blob: f5b5c1f3dcd755ae313bba1404f2c9b079d5c18f [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>
Vladimir Davydov33c3fc72015-09-09 15:35:45 -070062#include <linux/page_idle.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070063
64#include <asm/tlbflush.h>
65
Mel Gorman72b252a2015-09-04 15:47:32 -070066#include <trace/events/tlb.h>
67
Nick Pigginb291f002008-10-18 20:26:44 -070068#include "internal.h"
69
Adrian Bunkfdd2e5f2008-10-18 20:28:38 -070070static struct kmem_cache *anon_vma_cachep;
Rik van Riel5beb4932010-03-05 13:42:07 -080071static struct kmem_cache *anon_vma_chain_cachep;
Adrian Bunkfdd2e5f2008-10-18 20:28:38 -070072
73static inline struct anon_vma *anon_vma_alloc(void)
74{
Peter Zijlstra01d8b202011-03-22 16:32:49 -070075 struct anon_vma *anon_vma;
76
77 anon_vma = kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL);
78 if (anon_vma) {
79 atomic_set(&anon_vma->refcount, 1);
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -080080 anon_vma->degree = 1; /* Reference for first vma */
81 anon_vma->parent = anon_vma;
Peter Zijlstra01d8b202011-03-22 16:32:49 -070082 /*
83 * Initialise the anon_vma root to point to itself. If called
84 * from fork, the root will be reset to the parents anon_vma.
85 */
86 anon_vma->root = anon_vma;
87 }
88
89 return anon_vma;
Adrian Bunkfdd2e5f2008-10-18 20:28:38 -070090}
91
Peter Zijlstra01d8b202011-03-22 16:32:49 -070092static inline void anon_vma_free(struct anon_vma *anon_vma)
Adrian Bunkfdd2e5f2008-10-18 20:28:38 -070093{
Peter Zijlstra01d8b202011-03-22 16:32:49 -070094 VM_BUG_ON(atomic_read(&anon_vma->refcount));
Peter Zijlstra88c22082011-05-24 17:12:13 -070095
96 /*
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +000097 * Synchronize against page_lock_anon_vma_read() such that
Peter Zijlstra88c22082011-05-24 17:12:13 -070098 * we can safely hold the lock without the anon_vma getting
99 * freed.
100 *
101 * Relies on the full mb implied by the atomic_dec_and_test() from
102 * put_anon_vma() against the acquire barrier implied by
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000103 * down_read_trylock() from page_lock_anon_vma_read(). This orders:
Peter Zijlstra88c22082011-05-24 17:12:13 -0700104 *
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000105 * page_lock_anon_vma_read() VS put_anon_vma()
106 * down_read_trylock() atomic_dec_and_test()
Peter Zijlstra88c22082011-05-24 17:12:13 -0700107 * LOCK MB
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000108 * atomic_read() rwsem_is_locked()
Peter Zijlstra88c22082011-05-24 17:12:13 -0700109 *
110 * LOCK should suffice since the actual taking of the lock must
111 * happen _before_ what follows.
112 */
Hugh Dickins7f39dda2014-06-04 16:05:33 -0700113 might_sleep();
Ingo Molnar5a505082012-12-02 19:56:46 +0000114 if (rwsem_is_locked(&anon_vma->root->rwsem)) {
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000115 anon_vma_lock_write(anon_vma);
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800116 anon_vma_unlock_write(anon_vma);
Peter Zijlstra88c22082011-05-24 17:12:13 -0700117 }
118
Adrian Bunkfdd2e5f2008-10-18 20:28:38 -0700119 kmem_cache_free(anon_vma_cachep, anon_vma);
120}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121
Linus Torvaldsdd347392011-06-17 19:05:36 -0700122static inline struct anon_vma_chain *anon_vma_chain_alloc(gfp_t gfp)
Rik van Riel5beb4932010-03-05 13:42:07 -0800123{
Linus Torvaldsdd347392011-06-17 19:05:36 -0700124 return kmem_cache_alloc(anon_vma_chain_cachep, gfp);
Rik van Riel5beb4932010-03-05 13:42:07 -0800125}
126
Namhyung Kime574b5f2010-10-26 14:22:02 -0700127static void anon_vma_chain_free(struct anon_vma_chain *anon_vma_chain)
Rik van Riel5beb4932010-03-05 13:42:07 -0800128{
129 kmem_cache_free(anon_vma_chain_cachep, anon_vma_chain);
130}
131
Kautuk Consul6583a842012-03-21 16:34:01 -0700132static void anon_vma_chain_link(struct vm_area_struct *vma,
133 struct anon_vma_chain *avc,
134 struct anon_vma *anon_vma)
135{
136 avc->vma = vma;
137 avc->anon_vma = anon_vma;
138 list_add(&avc->same_vma, &vma->anon_vma_chain);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700139 anon_vma_interval_tree_insert(avc, &anon_vma->rb_root);
Kautuk Consul6583a842012-03-21 16:34:01 -0700140}
141
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700142/**
143 * anon_vma_prepare - attach an anon_vma to a memory region
144 * @vma: the memory region in question
145 *
146 * This makes sure the memory mapping described by 'vma' has
147 * an 'anon_vma' attached to it, so that we can associate the
148 * anonymous pages mapped into it with that anon_vma.
149 *
150 * The common case will be that we already have one, but if
Figo.zhang23a07902010-12-27 15:14:06 +0100151 * not we either need to find an adjacent mapping that we
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700152 * can re-use the anon_vma from (very common when the only
153 * reason for splitting a vma has been mprotect()), or we
154 * allocate a new one.
155 *
156 * Anon-vma allocations are very subtle, because we may have
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000157 * optimistically looked up an anon_vma in page_lock_anon_vma_read()
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700158 * and that may actually touch the spinlock even in the newly
159 * allocated vma (it depends on RCU to make sure that the
160 * anon_vma isn't actually destroyed).
161 *
162 * As a result, we need to do proper anon_vma locking even
163 * for the new allocation. At the same time, we do not want
164 * to do any locking for the common case of already having
165 * an anon_vma.
166 *
167 * This must be called with the mmap_sem held for reading.
168 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169int anon_vma_prepare(struct vm_area_struct *vma)
170{
171 struct anon_vma *anon_vma = vma->anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -0800172 struct anon_vma_chain *avc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173
174 might_sleep();
175 if (unlikely(!anon_vma)) {
176 struct mm_struct *mm = vma->vm_mm;
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700177 struct anon_vma *allocated;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178
Linus Torvaldsdd347392011-06-17 19:05:36 -0700179 avc = anon_vma_chain_alloc(GFP_KERNEL);
Rik van Riel5beb4932010-03-05 13:42:07 -0800180 if (!avc)
181 goto out_enomem;
182
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 anon_vma = find_mergeable_anon_vma(vma);
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700184 allocated = NULL;
185 if (!anon_vma) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 anon_vma = anon_vma_alloc();
187 if (unlikely(!anon_vma))
Rik van Riel5beb4932010-03-05 13:42:07 -0800188 goto out_enomem_free_avc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 allocated = anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 }
191
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000192 anon_vma_lock_write(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 /* page_table_lock to protect against threads */
194 spin_lock(&mm->page_table_lock);
195 if (likely(!vma->anon_vma)) {
196 vma->anon_vma = anon_vma;
Kautuk Consul6583a842012-03-21 16:34:01 -0700197 anon_vma_chain_link(vma, avc, anon_vma);
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800198 /* vma reference or self-parent link for new root */
199 anon_vma->degree++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 allocated = NULL;
Oleg Nesterov31f2b0e2010-04-23 13:18:01 -0400201 avc = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 }
203 spin_unlock(&mm->page_table_lock);
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800204 anon_vma_unlock_write(anon_vma);
Oleg Nesterov31f2b0e2010-04-23 13:18:01 -0400205
206 if (unlikely(allocated))
Peter Zijlstra01d8b202011-03-22 16:32:49 -0700207 put_anon_vma(allocated);
Oleg Nesterov31f2b0e2010-04-23 13:18:01 -0400208 if (unlikely(avc))
Rik van Riel5beb4932010-03-05 13:42:07 -0800209 anon_vma_chain_free(avc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 }
211 return 0;
Rik van Riel5beb4932010-03-05 13:42:07 -0800212
213 out_enomem_free_avc:
214 anon_vma_chain_free(avc);
215 out_enomem:
216 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217}
218
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700219/*
220 * This is a useful helper function for locking the anon_vma root as
221 * we traverse the vma->anon_vma_chain, looping over anon_vma's that
222 * have the same vma.
223 *
224 * Such anon_vma's should have the same root, so you'd expect to see
225 * just a single mutex_lock for the whole traversal.
226 */
227static inline struct anon_vma *lock_anon_vma_root(struct anon_vma *root, struct anon_vma *anon_vma)
228{
229 struct anon_vma *new_root = anon_vma->root;
230 if (new_root != root) {
231 if (WARN_ON_ONCE(root))
Ingo Molnar5a505082012-12-02 19:56:46 +0000232 up_write(&root->rwsem);
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700233 root = new_root;
Ingo Molnar5a505082012-12-02 19:56:46 +0000234 down_write(&root->rwsem);
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700235 }
236 return root;
237}
238
239static inline void unlock_anon_vma_root(struct anon_vma *root)
240{
241 if (root)
Ingo Molnar5a505082012-12-02 19:56:46 +0000242 up_write(&root->rwsem);
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700243}
244
Rik van Riel5beb4932010-03-05 13:42:07 -0800245/*
246 * Attach the anon_vmas from src to dst.
247 * Returns 0 on success, -ENOMEM on failure.
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800248 *
249 * If dst->anon_vma is NULL this function tries to find and reuse existing
250 * anon_vma which has no vmas and only one child anon_vma. This prevents
251 * degradation of anon_vma hierarchy to endless linear chain in case of
252 * constantly forking task. On the other hand, an anon_vma with more than one
253 * child isn't reused even if there was no alive vma, thus rmap walker has a
254 * good chance of avoiding scanning the whole hierarchy when it searches where
255 * page is mapped.
