blob: 31d8866fb5629cb35033ac414b7a95625c688722 [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),
Vladimir Davydov5d097052016-01-14 15:18:21 -0800431 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC|SLAB_ACCOUNT,
432 anon_vma_ctor);
433 anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain,
434 SLAB_PANIC|SLAB_ACCOUNT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435}
436
437/*
Peter Zijlstra6111e4c2011-05-24 17:12:08 -0700438 * Getting a lock on a stable anon_vma from a page off the LRU is tricky!
439 *
440 * Since there is no serialization what so ever against page_remove_rmap()
441 * the best this function can do is return a locked anon_vma that might
442 * have been relevant to this page.
443 *
444 * The page might have been remapped to a different anon_vma or the anon_vma
445 * returned may already be freed (and even reused).
446 *
Peter Zijlstrabc658c92011-05-29 10:33:44 +0200447 * In case it was remapped to a different anon_vma, the new anon_vma will be a
448 * child of the old anon_vma, and the anon_vma lifetime rules will therefore
449 * ensure that any anon_vma obtained from the page will still be valid for as
450 * long as we observe page_mapped() [ hence all those page_mapped() tests ].
451 *
Peter Zijlstra6111e4c2011-05-24 17:12:08 -0700452 * All users of this function must be very careful when walking the anon_vma
453 * chain and verify that the page in question is indeed mapped in it
454 * [ something equivalent to page_mapped_in_vma() ].
455 *
456 * Since anon_vma's slab is DESTROY_BY_RCU and we know from page_remove_rmap()
457 * that the anon_vma pointer from page->mapping is valid if there is a
458 * mapcount, we can dereference the anon_vma after observing those.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 */
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700460struct anon_vma *page_get_anon_vma(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461{
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700462 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 unsigned long anon_mapping;
464
465 rcu_read_lock();
Jason Low4db0c3c2015-04-15 16:14:08 -0700466 anon_mapping = (unsigned long)READ_ONCE(page->mapping);
Hugh Dickins3ca7b3c2009-12-14 17:58:57 -0800467 if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 goto out;
469 if (!page_mapped(page))
470 goto out;
471
472 anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700473 if (!atomic_inc_not_zero(&anon_vma->refcount)) {
474 anon_vma = NULL;
475 goto out;
476 }
Hugh Dickinsf18194272010-08-25 23:12:54 -0700477
478 /*
479 * If this page is still mapped, then its anon_vma cannot have been
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700480 * freed. But if it has been unmapped, we have no security against the
481 * anon_vma structure being freed and reused (for another anon_vma:
482 * SLAB_DESTROY_BY_RCU guarantees that - so the atomic_inc_not_zero()
483 * above cannot corrupt).
Hugh Dickinsf18194272010-08-25 23:12:54 -0700484 */
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700485 if (!page_mapped(page)) {
Hugh Dickins7f39dda2014-06-04 16:05:33 -0700486 rcu_read_unlock();
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700487 put_anon_vma(anon_vma);
Hugh Dickins7f39dda2014-06-04 16:05:33 -0700488 return NULL;
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700489 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490out:
491 rcu_read_unlock();
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700492
493 return anon_vma;
494}
495
Peter Zijlstra88c22082011-05-24 17:12:13 -0700496/*
497 * Similar to page_get_anon_vma() except it locks the anon_vma.
498 *
499 * Its a little more complex as it tries to keep the fast path to a single
500 * atomic op -- the trylock. If we fail the trylock, we fall back to getting a
501 * reference like with page_get_anon_vma() and then block on the mutex.
502 */
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000503struct anon_vma *page_lock_anon_vma_read(struct page *page)
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700504{
Peter Zijlstra88c22082011-05-24 17:12:13 -0700505 struct anon_vma *anon_vma = NULL;
Hugh Dickinseee0f252011-05-28 13:20:21 -0700506 struct anon_vma *root_anon_vma;
Peter Zijlstra88c22082011-05-24 17:12:13 -0700507 unsigned long anon_mapping;
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700508
Peter Zijlstra88c22082011-05-24 17:12:13 -0700509 rcu_read_lock();
Jason Low4db0c3c2015-04-15 16:14:08 -0700510 anon_mapping = (unsigned long)READ_ONCE(page->mapping);
Peter Zijlstra88c22082011-05-24 17:12:13 -0700511 if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
512 goto out;
513 if (!page_mapped(page))
514 goto out;
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700515
Peter Zijlstra88c22082011-05-24 17:12:13 -0700516 anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
Jason Low4db0c3c2015-04-15 16:14:08 -0700517 root_anon_vma = READ_ONCE(anon_vma->root);
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000518 if (down_read_trylock(&root_anon_vma->rwsem)) {
Peter Zijlstra88c22082011-05-24 17:12:13 -0700519 /*
Hugh Dickinseee0f252011-05-28 13:20:21 -0700520 * If the page is still mapped, then this anon_vma is still
521 * its anon_vma, and holding the mutex ensures that it will
Peter Zijlstrabc658c92011-05-29 10:33:44 +0200522 * not go away, see anon_vma_free().
