmas,
323 * so rmap can find non-COWed pages in child processes.
324 */
Daniel Forrestc4ea95d2014-12-02 15:59:42 -0800325 error = anon_vma_clone(vma, pvma);
326 if (error)
327 return error;
Rik van Riel5beb4932010-03-05 13:42:07 -0800328
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800329 /* An existing anon_vma has been reused, all done then. */
330 if (vma->anon_vma)
331 return 0;
332
Rik van Riel5beb4932010-03-05 13:42:07 -0800333 /* Then add our own anon_vma. */
334 anon_vma = anon_vma_alloc();
335 if (!anon_vma)
336 goto out_error;
Linus Torvaldsdd347392011-06-17 19:05:36 -0700337 avc = anon_vma_chain_alloc(GFP_KERNEL);
Rik van Riel5beb4932010-03-05 13:42:07 -0800338 if (!avc)
339 goto out_error_free_anon_vma;
Rik van Riel5c341ee12010-08-09 17:18:39 -0700340
341 /*
342 * The root anon_vma's spinlock is the lock actually used when we
343 * lock any of the anon_vmas in this anon_vma tree.
344 */
345 anon_vma->root = pvma->anon_vma->root;
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800346 anon_vma->parent = pvma->anon_vma;
Rik van Riel76545062010-08-09 17:18:41 -0700347 /*
Peter Zijlstra01d8b202011-03-22 16:32:49 -0700348 * With refcounts, an anon_vma can stay around longer than the
349 * process it belongs to. The root anon_vma needs to be pinned until
350 * this anon_vma is freed, because the lock lives in the root.
Rik van Riel76545062010-08-09 17:18:41 -0700351 */
352 get_anon_vma(anon_vma->root);
Rik van Riel5beb4932010-03-05 13:42:07 -0800353 /* Mark this anon_vma as the one where our new (COWed) pages go. */
354 vma->anon_vma = anon_vma;
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000355 anon_vma_lock_write(anon_vma);
Rik van Riel5c341ee12010-08-09 17:18:39 -0700356 anon_vma_chain_link(vma, avc, anon_vma);
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800357 anon_vma->parent->degree++;
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800358 anon_vma_unlock_write(anon_vma);
Rik van Riel5beb4932010-03-05 13:42:07 -0800359
360 return 0;
361
362 out_error_free_anon_vma:
Peter Zijlstra01d8b202011-03-22 16:32:49 -0700363 put_anon_vma(anon_vma);
Rik van Riel5beb4932010-03-05 13:42:07 -0800364 out_error:
Rik van Riel4946d542010-04-05 12:13:33 -0400365 unlink_anon_vmas(vma);
Rik van Riel5beb4932010-03-05 13:42:07 -0800366 return -ENOMEM;
367}
368
Rik van Riel5beb4932010-03-05 13:42:07 -0800369void unlink_anon_vmas(struct vm_area_struct *vma)
370{
371 struct anon_vma_chain *avc, *next;
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200372 struct anon_vma *root = NULL;
Rik van Riel5beb4932010-03-05 13:42:07 -0800373
Rik van Riel5c341ee12010-08-09 17:18:39 -0700374 /*
375 * Unlink each anon_vma chained to the VMA. This list is ordered
376 * from newest to oldest, ensuring the root anon_vma gets freed last.
377 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800378 list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) {
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200379 struct anon_vma *anon_vma = avc->anon_vma;
380
381 root = lock_anon_vma_root(root, anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700382 anon_vma_interval_tree_remove(avc, &anon_vma->rb_root);
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200383
384 /*
385 * Leave empty anon_vmas on the list - we'll need
386 * to free them outside the lock.
387 */
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800388 if (RB_EMPTY_ROOT(&anon_vma->rb_root)) {
389 anon_vma->parent->degree--;
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200390 continue;
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800391 }
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200392
393 list_del(&avc->same_vma);
394 anon_vma_chain_free(avc);
395 }
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800396 if (vma->anon_vma)
397 vma->anon_vma->degree--;
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200398 unlock_anon_vma_root(root);
399
400 /*
401 * Iterate the list once more, it now only contains empty and unlinked
402 * anon_vmas, destroy them. Could not do before due to __put_anon_vma()
Ingo Molnar5a505082012-12-02 19:56:46 +0000403 * needing to write-acquire the anon_vma->root->rwsem.
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200404 */
405 list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) {
406 struct anon_vma *anon_vma = avc->anon_vma;
407
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800408 BUG_ON(anon_vma->degree);
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200409 put_anon_vma(anon_vma);
410
Rik van Riel5beb4932010-03-05 13:42:07 -0800411 list_del(&avc->same_vma);
412 anon_vma_chain_free(avc);
413 }
414}
415
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700416static void anon_vma_ctor(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417{
Christoph Lametera35afb82007-05-16 22:10:57 -0700418 struct anon_vma *anon_vma = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419
Ingo Molnar5a505082012-12-02 19:56:46 +0000420 init_rwsem(&anon_vma->rwsem);
Peter Zijlstra83813262011-03-22 16:32:48 -0700421 atomic_set(&anon_vma->refcount, 0);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700422 anon_vma->rb_root = RB_ROOT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423}
424
425void __init anon_vma_init(void)
426{
427 anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma),
Paul Mundt20c2df82007-07-20 10:11:58 +0900428 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor);
Rik van Riel5beb4932010-03-05 13:42:07 -0800429 anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain, SLAB_PANIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430}
431
432/*
Peter Zijlstra6111e4c2011-05-24 17:12:08 -0700433 * Getting a lock on a stable anon_vma from a page off the LRU is tricky!
434 *
435 * Since there is no serialization what so ever against page_remove_rmap()
436 * the best this function can do is return a locked anon_vma that might
437 * have been relevant to this page.
438 *
439 * The page might have been remapped to a different anon_vma or the anon_vma
440 * returned may already be freed (and even reused).
441 *
Peter Zijlstrabc658c92011-05-29 10:33:44 +0200442 * In case it was remapped to a different anon_vma, the new anon_vma will be a
443 * child of the old anon_vma, and the anon_vma lifetime rules will therefore
444 * ensure that any anon_vma obtained from the page will still be valid for as
445 * long as we observe page_mapped() [ hence all those page_mapped() tests ].
446 *
Peter Zijlstra6111e4c2011-05-24 17:12:08 -0700447 * All users of this function must be very careful when walking the anon_vma
448 * chain and verify that the page in question is indeed mapped in it
449 * [ something equivalent to page_mapped_in_vma() ].
450 *
451 * Since anon_vma's slab is DESTROY_BY_RCU and we know from page_remove_rmap()
452 * that the anon_vma pointer from page->mapping is valid if there is a
453 * mapcount, we can dereference the anon_vma after observing those.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 */
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700455struct anon_vma *page_get_anon_vma(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456{
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700457 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 unsigned long anon_mapping;
459
460 rcu_read_lock();
Jason Low4db0c3c2015-04-15 16:14:08 -0700461 anon_mapping = (unsigned long)READ_ONCE(page->mapping);
Hugh Dickins3ca7b3c2009-12-14 17:58:57 -0800462 if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 goto out;
464 if (!page_mapped(page))
465 goto out;
466
467 anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700468 if (!atomic_inc_not_zero(&anon_vma->refcount)) {
469 anon_vma = NULL;
470 goto out;
471 }
Hugh Dickinsf18194272010-08-25 23:12:54 -0700472
473 /*
474 * If this page is still mapped, then its anon_vma cannot have been
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700475 * freed. But if it has been unmapped, we have no security against the
476 * anon_vma structure being freed and reused (for another anon_vma:
477 * SLAB_DESTROY_BY_RCU guarantees that - so the atomic_inc_not_zero()
478 * above cannot corrupt).
Hugh Dickinsf18194272010-08-25 23:12:54 -0700479 */
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700480 if (!page_mapped(page)) {
Hugh Dickins7f39dda2014-06-04 16:05:33 -0700481 rcu_read_unlock();
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700482 put_anon_vma(anon_vma);
Hugh Dickins7f39dda2014-06-04 16:05:33 -0700483 return NULL;
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700484 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485out:
486 rcu_read_unlock();
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700487
488 return anon_vma;
489}
490
Peter Zijlstra88c22082011-05-24 17:12:13 -0700491/*
492 * Similar to page_get_anon_vma() except it locks the anon_vma.