Rik van Riel5beb4932010-03-05 13:42:07 -0800256 */
257int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258{
Rik van Riel5beb4932010-03-05 13:42:07 -0800259 struct anon_vma_chain *avc, *pavc;
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700260 struct anon_vma *root = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261
Linus Torvalds646d87b2010-04-11 17:15:03 -0700262 list_for_each_entry_reverse(pavc, &src->anon_vma_chain, same_vma) {
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700263 struct anon_vma *anon_vma;
264
Linus Torvaldsdd347392011-06-17 19:05:36 -0700265 avc = anon_vma_chain_alloc(GFP_NOWAIT | __GFP_NOWARN);
266 if (unlikely(!avc)) {
267 unlock_anon_vma_root(root);
268 root = NULL;
269 avc = anon_vma_chain_alloc(GFP_KERNEL);
270 if (!avc)
271 goto enomem_failure;
272 }
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700273 anon_vma = pavc->anon_vma;
274 root = lock_anon_vma_root(root, anon_vma);
275 anon_vma_chain_link(dst, avc, anon_vma);
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800276
277 /*
278 * Reuse existing anon_vma if its degree lower than two,
279 * that means it has no vma and only one anon_vma child.
280 *
281 * Do not chose parent anon_vma, otherwise first child
282 * will always reuse it. Root anon_vma is never reused:
283 * it has self-parent reference and at least one child.
284 */
285 if (!dst->anon_vma && anon_vma != src->anon_vma &&
286 anon_vma->degree < 2)
287 dst->anon_vma = anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 }
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800289 if (dst->anon_vma)
290 dst->anon_vma->degree++;
Linus Torvaldsbb4aa392011-06-16 20:44:51 -0700291 unlock_anon_vma_root(root);
Rik van Riel5beb4932010-03-05 13:42:07 -0800292 return 0;
293
294 enomem_failure:
Leon Yu3fe89b32015-03-25 15:55:11 -0700295 /*
296 * dst->anon_vma is dropped here otherwise its degree can be incorrectly
297 * decremented in unlink_anon_vmas().
298 * We can safely do this because callers of anon_vma_clone() don't care
299 * about dst->anon_vma if anon_vma_clone() failed.
300 */
301 dst->anon_vma = NULL;
Rik van Riel5beb4932010-03-05 13:42:07 -0800302 unlink_anon_vmas(dst);
303 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304}
305
Rik van Riel5beb4932010-03-05 13:42:07 -0800306/*
307 * Attach vma to its own anon_vma, as well as to the anon_vmas that
308 * the corresponding VMA in the parent process is attached to.
309 * Returns 0 on success, non-zero on failure.
310 */
311int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312{
Rik van Riel5beb4932010-03-05 13:42:07 -0800313 struct anon_vma_chain *avc;
314 struct anon_vma *anon_vma;
Daniel Forrestc4ea95d2014-12-02 15:59:42 -0800315 int error;
Rik van Riel5beb4932010-03-05 13:42:07 -0800316
317 /* Don't bother if the parent process has no anon_vma here. */
318 if (!pvma->anon_vma)
319 return 0;
320
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800321 /* Drop inherited anon_vma, we'll reuse existing or allocate new. */
322 vma->anon_vma = NULL;
323
Rik van Riel5beb4932010-03-05 13:42:07 -0800324 /*
325 * First, attach the new VMA to the parent VMA's anon_vmas,
326 * so rmap can find non-COWed pages in child processes.
327 */
Daniel Forrestc4ea95d2014-12-02 15:59:42 -0800328 error = anon_vma_clone(vma, pvma);
329 if (error)
330 return error;
Rik van Riel5beb4932010-03-05 13:42:07 -0800331
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800332 /* An existing anon_vma has been reused, all done then. */
333 if (vma->anon_vma)
334 return 0;
335
Rik van Riel5beb4932010-03-05 13:42:07 -0800336 /* Then add our own anon_vma. */
337 anon_vma = anon_vma_alloc();
338 if (!anon_vma)
339 goto out_error;
Linus Torvaldsdd347392011-06-17 19:05:36 -0700340 avc = anon_vma_chain_alloc(GFP_KERNEL);
Rik van Riel5beb4932010-03-05 13:42:07 -0800341 if (!avc)
342 goto out_error_free_anon_vma;
Rik van Riel5c341ee12010-08-09 17:18:39 -0700343
344 /*
345 * The root anon_vma's spinlock is the lock actually used when we
346 * lock any of the anon_vmas in this anon_vma tree.
347 */
348 anon_vma->root = pvma->anon_vma->root;
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800349 anon_vma->parent = pvma->anon_vma;
Rik van Riel76545062010-08-09 17:18:41 -0700350 /*
Peter Zijlstra01d8b202011-03-22 16:32:49 -0700351 * With refcounts, an anon_vma can stay around longer than the
352 * process it belongs to. The root anon_vma needs to be pinned until
353 * this anon_vma is freed, because the lock lives in the root.
Rik van Riel76545062010-08-09 17:18:41 -0700354 */
355 get_anon_vma(anon_vma->root);
Rik van Riel5beb4932010-03-05 13:42:07 -0800356 /* Mark this anon_vma as the one where our new (COWed) pages go. */
357 vma->anon_vma = anon_vma;
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000358 anon_vma_lock_write(anon_vma);
Rik van Riel5c341ee12010-08-09 17:18:39 -0700359 anon_vma_chain_link(vma, avc, anon_vma);
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800360 anon_vma->parent->degree++;
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800361 anon_vma_unlock_write(anon_vma);
Rik van Riel5beb4932010-03-05 13:42:07 -0800362
363 return 0;
364
365 out_error_free_anon_vma:
Peter Zijlstra01d8b202011-03-22 16:32:49 -0700366 put_anon_vma(anon_vma);
Rik van Riel5beb4932010-03-05 13:42:07 -0800367 out_error:
Rik van Riel4946d542010-04-05 12:13:33 -0400368 unlink_anon_vmas(vma);
Rik van Riel5beb4932010-03-05 13:42:07 -0800369 return -ENOMEM;
370}
371
Rik van Riel5beb4932010-03-05 13:42:07 -0800372void unlink_anon_vmas(struct vm_area_struct *vma)
373{
374 struct anon_vma_chain *avc, *next;
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200375 struct anon_vma *root = NULL;
Rik van Riel5beb4932010-03-05 13:42:07 -0800376
Rik van Riel5c341ee12010-08-09 17:18:39 -0700377 /*
378 * Unlink each anon_vma chained to the VMA. This list is ordered
379 * from newest to oldest, ensuring the root anon_vma gets freed last.
380 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800381 list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) {
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200382 struct anon_vma *anon_vma = avc->anon_vma;
383
384 root = lock_anon_vma_root(root, anon_vma);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700385 anon_vma_interval_tree_remove(avc, &anon_vma->rb_root);
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200386
387 /*
388 * Leave empty anon_vmas on the list - we'll need
389 * to free them outside the lock.
390 */
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800391 if (RB_EMPTY_ROOT(&anon_vma->rb_root)) {
392 anon_vma->parent->degree--;
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200393 continue;
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800394 }
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200395
396 list_del(&avc->same_vma);
397 anon_vma_chain_free(avc);
398 }
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800399 if (vma->anon_vma)
400 vma->anon_vma->degree--;
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200401 unlock_anon_vma_root(root);
402
403 /*
404 * Iterate the list once more, it now only contains empty and unlinked
405 * anon_vmas, destroy them. Could not do before due to __put_anon_vma()
Ingo Molnar5a505082012-12-02 19:56:46 +0000406 * needing to write-acquire the anon_vma->root->rwsem.
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200407 */
408 list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) {
409 struct anon_vma *anon_vma = avc->anon_vma;
410
Konstantin Khlebnikov7a3ef202015-01-08 14:32:15 -0800411 BUG_ON(anon_vma->degree);
Peter Zijlstraeee2acb2011-06-17 13:54:23 +0200412 put_anon_vma(anon_vma);
413
Rik van Riel5beb4932010-03-05 13:42:07 -0800414 list_del(&avc->same_vma);
415 anon_vma_chain_free(avc);
416 }
417}
418
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700419static void anon_vma_ctor(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420{
Christoph Lametera35afb82007-05-16 22:10:57 -0700421 struct anon_vma *anon_vma = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422
Ingo Molnar5a505082012-12-02 19:56:46 +0000423 init_rwsem(&anon_vma->rwsem);
Peter Zijlstra83813262011-03-22 16:32:48 -0700424 atomic_set(&anon_vma->refcount, 0);
Michel Lespinassebf181b92012-10-08 16:31:39 -0700425 anon_vma->rb_root = RB_ROOT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426}
427
428void __init anon_vma_init(void)
429{
430 anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma),
Paul Mundt20c2df82007-07-20 10:11:58 +0900431 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor);
Rik van Riel5beb4932010-03-05 13:42:07 -0800432 anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain, SLAB_PANIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433}
434
435/*
Peter Zijlstra6111e4c2011-05-24 17:12:08 -0700436 * Getting a lock on a stable anon_vma from a page off the LRU is tricky!
437 *
438 * Since there is no serialization what so ever against page_remove_rmap()
439 * the best this function can do is return a locked anon_vma that might
440 * have been relevant to this page.
441 *
442 * The page might have been remapped to a different anon_vma or the anon_vma
443 * returned may already be freed (and even reused).
444 *
Peter Zijlstrabc658c92011-05-29 10:33:44 +0200445 * In case it was remapped to a different anon_vma, the new anon_vma will be a
446 * child of the old anon_vma, and the anon_vma lifetime rules will therefore
447 * ensure that any anon_vma obtained from the page will still be valid for as
448 * long as we observe page_mapped() [ hence all those page_mapped() tests ].
449 *
Peter Zijlstra6111e4c2011-05-24 17:12:08 -0700450 * All users of this function must be very careful when walking the anon_vma
451 * chain and verify that the page in question is indeed mapped in it
452 * [ something equivalent to page_mapped_in_vma() ].
453 *
454 * Since anon_vma's slab is DESTROY_BY_RCU and we know from page_remove_rmap()
455 * that the anon_vma pointer from page->mapping is valid if there is a
456 * mapcount, we can dereference the anon_vma after observing those.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 */
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700458struct anon_vma *page_get_anon_vma(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459{
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700460 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 unsigned long anon_mapping;
462
463 rcu_read_lock();
Jason Low4db0c3c2015-04-15 16:14:08 -0700464 anon_mapping = (unsigned long)READ_ONCE(page->mapping);
Hugh Dickins3ca7b3c2009-12-14 17:58:57 -0800465 if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 goto out;
467 if (!page_mapped(page))
468 goto out;
469
470 anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700471 if (!atomic_inc_not_zero(&anon_vma->refcount)) {
472 anon_vma = NULL;
473 goto out;
474 }
Hugh Dickinsf18194272010-08-25 23:12:54 -0700475
476 /*
477 * If this page is still mapped, then its anon_vma cannot have been
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700478 * freed. But if it has been unmapped, we have no security against the
479 * anon_vma structure being freed and reused (for another anon_vma:
480 * SLAB_DESTROY_BY_RCU guarantees that - so the atomic_inc_not_zero()
481 * above cannot corrupt).