Peter Zijlstra88c22082011-05-24 17:12:13 -0700523 */
Hugh Dickinseee0f252011-05-28 13:20:21 -0700524 if (!page_mapped(page)) {
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000525 up_read(&root_anon_vma->rwsem);
Peter Zijlstra88c22082011-05-24 17:12:13 -0700526 anon_vma = NULL;
527 }
528 goto out;
529 }
530
531 /* trylock failed, we got to sleep */
532 if (!atomic_inc_not_zero(&anon_vma->refcount)) {
533 anon_vma = NULL;
534 goto out;
535 }
536
537 if (!page_mapped(page)) {
Hugh Dickins7f39dda2014-06-04 16:05:33 -0700538 rcu_read_unlock();
Peter Zijlstra88c22082011-05-24 17:12:13 -0700539 put_anon_vma(anon_vma);
Hugh Dickins7f39dda2014-06-04 16:05:33 -0700540 return NULL;
Peter Zijlstra88c22082011-05-24 17:12:13 -0700541 }
542
543 /* we pinned the anon_vma, its safe to sleep */
544 rcu_read_unlock();
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000545 anon_vma_lock_read(anon_vma);
Peter Zijlstra88c22082011-05-24 17:12:13 -0700546
547 if (atomic_dec_and_test(&anon_vma->refcount)) {
548 /*
549 * Oops, we held the last refcount, release the lock
550 * and bail -- can't simply use put_anon_vma() because
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000551 * we'll deadlock on the anon_vma_lock_write() recursion.
Peter Zijlstra88c22082011-05-24 17:12:13 -0700552 */
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000553 anon_vma_unlock_read(anon_vma);
Peter Zijlstra88c22082011-05-24 17:12:13 -0700554 __put_anon_vma(anon_vma);
555 anon_vma = NULL;
556 }
557
558 return anon_vma;
559
560out:
561 rcu_read_unlock();
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700562 return anon_vma;
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800563}
564
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000565void page_unlock_anon_vma_read(struct anon_vma *anon_vma)
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800566{
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000567 anon_vma_unlock_read(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568}
569
Mel Gorman72b252a2015-09-04 15:47:32 -0700570#ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
571static void percpu_flush_tlb_batch_pages(void *data)
572{
573 /*
574 * All TLB entries are flushed on the assumption that it is
575 * cheaper to flush all TLBs and let them be refilled than
576 * flushing individual PFNs. Note that we do not track mm's
577 * to flush as that might simply be multiple full TLB flushes
578 * for no gain.
579 */
580 count_vm_tlb_event(NR_TLB_REMOTE_FLUSH_RECEIVED);
581 flush_tlb_local();
582}
583
584/*
585 * Flush TLB entries for recently unmapped pages from remote CPUs. It is
586 * important if a PTE was dirty when it was unmapped that it's flushed
587 * before any IO is initiated on the page to prevent lost writes. Similarly,
588 * it must be flushed before freeing to prevent data leakage.
589 */
590void try_to_unmap_flush(void)
591{
592 struct tlbflush_unmap_batch *tlb_ubc = &current->tlb_ubc;
593 int cpu;
594
595 if (!tlb_ubc->flush_required)
596 return;
597
598 cpu = get_cpu();
599
600 trace_tlb_flush(TLB_REMOTE_SHOOTDOWN, -1UL);
601
602 if (cpumask_test_cpu(cpu, &tlb_ubc->cpumask))
603 percpu_flush_tlb_batch_pages(&tlb_ubc->cpumask);
604
605 if (cpumask_any_but(&tlb_ubc->cpumask, cpu) < nr_cpu_ids) {
606 smp_call_function_many(&tlb_ubc->cpumask,
607 percpu_flush_tlb_batch_pages, (void *)tlb_ubc, true);
608 }
609 cpumask_clear(&tlb_ubc->cpumask);
610 tlb_ubc->flush_required = false;
Mel Gormand950c942015-09-04 15:47:35 -0700611 tlb_ubc->writable = false;
Mel Gorman72b252a2015-09-04 15:47:32 -0700612 put_cpu();
613}
614
Mel Gormand950c942015-09-04 15:47:35 -0700615/* Flush iff there are potentially writable TLB entries that can race with IO */
616void try_to_unmap_flush_dirty(void)
617{
618 struct tlbflush_unmap_batch *tlb_ubc = &current->tlb_ubc;
619
620 if (tlb_ubc->writable)
621 try_to_unmap_flush();
622}
623
Mel Gorman72b252a2015-09-04 15:47:32 -0700624static void set_tlb_ubc_flush_pending(struct mm_struct *mm,
Mel Gormand950c942015-09-04 15:47:35 -0700625 struct page *page, bool writable)
Mel Gorman72b252a2015-09-04 15:47:32 -0700626{
627 struct tlbflush_unmap_batch *tlb_ubc = &current->tlb_ubc;
628
629 cpumask_or(&tlb_ubc->cpumask, &tlb_ubc->cpumask, mm_cpumask(mm));
630 tlb_ubc->flush_required = true;
Mel Gormand950c942015-09-04 15:47:35 -0700631
632 /*
633 * If the PTE was dirty then it's best to assume it's writable. The
634 * caller must use try_to_unmap_flush_dirty() or try_to_unmap_flush()
635 * before the page is queued for IO.
636 */
637 if (writable)
638 tlb_ubc->writable = true;
Mel Gorman72b252a2015-09-04 15:47:32 -0700639}
640
641/*
642 * Returns true if the TLB flush should be deferred to the end of a batch of
643 * unmap operations to reduce IPIs.
644 */
645static bool should_defer_flush(struct mm_struct *mm, enum ttu_flags flags)
646{
647 bool should_defer = false;
648
649 if (!(flags & TTU_BATCH_FLUSH))
650 return false;
651
652 /* If remote CPUs need to be flushed then defer batch the flush */
653 if (cpumask_any_but(mm_cpumask(mm), get_cpu()) < nr_cpu_ids)
654 should_defer = true;
655 put_cpu();
656
657 return should_defer;
658}
659#else
660static void set_tlb_ubc_flush_pending(struct mm_struct *mm,
Mel Gormand950c942015-09-04 15:47:35 -0700661 struct page *page, bool writable)
Mel Gorman72b252a2015-09-04 15:47:32 -0700662{
663}
664
665static bool should_defer_flush(struct mm_struct *mm, enum ttu_flags flags)
666{
667 return false;
668}
669#endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
670
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671/*
Huang Shijiebf89c8c2009-10-01 15:44:04 -0700672 * At what user virtual address is page expected in vma?