493 *
494 * Its a little more complex as it tries to keep the fast path to a single
495 * atomic op -- the trylock. If we fail the trylock, we fall back to getting a
496 * reference like with page_get_anon_vma() and then block on the mutex.
497 */
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000498struct anon_vma *page_lock_anon_vma_read(struct page *page)
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700499{
Peter Zijlstra88c22082011-05-24 17:12:13 -0700500 struct anon_vma *anon_vma = NULL;
Hugh Dickinseee0f252011-05-28 13:20:21 -0700501 struct anon_vma *root_anon_vma;
Peter Zijlstra88c22082011-05-24 17:12:13 -0700502 unsigned long anon_mapping;
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700503
Peter Zijlstra88c22082011-05-24 17:12:13 -0700504 rcu_read_lock();
Jason Low4db0c3c2015-04-15 16:14:08 -0700505 anon_mapping = (unsigned long)READ_ONCE(page->mapping);
Peter Zijlstra88c22082011-05-24 17:12:13 -0700506 if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
507 goto out;
508 if (!page_mapped(page))
509 goto out;
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700510
Peter Zijlstra88c22082011-05-24 17:12:13 -0700511 anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
Jason Low4db0c3c2015-04-15 16:14:08 -0700512 root_anon_vma = READ_ONCE(anon_vma->root);
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000513 if (down_read_trylock(&root_anon_vma->rwsem)) {
Peter Zijlstra88c22082011-05-24 17:12:13 -0700514 /*
Hugh Dickinseee0f252011-05-28 13:20:21 -0700515 * If the page is still mapped, then this anon_vma is still
516 * its anon_vma, and holding the mutex ensures that it will
Peter Zijlstrabc658c92011-05-29 10:33:44 +0200517 * not go away, see anon_vma_free().
Peter Zijlstra88c22082011-05-24 17:12:13 -0700518 */
Hugh Dickinseee0f252011-05-28 13:20:21 -0700519 if (!page_mapped(page)) {
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000520 up_read(&root_anon_vma->rwsem);
Peter Zijlstra88c22082011-05-24 17:12:13 -0700521 anon_vma = NULL;
522 }
523 goto out;
524 }
525
526 /* trylock failed, we got to sleep */
527 if (!atomic_inc_not_zero(&anon_vma->refcount)) {
528 anon_vma = NULL;
529 goto out;
530 }
531
532 if (!page_mapped(page)) {
Hugh Dickins7f39dda2014-06-04 16:05:33 -0700533 rcu_read_unlock();
Peter Zijlstra88c22082011-05-24 17:12:13 -0700534 put_anon_vma(anon_vma);
Hugh Dickins7f39dda2014-06-04 16:05:33 -0700535 return NULL;
Peter Zijlstra88c22082011-05-24 17:12:13 -0700536 }
537
538 /* we pinned the anon_vma, its safe to sleep */
539 rcu_read_unlock();
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000540 anon_vma_lock_read(anon_vma);
Peter Zijlstra88c22082011-05-24 17:12:13 -0700541
542 if (atomic_dec_and_test(&anon_vma->refcount)) {
543 /*
544 * Oops, we held the last refcount, release the lock
545 * and bail -- can't simply use put_anon_vma() because
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000546 * we'll deadlock on the anon_vma_lock_write() recursion.
Peter Zijlstra88c22082011-05-24 17:12:13 -0700547 */
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000548 anon_vma_unlock_read(anon_vma);
Peter Zijlstra88c22082011-05-24 17:12:13 -0700549 __put_anon_vma(anon_vma);
550 anon_vma = NULL;
551 }
552
553 return anon_vma;
554
555out:
556 rcu_read_unlock();
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700557 return anon_vma;
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800558}
559
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000560void page_unlock_anon_vma_read(struct anon_vma *anon_vma)
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800561{
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000562 anon_vma_unlock_read(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563}
564
565/*
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -0800566 * At what user virtual address is page expected in @vma?
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 */
Michel Lespinasse86c2ad12012-10-08 16:31:42 -0700568static inline unsigned long
569__vma_address(struct page *page, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570{
Naoya Horiguchia0f7a752014-07-23 14:00:01 -0700571 pgoff_t pgoff = page_to_pgoff(page);
Michel Lespinasse86c2ad12012-10-08 16:31:42 -0700572 return vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
573}
574
575inline unsigned long
576vma_address(struct page *page, struct vm_area_struct *vma)
577{
578 unsigned long address = __vma_address(page, vma);
579
580 /* page should be within @vma mapping range */
Sasha Levin81d1b092014-10-09 15:28:10 -0700581 VM_BUG_ON_VMA(address < vma->vm_start || address >= vma->vm_end, vma);
Michel Lespinasse86c2ad12012-10-08 16:31:42 -0700582
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 return address;
584}
585
586/*
Huang Shijiebf89c8c2009-10-01 15:44:04 -0700587 * At what user virtual address is page expected in vma?
Naoya Horiguchiab941e02010-05-11 14:06:55 -0700588 * Caller should check the page is actually part of the vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 */
590unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma)
591{
Michel Lespinasse86c2ad12012-10-08 16:31:42 -0700592 unsigned long address;
Andrea Arcangeli21d0d442010-08-09 17:19:10 -0700593 if (PageAnon(page)) {
Hugh Dickins4829b902010-10-02 17:46:06 -0700594 struct anon_vma *page__anon_vma = page_anon_vma(page);
595 /*
596 * Note: swapoff's unuse_vma() is more efficient with this
597 * check, and needs it to match anon_vma when KSM is active.
598 */
599 if (!vma->anon_vma || !page__anon_vma ||
600 vma->anon_vma->root != page__anon_vma->root)
Andrea Arcangeli21d0d442010-08-09 17:19:10 -0700601 return -EFAULT;
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800602 } else if (page->mapping) {
603 if (!vma->vm_file || vma->vm_file->f_mapping != page->mapping)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 return -EFAULT;
605 } else
606 return -EFAULT;
Michel Lespinasse86c2ad12012-10-08 16:31:42 -0700607 address = __vma_address(page, vma);
608 if (unlikely(address < vma->vm_start || address >= vma->vm_end))
609 return -EFAULT;
610 return address;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611}
612
Bob Liu62190492012-12-11 16:00:37 -0800613pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address)
614{
615 pgd_t *pgd;
616 pud_t *pud;
617 pmd_t *pmd = NULL;
Hugh Dickinsf72e7dc2014-06-23 13:22:05 -0700618 pmd_t pmde;
Bob Liu62190492012-12-11 16:00:37 -0800619
620 pgd = pgd_offset(mm, address);
621 if (!pgd_present(*pgd))
622 goto out;
623
624 pud = pud_offset(pgd, address);
625 if (!pud_present(*pud))
626 goto out;
627
628 pmd = pmd_offset(pud, address);
Hugh Dickinsf72e7dc2014-06-23 13:22:05 -0700629 /*
Aneesh Kumar K.V8809aa22015-06-24 16:57:44 -0700630 * Some THP functions use the sequence pmdp_huge_clear_flush(), set_pmd_at()
Hugh Dickinsf72e7dc2014-06-23 13:22:05 -0700631 * without holding anon_vma lock for write. So when looking for a
632 * genuine pmde (in which to find pte), test present and !THP together.
633 */
Christian Borntraegere37c6982014-12-07 21:41:33 +0100634 pmde = *pmd;
635 barrier();
Hugh Dickinsf72e7dc2014-06-23 13:22:05 -0700636 if (!pmd_present(pmde) || pmd_trans_huge(pmde))
Bob Liu62190492012-12-11 16:00:37 -0800637 pmd = NULL;
638out:
639 return pmd;
640}
641
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642/*
Nikita Danilov81b40822005-05-01 08:58:36 -0700643 * Check that @page is mapped at @address into @mm.
644 *
Nick Piggin479db0b2008-08-20 14:09:18 -0700645 * If @sync is false, page_check_address may perform a racy check to avoid
646 * the page table lock when the pte is not present (helpful when reclaiming
647 * highly shared pages).
648 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -0700649 * On success returns with pte mapped and locked.