Hugh Dickinsf18194272010-08-25 23:12:54 -0700482 */
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700483 if (!page_mapped(page)) {
Hugh Dickins7f39dda2014-06-04 16:05:33 -0700484 rcu_read_unlock();
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700485 put_anon_vma(anon_vma);
Hugh Dickins7f39dda2014-06-04 16:05:33 -0700486 return NULL;
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700487 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488out:
489 rcu_read_unlock();
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700490
491 return anon_vma;
492}
493
Peter Zijlstra88c22082011-05-24 17:12:13 -0700494/*
495 * Similar to page_get_anon_vma() except it locks the anon_vma.
496 *
497 * Its a little more complex as it tries to keep the fast path to a single
498 * atomic op -- the trylock. If we fail the trylock, we fall back to getting a
499 * reference like with page_get_anon_vma() and then block on the mutex.
500 */
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000501struct anon_vma *page_lock_anon_vma_read(struct page *page)
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700502{
Peter Zijlstra88c22082011-05-24 17:12:13 -0700503 struct anon_vma *anon_vma = NULL;
Hugh Dickinseee0f252011-05-28 13:20:21 -0700504 struct anon_vma *root_anon_vma;
Peter Zijlstra88c22082011-05-24 17:12:13 -0700505 unsigned long anon_mapping;
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700506
Peter Zijlstra88c22082011-05-24 17:12:13 -0700507 rcu_read_lock();
Jason Low4db0c3c2015-04-15 16:14:08 -0700508 anon_mapping = (unsigned long)READ_ONCE(page->mapping);
Peter Zijlstra88c22082011-05-24 17:12:13 -0700509 if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
510 goto out;
511 if (!page_mapped(page))
512 goto out;
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700513
Peter Zijlstra88c22082011-05-24 17:12:13 -0700514 anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
Jason Low4db0c3c2015-04-15 16:14:08 -0700515 root_anon_vma = READ_ONCE(anon_vma->root);
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000516 if (down_read_trylock(&root_anon_vma->rwsem)) {
Peter Zijlstra88c22082011-05-24 17:12:13 -0700517 /*
Hugh Dickinseee0f252011-05-28 13:20:21 -0700518 * If the page is still mapped, then this anon_vma is still
519 * its anon_vma, and holding the mutex ensures that it will
Peter Zijlstrabc658c92011-05-29 10:33:44 +0200520 * not go away, see anon_vma_free().
Peter Zijlstra88c22082011-05-24 17:12:13 -0700521 */
Hugh Dickinseee0f252011-05-28 13:20:21 -0700522 if (!page_mapped(page)) {
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000523 up_read(&root_anon_vma->rwsem);
Peter Zijlstra88c22082011-05-24 17:12:13 -0700524 anon_vma = NULL;
525 }
526 goto out;
527 }
528
529 /* trylock failed, we got to sleep */
530 if (!atomic_inc_not_zero(&anon_vma->refcount)) {
531 anon_vma = NULL;
532 goto out;
533 }
534
535 if (!page_mapped(page)) {
Hugh Dickins7f39dda2014-06-04 16:05:33 -0700536 rcu_read_unlock();
Peter Zijlstra88c22082011-05-24 17:12:13 -0700537 put_anon_vma(anon_vma);
Hugh Dickins7f39dda2014-06-04 16:05:33 -0700538 return NULL;
Peter Zijlstra88c22082011-05-24 17:12:13 -0700539 }
540
541 /* we pinned the anon_vma, its safe to sleep */
542 rcu_read_unlock();
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000543 anon_vma_lock_read(anon_vma);
Peter Zijlstra88c22082011-05-24 17:12:13 -0700544
545 if (atomic_dec_and_test(&anon_vma->refcount)) {
546 /*
547 * Oops, we held the last refcount, release the lock
548 * and bail -- can't simply use put_anon_vma() because
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000549 * we'll deadlock on the anon_vma_lock_write() recursion.
Peter Zijlstra88c22082011-05-24 17:12:13 -0700550 */
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000551 anon_vma_unlock_read(anon_vma);
Peter Zijlstra88c22082011-05-24 17:12:13 -0700552 __put_anon_vma(anon_vma);
553 anon_vma = NULL;
554 }
555
556 return anon_vma;
557
558out:
559 rcu_read_unlock();
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700560 return anon_vma;
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800561}
562
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000563void page_unlock_anon_vma_read(struct anon_vma *anon_vma)
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800564{
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000565 anon_vma_unlock_read(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566}
567
568/*
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -0800569 * At what user virtual address is page expected in @vma?
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 */
Michel Lespinasse86c2ad12012-10-08 16:31:42 -0700571static inline unsigned long
572__vma_address(struct page *page, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573{
Naoya Horiguchia0f7a752014-07-23 14:00:01 -0700574 pgoff_t pgoff = page_to_pgoff(page);
Michel Lespinasse86c2ad12012-10-08 16:31:42 -0700575 return vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
576}
577
578inline unsigned long
579vma_address(struct page *page, struct vm_area_struct *vma)
580{
581 unsigned long address = __vma_address(page, vma);
582
583 /* page should be within @vma mapping range */
Sasha Levin81d1b092014-10-09 15:28:10 -0700584 VM_BUG_ON_VMA(address < vma->vm_start || address >= vma->vm_end, vma);
Michel Lespinasse86c2ad12012-10-08 16:31:42 -0700585
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 return address;
587}
588
Mel Gorman72b252a2015-09-04 15:47:32 -0700589#ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
590static void percpu_flush_tlb_batch_pages(void *data)
591{
592 /*
593 * All TLB entries are flushed on the assumption that it is
594 * cheaper to flush all TLBs and let them be refilled than
595 * flushing individual PFNs. Note that we do not track mm's
596 * to flush as that might simply be multiple full TLB flushes
597 * for no gain.
598 */
599 count_vm_tlb_event(NR_TLB_REMOTE_FLUSH_RECEIVED);
600 flush_tlb_local();
601}
602
603/*
604 * Flush TLB entries for recently unmapped pages from remote CPUs. It is
605 * important if a PTE was dirty when it was unmapped that it's flushed
606 * before any IO is initiated on the page to prevent lost writes. Similarly,
607 * it must be flushed before freeing to prevent data leakage.
608 */
609void try_to_unmap_flush(void)
610{
611 struct tlbflush_unmap_batch *tlb_ubc = &current->tlb_ubc;
612 int cpu;
613
614 if (!tlb_ubc->flush_required)
615 return;
616
617 cpu = get_cpu();
618
619 trace_tlb_flush(TLB_REMOTE_SHOOTDOWN, -1UL);
620
621 if (cpumask_test_cpu(cpu, &tlb_ubc->cpumask))
622 percpu_flush_tlb_batch_pages(&tlb_ubc->cpumask);
623
624 if (cpumask_any_but(&tlb_ubc->cpumask, cpu) < nr_cpu_ids) {
625 smp_call_function_many(&tlb_ubc->cpumask,
626 percpu_flush_tlb_batch_pages, (void *)tlb_ubc, true);
627 }
628 cpumask_clear(&tlb_ubc->cpumask);
629 tlb_ubc->flush_required = false;
Mel Gormand950c942015-09-04 15:47:35 -0700630 tlb_ubc->writable = false;
Mel Gorman72b252a2015-09-04 15:47:32 -0700631 put_cpu();
632}
633
Mel Gormand950c942015-09-04 15:47:35 -0700634/* Flush iff there are potentially writable TLB entries that can race with IO */
635void try_to_unmap_flush_dirty(void)
636{
637 struct tlbflush_unmap_batch *tlb_ubc = &current->tlb_ubc;
638
639 if (tlb_ubc->writable)
640 try_to_unmap_flush();
641}
642
Mel Gorman72b252a2015-09-04 15:47:32 -0700643static void set_tlb_ubc_flush_pending(struct mm_struct *mm,
Mel Gormand950c942015-09-04 15:47:35 -0700644 struct page *page, bool writable)
Mel Gorman72b252a2015-09-04 15:47:32 -0700645{
646 struct tlbflush_unmap_batch *tlb_ubc = &current->tlb_ubc;
647
648 cpumask_or(&tlb_ubc->cpumask, &tlb_ubc->cpumask, mm_cpumask(mm));
649 tlb_ubc->flush_required = true;
Mel Gormand950c942015-09-04 15:47:35 -0700650
651 /*
652 * If the PTE was dirty then it's best to assume it's writable. The
653 * caller must use try_to_unmap_flush_dirty() or try_to_unmap_flush()
654 * before the page is queued for IO.
655 */
656 if (writable)
657 tlb_ubc->writable = true;
Mel Gorman72b252a2015-09-04 15:47:32 -0700658}
659
660/*
661 * Returns true if the TLB flush should be deferred to the end of a batch of
662 * unmap operations to reduce IPIs.
663 */
664static bool should_defer_flush(struct mm_struct *mm, enum ttu_flags flags)
665{
666 bool should_defer = false;
667
668 if (!(flags & TTU_BATCH_FLUSH))
669 return false;
670
671 /* If remote CPUs need to be flushed then defer batch the flush */
672 if (cpumask_any_but(mm_cpumask(mm), get_cpu()) < nr_cpu_ids)
673 should_defer = true;
674 put_cpu();
675
676 return should_defer;
677}
678#else
679static void set_tlb_ubc_flush_pending(struct mm_struct *mm,
Mel Gormand950c942015-09-04 15:47:35 -0700680 struct page *page, bool writable)
Mel Gorman72b252a2015-09-04 15:47:32 -0700681{
682}
683
684static bool should_defer_flush(struct mm_struct *mm, enum ttu_flags flags)
685{
686 return false;
687}
688#endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
689
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690/*
Huang Shijiebf89c8c2009-10-01 15:44:04 -0700691 * At what user virtual address is page expected in vma?
Naoya Horiguchiab941e02010-05-11 14:06:55 -0700692 * Caller should check the page is actually part of the vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 */
694unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma)
695{
Michel Lespinasse86c2ad12012-10-08 16:31:42 -0700696 unsigned long address;
Andrea Arcangeli21d0d442010-08-09 17:19:10 -0700697 if (PageAnon(page)) {
Hugh Dickins4829b902010-10-02 17:46:06 -0700698 struct anon_vma *page__anon_vma = page_anon_vma(page);
699 /*
700 * Note: swapoff's unuse_vma() is more efficient with this
701 * check, and needs it to match anon_vma when KSM is active.