Naoya Horiguchiab941e02010-05-11 14:06:55 -0700673 * Caller should check the page is actually part of the vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 */
675unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma)
676{
Michel Lespinasse86c2ad12012-10-08 16:31:42 -0700677 unsigned long address;
Andrea Arcangeli21d0d442010-08-09 17:19:10 -0700678 if (PageAnon(page)) {
Hugh Dickins4829b902010-10-02 17:46:06 -0700679 struct anon_vma *page__anon_vma = page_anon_vma(page);
680 /*
681 * Note: swapoff's unuse_vma() is more efficient with this
682 * check, and needs it to match anon_vma when KSM is active.
683 */
684 if (!vma->anon_vma || !page__anon_vma ||
685 vma->anon_vma->root != page__anon_vma->root)
Andrea Arcangeli21d0d442010-08-09 17:19:10 -0700686 return -EFAULT;
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800687 } else if (page->mapping) {
688 if (!vma->vm_file || vma->vm_file->f_mapping != page->mapping)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 return -EFAULT;
690 } else
691 return -EFAULT;
Michel Lespinasse86c2ad12012-10-08 16:31:42 -0700692 address = __vma_address(page, vma);
693 if (unlikely(address < vma->vm_start || address >= vma->vm_end))
694 return -EFAULT;
695 return address;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696}
697
Bob Liu62190492012-12-11 16:00:37 -0800698pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address)
699{
700 pgd_t *pgd;
701 pud_t *pud;
702 pmd_t *pmd = NULL;
Hugh Dickinsf72e7dc2014-06-23 13:22:05 -0700703 pmd_t pmde;
Bob Liu62190492012-12-11 16:00:37 -0800704
705 pgd = pgd_offset(mm, address);
706 if (!pgd_present(*pgd))
707 goto out;
708
709 pud = pud_offset(pgd, address);
710 if (!pud_present(*pud))
711 goto out;
712
713 pmd = pmd_offset(pud, address);
Hugh Dickinsf72e7dc2014-06-23 13:22:05 -0700714 /*
Aneesh Kumar K.V8809aa22015-06-24 16:57:44 -0700715 * Some THP functions use the sequence pmdp_huge_clear_flush(), set_pmd_at()
Hugh Dickinsf72e7dc2014-06-23 13:22:05 -0700716 * without holding anon_vma lock for write. So when looking for a
717 * genuine pmde (in which to find pte), test present and !THP together.
718 */
Christian Borntraegere37c6982014-12-07 21:41:33 +0100719 pmde = *pmd;
720 barrier();
Hugh Dickinsf72e7dc2014-06-23 13:22:05 -0700721 if (!pmd_present(pmde) || pmd_trans_huge(pmde))
Bob Liu62190492012-12-11 16:00:37 -0800722 pmd = NULL;
723out:
724 return pmd;
725}
726
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727/*
Nikita Danilov81b40822005-05-01 08:58:36 -0700728 * Check that @page is mapped at @address into @mm.
729 *
Nick Piggin479db0b2008-08-20 14:09:18 -0700730 * If @sync is false, page_check_address may perform a racy check to avoid
731 * the page table lock when the pte is not present (helpful when reclaiming
732 * highly shared pages).
733 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -0700734 * On success returns with pte mapped and locked.
Nikita Danilov81b40822005-05-01 08:58:36 -0700735 */
Namhyung Kime9a81a82010-10-26 14:22:01 -0700736pte_t *__page_check_address(struct page *page, struct mm_struct *mm,
Nick Piggin479db0b2008-08-20 14:09:18 -0700737 unsigned long address, spinlock_t **ptlp, int sync)
Nikita Danilov81b40822005-05-01 08:58:36 -0700738{
Nikita Danilov81b40822005-05-01 08:58:36 -0700739 pmd_t *pmd;
740 pte_t *pte;
Hugh Dickinsc0718802005-10-29 18:16:31 -0700741 spinlock_t *ptl;
Nikita Danilov81b40822005-05-01 08:58:36 -0700742
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +0900743 if (unlikely(PageHuge(page))) {
Jianguo Wu98398c32013-12-18 17:08:59 -0800744 /* when pud is not present, pte will be NULL */
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +0900745 pte = huge_pte_offset(mm, address);
Jianguo Wu98398c32013-12-18 17:08:59 -0800746 if (!pte)
747 return NULL;
748
Kirill A. Shutemovcb900f42013-11-14 14:31:02 -0800749 ptl = huge_pte_lockptr(page_hstate(page), mm, pte);
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +0900750 goto check;
751 }
752
Bob Liu62190492012-12-11 16:00:37 -0800753 pmd = mm_find_pmd(mm, address);
754 if (!pmd)
Hugh Dickinsc0718802005-10-29 18:16:31 -0700755 return NULL;
756
Hugh Dickinsc0718802005-10-29 18:16:31 -0700757 pte = pte_offset_map(pmd, address);
758 /* Make a quick check before getting the lock */
Nick Piggin479db0b2008-08-20 14:09:18 -0700759 if (!sync && !pte_present(*pte)) {
Hugh Dickinsc0718802005-10-29 18:16:31 -0700760 pte_unmap(pte);
761 return NULL;
Nikita Danilov81b40822005-05-01 08:58:36 -0700762 }
Hugh Dickinsc0718802005-10-29 18:16:31 -0700763
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700764 ptl = pte_lockptr(mm, pmd);
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +0900765check:
Hugh Dickinsc0718802005-10-29 18:16:31 -0700766 spin_lock(ptl);
767 if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) {
768 *ptlp = ptl;
769 return pte;
770 }
771 pte_unmap_unlock(pte, ptl);
772 return NULL;
Nikita Danilov81b40822005-05-01 08:58:36 -0700773}
774
Nick Pigginb291f002008-10-18 20:26:44 -0700775/**
776 * page_mapped_in_vma - check whether a page is really mapped in a VMA
777 * @page: the page to test
778 * @vma: the VMA to test
779 *
780 * Returns 1 if the page is mapped into the page tables of the VMA, 0
781 * if the page is not mapped into the page tables of this VMA. Only
782 * valid for normal file or anonymous VMAs.