Nikita Danilov81b40822005-05-01 08:58:36 -0700650 */
Namhyung Kime9a81a82010-10-26 14:22:01 -0700651pte_t *__page_check_address(struct page *page, struct mm_struct *mm,
Nick Piggin479db0b2008-08-20 14:09:18 -0700652 unsigned long address, spinlock_t **ptlp, int sync)
Nikita Danilov81b40822005-05-01 08:58:36 -0700653{
Nikita Danilov81b40822005-05-01 08:58:36 -0700654 pmd_t *pmd;
655 pte_t *pte;
Hugh Dickinsc0718802005-10-29 18:16:31 -0700656 spinlock_t *ptl;
Nikita Danilov81b40822005-05-01 08:58:36 -0700657
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +0900658 if (unlikely(PageHuge(page))) {
Jianguo Wu98398c32013-12-18 17:08:59 -0800659 /* when pud is not present, pte will be NULL */
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +0900660 pte = huge_pte_offset(mm, address);
Jianguo Wu98398c32013-12-18 17:08:59 -0800661 if (!pte)
662 return NULL;
663
Kirill A. Shutemovcb900f42013-11-14 14:31:02 -0800664 ptl = huge_pte_lockptr(page_hstate(page), mm, pte);
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +0900665 goto check;
666 }
667
Bob Liu62190492012-12-11 16:00:37 -0800668 pmd = mm_find_pmd(mm, address);
669 if (!pmd)
Hugh Dickinsc0718802005-10-29 18:16:31 -0700670 return NULL;
671
Hugh Dickinsc0718802005-10-29 18:16:31 -0700672 pte = pte_offset_map(pmd, address);
673 /* Make a quick check before getting the lock */
Nick Piggin479db0b2008-08-20 14:09:18 -0700674 if (!sync && !pte_present(*pte)) {
Hugh Dickinsc0718802005-10-29 18:16:31 -0700675 pte_unmap(pte);
676 return NULL;
Nikita Danilov81b40822005-05-01 08:58:36 -0700677 }
Hugh Dickinsc0718802005-10-29 18:16:31 -0700678
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700679 ptl = pte_lockptr(mm, pmd);
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +0900680check:
Hugh Dickinsc0718802005-10-29 18:16:31 -0700681 spin_lock(ptl);
682 if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) {
683 *ptlp = ptl;
684 return pte;
685 }
686 pte_unmap_unlock(pte, ptl);
687 return NULL;
Nikita Danilov81b40822005-05-01 08:58:36 -0700688}
689
Nick Pigginb291f002008-10-18 20:26:44 -0700690/**
691 * page_mapped_in_vma - check whether a page is really mapped in a VMA
692 * @page: the page to test
693 * @vma: the VMA to test
694 *
695 * Returns 1 if the page is mapped into the page tables of the VMA, 0
696 * if the page is not mapped into the page tables of this VMA. Only
697 * valid for normal file or anonymous VMAs.
698 */
Andi Kleen6a460792009-09-16 11:50:15 +0200699int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma)
Nick Pigginb291f002008-10-18 20:26:44 -0700700{
701 unsigned long address;
702 pte_t *pte;
703 spinlock_t *ptl;
704
Michel Lespinasse86c2ad12012-10-08 16:31:42 -0700705 address = __vma_address(page, vma);
706 if (unlikely(address < vma->vm_start || address >= vma->vm_end))
Nick Pigginb291f002008-10-18 20:26:44 -0700707 return 0;
708 pte = page_check_address(page, vma->vm_mm, address, &ptl, 1);
709 if (!pte) /* the page is not in this mm */
710 return 0;
711 pte_unmap_unlock(pte, ptl);
712
713 return 1;
714}
715
Joonsoo Kim9f326242014-01-21 15:49:53 -0800716struct page_referenced_arg {
717 int mapcount;
718 int referenced;
719 unsigned long vm_flags;
720 struct mem_cgroup *memcg;
721};
Nikita Danilov81b40822005-05-01 08:58:36 -0700722/*
Joonsoo Kim9f326242014-01-21 15:49:53 -0800723 * arg: page_referenced_arg will be passed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 */
Kirill A. Shutemovac769502014-06-04 16:08:17 -0700725static int page_referenced_one(struct page *page, struct vm_area_struct *vma,
Joonsoo Kim9f326242014-01-21 15:49:53 -0800726 unsigned long address, void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727{
728 struct mm_struct *mm = vma->vm_mm;
Kirill A. Shutemov117b0792013-11-14 14:30:56 -0800729 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 int referenced = 0;
Joonsoo Kim9f326242014-01-21 15:49:53 -0800731 struct page_referenced_arg *pra = arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800733 if (unlikely(PageTransHuge(page))) {
734 pmd_t *pmd;
735
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800736 /*
737 * rmap might return false positives; we must filter
738 * these out using page_check_address_pmd().
739 */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800740 pmd = page_check_address_pmd(page, mm, address,
Kirill A. Shutemov117b0792013-11-14 14:30:56 -0800741 PAGE_CHECK_ADDRESS_PMD_FLAG, &ptl);
742 if (!pmd)
Joonsoo Kim9f326242014-01-21 15:49:53 -0800743 return SWAP_AGAIN;
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800744
745 if (vma->vm_flags & VM_LOCKED) {
Kirill A. Shutemov117b0792013-11-14 14:30:56 -0800746 spin_unlock(ptl);
Joonsoo Kim9f326242014-01-21 15:49:53 -0800747 pra->vm_flags |= VM_LOCKED;
748 return SWAP_FAIL; /* To break the loop */
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800749 }
750
751 /* go ahead even if the pmd is pmd_trans_splitting() */
752 if (pmdp_clear_flush_young_notify(vma, address, pmd))
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800753 referenced++;
Kirill A. Shutemov117b0792013-11-14 14:30:56 -0800754 spin_unlock(ptl);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800755 } else {
756 pte_t *pte;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800757
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800758 /*
759 * rmap might return false positives; we must filter
760 * these out using page_check_address().
761 */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800762 pte = page_check_address(page, mm, address, &ptl, 0);
763 if (!pte)
Joonsoo Kim9f326242014-01-21 15:49:53 -0800764 return SWAP_AGAIN;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800765
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800766 if (vma->vm_flags & VM_LOCKED) {
767 pte_unmap_unlock(pte, ptl);
Joonsoo Kim9f326242014-01-21 15:49:53 -0800768 pra->vm_flags |= VM_LOCKED;
769 return SWAP_FAIL; /* To break the loop */
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800770 }
771
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800772 if (ptep_clear_flush_young_notify(vma, address, pte)) {
773 /*
774 * Don't treat a reference through a sequentially read
775 * mapping as such. If the page has been used in
776 * another mapping, we will catch it; if this other
777 * mapping is already gone, the unmap path will have
778 * set PG_referenced or activated the page.
779 */
Joe Perches64363aa2013-07-08 16:00:18 -0700780 if (likely(!(vma->vm_flags & VM_SEQ_READ)))
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800781 referenced++;
782 }
783 pte_unmap_unlock(pte, ptl);
784 }
785
Joonsoo Kim9f326242014-01-21 15:49:53 -0800786 if (referenced) {
787 pra->referenced++;
788 pra->vm_flags |= vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 }
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800790
Joonsoo Kim9f326242014-01-21 15:49:53 -0800791 pra->mapcount--;
792 if (!pra->mapcount)
793 return SWAP_SUCCESS; /* To break the loop */
794
795 return SWAP_AGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796}
797
Joonsoo Kim9f326242014-01-21 15:49:53 -0800798static bool invalid_page_referenced_vma(struct vm_area_struct *vma, void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799{
Joonsoo Kim9f326242014-01-21 15:49:53 -0800800 struct page_referenced_arg *pra = arg;
801 struct mem_cgroup *memcg = pra->memcg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802
Joonsoo Kim9f326242014-01-21 15:49:53 -0800803 if (!mm_match_cgroup(vma->vm_mm, memcg))
804 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805
Joonsoo Kim9f326242014-01-21 15:49:53 -0800806 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807}
808
809/**
810 * page_referenced - test if the page was referenced
811 * @page: the page to test
812 * @is_locked: caller holds lock on the page
Johannes Weiner72835c82012-01-12 17:18:32 -0800813 * @memcg: target memory cgroup
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700814 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 *
816 * Quick test_and_clear_referenced for all mappings to a page,
817 * returns the number of ptes which referenced the page.