702 */
703 if (!vma->anon_vma || !page__anon_vma ||
704 vma->anon_vma->root != page__anon_vma->root)
Andrea Arcangeli21d0d442010-08-09 17:19:10 -0700705 return -EFAULT;
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800706 } else if (page->mapping) {
707 if (!vma->vm_file || vma->vm_file->f_mapping != page->mapping)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 return -EFAULT;
709 } else
710 return -EFAULT;
Michel Lespinasse86c2ad12012-10-08 16:31:42 -0700711 address = __vma_address(page, vma);
712 if (unlikely(address < vma->vm_start || address >= vma->vm_end))
713 return -EFAULT;
714 return address;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715}
716
Bob Liu62190492012-12-11 16:00:37 -0800717pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address)
718{
719 pgd_t *pgd;
720 pud_t *pud;
721 pmd_t *pmd = NULL;
Hugh Dickinsf72e7dc2014-06-23 13:22:05 -0700722 pmd_t pmde;
Bob Liu62190492012-12-11 16:00:37 -0800723
724 pgd = pgd_offset(mm, address);
725 if (!pgd_present(*pgd))
726 goto out;
727
728 pud = pud_offset(pgd, address);
729 if (!pud_present(*pud))
730 goto out;
731
732 pmd = pmd_offset(pud, address);
Hugh Dickinsf72e7dc2014-06-23 13:22:05 -0700733 /*
Aneesh Kumar K.V8809aa22015-06-24 16:57:44 -0700734 * Some THP functions use the sequence pmdp_huge_clear_flush(), set_pmd_at()
Hugh Dickinsf72e7dc2014-06-23 13:22:05 -0700735 * without holding anon_vma lock for write. So when looking for a
736 * genuine pmde (in which to find pte), test present and !THP together.
737 */
Christian Borntraegere37c6982014-12-07 21:41:33 +0100738 pmde = *pmd;
739 barrier();
Hugh Dickinsf72e7dc2014-06-23 13:22:05 -0700740 if (!pmd_present(pmde) || pmd_trans_huge(pmde))
Bob Liu62190492012-12-11 16:00:37 -0800741 pmd = NULL;
742out:
743 return pmd;
744}
745
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746/*
Nikita Danilov81b40822005-05-01 08:58:36 -0700747 * Check that @page is mapped at @address into @mm.
748 *
Nick Piggin479db0b2008-08-20 14:09:18 -0700749 * If @sync is false, page_check_address may perform a racy check to avoid
750 * the page table lock when the pte is not present (helpful when reclaiming
751 * highly shared pages).
752 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -0700753 * On success returns with pte mapped and locked.
Nikita Danilov81b40822005-05-01 08:58:36 -0700754 */
Namhyung Kime9a81a82010-10-26 14:22:01 -0700755pte_t *__page_check_address(struct page *page, struct mm_struct *mm,
Nick Piggin479db0b2008-08-20 14:09:18 -0700756 unsigned long address, spinlock_t **ptlp, int sync)
Nikita Danilov81b40822005-05-01 08:58:36 -0700757{
Nikita Danilov81b40822005-05-01 08:58:36 -0700758 pmd_t *pmd;
759 pte_t *pte;
Hugh Dickinsc0718802005-10-29 18:16:31 -0700760 spinlock_t *ptl;
Nikita Danilov81b40822005-05-01 08:58:36 -0700761
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +0900762 if (unlikely(PageHuge(page))) {
Jianguo Wu98398c32013-12-18 17:08:59 -0800763 /* when pud is not present, pte will be NULL */
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +0900764 pte = huge_pte_offset(mm, address);
Jianguo Wu98398c32013-12-18 17:08:59 -0800765 if (!pte)
766 return NULL;
767
Kirill A. Shutemovcb900f42013-11-14 14:31:02 -0800768 ptl = huge_pte_lockptr(page_hstate(page), mm, pte);
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +0900769 goto check;
770 }
771
Bob Liu62190492012-12-11 16:00:37 -0800772 pmd = mm_find_pmd(mm, address);
773 if (!pmd)
Hugh Dickinsc0718802005-10-29 18:16:31 -0700774 return NULL;
775
Hugh Dickinsc0718802005-10-29 18:16:31 -0700776 pte = pte_offset_map(pmd, address);
777 /* Make a quick check before getting the lock */
Nick Piggin479db0b2008-08-20 14:09:18 -0700778 if (!sync && !pte_present(*pte)) {
Hugh Dickinsc0718802005-10-29 18:16:31 -0700779 pte_unmap(pte);
780 return NULL;
Nikita Danilov81b40822005-05-01 08:58:36 -0700781 }
Hugh Dickinsc0718802005-10-29 18:16:31 -0700782
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700783 ptl = pte_lockptr(mm, pmd);
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +0900784check:
Hugh Dickinsc0718802005-10-29 18:16:31 -0700785 spin_lock(ptl);
786 if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) {
787 *ptlp = ptl;
788 return pte;
789 }
790 pte_unmap_unlock(pte, ptl);
791 return NULL;
Nikita Danilov81b40822005-05-01 08:58:36 -0700792}
793
Nick Pigginb291f002008-10-18 20:26:44 -0700794/**
795 * page_mapped_in_vma - check whether a page is really mapped in a VMA
796 * @page: the page to test
797 * @vma: the VMA to test
798 *
799 * Returns 1 if the page is mapped into the page tables of the VMA, 0
800 * if the page is not mapped into the page tables of this VMA. Only
801 * valid for normal file or anonymous VMAs.
802 */
Andi Kleen6a460792009-09-16 11:50:15 +0200803int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma)
Nick Pigginb291f002008-10-18 20:26:44 -0700804{
805 unsigned long address;
806 pte_t *pte;
807 spinlock_t *ptl;
808
Michel Lespinasse86c2ad12012-10-08 16:31:42 -0700809 address = __vma_address(page, vma);
810 if (unlikely(address < vma->vm_start || address >= vma->vm_end))
Nick Pigginb291f002008-10-18 20:26:44 -0700811 return 0;
812 pte = page_check_address(page, vma->vm_mm, address, &ptl, 1);
813 if (!pte) /* the page is not in this mm */
814 return 0;
815 pte_unmap_unlock(pte, ptl);
816
817 return 1;
818}
819
Joonsoo Kim9f326242014-01-21 15:49:53 -0800820struct page_referenced_arg {
821 int mapcount;
822 int referenced;
823 unsigned long vm_flags;
824 struct mem_cgroup *memcg;
825};
Nikita Danilov81b40822005-05-01 08:58:36 -0700826/*
Joonsoo Kim9f326242014-01-21 15:49:53 -0800827 * arg: page_referenced_arg will be passed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 */
Kirill A. Shutemovac769502014-06-04 16:08:17 -0700829static int page_referenced_one(struct page *page, struct vm_area_struct *vma,
Joonsoo Kim9f326242014-01-21 15:49:53 -0800830 unsigned long address, void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831{
832 struct mm_struct *mm = vma->vm_mm;
Kirill A. Shutemov117b0792013-11-14 14:30:56 -0800833 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 int referenced = 0;
Joonsoo Kim9f326242014-01-21 15:49:53 -0800835 struct page_referenced_arg *pra = arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800837 if (unlikely(PageTransHuge(page))) {
838 pmd_t *pmd;
839
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800840 /*
841 * rmap might return false positives; we must filter
842 * these out using page_check_address_pmd().
843 */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800844 pmd = page_check_address_pmd(page, mm, address,
Kirill A. Shutemov117b0792013-11-14 14:30:56 -0800845 PAGE_CHECK_ADDRESS_PMD_FLAG, &ptl);
846 if (!pmd)
Joonsoo Kim9f326242014-01-21 15:49:53 -0800847 return SWAP_AGAIN;
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800848
849 if (vma->vm_flags & VM_LOCKED) {
Kirill A. Shutemov117b0792013-11-14 14:30:56 -0800850 spin_unlock(ptl);
Joonsoo Kim9f326242014-01-21 15:49:53 -0800851 pra->vm_flags |= VM_LOCKED;
852 return SWAP_FAIL; /* To break the loop */
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800853 }
854
855 /* go ahead even if the pmd is pmd_trans_splitting() */
856 if (pmdp_clear_flush_young_notify(vma, address, pmd))
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800857 referenced++;
Kirill A. Shutemov117b0792013-11-14 14:30:56 -0800858 spin_unlock(ptl);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800859 } else {
860 pte_t *pte;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800861
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800862 /*
863 * rmap might return false positives; we must filter
864 * these out using page_check_address().
865 */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800866 pte = page_check_address(page, mm, address, &ptl, 0);
867 if (!pte)
Joonsoo Kim9f326242014-01-21 15:49:53 -0800868 return SWAP_AGAIN;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800869
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800870 if (vma->vm_flags & VM_LOCKED) {
871 pte_unmap_unlock(pte, ptl);
Joonsoo Kim9f326242014-01-21 15:49:53 -0800872 pra->vm_flags |= VM_LOCKED;
873 return SWAP_FAIL; /* To break the loop */
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800874 }
875
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800876 if (ptep_clear_flush_young_notify(vma, address, pte)) {
877 /*
878 * Don't treat a reference through a sequentially read
879 * mapping as such. If the page has been used in
880 * another mapping, we will catch it; if this other
881 * mapping is already gone, the unmap path will have
882 * set PG_referenced or activated the page.
883 */
Joe Perches64363aa2013-07-08 16:00:18 -0700884 if (likely(!(vma->vm_flags & VM_SEQ_READ)))
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800885 referenced++;
886 }
887 pte_unmap_unlock(pte, ptl);
888 }
889
Vladimir Davydov33c3fc72015-09-09 15:35:45 -0700890 if (referenced)
891 clear_page_idle(page);
892 if (test_and_clear_page_young(page))
893 referenced++;
894
Joonsoo Kim9f326242014-01-21 15:49:53 -0800895 if (referenced) {
896 pra->referenced++;
897 pra->vm_flags |= vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 }
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800899
Joonsoo Kim9f326242014-01-21 15:49:53 -0800900 pra->mapcount--;
901 if (!pra->mapcount)
902 return SWAP_SUCCESS; /* To break the loop */
903
904 return SWAP_AGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905}
906
Joonsoo Kim9f326242014-01-21 15:49:53 -0800907static bool invalid_page_referenced_vma(struct vm_area_struct *vma, void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908{
Joonsoo Kim9f326242014-01-21 15:49:53 -0800909 struct page_referenced_arg *pra = arg;
910 struct mem_cgroup *memcg = pra->memcg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911
Joonsoo Kim9f326242014-01-21 15:49:53 -0800912 if (!mm_match_cgroup(vma->vm_mm, memcg))
913 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914
Joonsoo Kim9f326242014-01-21 15:49:53 -0800915 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916}
917
918/**
919 * page_referenced - test if the page was referenced
920 * @page: the page to test
921 * @is_locked: caller holds lock on the page
Johannes Weiner72835c82012-01-12 17:18:32 -0800922 * @memcg: target memory cgroup
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700923 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 *
925 * Quick test_and_clear_referenced for all mappings to a page,
926 * returns the number of ptes which referenced the page.