783 */
Andi Kleen6a460792009-09-16 11:50:15 +0200784int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma)
Nick Pigginb291f002008-10-18 20:26:44 -0700785{
786 unsigned long address;
787 pte_t *pte;
788 spinlock_t *ptl;
789
Michel Lespinasse86c2ad12012-10-08 16:31:42 -0700790 address = __vma_address(page, vma);
791 if (unlikely(address < vma->vm_start || address >= vma->vm_end))
Nick Pigginb291f002008-10-18 20:26:44 -0700792 return 0;
793 pte = page_check_address(page, vma->vm_mm, address, &ptl, 1);
794 if (!pte) /* the page is not in this mm */
795 return 0;
796 pte_unmap_unlock(pte, ptl);
797
798 return 1;
799}
800
Joonsoo Kim9f326242014-01-21 15:49:53 -0800801struct page_referenced_arg {
802 int mapcount;
803 int referenced;
804 unsigned long vm_flags;
805 struct mem_cgroup *memcg;
806};
Nikita Danilov81b40822005-05-01 08:58:36 -0700807/*
Joonsoo Kim9f326242014-01-21 15:49:53 -0800808 * arg: page_referenced_arg will be passed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 */
Kirill A. Shutemovac769502014-06-04 16:08:17 -0700810static int page_referenced_one(struct page *page, struct vm_area_struct *vma,
Joonsoo Kim9f326242014-01-21 15:49:53 -0800811 unsigned long address, void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812{
813 struct mm_struct *mm = vma->vm_mm;
Kirill A. Shutemov117b0792013-11-14 14:30:56 -0800814 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 int referenced = 0;
Joonsoo Kim9f326242014-01-21 15:49:53 -0800816 struct page_referenced_arg *pra = arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800818 if (unlikely(PageTransHuge(page))) {
819 pmd_t *pmd;
820
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800821 /*
822 * rmap might return false positives; we must filter
823 * these out using page_check_address_pmd().
824 */
Kirill A. Shutemov4b471e82016-01-15 16:53:39 -0800825 pmd = page_check_address_pmd(page, mm, address, &ptl);
Kirill A. Shutemov117b0792013-11-14 14:30:56 -0800826 if (!pmd)
Joonsoo Kim9f326242014-01-21 15:49:53 -0800827 return SWAP_AGAIN;
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800828
829 if (vma->vm_flags & VM_LOCKED) {
Kirill A. Shutemov117b0792013-11-14 14:30:56 -0800830 spin_unlock(ptl);
Joonsoo Kim9f326242014-01-21 15:49:53 -0800831 pra->vm_flags |= VM_LOCKED;
832 return SWAP_FAIL; /* To break the loop */
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800833 }
834
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800835 if (pmdp_clear_flush_young_notify(vma, address, pmd))
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800836 referenced++;
Kirill A. Shutemov117b0792013-11-14 14:30:56 -0800837 spin_unlock(ptl);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800838 } else {
839 pte_t *pte;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800840
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800841 /*
842 * rmap might return false positives; we must filter
843 * these out using page_check_address().
844 */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800845 pte = page_check_address(page, mm, address, &ptl, 0);
846 if (!pte)
Joonsoo Kim9f326242014-01-21 15:49:53 -0800847 return SWAP_AGAIN;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800848
Andrea Arcangeli2da28bf2011-03-11 14:58:29 -0800849 if (vma->vm_flags & VM_LOCKED) {
850 pte_unmap_unlock(pte, 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
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800855 if (ptep_clear_flush_young_notify(vma, address, pte)) {
856 /*
857 * Don't treat a reference through a sequentially read
858 * mapping as such. If the page has been used in
859 * another mapping, we will catch it; if this other
860 * mapping is already gone, the unmap path will have
861 * set PG_referenced or activated the page.
862 */
Joe Perches64363aa2013-07-08 16:00:18 -0700863 if (likely(!(vma->vm_flags & VM_SEQ_READ)))
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800864 referenced++;
865 }
866 pte_unmap_unlock(pte, ptl);
867 }
868
Vladimir Davydov33c3fc72015-09-09 15:35:45 -0700869 if (referenced)
870 clear_page_idle(page);
871 if (test_and_clear_page_young(page))
872 referenced++;
873
Joonsoo Kim9f326242014-01-21 15:49:53 -0800874 if (referenced) {
875 pra->referenced++;
876 pra->vm_flags |= vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 }
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800878
Joonsoo Kim9f326242014-01-21 15:49:53 -0800879 pra->mapcount--;
880 if (!pra->mapcount)
881 return SWAP_SUCCESS; /* To break the loop */
882
883 return SWAP_AGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884}
885
Joonsoo Kim9f326242014-01-21 15:49:53 -0800886static bool invalid_page_referenced_vma(struct vm_area_struct *vma, void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887{
Joonsoo Kim9f326242014-01-21 15:49:53 -0800888 struct page_referenced_arg *pra = arg;
889 struct mem_cgroup *memcg = pra->memcg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890
Joonsoo Kim9f326242014-01-21 15:49:53 -0800891 if (!mm_match_cgroup(vma->vm_mm, memcg))
892 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893
Joonsoo Kim9f326242014-01-21 15:49:53 -0800894 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895}
896
897/**
898 * page_referenced - test if the page was referenced
899 * @page: the page to test
900 * @is_locked: caller holds lock on the page
Johannes Weiner72835c82012-01-12 17:18:32 -0800901 * @memcg: target memory cgroup
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700902 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 *
904 * Quick test_and_clear_referenced for all mappings to a page,
905 * returns the number of ptes which referenced the page.