818 */
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700819int page_referenced(struct page *page,
820 int is_locked,
Johannes Weiner72835c82012-01-12 17:18:32 -0800821 struct mem_cgroup *memcg,
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700822 unsigned long *vm_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823{
Joonsoo Kim9f326242014-01-21 15:49:53 -0800824 int ret;
Hugh Dickins5ad64682009-12-14 17:59:24 -0800825 int we_locked = 0;
Joonsoo Kim9f326242014-01-21 15:49:53 -0800826 struct page_referenced_arg pra = {
827 .mapcount = page_mapcount(page),
828 .memcg = memcg,
829 };
830 struct rmap_walk_control rwc = {
831 .rmap_one = page_referenced_one,
832 .arg = (void *)&pra,
833 .anon_lock = page_lock_anon_vma_read,
834 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700836 *vm_flags = 0;
Joonsoo Kim9f326242014-01-21 15:49:53 -0800837 if (!page_mapped(page))
838 return 0;
839
840 if (!page_rmapping(page))
841 return 0;
842
843 if (!is_locked && (!PageAnon(page) || PageKsm(page))) {
844 we_locked = trylock_page(page);
845 if (!we_locked)
846 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 }
Joonsoo Kim9f326242014-01-21 15:49:53 -0800848
849 /*
850 * If we are reclaiming on behalf of a cgroup, skip
851 * counting on behalf of references from different
852 * cgroups
853 */
854 if (memcg) {
855 rwc.invalid_vma = invalid_page_referenced_vma;
856 }
857
858 ret = rmap_walk(page, &rwc);
859 *vm_flags = pra.vm_flags;
860
861 if (we_locked)
862 unlock_page(page);
863
864 return pra.referenced;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865}
866
Hugh Dickins1cb17292009-12-14 17:59:01 -0800867static int page_mkclean_one(struct page *page, struct vm_area_struct *vma,
Joonsoo Kim9853a402014-01-21 15:49:55 -0800868 unsigned long address, void *arg)
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700869{
870 struct mm_struct *mm = vma->vm_mm;
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100871 pte_t *pte;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700872 spinlock_t *ptl;
873 int ret = 0;
Joonsoo Kim9853a402014-01-21 15:49:55 -0800874 int *cleaned = arg;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700875
Nick Piggin479db0b2008-08-20 14:09:18 -0700876 pte = page_check_address(page, mm, address, &ptl, 1);
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700877 if (!pte)
878 goto out;
879
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100880 if (pte_dirty(*pte) || pte_write(*pte)) {
881 pte_t entry;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700882
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100883 flush_cache_page(vma, address, pte_pfn(*pte));
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700884 entry = ptep_clear_flush(vma, address, pte);
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100885 entry = pte_wrprotect(entry);
886 entry = pte_mkclean(entry);
Al Virod6e88e62006-12-29 16:48:35 -0800887 set_pte_at(mm, address, pte, entry);
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100888 ret = 1;
889 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700890
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700891 pte_unmap_unlock(pte, ptl);
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700892
Joonsoo Kim9853a402014-01-21 15:49:55 -0800893 if (ret) {
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700894 mmu_notifier_invalidate_page(mm, address);
Joonsoo Kim9853a402014-01-21 15:49:55 -0800895 (*cleaned)++;
896 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700897out:
Joonsoo Kim9853a402014-01-21 15:49:55 -0800898 return SWAP_AGAIN;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700899}
900
Joonsoo Kim9853a402014-01-21 15:49:55 -0800901static bool invalid_mkclean_vma(struct vm_area_struct *vma, void *arg)
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700902{
Joonsoo Kim9853a402014-01-21 15:49:55 -0800903 if (vma->vm_flags & VM_SHARED)
Fengguang Wu871beb82014-01-23 15:53:41 -0800904 return false;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700905
Fengguang Wu871beb82014-01-23 15:53:41 -0800906 return true;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700907}
908
909int page_mkclean(struct page *page)
910{
Joonsoo Kim9853a402014-01-21 15:49:55 -0800911 int cleaned = 0;
912 struct address_space *mapping;
913 struct rmap_walk_control rwc = {
914 .arg = (void *)&cleaned,
915 .rmap_one = page_mkclean_one,
916 .invalid_vma = invalid_mkclean_vma,
917 };
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700918
919 BUG_ON(!PageLocked(page));
920
Joonsoo Kim9853a402014-01-21 15:49:55 -0800921 if (!page_mapped(page))
922 return 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700923
Joonsoo Kim9853a402014-01-21 15:49:55 -0800924 mapping = page_mapping(page);
925 if (!mapping)
926 return 0;
927
928 rmap_walk(page, &rwc);
929
930 return cleaned;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700931}
Jaya Kumar60b59be2007-05-08 00:37:37 -0700932EXPORT_SYMBOL_GPL(page_mkclean);
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700933
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934/**
Rik van Rielc44b6742010-03-05 13:42:09 -0800935 * page_move_anon_rmap - move a page to our anon_vma
936 * @page: the page to move to our anon_vma
937 * @vma: the vma the page belongs to
938 * @address: the user virtual address mapped
939 *
940 * When a page belongs exclusively to one process after a COW event,
941 * that page can be moved into the anon_vma that belongs to just that
942 * process, so the rmap code will not search the parent or sibling
943 * processes.
944 */
945void page_move_anon_rmap(struct page *page,
946 struct vm_area_struct *vma, unsigned long address)
947{
948 struct anon_vma *anon_vma = vma->anon_vma;
949
Sasha Levin309381fea2014-01-23 15:52:54 -0800950 VM_BUG_ON_PAGE(!PageLocked(page), page);
Sasha Levin81d1b092014-10-09 15:28:10 -0700951 VM_BUG_ON_VMA(!anon_vma, vma);
Sasha Levin309381fea2014-01-23 15:52:54 -0800952 VM_BUG_ON_PAGE(page->index != linear_page_index(vma, address), page);
Rik van Rielc44b6742010-03-05 13:42:09 -0800953
954 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
Vladimir Davydov414e2fb2015-06-24 16:56:56 -0700955 /*
956 * Ensure that anon_vma and the PAGE_MAPPING_ANON bit are written
957 * simultaneously, so a concurrent reader (eg page_referenced()'s
958 * PageAnon()) will not see one without the other.
959 */
960 WRITE_ONCE(page->mapping, (struct address_space *) anon_vma);
Rik van Rielc44b6742010-03-05 13:42:09 -0800961}
962
963/**
Andi Kleen4e1c1972010-09-22 12:43:56 +0200964 * __page_set_anon_rmap - set up new anonymous rmap
965 * @page: Page to add to rmap
966 * @vma: VM area to add page to.
967 * @address: User virtual address of the mapping
Rik van Riele8a03fe2010-04-14 17:59:28 -0400968 * @exclusive: the page is exclusively owned by the current process
Nick Piggin9617d952006-01-06 00:11:12 -0800969 */
970static void __page_set_anon_rmap(struct page *page,
Rik van Riele8a03fe2010-04-14 17:59:28 -0400971 struct vm_area_struct *vma, unsigned long address, int exclusive)
Nick Piggin9617d952006-01-06 00:11:12 -0800972{
Rik van Riele8a03fe2010-04-14 17:59:28 -0400973 struct anon_vma *anon_vma = vma->anon_vma;
Nick Piggin9617d952006-01-06 00:11:12 -0800974
Rik van Riele8a03fe2010-04-14 17:59:28 -0400975 BUG_ON(!anon_vma);
Linus Torvaldsea900022010-04-12 12:44:29 -0700976
Andi Kleen4e1c1972010-09-22 12:43:56 +0200977 if (PageAnon(page))
978 return;
979
Linus Torvaldsea900022010-04-12 12:44:29 -0700980 /*
Rik van Riele8a03fe2010-04-14 17:59:28 -0400981 * If the page isn't exclusively mapped into this vma,
982 * we must use the _oldest_ possible anon_vma for the
983 * page mapping!