927 */
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700928int page_referenced(struct page *page,
929 int is_locked,
Johannes Weiner72835c82012-01-12 17:18:32 -0800930 struct mem_cgroup *memcg,
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700931 unsigned long *vm_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932{
Joonsoo Kim9f326242014-01-21 15:49:53 -0800933 int ret;
Hugh Dickins5ad64682009-12-14 17:59:24 -0800934 int we_locked = 0;
Joonsoo Kim9f326242014-01-21 15:49:53 -0800935 struct page_referenced_arg pra = {
936 .mapcount = page_mapcount(page),
937 .memcg = memcg,
938 };
939 struct rmap_walk_control rwc = {
940 .rmap_one = page_referenced_one,
941 .arg = (void *)&pra,
942 .anon_lock = page_lock_anon_vma_read,
943 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700945 *vm_flags = 0;
Joonsoo Kim9f326242014-01-21 15:49:53 -0800946 if (!page_mapped(page))
947 return 0;
948
949 if (!page_rmapping(page))
950 return 0;
951
952 if (!is_locked && (!PageAnon(page) || PageKsm(page))) {
953 we_locked = trylock_page(page);
954 if (!we_locked)
955 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 }
Joonsoo Kim9f326242014-01-21 15:49:53 -0800957
958 /*
959 * If we are reclaiming on behalf of a cgroup, skip
960 * counting on behalf of references from different
961 * cgroups
962 */
963 if (memcg) {
964 rwc.invalid_vma = invalid_page_referenced_vma;
965 }
966
967 ret = rmap_walk(page, &rwc);
968 *vm_flags = pra.vm_flags;
969
970 if (we_locked)
971 unlock_page(page);
972
973 return pra.referenced;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974}
975
Hugh Dickins1cb17292009-12-14 17:59:01 -0800976static int page_mkclean_one(struct page *page, struct vm_area_struct *vma,
Joonsoo Kim9853a402014-01-21 15:49:55 -0800977 unsigned long address, void *arg)
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700978{
979 struct mm_struct *mm = vma->vm_mm;
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100980 pte_t *pte;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700981 spinlock_t *ptl;
982 int ret = 0;
Joonsoo Kim9853a402014-01-21 15:49:55 -0800983 int *cleaned = arg;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700984
Nick Piggin479db0b2008-08-20 14:09:18 -0700985 pte = page_check_address(page, mm, address, &ptl, 1);
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700986 if (!pte)
987 goto out;
988
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100989 if (pte_dirty(*pte) || pte_write(*pte)) {
990 pte_t entry;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700991
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100992 flush_cache_page(vma, address, pte_pfn(*pte));
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700993 entry = ptep_clear_flush(vma, address, pte);
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100994 entry = pte_wrprotect(entry);
995 entry = pte_mkclean(entry);
Al Virod6e88e62006-12-29 16:48:35 -0800996 set_pte_at(mm, address, pte, entry);
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100997 ret = 1;
998 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700999
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001000 pte_unmap_unlock(pte, ptl);
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001001
Joonsoo Kim9853a402014-01-21 15:49:55 -08001002 if (ret) {
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001003 mmu_notifier_invalidate_page(mm, address);
Joonsoo Kim9853a402014-01-21 15:49:55 -08001004 (*cleaned)++;
1005 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001006out:
Joonsoo Kim9853a402014-01-21 15:49:55 -08001007 return SWAP_AGAIN;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001008}
1009
Joonsoo Kim9853a402014-01-21 15:49:55 -08001010static bool invalid_mkclean_vma(struct vm_area_struct *vma, void *arg)
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001011{
Joonsoo Kim9853a402014-01-21 15:49:55 -08001012 if (vma->vm_flags & VM_SHARED)
Fengguang Wu871beb82014-01-23 15:53:41 -08001013 return false;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001014
Fengguang Wu871beb82014-01-23 15:53:41 -08001015 return true;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001016}
1017
1018int page_mkclean(struct page *page)
1019{
Joonsoo Kim9853a402014-01-21 15:49:55 -08001020 int cleaned = 0;
1021 struct address_space *mapping;
1022 struct rmap_walk_control rwc = {
1023 .arg = (void *)&cleaned,
1024 .rmap_one = page_mkclean_one,
1025 .invalid_vma = invalid_mkclean_vma,
1026 };
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001027
1028 BUG_ON(!PageLocked(page));
1029
Joonsoo Kim9853a402014-01-21 15:49:55 -08001030 if (!page_mapped(page))
1031 return 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001032
Joonsoo Kim9853a402014-01-21 15:49:55 -08001033 mapping = page_mapping(page);
1034 if (!mapping)
1035 return 0;
1036
1037 rmap_walk(page, &rwc);
1038
1039 return cleaned;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001040}
Jaya Kumar60b59be2007-05-08 00:37:37 -07001041EXPORT_SYMBOL_GPL(page_mkclean);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001042
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043/**
Rik van Rielc44b6742010-03-05 13:42:09 -08001044 * page_move_anon_rmap - move a page to our anon_vma
1045 * @page: the page to move to our anon_vma
1046 * @vma: the vma the page belongs to
1047 * @address: the user virtual address mapped
1048 *
1049 * When a page belongs exclusively to one process after a COW event,
1050 * that page can be moved into the anon_vma that belongs to just that
1051 * process, so the rmap code will not search the parent or sibling
1052 * processes.
1053 */
1054void page_move_anon_rmap(struct page *page,
1055 struct vm_area_struct *vma, unsigned long address)
1056{
1057 struct anon_vma *anon_vma = vma->anon_vma;
1058
Sasha Levin309381fea2014-01-23 15:52:54 -08001059 VM_BUG_ON_PAGE(!PageLocked(page), page);
Sasha Levin81d1b092014-10-09 15:28:10 -07001060 VM_BUG_ON_VMA(!anon_vma, vma);
Sasha Levin309381fea2014-01-23 15:52:54 -08001061 VM_BUG_ON_PAGE(page->index != linear_page_index(vma, address), page);
Rik van Rielc44b6742010-03-05 13:42:09 -08001062
1063 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
Vladimir Davydov414e2fb2015-06-24 16:56:56 -07001064 /*
1065 * Ensure that anon_vma and the PAGE_MAPPING_ANON bit are written
1066 * simultaneously, so a concurrent reader (eg page_referenced()'s
1067 * PageAnon()) will not see one without the other.
1068 */
1069 WRITE_ONCE(page->mapping, (struct address_space *) anon_vma);
Rik van Rielc44b6742010-03-05 13:42:09 -08001070}
1071
1072/**
Andi Kleen4e1c1972010-09-22 12:43:56 +02001073 * __page_set_anon_rmap - set up new anonymous rmap
1074 * @page: Page to add to rmap
1075 * @vma: VM area to add page to.
1076 * @address: User virtual address of the mapping
Rik van Riele8a03fe2010-04-14 17:59:28 -04001077 * @exclusive: the page is exclusively owned by the current process
Nick Piggin9617d952006-01-06 00:11:12 -08001078 */
1079static void __page_set_anon_rmap(struct page *page,
Rik van Riele8a03fe2010-04-14 17:59:28 -04001080 struct vm_area_struct *vma, unsigned long address, int exclusive)
Nick Piggin9617d952006-01-06 00:11:12 -08001081{
Rik van Riele8a03fe2010-04-14 17:59:28 -04001082 struct anon_vma *anon_vma = vma->anon_vma;
Nick Piggin9617d952006-01-06 00:11:12 -08001083
Rik van Riele8a03fe2010-04-14 17:59:28 -04001084 BUG_ON(!anon_vma);
Linus Torvaldsea900022010-04-12 12:44:29 -07001085
Andi Kleen4e1c1972010-09-22 12:43:56 +02001086 if (PageAnon(page))
1087 return;
1088
Linus Torvaldsea900022010-04-12 12:44:29 -07001089 /*
Rik van Riele8a03fe2010-04-14 17:59:28 -04001090 * If the page isn't exclusively mapped into this vma,
1091 * we must use the _oldest_ possible anon_vma for the
1092 * page mapping!
Linus Torvaldsea900022010-04-12 12:44:29 -07001093 */
Andi Kleen4e1c1972010-09-22 12:43:56 +02001094 if (!exclusive)
Andrea Arcangeli288468c2010-08-09 17:19:09 -07001095 anon_vma = anon_vma->root;
Linus Torvaldsea900022010-04-12 12:44:29 -07001096
Nick Piggin9617d952006-01-06 00:11:12 -08001097 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
1098 page->mapping = (struct address_space *) anon_vma;
Nick Piggin9617d952006-01-06 00:11:12 -08001099 page->index = linear_page_index(vma, address);
Nick Piggin9617d952006-01-06 00:11:12 -08001100}
1101
1102/**
Randy Dunlap43d8eac2008-03-19 17:00:43 -07001103 * __page_check_anon_rmap - sanity check anonymous rmap addition
Nick Pigginc97a9e12007-05-16 22:11:21 -07001104 * @page: the page to add the mapping to
1105 * @vma: the vm area in which the mapping is added
1106 * @address: the user virtual address mapped
1107 */
1108static void __page_check_anon_rmap(struct page *page,
1109 struct vm_area_struct *vma, unsigned long address)
1110{
1111#ifdef CONFIG_DEBUG_VM
1112 /*
1113 * The page's anon-rmap details (mapping and index) are guaranteed to
1114 * be set up correctly at this point.
1115 *
1116 * We have exclusion against page_add_anon_rmap because the caller
1117 * always holds the page locked, except if called from page_dup_rmap,
1118 * in which case the page is already known to be setup.
1119 *
1120 * We have exclusion against page_add_new_anon_rmap because those pages
1121 * are initially only visible via the pagetables, and the pte is locked
1122 * over the call to page_add_new_anon_rmap.