906 */
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700907int page_referenced(struct page *page,
908 int is_locked,
Johannes Weiner72835c82012-01-12 17:18:32 -0800909 struct mem_cgroup *memcg,
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700910 unsigned long *vm_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911{
Joonsoo Kim9f326242014-01-21 15:49:53 -0800912 int ret;
Hugh Dickins5ad64682009-12-14 17:59:24 -0800913 int we_locked = 0;
Joonsoo Kim9f326242014-01-21 15:49:53 -0800914 struct page_referenced_arg pra = {
915 .mapcount = page_mapcount(page),
916 .memcg = memcg,
917 };
918 struct rmap_walk_control rwc = {
919 .rmap_one = page_referenced_one,
920 .arg = (void *)&pra,
921 .anon_lock = page_lock_anon_vma_read,
922 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700924 *vm_flags = 0;
Joonsoo Kim9f326242014-01-21 15:49:53 -0800925 if (!page_mapped(page))
926 return 0;
927
928 if (!page_rmapping(page))
929 return 0;
930
931 if (!is_locked && (!PageAnon(page) || PageKsm(page))) {
932 we_locked = trylock_page(page);
933 if (!we_locked)
934 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 }
Joonsoo Kim9f326242014-01-21 15:49:53 -0800936
937 /*
938 * If we are reclaiming on behalf of a cgroup, skip
939 * counting on behalf of references from different
940 * cgroups
941 */
942 if (memcg) {
943 rwc.invalid_vma = invalid_page_referenced_vma;
944 }
945
946 ret = rmap_walk(page, &rwc);
947 *vm_flags = pra.vm_flags;
948
949 if (we_locked)
950 unlock_page(page);
951
952 return pra.referenced;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953}
954
Hugh Dickins1cb17292009-12-14 17:59:01 -0800955static int page_mkclean_one(struct page *page, struct vm_area_struct *vma,
Joonsoo Kim9853a402014-01-21 15:49:55 -0800956 unsigned long address, void *arg)
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700957{
958 struct mm_struct *mm = vma->vm_mm;
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100959 pte_t *pte;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700960 spinlock_t *ptl;
961 int ret = 0;
Joonsoo Kim9853a402014-01-21 15:49:55 -0800962 int *cleaned = arg;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700963
Nick Piggin479db0b2008-08-20 14:09:18 -0700964 pte = page_check_address(page, mm, address, &ptl, 1);
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700965 if (!pte)
966 goto out;
967
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100968 if (pte_dirty(*pte) || pte_write(*pte)) {
969 pte_t entry;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700970
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100971 flush_cache_page(vma, address, pte_pfn(*pte));
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700972 entry = ptep_clear_flush(vma, address, pte);
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100973 entry = pte_wrprotect(entry);
974 entry = pte_mkclean(entry);
Al Virod6e88e62006-12-29 16:48:35 -0800975 set_pte_at(mm, address, pte, entry);
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100976 ret = 1;
977 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700978
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700979 pte_unmap_unlock(pte, ptl);
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700980
Joonsoo Kim9853a402014-01-21 15:49:55 -0800981 if (ret) {
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700982 mmu_notifier_invalidate_page(mm, address);
Joonsoo Kim9853a402014-01-21 15:49:55 -0800983 (*cleaned)++;
984 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700985out:
Joonsoo Kim9853a402014-01-21 15:49:55 -0800986 return SWAP_AGAIN;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700987}
988
Joonsoo Kim9853a402014-01-21 15:49:55 -0800989static bool invalid_mkclean_vma(struct vm_area_struct *vma, void *arg)
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700990{
Joonsoo Kim9853a402014-01-21 15:49:55 -0800991 if (vma->vm_flags & VM_SHARED)
Fengguang Wu871beb82014-01-23 15:53:41 -0800992 return false;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700993
Fengguang Wu871beb82014-01-23 15:53:41 -0800994 return true;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700995}
996
997int page_mkclean(struct page *page)
998{
Joonsoo Kim9853a402014-01-21 15:49:55 -0800999 int cleaned = 0;
1000 struct address_space *mapping;
1001 struct rmap_walk_control rwc = {
1002 .arg = (void *)&cleaned,
1003 .rmap_one = page_mkclean_one,
1004 .invalid_vma = invalid_mkclean_vma,
1005 };
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001006
1007 BUG_ON(!PageLocked(page));
1008
Joonsoo Kim9853a402014-01-21 15:49:55 -08001009 if (!page_mapped(page))
1010 return 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001011
Joonsoo Kim9853a402014-01-21 15:49:55 -08001012 mapping = page_mapping(page);
1013 if (!mapping)
1014 return 0;
1015
1016 rmap_walk(page, &rwc);
1017
1018 return cleaned;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001019}
Jaya Kumar60b59be2007-05-08 00:37:37 -07001020EXPORT_SYMBOL_GPL(page_mkclean);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001021
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022/**
Rik van Rielc44b6742010-03-05 13:42:09 -08001023 * page_move_anon_rmap - move a page to our anon_vma
1024 * @page: the page to move to our anon_vma
1025 * @vma: the vma the page belongs to
1026 * @address: the user virtual address mapped
1027 *
1028 * When a page belongs exclusively to one process after a COW event,
1029 * that page can be moved into the anon_vma that belongs to just that
1030 * process, so the rmap code will not search the parent or sibling
1031 * processes.