Linus Torvaldsea900022010-04-12 12:44:29 -0700984 */
Andi Kleen4e1c1972010-09-22 12:43:56 +0200985 if (!exclusive)
Andrea Arcangeli288468c2010-08-09 17:19:09 -0700986 anon_vma = anon_vma->root;
Linus Torvaldsea900022010-04-12 12:44:29 -0700987
Nick Piggin9617d952006-01-06 00:11:12 -0800988 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
989 page->mapping = (struct address_space *) anon_vma;
Nick Piggin9617d952006-01-06 00:11:12 -0800990 page->index = linear_page_index(vma, address);
Nick Piggin9617d952006-01-06 00:11:12 -0800991}
992
993/**
Randy Dunlap43d8eac2008-03-19 17:00:43 -0700994 * __page_check_anon_rmap - sanity check anonymous rmap addition
Nick Pigginc97a9e12007-05-16 22:11:21 -0700995 * @page: the page to add the mapping to
996 * @vma: the vm area in which the mapping is added
997 * @address: the user virtual address mapped
998 */
999static void __page_check_anon_rmap(struct page *page,
1000 struct vm_area_struct *vma, unsigned long address)
1001{
1002#ifdef CONFIG_DEBUG_VM
1003 /*
1004 * The page's anon-rmap details (mapping and index) are guaranteed to
1005 * be set up correctly at this point.
1006 *
1007 * We have exclusion against page_add_anon_rmap because the caller
1008 * always holds the page locked, except if called from page_dup_rmap,
1009 * in which case the page is already known to be setup.
1010 *
1011 * We have exclusion against page_add_new_anon_rmap because those pages
1012 * are initially only visible via the pagetables, and the pte is locked
1013 * over the call to page_add_new_anon_rmap.
1014 */
Andrea Arcangeli44ab57a2010-08-09 17:19:10 -07001015 BUG_ON(page_anon_vma(page)->root != vma->anon_vma->root);
Nick Pigginc97a9e12007-05-16 22:11:21 -07001016 BUG_ON(page->index != linear_page_index(vma, address));
1017#endif
1018}
1019
1020/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 * page_add_anon_rmap - add pte mapping to an anonymous page
1022 * @page: the page to add the mapping to
1023 * @vma: the vm area in which the mapping is added
1024 * @address: the user virtual address mapped
1025 *
Hugh Dickins5ad64682009-12-14 17:59:24 -08001026 * The caller needs to hold the pte lock, and the page must be locked in
Hugh Dickins80e148222009-12-14 17:59:29 -08001027 * the anon_vma case: to serialize mapping,index checking after setting,
1028 * and to ensure that PageAnon is not being upgraded racily to PageKsm
1029 * (but PageKsm is never downgraded to PageAnon).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 */
1031void page_add_anon_rmap(struct page *page,
1032 struct vm_area_struct *vma, unsigned long address)
1033{
Rik van Rielad8c2ee2010-08-09 17:19:48 -07001034 do_page_add_anon_rmap(page, vma, address, 0);
1035}
1036
1037/*
1038 * Special version of the above for do_swap_page, which often runs
1039 * into pages that are exclusively owned by the current process.
1040 * Everybody else should continue to use page_add_anon_rmap above.
1041 */
1042void do_page_add_anon_rmap(struct page *page,
1043 struct vm_area_struct *vma, unsigned long address, int exclusive)
1044{
Hugh Dickins5ad64682009-12-14 17:59:24 -08001045 int first = atomic_inc_and_test(&page->_mapcount);
Andrea Arcangeli79134172011-01-13 15:46:58 -08001046 if (first) {
Jianyu Zhanbea04b02014-06-04 16:09:51 -07001047 /*
1048 * We use the irq-unsafe __{inc|mod}_zone_page_stat because
1049 * these counters are not modified in interrupt context, and
1050 * pte lock(a spinlock) is held, which implies preemption
1051 * disabled.
1052 */
Kirill A. Shutemov3cd14fc2013-09-12 15:13:58 -07001053 if (PageTransHuge(page))
Andrea Arcangeli79134172011-01-13 15:46:58 -08001054 __inc_zone_page_state(page,
1055 NR_ANON_TRANSPARENT_HUGEPAGES);
Kirill A. Shutemov3cd14fc2013-09-12 15:13:58 -07001056 __mod_zone_page_state(page_zone(page), NR_ANON_PAGES,
1057 hpage_nr_pages(page));
Andrea Arcangeli79134172011-01-13 15:46:58 -08001058 }
Hugh Dickins5ad64682009-12-14 17:59:24 -08001059 if (unlikely(PageKsm(page)))
1060 return;
1061
Sasha Levin309381fea2014-01-23 15:52:54 -08001062 VM_BUG_ON_PAGE(!PageLocked(page), page);
Hugh Dickins5dbe0af2011-05-28 13:17:04 -07001063 /* address might be in next vma when migration races vma_adjust */
Hugh Dickins5ad64682009-12-14 17:59:24 -08001064 if (first)
Rik van Rielad8c2ee2010-08-09 17:19:48 -07001065 __page_set_anon_rmap(page, vma, address, exclusive);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001066 else
Nick Pigginc97a9e12007-05-16 22:11:21 -07001067 __page_check_anon_rmap(page, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068}
1069
Randy Dunlap43d8eac2008-03-19 17:00:43 -07001070/**
Nick Piggin9617d952006-01-06 00:11:12 -08001071 * page_add_new_anon_rmap - add pte mapping to a new anonymous page
1072 * @page: the page to add the mapping to
1073 * @vma: the vm area in which the mapping is added
1074 * @address: the user virtual address mapped
1075 *
1076 * Same as page_add_anon_rmap but must only be called on *new* pages.
1077 * This means the inc-and-test can be bypassed.
Nick Pigginc97a9e12007-05-16 22:11:21 -07001078 * Page does not have to be locked.
Nick Piggin9617d952006-01-06 00:11:12 -08001079 */
1080void page_add_new_anon_rmap(struct page *page,
1081 struct vm_area_struct *vma, unsigned long address)
1082{
Sasha Levin81d1b092014-10-09 15:28:10 -07001083 VM_BUG_ON_VMA(address < vma->vm_start || address >= vma->vm_end, vma);
Hugh Dickinscbf84b72009-01-06 14:39:27 -08001084 SetPageSwapBacked(page);
1085 atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */
Kirill A. Shutemov3cd14fc2013-09-12 15:13:58 -07001086 if (PageTransHuge(page))
Andrea Arcangeli79134172011-01-13 15:46:58 -08001087 __inc_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES);
Kirill A. Shutemov3cd14fc2013-09-12 15:13:58 -07001088 __mod_zone_page_state(page_zone(page), NR_ANON_PAGES,
1089 hpage_nr_pages(page));
Rik van Riele8a03fe2010-04-14 17:59:28 -04001090 __page_set_anon_rmap(page, vma, address, 1);
Nick Piggin9617d952006-01-06 00:11:12 -08001091}
1092
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093/**
1094 * page_add_file_rmap - add pte mapping to a file page
1095 * @page: the page to add the mapping to
1096 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -07001097 * The caller needs to hold the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 */
1099void page_add_file_rmap(struct page *page)
1100{
Johannes Weinerd7365e72014-10-29 14:50:48 -07001101 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001102
Johannes Weiner6de22612015-02-11 15:25:01 -08001103 memcg = mem_cgroup_begin_page_stat(page);
Balbir Singhd69b0422009-06-17 16:26:34 -07001104 if (atomic_inc_and_test(&page->_mapcount)) {
Christoph Lameter65ba55f2006-06-30 01:55:34 -07001105 __inc_zone_page_state(page, NR_FILE_MAPPED);
Johannes Weinerd7365e72014-10-29 14:50:48 -07001106 mem_cgroup_inc_page_stat(memcg, MEM_CGROUP_STAT_FILE_MAPPED);
Balbir Singhd69b0422009-06-17 16:26:34 -07001107 }
Johannes Weiner6de22612015-02-11 15:25:01 -08001108 mem_cgroup_end_page_stat(memcg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109}
1110
Johannes Weiner8186eb62014-10-29 14:50:51 -07001111static void page_remove_file_rmap(struct page *page)
1112{
1113 struct mem_cgroup *memcg;
Johannes Weiner8186eb62014-10-29 14:50:51 -07001114
Johannes Weiner6de22612015-02-11 15:25:01 -08001115 memcg = mem_cgroup_begin_page_stat(page);
Johannes Weiner8186eb62014-10-29 14:50:51 -07001116
1117 /* page still mapped by someone else? */
1118 if (!atomic_add_negative(-1, &page->_mapcount))
1119 goto out;
1120
1121 /* Hugepages are not counted in NR_FILE_MAPPED for now. */
1122 if (unlikely(PageHuge(page)))
1123 goto out;
1124
1125 /*
1126 * We use the irq-unsafe __{inc|mod}_zone_page_stat because
1127 * these counters are not modified in interrupt context, and
1128 * pte lock(a spinlock) is held, which implies preemption disabled.