1123 */
Andrea Arcangeli44ab57a2010-08-09 17:19:10 -07001124 BUG_ON(page_anon_vma(page)->root != vma->anon_vma->root);
Nick Pigginc97a9e12007-05-16 22:11:21 -07001125 BUG_ON(page->index != linear_page_index(vma, address));
1126#endif
1127}
1128
1129/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 * page_add_anon_rmap - add pte mapping to an anonymous page
1131 * @page: the page to add the mapping to
1132 * @vma: the vm area in which the mapping is added
1133 * @address: the user virtual address mapped
1134 *
Hugh Dickins5ad64682009-12-14 17:59:24 -08001135 * The caller needs to hold the pte lock, and the page must be locked in
Hugh Dickins80e148222009-12-14 17:59:29 -08001136 * the anon_vma case: to serialize mapping,index checking after setting,
1137 * and to ensure that PageAnon is not being upgraded racily to PageKsm
1138 * (but PageKsm is never downgraded to PageAnon).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 */
1140void page_add_anon_rmap(struct page *page,
1141 struct vm_area_struct *vma, unsigned long address)
1142{
Rik van Rielad8c2ee2010-08-09 17:19:48 -07001143 do_page_add_anon_rmap(page, vma, address, 0);
1144}
1145
1146/*
1147 * Special version of the above for do_swap_page, which often runs
1148 * into pages that are exclusively owned by the current process.
1149 * Everybody else should continue to use page_add_anon_rmap above.
1150 */
1151void do_page_add_anon_rmap(struct page *page,
1152 struct vm_area_struct *vma, unsigned long address, int exclusive)
1153{
Hugh Dickins5ad64682009-12-14 17:59:24 -08001154 int first = atomic_inc_and_test(&page->_mapcount);
Andrea Arcangeli79134172011-01-13 15:46:58 -08001155 if (first) {
Jianyu Zhanbea04b02014-06-04 16:09:51 -07001156 /*
1157 * We use the irq-unsafe __{inc|mod}_zone_page_stat because
1158 * these counters are not modified in interrupt context, and
1159 * pte lock(a spinlock) is held, which implies preemption
1160 * disabled.
1161 */
Kirill A. Shutemov3cd14fc2013-09-12 15:13:58 -07001162 if (PageTransHuge(page))
Andrea Arcangeli79134172011-01-13 15:46:58 -08001163 __inc_zone_page_state(page,
1164 NR_ANON_TRANSPARENT_HUGEPAGES);
Kirill A. Shutemov3cd14fc2013-09-12 15:13:58 -07001165 __mod_zone_page_state(page_zone(page), NR_ANON_PAGES,
1166 hpage_nr_pages(page));
Andrea Arcangeli79134172011-01-13 15:46:58 -08001167 }
Hugh Dickins5ad64682009-12-14 17:59:24 -08001168 if (unlikely(PageKsm(page)))
1169 return;
1170
Sasha Levin309381fea2014-01-23 15:52:54 -08001171 VM_BUG_ON_PAGE(!PageLocked(page), page);
Hugh Dickins5dbe0af2011-05-28 13:17:04 -07001172 /* address might be in next vma when migration races vma_adjust */
Hugh Dickins5ad64682009-12-14 17:59:24 -08001173 if (first)
Rik van Rielad8c2ee2010-08-09 17:19:48 -07001174 __page_set_anon_rmap(page, vma, address, exclusive);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001175 else
Nick Pigginc97a9e12007-05-16 22:11:21 -07001176 __page_check_anon_rmap(page, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177}
1178
Randy Dunlap43d8eac2008-03-19 17:00:43 -07001179/**
Nick Piggin9617d952006-01-06 00:11:12 -08001180 * page_add_new_anon_rmap - add pte mapping to a new anonymous page
1181 * @page: the page to add the mapping to
1182 * @vma: the vm area in which the mapping is added
1183 * @address: the user virtual address mapped
1184 *
1185 * Same as page_add_anon_rmap but must only be called on *new* pages.
1186 * This means the inc-and-test can be bypassed.
Nick Pigginc97a9e12007-05-16 22:11:21 -07001187 * Page does not have to be locked.
Nick Piggin9617d952006-01-06 00:11:12 -08001188 */
1189void page_add_new_anon_rmap(struct page *page,
1190 struct vm_area_struct *vma, unsigned long address)
1191{
Sasha Levin81d1b092014-10-09 15:28:10 -07001192 VM_BUG_ON_VMA(address < vma->vm_start || address >= vma->vm_end, vma);
Hugh Dickinscbf84b72009-01-06 14:39:27 -08001193 SetPageSwapBacked(page);
1194 atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */
Kirill A. Shutemov3cd14fc2013-09-12 15:13:58 -07001195 if (PageTransHuge(page))
Andrea Arcangeli79134172011-01-13 15:46:58 -08001196 __inc_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES);
Kirill A. Shutemov3cd14fc2013-09-12 15:13:58 -07001197 __mod_zone_page_state(page_zone(page), NR_ANON_PAGES,
1198 hpage_nr_pages(page));
Rik van Riele8a03fe2010-04-14 17:59:28 -04001199 __page_set_anon_rmap(page, vma, address, 1);
Nick Piggin9617d952006-01-06 00:11:12 -08001200}
1201
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202/**
1203 * page_add_file_rmap - add pte mapping to a file page
1204 * @page: the page to add the mapping to
1205 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -07001206 * The caller needs to hold the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 */
1208void page_add_file_rmap(struct page *page)
1209{
Johannes Weinerd7365e72014-10-29 14:50:48 -07001210 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001211
Johannes Weiner6de22612015-02-11 15:25:01 -08001212 memcg = mem_cgroup_begin_page_stat(page);
Balbir Singhd69b0422009-06-17 16:26:34 -07001213 if (atomic_inc_and_test(&page->_mapcount)) {
Christoph Lameter65ba55f2006-06-30 01:55:34 -07001214 __inc_zone_page_state(page, NR_FILE_MAPPED);
Johannes Weinerd7365e72014-10-29 14:50:48 -07001215 mem_cgroup_inc_page_stat(memcg, MEM_CGROUP_STAT_FILE_MAPPED);
Balbir Singhd69b0422009-06-17 16:26:34 -07001216 }
Johannes Weiner6de22612015-02-11 15:25:01 -08001217 mem_cgroup_end_page_stat(memcg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218}
1219
Johannes Weiner8186eb62014-10-29 14:50:51 -07001220static void page_remove_file_rmap(struct page *page)
1221{
1222 struct mem_cgroup *memcg;
Johannes Weiner8186eb62014-10-29 14:50:51 -07001223
Johannes Weiner6de22612015-02-11 15:25:01 -08001224 memcg = mem_cgroup_begin_page_stat(page);
Johannes Weiner8186eb62014-10-29 14:50:51 -07001225
1226 /* page still mapped by someone else? */
1227 if (!atomic_add_negative(-1, &page->_mapcount))
1228 goto out;
1229
1230 /* Hugepages are not counted in NR_FILE_MAPPED for now. */
1231 if (unlikely(PageHuge(page)))
1232 goto out;
1233
1234 /*
1235 * We use the irq-unsafe __{inc|mod}_zone_page_stat because
1236 * these counters are not modified in interrupt context, and
1237 * pte lock(a spinlock) is held, which implies preemption disabled.
1238 */
1239 __dec_zone_page_state(page, NR_FILE_MAPPED);
1240 mem_cgroup_dec_page_stat(memcg, MEM_CGROUP_STAT_FILE_MAPPED);
1241
1242 if (unlikely(PageMlocked(page)))
1243 clear_page_mlock(page);
1244out:
Johannes Weiner6de22612015-02-11 15:25:01 -08001245 mem_cgroup_end_page_stat(memcg);
Johannes Weiner8186eb62014-10-29 14:50:51 -07001246}
1247
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248/**
1249 * page_remove_rmap - take down pte mapping from a page
1250 * @page: page to remove mapping from
1251 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -07001252 * The caller needs to hold the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 */
Hugh Dickinsedc315f2009-01-06 14:40:11 -08001254void page_remove_rmap(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255{
Johannes Weiner8186eb62014-10-29 14:50:51 -07001256 if (!PageAnon(page)) {
1257 page_remove_file_rmap(page);
1258 return;
1259 }
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001260
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001261 /* page still mapped by someone else? */
1262 if (!atomic_add_negative(-1, &page->_mapcount))
Johannes Weiner8186eb62014-10-29 14:50:51 -07001263 return;
1264
1265 /* Hugepages are not counted in NR_ANON_PAGES for now. */
1266 if (unlikely(PageHuge(page)))
1267 return;
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001268
1269 /*
Jianyu Zhanbea04b02014-06-04 16:09:51 -07001270 * We use the irq-unsafe __{inc|mod}_zone_page_stat because
1271 * these counters are not modified in interrupt context, and
Jianyu Zhanbea04b02014-06-04 16:09:51 -07001272 * pte lock(a spinlock) is held, which implies preemption disabled.
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001273 */
Johannes Weiner8186eb62014-10-29 14:50:51 -07001274 if (PageTransHuge(page))
1275 __dec_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES);
1276
1277 __mod_zone_page_state(page_zone(page), NR_ANON_PAGES,
1278 -hpage_nr_pages(page));
1279
Hugh Dickinse6c509f2012-10-08 16:33:19 -07001280 if (unlikely(PageMlocked(page)))
1281 clear_page_mlock(page);
Johannes Weiner8186eb62014-10-29 14:50:51 -07001282
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001283 /*
1284 * It would be tidy to reset the PageAnon mapping here,
1285 * but that might overwrite a racing page_add_anon_rmap
1286 * which increments mapcount after us but sets mapping
1287 * before us: so leave the reset to free_hot_cold_page,
1288 * and remember that it's only reliable while mapped.
1289 * Leaving it set also helps swapoff to reinstate ptes
1290 * faster for those pages still in swapcache.
1291 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292}
1293
1294/*
Joonsoo Kim52629502014-01-21 15:49:50 -08001295 * @arg: enum ttu_flags will be passed to this argument
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 */
Kirill A. Shutemovac769502014-06-04 16:08:17 -07001297static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
Joonsoo Kim52629502014-01-21 15:49:50 -08001298 unsigned long address, void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299{
1300 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 pte_t *pte;
1302 pte_t pteval;
Hugh Dickinsc0718802005-10-29 18:16:31 -07001303 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 int ret = SWAP_AGAIN;
Joonsoo Kim52629502014-01-21 15:49:50 -08001305 enum ttu_flags flags = (enum ttu_flags)arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306
Nick Piggin479db0b2008-08-20 14:09:18 -07001307 pte = page_check_address(page, mm, address, &ptl, 0);
Hugh Dickinsc0718802005-10-29 18:16:31 -07001308 if (!pte)
Nikita Danilov81b40822005-05-01 08:58:36 -07001309 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310
1311 /*
1312 * If the page is mlock()d, we cannot swap it out.