1032 */
1033void page_move_anon_rmap(struct page *page,
1034 struct vm_area_struct *vma, unsigned long address)
1035{
1036 struct anon_vma *anon_vma = vma->anon_vma;
1037
Sasha Levin309381fea2014-01-23 15:52:54 -08001038 VM_BUG_ON_PAGE(!PageLocked(page), page);
Sasha Levin81d1b092014-10-09 15:28:10 -07001039 VM_BUG_ON_VMA(!anon_vma, vma);
Sasha Levin309381fea2014-01-23 15:52:54 -08001040 VM_BUG_ON_PAGE(page->index != linear_page_index(vma, address), page);
Rik van Rielc44b6742010-03-05 13:42:09 -08001041
1042 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
Vladimir Davydov414e2fb2015-06-24 16:56:56 -07001043 /*
1044 * Ensure that anon_vma and the PAGE_MAPPING_ANON bit are written
1045 * simultaneously, so a concurrent reader (eg page_referenced()'s
1046 * PageAnon()) will not see one without the other.
1047 */
1048 WRITE_ONCE(page->mapping, (struct address_space *) anon_vma);
Rik van Rielc44b6742010-03-05 13:42:09 -08001049}
1050
1051/**
Andi Kleen4e1c1972010-09-22 12:43:56 +02001052 * __page_set_anon_rmap - set up new anonymous rmap
1053 * @page: Page to add to rmap
1054 * @vma: VM area to add page to.
1055 * @address: User virtual address of the mapping
Rik van Riele8a03fe2010-04-14 17:59:28 -04001056 * @exclusive: the page is exclusively owned by the current process
Nick Piggin9617d952006-01-06 00:11:12 -08001057 */
1058static void __page_set_anon_rmap(struct page *page,
Rik van Riele8a03fe2010-04-14 17:59:28 -04001059 struct vm_area_struct *vma, unsigned long address, int exclusive)
Nick Piggin9617d952006-01-06 00:11:12 -08001060{
Rik van Riele8a03fe2010-04-14 17:59:28 -04001061 struct anon_vma *anon_vma = vma->anon_vma;
Nick Piggin9617d952006-01-06 00:11:12 -08001062
Rik van Riele8a03fe2010-04-14 17:59:28 -04001063 BUG_ON(!anon_vma);
Linus Torvaldsea900022010-04-12 12:44:29 -07001064
Andi Kleen4e1c1972010-09-22 12:43:56 +02001065 if (PageAnon(page))
1066 return;
1067
Linus Torvaldsea900022010-04-12 12:44:29 -07001068 /*
Rik van Riele8a03fe2010-04-14 17:59:28 -04001069 * If the page isn't exclusively mapped into this vma,
1070 * we must use the _oldest_ possible anon_vma for the
1071 * page mapping!
Linus Torvaldsea900022010-04-12 12:44:29 -07001072 */
Andi Kleen4e1c1972010-09-22 12:43:56 +02001073 if (!exclusive)
Andrea Arcangeli288468c2010-08-09 17:19:09 -07001074 anon_vma = anon_vma->root;
Linus Torvaldsea900022010-04-12 12:44:29 -07001075
Nick Piggin9617d952006-01-06 00:11:12 -08001076 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
1077 page->mapping = (struct address_space *) anon_vma;
Nick Piggin9617d952006-01-06 00:11:12 -08001078 page->index = linear_page_index(vma, address);
Nick Piggin9617d952006-01-06 00:11:12 -08001079}
1080
1081/**
Randy Dunlap43d8eac2008-03-19 17:00:43 -07001082 * __page_check_anon_rmap - sanity check anonymous rmap addition
Nick Pigginc97a9e12007-05-16 22:11:21 -07001083 * @page: the page to add the mapping to
1084 * @vma: the vm area in which the mapping is added
1085 * @address: the user virtual address mapped
1086 */
1087static void __page_check_anon_rmap(struct page *page,
1088 struct vm_area_struct *vma, unsigned long address)
1089{
1090#ifdef CONFIG_DEBUG_VM
1091 /*
1092 * The page's anon-rmap details (mapping and index) are guaranteed to
1093 * be set up correctly at this point.
1094 *
1095 * We have exclusion against page_add_anon_rmap because the caller
1096 * always holds the page locked, except if called from page_dup_rmap,
1097 * in which case the page is already known to be setup.
1098 *
1099 * We have exclusion against page_add_new_anon_rmap because those pages
1100 * are initially only visible via the pagetables, and the pte is locked
1101 * over the call to page_add_new_anon_rmap.
1102 */
Andrea Arcangeli44ab57a2010-08-09 17:19:10 -07001103 BUG_ON(page_anon_vma(page)->root != vma->anon_vma->root);
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08001104 BUG_ON(page_to_pgoff(page) != linear_page_index(vma, address));
Nick Pigginc97a9e12007-05-16 22:11:21 -07001105#endif
1106}
1107
1108/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 * page_add_anon_rmap - add pte mapping to an anonymous page
1110 * @page: the page to add the mapping to
1111 * @vma: the vm area in which the mapping is added
1112 * @address: the user virtual address mapped
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08001113 * @compound: charge the page as compound or small page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 *
Hugh Dickins5ad64682009-12-14 17:59:24 -08001115 * The caller needs to hold the pte lock, and the page must be locked in
Hugh Dickins80e148222009-12-14 17:59:29 -08001116 * the anon_vma case: to serialize mapping,index checking after setting,
1117 * and to ensure that PageAnon is not being upgraded racily to PageKsm
1118 * (but PageKsm is never downgraded to PageAnon).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 */
1120void page_add_anon_rmap(struct page *page,
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08001121 struct vm_area_struct *vma, unsigned long address, bool compound)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122{
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08001123 do_page_add_anon_rmap(page, vma, address, compound ? RMAP_COMPOUND : 0);
Rik van Rielad8c2ee2010-08-09 17:19:48 -07001124}
1125
1126/*
1127 * Special version of the above for do_swap_page, which often runs
1128 * into pages that are exclusively owned by the current process.