1129 */
1130 __dec_zone_page_state(page, NR_FILE_MAPPED);
1131 mem_cgroup_dec_page_stat(memcg, MEM_CGROUP_STAT_FILE_MAPPED);
1132
1133 if (unlikely(PageMlocked(page)))
1134 clear_page_mlock(page);
1135out:
Johannes Weiner6de22612015-02-11 15:25:01 -08001136 mem_cgroup_end_page_stat(memcg);
Johannes Weiner8186eb62014-10-29 14:50:51 -07001137}
1138
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139/**
1140 * page_remove_rmap - take down pte mapping from a page
1141 * @page: page to remove mapping from
1142 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -07001143 * The caller needs to hold the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 */
Hugh Dickinsedc315f2009-01-06 14:40:11 -08001145void page_remove_rmap(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146{
Johannes Weiner8186eb62014-10-29 14:50:51 -07001147 if (!PageAnon(page)) {
1148 page_remove_file_rmap(page);
1149 return;
1150 }
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001151
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001152 /* page still mapped by someone else? */
1153 if (!atomic_add_negative(-1, &page->_mapcount))
Johannes Weiner8186eb62014-10-29 14:50:51 -07001154 return;
1155
1156 /* Hugepages are not counted in NR_ANON_PAGES for now. */
1157 if (unlikely(PageHuge(page)))
1158 return;
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001159
1160 /*
Jianyu Zhanbea04b02014-06-04 16:09:51 -07001161 * We use the irq-unsafe __{inc|mod}_zone_page_stat because
1162 * these counters are not modified in interrupt context, and
Jianyu Zhanbea04b02014-06-04 16:09:51 -07001163 * pte lock(a spinlock) is held, which implies preemption disabled.
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001164 */
Johannes Weiner8186eb62014-10-29 14:50:51 -07001165 if (PageTransHuge(page))
1166 __dec_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES);
1167
1168 __mod_zone_page_state(page_zone(page), NR_ANON_PAGES,
1169 -hpage_nr_pages(page));
1170
Hugh Dickinse6c509f2012-10-08 16:33:19 -07001171 if (unlikely(PageMlocked(page)))
1172 clear_page_mlock(page);
Johannes Weiner8186eb62014-10-29 14:50:51 -07001173
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001174 /*
1175 * It would be tidy to reset the PageAnon mapping here,
1176 * but that might overwrite a racing page_add_anon_rmap
1177 * which increments mapcount after us but sets mapping
1178 * before us: so leave the reset to free_hot_cold_page,
1179 * and remember that it's only reliable while mapped.
1180 * Leaving it set also helps swapoff to reinstate ptes
1181 * faster for those pages still in swapcache.
1182 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183}
1184
1185/*
Joonsoo Kim52629502014-01-21 15:49:50 -08001186 * @arg: enum ttu_flags will be passed to this argument
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 */
Kirill A. Shutemovac769502014-06-04 16:08:17 -07001188static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
Joonsoo Kim52629502014-01-21 15:49:50 -08001189 unsigned long address, void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190{
1191 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 pte_t *pte;
1193 pte_t pteval;
Hugh Dickinsc0718802005-10-29 18:16:31 -07001194 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 int ret = SWAP_AGAIN;
Joonsoo Kim52629502014-01-21 15:49:50 -08001196 enum ttu_flags flags = (enum ttu_flags)arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197
Nick Piggin479db0b2008-08-20 14:09:18 -07001198 pte = page_check_address(page, mm, address, &ptl, 0);
Hugh Dickinsc0718802005-10-29 18:16:31 -07001199 if (!pte)
Nikita Danilov81b40822005-05-01 08:58:36 -07001200 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201
1202 /*
1203 * If the page is mlock()d, we cannot swap it out.
1204 * If it's recently referenced (perhaps page_referenced
1205 * skipped over this mm) then we should reactivate it.
1206 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001207 if (!(flags & TTU_IGNORE_MLOCK)) {
KOSAKI Motohirocaed0f482009-12-14 17:59:45 -08001208 if (vma->vm_flags & VM_LOCKED)
1209 goto out_mlock;
1210
Konstantin Khlebnikovdaa5ba72014-06-04 16:10:52 -07001211 if (flags & TTU_MUNLOCK)
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001212 goto out_unmap;
Andi Kleen14fa31b2009-09-16 11:50:10 +02001213 }
1214 if (!(flags & TTU_IGNORE_ACCESS)) {
Nick Pigginb291f002008-10-18 20:26:44 -07001215 if (ptep_clear_flush_young_notify(vma, address, pte)) {
1216 ret = SWAP_FAIL;
1217 goto out_unmap;
1218 }
1219 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 /* Nuke the page table entry. */
1222 flush_cache_page(vma, address, page_to_pfn(page));
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001223 pteval = ptep_clear_flush(vma, address, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224
1225 /* Move the dirty bit to the physical page now the pte is gone. */
1226 if (pte_dirty(pteval))
1227 set_page_dirty(page);
1228
Hugh Dickins365e9c872005-10-29 18:16:18 -07001229 /* Update high watermark before we lower rss */
1230 update_hiwater_rss(mm);
1231
Andi Kleen888b9f72009-09-16 11:50:11 +02001232 if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) {
Naoya Horiguchi5f24ae52012-12-12 13:52:30 -08001233 if (!PageHuge(page)) {
1234 if (PageAnon(page))
1235 dec_mm_counter(mm, MM_ANONPAGES);
1236 else
1237 dec_mm_counter(mm, MM_FILEPAGES);
1238 }
Andi Kleen888b9f72009-09-16 11:50:11 +02001239 set_pte_at(mm, address, pte,
Naoya Horiguchi5f24ae52012-12-12 13:52:30 -08001240 swp_entry_to_pte(make_hwpoison_entry(page)));
Konstantin Weitz45961722013-04-17 13:59:32 +02001241 } else if (pte_unused(pteval)) {
1242 /*
1243 * The guest indicated that the page content is of no
1244 * interest anymore. Simply discard the pte, vmscan
1245 * will take care of the rest.
1246 */
1247 if (PageAnon(page))
1248 dec_mm_counter(mm, MM_ANONPAGES);
1249 else
1250 dec_mm_counter(mm, MM_FILEPAGES);
Andi Kleen888b9f72009-09-16 11:50:11 +02001251 } else if (PageAnon(page)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001252 swp_entry_t entry = { .val = page_private(page) };
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001253 pte_t swp_pte;
Christoph Lameter06972122006-06-23 02:03:35 -07001254
1255 if (PageSwapCache(page)) {
1256 /*
1257 * Store the swap location in the pte.
1258 * See handle_pte_fault() ...
1259 */
Hugh Dickins570a335b2009-12-14 17:58:46 -08001260 if (swap_duplicate(entry) < 0) {
1261 set_pte_at(mm, address, pte, pteval);
1262 ret = SWAP_FAIL;
1263 goto out_unmap;
1264 }
Christoph Lameter06972122006-06-23 02:03:35 -07001265 if (list_empty(&mm->mmlist)) {
1266 spin_lock(&mmlist_lock);
1267 if (list_empty(&mm->mmlist))
1268 list_add(&mm->mmlist, &init_mm.mmlist);
1269 spin_unlock(&mmlist_lock);
1270 }
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001271 dec_mm_counter(mm, MM_ANONPAGES);
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08001272 inc_mm_counter(mm, MM_SWAPENTS);
Konstantin Khlebnikovce1744f2012-03-21 16:33:59 -07001273 } else if (IS_ENABLED(CONFIG_MIGRATION)) {
Christoph Lameter06972122006-06-23 02:03:35 -07001274 /*
1275 * Store the pfn of the page in a special migration
1276 * pte. do_swap_page() will wait until the migration
1277 * pte is removed and then restart fault handling.