1313 * If it's recently referenced (perhaps page_referenced
1314 * skipped over this mm) then we should reactivate it.
1315 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001316 if (!(flags & TTU_IGNORE_MLOCK)) {
KOSAKI Motohirocaed0f482009-12-14 17:59:45 -08001317 if (vma->vm_flags & VM_LOCKED)
1318 goto out_mlock;
1319
Konstantin Khlebnikovdaa5ba72014-06-04 16:10:52 -07001320 if (flags & TTU_MUNLOCK)
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001321 goto out_unmap;
Andi Kleen14fa31b2009-09-16 11:50:10 +02001322 }
1323 if (!(flags & TTU_IGNORE_ACCESS)) {
Nick Pigginb291f002008-10-18 20:26:44 -07001324 if (ptep_clear_flush_young_notify(vma, address, pte)) {
1325 ret = SWAP_FAIL;
1326 goto out_unmap;
1327 }
1328 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 /* Nuke the page table entry. */
1331 flush_cache_page(vma, address, page_to_pfn(page));
Mel Gorman72b252a2015-09-04 15:47:32 -07001332 if (should_defer_flush(mm, flags)) {
1333 /*
1334 * We clear the PTE but do not flush so potentially a remote
1335 * CPU could still be writing to the page. If the entry was
1336 * previously clean then the architecture must guarantee that
1337 * a clear->dirty transition on a cached TLB entry is written
1338 * through and traps if the PTE is unmapped.
1339 */
1340 pteval = ptep_get_and_clear(mm, address, pte);
1341
Mel Gormand950c942015-09-04 15:47:35 -07001342 set_tlb_ubc_flush_pending(mm, page, pte_dirty(pteval));
Mel Gorman72b252a2015-09-04 15:47:32 -07001343 } else {
1344 pteval = ptep_clear_flush(vma, address, pte);
1345 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346
1347 /* Move the dirty bit to the physical page now the pte is gone. */
1348 if (pte_dirty(pteval))
1349 set_page_dirty(page);
1350
Hugh Dickins365e9c872005-10-29 18:16:18 -07001351 /* Update high watermark before we lower rss */
1352 update_hiwater_rss(mm);
1353
Andi Kleen888b9f72009-09-16 11:50:11 +02001354 if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) {
Naoya Horiguchi5f24ae52012-12-12 13:52:30 -08001355 if (!PageHuge(page)) {
1356 if (PageAnon(page))
1357 dec_mm_counter(mm, MM_ANONPAGES);
1358 else
1359 dec_mm_counter(mm, MM_FILEPAGES);
1360 }
Andi Kleen888b9f72009-09-16 11:50:11 +02001361 set_pte_at(mm, address, pte,
Naoya Horiguchi5f24ae52012-12-12 13:52:30 -08001362 swp_entry_to_pte(make_hwpoison_entry(page)));
Konstantin Weitz45961722013-04-17 13:59:32 +02001363 } else if (pte_unused(pteval)) {
1364 /*
1365 * The guest indicated that the page content is of no
1366 * interest anymore. Simply discard the pte, vmscan
1367 * will take care of the rest.
1368 */
1369 if (PageAnon(page))
1370 dec_mm_counter(mm, MM_ANONPAGES);
1371 else
1372 dec_mm_counter(mm, MM_FILEPAGES);
Andi Kleen888b9f72009-09-16 11:50:11 +02001373 } else if (PageAnon(page)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001374 swp_entry_t entry = { .val = page_private(page) };
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001375 pte_t swp_pte;
Christoph Lameter06972122006-06-23 02:03:35 -07001376
1377 if (PageSwapCache(page)) {
1378 /*
1379 * Store the swap location in the pte.
1380 * See handle_pte_fault() ...
1381 */
Hugh Dickins570a335b2009-12-14 17:58:46 -08001382 if (swap_duplicate(entry) < 0) {
1383 set_pte_at(mm, address, pte, pteval);
1384 ret = SWAP_FAIL;
1385 goto out_unmap;
1386 }
Christoph Lameter06972122006-06-23 02:03:35 -07001387 if (list_empty(&mm->mmlist)) {
1388 spin_lock(&mmlist_lock);
1389 if (list_empty(&mm->mmlist))
1390 list_add(&mm->mmlist, &init_mm.mmlist);
1391 spin_unlock(&mmlist_lock);
1392 }
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001393 dec_mm_counter(mm, MM_ANONPAGES);
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08001394 inc_mm_counter(mm, MM_SWAPENTS);
Konstantin Khlebnikovce1744f2012-03-21 16:33:59 -07001395 } else if (IS_ENABLED(CONFIG_MIGRATION)) {
Christoph Lameter06972122006-06-23 02:03:35 -07001396 /*
1397 * Store the pfn of the page in a special migration
1398 * pte. do_swap_page() will wait until the migration
1399 * pte is removed and then restart fault handling.
1400 */
Konstantin Khlebnikovdaa5ba72014-06-04 16:10:52 -07001401 BUG_ON(!(flags & TTU_MIGRATION));
Christoph Lameter06972122006-06-23 02:03:35 -07001402 entry = make_migration_entry(page, pte_write(pteval));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 }
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001404 swp_pte = swp_entry_to_pte(entry);
1405 if (pte_soft_dirty(pteval))
1406 swp_pte = pte_swp_mksoft_dirty(swp_pte);
1407 set_pte_at(mm, address, pte, swp_pte);
Konstantin Khlebnikovce1744f2012-03-21 16:33:59 -07001408 } else if (IS_ENABLED(CONFIG_MIGRATION) &&
Konstantin Khlebnikovdaa5ba72014-06-04 16:10:52 -07001409 (flags & TTU_MIGRATION)) {
Christoph Lameter04e62a22006-06-23 02:03:38 -07001410 /* Establish migration entry for a file page */
1411 swp_entry_t entry;
1412 entry = make_migration_entry(page, pte_write(pteval));
1413 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
1414 } else
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001415 dec_mm_counter(mm, MM_FILEPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416
Hugh Dickinsedc315f2009-01-06 14:40:11 -08001417 page_remove_rmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 page_cache_release(page);
1419
1420out_unmap:
Hugh Dickinsc0718802005-10-29 18:16:31 -07001421 pte_unmap_unlock(pte, ptl);
Konstantin Khlebnikovdaa5ba72014-06-04 16:10:52 -07001422 if (ret != SWAP_FAIL && !(flags & TTU_MUNLOCK))
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001423 mmu_notifier_invalidate_page(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424out:
1425 return ret;
KOSAKI Motohirocaed0f482009-12-14 17:59:45 -08001426
1427out_mlock:
1428 pte_unmap_unlock(pte, ptl);
1429
1430
1431 /*
1432 * We need mmap_sem locking, Otherwise VM_LOCKED check makes
1433 * unstable result and race. Plus, We can't wait here because
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08001434 * we now hold anon_vma->rwsem or mapping->i_mmap_rwsem.
KOSAKI Motohirocaed0f482009-12-14 17:59:45 -08001435 * if trylock failed, the page remain in evictable lru and later
1436 * vmscan could retry to move the page to unevictable lru if the
1437 * page is actually mlocked.
1438 */
1439 if (down_read_trylock(&vma->vm_mm->mmap_sem)) {
1440 if (vma->vm_flags & VM_LOCKED) {
1441 mlock_vma_page(page);
1442 ret = SWAP_MLOCK;
1443 }
1444 up_read(&vma->vm_mm->mmap_sem);
1445 }
1446 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447}
1448
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001449bool is_vma_temporary_stack(struct vm_area_struct *vma)
Mel Gormana8bef8ff2010-05-24 14:32:24 -07001450{
1451 int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP);
1452
1453 if (!maybe_stack)
1454 return false;
1455
1456 if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) ==
1457 VM_STACK_INCOMPLETE_SETUP)
1458 return true;
1459
1460 return false;
1461}
1462
Joonsoo Kim52629502014-01-21 15:49:50 -08001463static bool invalid_migration_vma(struct vm_area_struct *vma, void *arg)
1464{
1465 return is_vma_temporary_stack(vma);
1466}
1467
Joonsoo Kim52629502014-01-21 15:49:50 -08001468static int page_not_mapped(struct page *page)
1469{
1470 return !page_mapped(page);
1471};
1472
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473/**
1474 * try_to_unmap - try to remove all page table mappings to a page
1475 * @page: the page to get unmapped
Andi Kleen14fa31b2009-09-16 11:50:10 +02001476 * @flags: action and flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 *
1478 * Tries to remove all the page table entries which are mapping this
1479 * page, used in the pageout path. Caller must hold the page lock.
1480 * Return values are:
1481 *
1482 * SWAP_SUCCESS - we succeeded in removing all mappings
1483 * SWAP_AGAIN - we missed a mapping, try again later
1484 * SWAP_FAIL - the page is unswappable
Nick Pigginb291f002008-10-18 20:26:44 -07001485 * SWAP_MLOCK - page is mlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001487int try_to_unmap(struct page *page, enum ttu_flags flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488{
1489 int ret;
Joonsoo Kim52629502014-01-21 15:49:50 -08001490 struct rmap_walk_control rwc = {
1491 .rmap_one = try_to_unmap_one,
1492 .arg = (void *)flags,
1493 .done = page_not_mapped,
Joonsoo Kim52629502014-01-21 15:49:50 -08001494 .anon_lock = page_lock_anon_vma_read,
1495 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496
Sasha Levin309381fea2014-01-23 15:52:54 -08001497 VM_BUG_ON_PAGE(!PageHuge(page) && PageTransHuge(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498
Joonsoo Kim52629502014-01-21 15:49:50 -08001499 /*
1500 * During exec, a temporary VMA is setup and later moved.
1501 * The VMA is moved under the anon_vma lock but not the
1502 * page tables leading to a race where migration cannot
1503 * find the migration ptes. Rather than increasing the
1504 * locking requirements of exec(), migration skips
1505 * temporary VMAs until after exec() completes.
1506 */
Konstantin Khlebnikovdaa5ba72014-06-04 16:10:52 -07001507 if ((flags & TTU_MIGRATION) && !PageKsm(page) && PageAnon(page))
Joonsoo Kim52629502014-01-21 15:49:50 -08001508 rwc.invalid_vma = invalid_migration_vma;
1509
1510 ret = rmap_walk(page, &rwc);
1511
Nick Pigginb291f002008-10-18 20:26:44 -07001512 if (ret != SWAP_MLOCK && !page_mapped(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513 ret = SWAP_SUCCESS;
1514 return ret;
1515}
Nikita Danilov81b40822005-05-01 08:58:36 -07001516
Nick Pigginb291f002008-10-18 20:26:44 -07001517/**
1518 * try_to_munlock - try to munlock a page
1519 * @page: the page to be munlocked
1520 *
1521 * Called from munlock code. Checks all of the VMAs mapping the page
1522 * to make sure nobody else has this page mlocked. The page will be
1523 * returned with PG_mlocked cleared if no other vmas have it mlocked.