1129 * Everybody else should continue to use page_add_anon_rmap above.
1130 */
1131void do_page_add_anon_rmap(struct page *page,
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08001132 struct vm_area_struct *vma, unsigned long address, int flags)
Rik van Rielad8c2ee2010-08-09 17:19:48 -07001133{
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08001134 bool compound = flags & RMAP_COMPOUND;
1135 bool first;
1136
Kirill A. Shutemove9b61f12016-01-15 16:54:10 -08001137 if (compound) {
1138 atomic_t *mapcount;
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08001139 VM_BUG_ON_PAGE(!PageLocked(page), page);
Kirill A. Shutemove9b61f12016-01-15 16:54:10 -08001140 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
1141 mapcount = compound_mapcount_ptr(page);
1142 first = atomic_inc_and_test(mapcount);
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08001143 } else {
1144 first = atomic_inc_and_test(&page->_mapcount);
1145 }
1146
Andrea Arcangeli79134172011-01-13 15:46:58 -08001147 if (first) {
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08001148 int nr = compound ? hpage_nr_pages(page) : 1;
Jianyu Zhanbea04b02014-06-04 16:09:51 -07001149 /*
1150 * We use the irq-unsafe __{inc|mod}_zone_page_stat because
1151 * these counters are not modified in interrupt context, and
1152 * pte lock(a spinlock) is held, which implies preemption
1153 * disabled.
1154 */
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08001155 if (compound) {
Andrea Arcangeli79134172011-01-13 15:46:58 -08001156 __inc_zone_page_state(page,
1157 NR_ANON_TRANSPARENT_HUGEPAGES);
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08001158 }
1159 __mod_zone_page_state(page_zone(page), NR_ANON_PAGES, nr);
Andrea Arcangeli79134172011-01-13 15:46:58 -08001160 }
Hugh Dickins5ad64682009-12-14 17:59:24 -08001161 if (unlikely(PageKsm(page)))
1162 return;
1163
Sasha Levin309381fea2014-01-23 15:52:54 -08001164 VM_BUG_ON_PAGE(!PageLocked(page), page);
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08001165
Hugh Dickins5dbe0af2011-05-28 13:17:04 -07001166 /* address might be in next vma when migration races vma_adjust */
Hugh Dickins5ad64682009-12-14 17:59:24 -08001167 if (first)
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08001168 __page_set_anon_rmap(page, vma, address,
1169 flags & RMAP_EXCLUSIVE);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001170 else
Nick Pigginc97a9e12007-05-16 22:11:21 -07001171 __page_check_anon_rmap(page, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172}
1173
Randy Dunlap43d8eac2008-03-19 17:00:43 -07001174/**
Nick Piggin9617d952006-01-06 00:11:12 -08001175 * page_add_new_anon_rmap - add pte mapping to a new anonymous page
1176 * @page: the page to add the mapping to
1177 * @vma: the vm area in which the mapping is added
1178 * @address: the user virtual address mapped
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08001179 * @compound: charge the page as compound or small page
Nick Piggin9617d952006-01-06 00:11:12 -08001180 *
1181 * Same as page_add_anon_rmap but must only be called on *new* pages.
1182 * This means the inc-and-test can be bypassed.
Nick Pigginc97a9e12007-05-16 22:11:21 -07001183 * Page does not have to be locked.
Nick Piggin9617d952006-01-06 00:11:12 -08001184 */
1185void page_add_new_anon_rmap(struct page *page,
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08001186 struct vm_area_struct *vma, unsigned long address, bool compound)
Nick Piggin9617d952006-01-06 00:11:12 -08001187{
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08001188 int nr = compound ? hpage_nr_pages(page) : 1;
1189
Sasha Levin81d1b092014-10-09 15:28:10 -07001190 VM_BUG_ON_VMA(address < vma->vm_start || address >= vma->vm_end, vma);
Hugh Dickinscbf84b72009-01-06 14:39:27 -08001191 SetPageSwapBacked(page);
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08001192 if (compound) {
1193 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08001194 /* increment count (starts at -1) */
1195 atomic_set(compound_mapcount_ptr(page), 0);
Andrea Arcangeli79134172011-01-13 15:46:58 -08001196 __inc_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES);
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08001197 } else {
1198 /* Anon THP always mapped first with PMD */
1199 VM_BUG_ON_PAGE(PageTransCompound(page), page);
1200 /* increment count (starts at -1) */
1201 atomic_set(&page->_mapcount, 0);
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08001202 }
1203 __mod_zone_page_state(page_zone(page), NR_ANON_PAGES, nr);
Rik van Riele8a03fe2010-04-14 17:59:28 -04001204 __page_set_anon_rmap(page, vma, address, 1);
Nick Piggin9617d952006-01-06 00:11:12 -08001205}
1206
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207/**
1208 * page_add_file_rmap - add pte mapping to a file page
1209 * @page: the page to add the mapping to
1210 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -07001211 * The caller needs to hold the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 */
1213void page_add_file_rmap(struct page *page)
1214{
Johannes Weinerd7365e72014-10-29 14:50:48 -07001215 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001216
Johannes Weiner6de22612015-02-11 15:25:01 -08001217 memcg = mem_cgroup_begin_page_stat(page);