1278 */
Konstantin Khlebnikovdaa5ba72014-06-04 16:10:52 -07001279 BUG_ON(!(flags & TTU_MIGRATION));
Christoph Lameter06972122006-06-23 02:03:35 -07001280 entry = make_migration_entry(page, pte_write(pteval));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 }
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001282 swp_pte = swp_entry_to_pte(entry);
1283 if (pte_soft_dirty(pteval))
1284 swp_pte = pte_swp_mksoft_dirty(swp_pte);
1285 set_pte_at(mm, address, pte, swp_pte);
Konstantin Khlebnikovce1744f2012-03-21 16:33:59 -07001286 } else if (IS_ENABLED(CONFIG_MIGRATION) &&
Konstantin Khlebnikovdaa5ba72014-06-04 16:10:52 -07001287 (flags & TTU_MIGRATION)) {
Christoph Lameter04e62a22006-06-23 02:03:38 -07001288 /* Establish migration entry for a file page */
1289 swp_entry_t entry;
1290 entry = make_migration_entry(page, pte_write(pteval));
1291 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
1292 } else
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001293 dec_mm_counter(mm, MM_FILEPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294
Hugh Dickinsedc315f2009-01-06 14:40:11 -08001295 page_remove_rmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 page_cache_release(page);
1297
1298out_unmap:
Hugh Dickinsc0718802005-10-29 18:16:31 -07001299 pte_unmap_unlock(pte, ptl);
Konstantin Khlebnikovdaa5ba72014-06-04 16:10:52 -07001300 if (ret != SWAP_FAIL && !(flags & TTU_MUNLOCK))
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001301 mmu_notifier_invalidate_page(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302out:
1303 return ret;
KOSAKI Motohirocaed0f482009-12-14 17:59:45 -08001304
1305out_mlock:
1306 pte_unmap_unlock(pte, ptl);
1307
1308
1309 /*
1310 * We need mmap_sem locking, Otherwise VM_LOCKED check makes
1311 * unstable result and race. Plus, We can't wait here because
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08001312 * we now hold anon_vma->rwsem or mapping->i_mmap_rwsem.
KOSAKI Motohirocaed0f482009-12-14 17:59:45 -08001313 * if trylock failed, the page remain in evictable lru and later
1314 * vmscan could retry to move the page to unevictable lru if the
1315 * page is actually mlocked.
1316 */
1317 if (down_read_trylock(&vma->vm_mm->mmap_sem)) {
1318 if (vma->vm_flags & VM_LOCKED) {
1319 mlock_vma_page(page);
1320 ret = SWAP_MLOCK;
1321 }
1322 up_read(&vma->vm_mm->mmap_sem);
1323 }
1324 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325}
1326
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001327bool is_vma_temporary_stack(struct vm_area_struct *vma)
Mel Gormana8bef8ff2010-05-24 14:32:24 -07001328{
1329 int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP);
1330
1331 if (!maybe_stack)
1332 return false;
1333
1334 if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) ==
1335 VM_STACK_INCOMPLETE_SETUP)
1336 return true;
1337
1338 return false;
1339}
1340
Joonsoo Kim52629502014-01-21 15:49:50 -08001341static bool invalid_migration_vma(struct vm_area_struct *vma, void *arg)
1342{
1343 return is_vma_temporary_stack(vma);
1344}
1345
Joonsoo Kim52629502014-01-21 15:49:50 -08001346static int page_not_mapped(struct page *page)
1347{
1348 return !page_mapped(page);
1349};
1350
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351/**
1352 * try_to_unmap - try to remove all page table mappings to a page
1353 * @page: the page to get unmapped
Andi Kleen14fa31b2009-09-16 11:50:10 +02001354 * @flags: action and flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 *
1356 * Tries to remove all the page table entries which are mapping this
1357 * page, used in the pageout path. Caller must hold the page lock.
1358 * Return values are:
1359 *
1360 * SWAP_SUCCESS - we succeeded in removing all mappings
1361 * SWAP_AGAIN - we missed a mapping, try again later
1362 * SWAP_FAIL - the page is unswappable
Nick Pigginb291f002008-10-18 20:26:44 -07001363 * SWAP_MLOCK - page is mlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001365int try_to_unmap(struct page *page, enum ttu_flags flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366{
1367 int ret;
Joonsoo Kim52629502014-01-21 15:49:50 -08001368 struct rmap_walk_control rwc = {
1369 .rmap_one = try_to_unmap_one,
1370 .arg = (void *)flags,
1371 .done = page_not_mapped,
Joonsoo Kim52629502014-01-21 15:49:50 -08001372 .anon_lock = page_lock_anon_vma_read,
1373 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374
Sasha Levin309381fea2014-01-23 15:52:54 -08001375 VM_BUG_ON_PAGE(!PageHuge(page) && PageTransHuge(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376
Joonsoo Kim52629502014-01-21 15:49:50 -08001377 /*
1378 * During exec, a temporary VMA is setup and later moved.
1379 * The VMA is moved under the anon_vma lock but not the
1380 * page tables leading to a race where migration cannot
1381 * find the migration ptes. Rather than increasing the
1382 * locking requirements of exec(), migration skips
1383 * temporary VMAs until after exec() completes.
1384 */
Konstantin Khlebnikovdaa5ba72014-06-04 16:10:52 -07001385 if ((flags & TTU_MIGRATION) && !PageKsm(page) && PageAnon(page))
Joonsoo Kim52629502014-01-21 15:49:50 -08001386 rwc.invalid_vma = invalid_migration_vma;
1387
1388 ret = rmap_walk(page, &rwc);
1389
Nick Pigginb291f002008-10-18 20:26:44 -07001390 if (ret != SWAP_MLOCK && !page_mapped(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 ret = SWAP_SUCCESS;
1392 return ret;
1393}
Nikita Danilov81b40822005-05-01 08:58:36 -07001394
Nick Pigginb291f002008-10-18 20:26:44 -07001395/**
1396 * try_to_munlock - try to munlock a page
1397 * @page: the page to be munlocked
1398 *
1399 * Called from munlock code. Checks all of the VMAs mapping the page
1400 * to make sure nobody else has this page mlocked. The page will be
1401 * returned with PG_mlocked cleared if no other vmas have it mlocked.
1402 *
1403 * Return values are:
1404 *
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001405 * SWAP_AGAIN - no vma is holding page mlocked, or,
Nick Pigginb291f002008-10-18 20:26:44 -07001406 * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem
Hugh Dickins5ad64682009-12-14 17:59:24 -08001407 * SWAP_FAIL - page cannot be located at present
Nick Pigginb291f002008-10-18 20:26:44 -07001408 * SWAP_MLOCK - page is now mlocked.
1409 */
1410int try_to_munlock(struct page *page)
1411{
Joonsoo Kime8351ac2014-01-21 15:49:52 -08001412 int ret;
1413 struct rmap_walk_control rwc = {
1414 .rmap_one = try_to_unmap_one,
1415 .arg = (void *)TTU_MUNLOCK,
1416 .done = page_not_mapped,
Joonsoo Kime8351ac2014-01-21 15:49:52 -08001417 .anon_lock = page_lock_anon_vma_read,
1418
1419 };
1420
Sasha Levin309381fea2014-01-23 15:52:54 -08001421 VM_BUG_ON_PAGE(!PageLocked(page) || PageLRU(page), page);
Nick Pigginb291f002008-10-18 20:26:44 -07001422
Joonsoo Kime8351ac2014-01-21 15:49:52 -08001423 ret = rmap_walk(page, &rwc);
1424 return ret;
Nick Pigginb291f002008-10-18 20:26:44 -07001425}
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001426
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001427void __put_anon_vma(struct anon_vma *anon_vma)
Rik van Riel76545062010-08-09 17:18:41 -07001428{
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001429 struct anon_vma *root = anon_vma->root;
Rik van Riel76545062010-08-09 17:18:41 -07001430
Andrey Ryabinin624483f2014-06-06 19:09:30 +04001431 anon_vma_free(anon_vma);
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001432 if (root != anon_vma && atomic_dec_and_test(&root->refcount))
1433 anon_vma_free(root);
Rik van Riel76545062010-08-09 17:18:41 -07001434}
Rik van Riel76545062010-08-09 17:18:41 -07001435
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001436static struct anon_vma *rmap_walk_anon_lock(struct page *page,
1437 struct rmap_walk_control *rwc)
Joonsoo Kimfaecd8d2014-01-21 15:49:46 -08001438{
1439 struct anon_vma *anon_vma;
1440
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001441 if (rwc->anon_lock)
1442 return rwc->anon_lock(page);
1443
Joonsoo Kimfaecd8d2014-01-21 15:49:46 -08001444 /*
1445 * Note: remove_migration_ptes() cannot use page_lock_anon_vma_read()
1446 * because that depends on page_mapped(); but not all its usages
1447 * are holding mmap_sem. Users without mmap_sem are required to
1448 * take a reference count to prevent the anon_vma disappearing
1449 */
1450 anon_vma = page_anon_vma(page);
1451 if (!anon_vma)
1452 return NULL;
1453
1454 anon_vma_lock_read(anon_vma);
1455 return anon_vma;
1456}
1457
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001458/*
Joonsoo Kime8351ac2014-01-21 15:49:52 -08001459 * rmap_walk_anon - do something to anonymous page using the object-based
1460 * rmap method
1461 * @page: the page to be handled
1462 * @rwc: control variable according to each walk type
1463 *
1464 * Find all the mappings of a page using the mapping pointer and the vma chains
1465 * contained in the anon_vma struct it points to.