1524 *
1525 * Return values are:
1526 *
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001527 * SWAP_AGAIN - no vma is holding page mlocked, or,
Nick Pigginb291f002008-10-18 20:26:44 -07001528 * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem
Hugh Dickins5ad64682009-12-14 17:59:24 -08001529 * SWAP_FAIL - page cannot be located at present
Nick Pigginb291f002008-10-18 20:26:44 -07001530 * SWAP_MLOCK - page is now mlocked.
1531 */
1532int try_to_munlock(struct page *page)
1533{
Joonsoo Kime8351ac2014-01-21 15:49:52 -08001534 int ret;
1535 struct rmap_walk_control rwc = {
1536 .rmap_one = try_to_unmap_one,
1537 .arg = (void *)TTU_MUNLOCK,
1538 .done = page_not_mapped,
Joonsoo Kime8351ac2014-01-21 15:49:52 -08001539 .anon_lock = page_lock_anon_vma_read,
1540
1541 };
1542
Sasha Levin309381fea2014-01-23 15:52:54 -08001543 VM_BUG_ON_PAGE(!PageLocked(page) || PageLRU(page), page);
Nick Pigginb291f002008-10-18 20:26:44 -07001544
Joonsoo Kime8351ac2014-01-21 15:49:52 -08001545 ret = rmap_walk(page, &rwc);
1546 return ret;
Nick Pigginb291f002008-10-18 20:26:44 -07001547}
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001548
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001549void __put_anon_vma(struct anon_vma *anon_vma)
Rik van Riel76545062010-08-09 17:18:41 -07001550{
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001551 struct anon_vma *root = anon_vma->root;
Rik van Riel76545062010-08-09 17:18:41 -07001552
Andrey Ryabinin624483f2014-06-06 19:09:30 +04001553 anon_vma_free(anon_vma);
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001554 if (root != anon_vma && atomic_dec_and_test(&root->refcount))
1555 anon_vma_free(root);
Rik van Riel76545062010-08-09 17:18:41 -07001556}
Rik van Riel76545062010-08-09 17:18:41 -07001557
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001558static struct anon_vma *rmap_walk_anon_lock(struct page *page,
1559 struct rmap_walk_control *rwc)
Joonsoo Kimfaecd8d2014-01-21 15:49:46 -08001560{
1561 struct anon_vma *anon_vma;
1562
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001563 if (rwc->anon_lock)
1564 return rwc->anon_lock(page);
1565
Joonsoo Kimfaecd8d2014-01-21 15:49:46 -08001566 /*
1567 * Note: remove_migration_ptes() cannot use page_lock_anon_vma_read()
1568 * because that depends on page_mapped(); but not all its usages
1569 * are holding mmap_sem. Users without mmap_sem are required to
1570 * take a reference count to prevent the anon_vma disappearing
1571 */
1572 anon_vma = page_anon_vma(page);
1573 if (!anon_vma)
1574 return NULL;
1575
1576 anon_vma_lock_read(anon_vma);
1577 return anon_vma;
1578}
1579
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001580/*
Joonsoo Kime8351ac2014-01-21 15:49:52 -08001581 * rmap_walk_anon - do something to anonymous page using the object-based
1582 * rmap method
1583 * @page: the page to be handled
1584 * @rwc: control variable according to each walk type
1585 *
1586 * Find all the mappings of a page using the mapping pointer and the vma chains
1587 * contained in the anon_vma struct it points to.
1588 *
1589 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1590 * where the page was found will be held for write. So, we won't recheck
1591 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1592 * LOCKED.
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001593 */
Joonsoo Kim051ac832014-01-21 15:49:48 -08001594static int rmap_walk_anon(struct page *page, struct rmap_walk_control *rwc)
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001595{
1596 struct anon_vma *anon_vma;
Davidlohr Buesob258d862014-12-12 16:55:04 -08001597 pgoff_t pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001598 struct anon_vma_chain *avc;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001599 int ret = SWAP_AGAIN;
1600
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001601 anon_vma = rmap_walk_anon_lock(page, rwc);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001602 if (!anon_vma)
1603 return ret;
Joonsoo Kimfaecd8d2014-01-21 15:49:46 -08001604
Davidlohr Buesob258d862014-12-12 16:55:04 -08001605 pgoff = page_to_pgoff(page);
Michel Lespinassebf181b92012-10-08 16:31:39 -07001606 anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) {
Rik van Riel5beb4932010-03-05 13:42:07 -08001607 struct vm_area_struct *vma = avc->vma;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001608 unsigned long address = vma_address(page, vma);
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001609
1610 if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg))
1611 continue;
1612
Joonsoo Kim051ac832014-01-21 15:49:48 -08001613 ret = rwc->rmap_one(page, vma, address, rwc->arg);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001614 if (ret != SWAP_AGAIN)
1615 break;
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001616 if (rwc->done && rwc->done(page))
1617 break;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001618 }
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +00001619 anon_vma_unlock_read(anon_vma);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001620 return ret;
1621}
1622
Joonsoo Kime8351ac2014-01-21 15:49:52 -08001623/*
1624 * rmap_walk_file - do something to file page using the object-based rmap method
1625 * @page: the page to be handled
1626 * @rwc: control variable according to each walk type
1627 *
1628 * Find all the mappings of a page using the mapping pointer and the vma chains
1629 * contained in the address_space struct it points to.
1630 *
1631 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1632 * where the page was found will be held for write. So, we won't recheck
1633 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1634 * LOCKED.
1635 */
Joonsoo Kim051ac832014-01-21 15:49:48 -08001636static int rmap_walk_file(struct page *page, struct rmap_walk_control *rwc)
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001637{
1638 struct address_space *mapping = page->mapping;
Davidlohr Buesob258d862014-12-12 16:55:04 -08001639 pgoff_t pgoff;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001640 struct vm_area_struct *vma;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001641 int ret = SWAP_AGAIN;
1642
Joonsoo Kim9f326242014-01-21 15:49:53 -08001643 /*
1644 * The page lock not only makes sure that page->mapping cannot
1645 * suddenly be NULLified by truncation, it makes sure that the
1646 * structure at mapping cannot be freed and reused yet,
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08001647 * so we can safely take mapping->i_mmap_rwsem.
Joonsoo Kim9f326242014-01-21 15:49:53 -08001648 */
Sasha Levin81d1b092014-10-09 15:28:10 -07001649 VM_BUG_ON_PAGE(!PageLocked(page), page);
Joonsoo Kim9f326242014-01-21 15:49:53 -08001650
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001651 if (!mapping)
1652 return ret;
Davidlohr Bueso3dec0ba2014-12-12 16:54:27 -08001653
Davidlohr Buesob258d862014-12-12 16:55:04 -08001654 pgoff = page_to_pgoff(page);
Davidlohr Bueso3dec0ba2014-12-12 16:54:27 -08001655 i_mmap_lock_read(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07001656 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001657 unsigned long address = vma_address(page, vma);
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001658
1659 if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg))
1660 continue;
1661
Joonsoo Kim051ac832014-01-21 15:49:48 -08001662 ret = rwc->rmap_one(page, vma, address, rwc->arg);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001663 if (ret != SWAP_AGAIN)
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001664 goto done;
1665 if (rwc->done && rwc->done(page))
1666 goto done;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001667 }
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001668
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001669done:
Davidlohr Bueso3dec0ba2014-12-12 16:54:27 -08001670 i_mmap_unlock_read(mapping);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001671 return ret;
1672}
1673
Joonsoo Kim051ac832014-01-21 15:49:48 -08001674int rmap_walk(struct page *page, struct rmap_walk_control *rwc)
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001675{
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001676 if (unlikely(PageKsm(page)))
Joonsoo Kim051ac832014-01-21 15:49:48 -08001677 return rmap_walk_ksm(page, rwc);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001678 else if (PageAnon(page))
Joonsoo Kim051ac832014-01-21 15:49:48 -08001679 return rmap_walk_anon(page, rwc);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001680 else
Joonsoo Kim051ac832014-01-21 15:49:48 -08001681 return rmap_walk_file(page, rwc);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001682}
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001683
Naoya Horiguchie3390f62010-06-15 13:18:13 +09001684#ifdef CONFIG_HUGETLB_PAGE
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001685/*
1686 * The following three functions are for anonymous (private mapped) hugepages.
1687 * Unlike common anonymous pages, anonymous hugepages have no accounting code
1688 * and no lru code, because we handle hugepages differently from common pages.
1689 */
1690static void __hugepage_set_anon_rmap(struct page *page,
1691 struct vm_area_struct *vma, unsigned long address, int exclusive)
1692{
1693 struct anon_vma *anon_vma = vma->anon_vma;
Naoya Horiguchi433abed2010-09-10 13:23:03 +09001694
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001695 BUG_ON(!anon_vma);
Naoya Horiguchi433abed2010-09-10 13:23:03 +09001696
1697 if (PageAnon(page))
1698 return;
1699 if (!exclusive)
1700 anon_vma = anon_vma->root;
1701
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001702 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
1703 page->mapping = (struct address_space *) anon_vma;
1704 page->index = linear_page_index(vma, address);
1705}
1706
1707void hugepage_add_anon_rmap(struct page *page,
1708 struct vm_area_struct *vma, unsigned long address)
1709{
1710 struct anon_vma *anon_vma = vma->anon_vma;
1711 int first;
Naoya Horiguchia850ea32010-09-10 13:23:06 +09001712
1713 BUG_ON(!PageLocked(page));
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001714 BUG_ON(!anon_vma);
Hugh Dickins5dbe0af2011-05-28 13:17:04 -07001715 /* address might be in next vma when migration races vma_adjust */
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001716 first = atomic_inc_and_test(&page->_mapcount);
1717 if (first)
1718 __hugepage_set_anon_rmap(page, vma, address, 0);
1719}
1720
1721void hugepage_add_new_anon_rmap(struct page *page,
1722 struct vm_area_struct *vma, unsigned long address)
1723{
1724 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
1725 atomic_set(&page->_mapcount, 0);
1726 __hugepage_set_anon_rmap(page, vma, address, 1);
1727}
Naoya Horiguchie3390f62010-06-15 13:18:13 +09001728#endif /* CONFIG_HUGETLB_PAGE */