1466 *
1467 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1468 * where the page was found will be held for write. So, we won't recheck
1469 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1470 * LOCKED.
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001471 */
Joonsoo Kim051ac832014-01-21 15:49:48 -08001472static int rmap_walk_anon(struct page *page, struct rmap_walk_control *rwc)
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001473{
1474 struct anon_vma *anon_vma;
Davidlohr Buesob258d862014-12-12 16:55:04 -08001475 pgoff_t pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001476 struct anon_vma_chain *avc;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001477 int ret = SWAP_AGAIN;
1478
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001479 anon_vma = rmap_walk_anon_lock(page, rwc);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001480 if (!anon_vma)
1481 return ret;
Joonsoo Kimfaecd8d2014-01-21 15:49:46 -08001482
Davidlohr Buesob258d862014-12-12 16:55:04 -08001483 pgoff = page_to_pgoff(page);
Michel Lespinassebf181b92012-10-08 16:31:39 -07001484 anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) {
Rik van Riel5beb4932010-03-05 13:42:07 -08001485 struct vm_area_struct *vma = avc->vma;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001486 unsigned long address = vma_address(page, vma);
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001487
1488 if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg))
1489 continue;
1490
Joonsoo Kim051ac832014-01-21 15:49:48 -08001491 ret = rwc->rmap_one(page, vma, address, rwc->arg);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001492 if (ret != SWAP_AGAIN)
1493 break;
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001494 if (rwc->done && rwc->done(page))
1495 break;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001496 }
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +00001497 anon_vma_unlock_read(anon_vma);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001498 return ret;
1499}
1500
Joonsoo Kime8351ac2014-01-21 15:49:52 -08001501/*
1502 * rmap_walk_file - do something to file page using the object-based rmap method
1503 * @page: the page to be handled
1504 * @rwc: control variable according to each walk type
1505 *
1506 * Find all the mappings of a page using the mapping pointer and the vma chains
1507 * contained in the address_space struct it points to.
1508 *
1509 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1510 * where the page was found will be held for write. So, we won't recheck
1511 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1512 * LOCKED.
1513 */
Joonsoo Kim051ac832014-01-21 15:49:48 -08001514static int rmap_walk_file(struct page *page, struct rmap_walk_control *rwc)
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001515{
1516 struct address_space *mapping = page->mapping;
Davidlohr Buesob258d862014-12-12 16:55:04 -08001517 pgoff_t pgoff;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001518 struct vm_area_struct *vma;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001519 int ret = SWAP_AGAIN;
1520
Joonsoo Kim9f326242014-01-21 15:49:53 -08001521 /*
1522 * The page lock not only makes sure that page->mapping cannot
1523 * suddenly be NULLified by truncation, it makes sure that the
1524 * structure at mapping cannot be freed and reused yet,
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08001525 * so we can safely take mapping->i_mmap_rwsem.
Joonsoo Kim9f326242014-01-21 15:49:53 -08001526 */
Sasha Levin81d1b092014-10-09 15:28:10 -07001527 VM_BUG_ON_PAGE(!PageLocked(page), page);
Joonsoo Kim9f326242014-01-21 15:49:53 -08001528
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001529 if (!mapping)
1530 return ret;
Davidlohr Bueso3dec0ba2014-12-12 16:54:27 -08001531
Davidlohr Buesob258d862014-12-12 16:55:04 -08001532 pgoff = page_to_pgoff(page);
Davidlohr Bueso3dec0ba2014-12-12 16:54:27 -08001533 i_mmap_lock_read(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07001534 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001535 unsigned long address = vma_address(page, vma);
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001536
1537 if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg))
1538 continue;
1539
Joonsoo Kim051ac832014-01-21 15:49:48 -08001540 ret = rwc->rmap_one(page, vma, address, rwc->arg);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001541 if (ret != SWAP_AGAIN)
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001542 goto done;
1543 if (rwc->done && rwc->done(page))
1544 goto done;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001545 }
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001546
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001547done:
Davidlohr Bueso3dec0ba2014-12-12 16:54:27 -08001548 i_mmap_unlock_read(mapping);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001549 return ret;
1550}
1551
Joonsoo Kim051ac832014-01-21 15:49:48 -08001552int rmap_walk(struct page *page, struct rmap_walk_control *rwc)
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001553{
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001554 if (unlikely(PageKsm(page)))
Joonsoo Kim051ac832014-01-21 15:49:48 -08001555 return rmap_walk_ksm(page, rwc);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001556 else if (PageAnon(page))
Joonsoo Kim051ac832014-01-21 15:49:48 -08001557 return rmap_walk_anon(page, rwc);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001558 else
Joonsoo Kim051ac832014-01-21 15:49:48 -08001559 return rmap_walk_file(page, rwc);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001560}
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001561
Naoya Horiguchie3390f62010-06-15 13:18:13 +09001562#ifdef CONFIG_HUGETLB_PAGE
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001563/*
1564 * The following three functions are for anonymous (private mapped) hugepages.
1565 * Unlike common anonymous pages, anonymous hugepages have no accounting code
1566 * and no lru code, because we handle hugepages differently from common pages.
1567 */
1568static void __hugepage_set_anon_rmap(struct page *page,
1569 struct vm_area_struct *vma, unsigned long address, int exclusive)
1570{
1571 struct anon_vma *anon_vma = vma->anon_vma;
Naoya Horiguchi433abed2010-09-10 13:23:03 +09001572
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001573 BUG_ON(!anon_vma);
Naoya Horiguchi433abed2010-09-10 13:23:03 +09001574
1575 if (PageAnon(page))
1576 return;
1577 if (!exclusive)
1578 anon_vma = anon_vma->root;
1579
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001580 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
1581 page->mapping = (struct address_space *) anon_vma;
1582 page->index = linear_page_index(vma, address);
1583}
1584
1585void hugepage_add_anon_rmap(struct page *page,
1586 struct vm_area_struct *vma, unsigned long address)
1587{
1588 struct anon_vma *anon_vma = vma->anon_vma;
1589 int first;
Naoya Horiguchia850ea32010-09-10 13:23:06 +09001590
1591 BUG_ON(!PageLocked(page));
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001592 BUG_ON(!anon_vma);
Hugh Dickins5dbe0af2011-05-28 13:17:04 -07001593 /* address might be in next vma when migration races vma_adjust */
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001594 first = atomic_inc_and_test(&page->_mapcount);
1595 if (first)
1596 __hugepage_set_anon_rmap(page, vma, address, 0);
1597}
1598
1599void hugepage_add_new_anon_rmap(struct page *page,
1600 struct vm_area_struct *vma, unsigned long address)
1601{
1602 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
1603 atomic_set(&page->_mapcount, 0);
1604 __hugepage_set_anon_rmap(page, vma, address, 1);
1605}
Naoya Horiguchie3390f62010-06-15 13:18:13 +09001606#endif /* CONFIG_HUGETLB_PAGE */