Balbir Singhd69b0422009-06-17 16:26:34 -07001218 if (atomic_inc_and_test(&page->_mapcount)) {
Christoph Lameter65ba55f2006-06-30 01:55:34 -07001219 __inc_zone_page_state(page, NR_FILE_MAPPED);
Johannes Weinerd7365e72014-10-29 14:50:48 -07001220 mem_cgroup_inc_page_stat(memcg, MEM_CGROUP_STAT_FILE_MAPPED);
Balbir Singhd69b0422009-06-17 16:26:34 -07001221 }
Johannes Weiner6de22612015-02-11 15:25:01 -08001222 mem_cgroup_end_page_stat(memcg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223}
1224
Johannes Weiner8186eb62014-10-29 14:50:51 -07001225static void page_remove_file_rmap(struct page *page)
1226{
1227 struct mem_cgroup *memcg;
Johannes Weiner8186eb62014-10-29 14:50:51 -07001228
Johannes Weiner6de22612015-02-11 15:25:01 -08001229 memcg = mem_cgroup_begin_page_stat(page);
Johannes Weiner8186eb62014-10-29 14:50:51 -07001230
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08001231 /* Hugepages are not counted in NR_FILE_MAPPED for now. */
1232 if (unlikely(PageHuge(page))) {
1233 /* hugetlb pages are always mapped with pmds */
1234 atomic_dec(compound_mapcount_ptr(page));
1235 goto out;
1236 }
1237
Johannes Weiner8186eb62014-10-29 14:50:51 -07001238 /* page still mapped by someone else? */
1239 if (!atomic_add_negative(-1, &page->_mapcount))
1240 goto out;
1241
Johannes Weiner8186eb62014-10-29 14:50:51 -07001242 /*
1243 * We use the irq-unsafe __{inc|mod}_zone_page_stat because
1244 * these counters are not modified in interrupt context, and
1245 * pte lock(a spinlock) is held, which implies preemption disabled.
1246 */
1247 __dec_zone_page_state(page, NR_FILE_MAPPED);
1248 mem_cgroup_dec_page_stat(memcg, MEM_CGROUP_STAT_FILE_MAPPED);
1249
1250 if (unlikely(PageMlocked(page)))
1251 clear_page_mlock(page);
1252out:
Johannes Weiner6de22612015-02-11 15:25:01 -08001253 mem_cgroup_end_page_stat(memcg);
Johannes Weiner8186eb62014-10-29 14:50:51 -07001254}
1255
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08001256static void page_remove_anon_compound_rmap(struct page *page)
1257{
1258 int i, nr;
1259
1260 if (!atomic_add_negative(-1, compound_mapcount_ptr(page)))
1261 return;
1262
1263 /* Hugepages are not counted in NR_ANON_PAGES for now. */
1264 if (unlikely(PageHuge(page)))
1265 return;
1266
1267 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
1268 return;
1269
1270 __dec_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES);
1271
1272 if (TestClearPageDoubleMap(page)) {
1273 /*
1274 * Subpages can be mapped with PTEs too. Check how many of
1275 * themi are still mapped.
1276 */
1277 for (i = 0, nr = 0; i < HPAGE_PMD_NR; i++) {
1278 if (atomic_add_negative(-1, &page[i]._mapcount))
1279 nr++;
1280 }
1281 } else {
1282 nr = HPAGE_PMD_NR;
1283 }
1284
Kirill A. Shutemove90309c2016-01-15 16:54:33 -08001285 if (unlikely(PageMlocked(page)))
1286 clear_page_mlock(page);
1287
Kirill A. Shutemov9a982252016-01-15 16:54:17 -08001288 if (nr) {
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08001289 __mod_zone_page_state(page_zone(page), NR_ANON_PAGES, -nr);
Kirill A. Shutemov9a982252016-01-15 16:54:17 -08001290 deferred_split_huge_page(page);
1291 }
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08001292}
1293
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294/**
1295 * page_remove_rmap - take down pte mapping from a page
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08001296 * @page: page to remove mapping from
1297 * @compound: uncharge the page as compound or small page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -07001299 * The caller needs to hold the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 */
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08001301void page_remove_rmap(struct page *page, bool compound)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302{
Johannes Weiner8186eb62014-10-29 14:50:51 -07001303 if (!PageAnon(page)) {
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08001304 VM_BUG_ON_PAGE(compound && !PageHuge(page), page);
Johannes Weiner8186eb62014-10-29 14:50:51 -07001305 page_remove_file_rmap(page);
1306 return;
1307 }
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001308
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08001309 if (compound)
1310 return page_remove_anon_compound_rmap(page);
1311
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001312 /* page still mapped by someone else? */
1313 if (!atomic_add_negative(-1, &page->_mapcount))
Johannes Weiner8186eb62014-10-29 14:50:51 -07001314 return;
1315
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001316 /*
Jianyu Zhanbea04b02014-06-04 16:09:51 -07001317 * We use the irq-unsafe __{inc|mod}_zone_page_stat because
1318 * these counters are not modified in interrupt context, and
Jianyu Zhanbea04b02014-06-04 16:09:51 -07001319 * pte lock(a spinlock) is held, which implies preemption disabled.
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001320 */
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08001321 __dec_zone_page_state(page, NR_ANON_PAGES);
Johannes Weiner8186eb62014-10-29 14:50:51 -07001322
Hugh Dickinse6c509f2012-10-08 16:33:19 -07001323 if (unlikely(PageMlocked(page)))
1324 clear_page_mlock(page);
Johannes Weiner8186eb62014-10-29 14:50:51 -07001325
Kirill A. Shutemov9a982252016-01-15 16:54:17 -08001326 if (PageTransCompound(page))
1327 deferred_split_huge_page(compound_head(page));
1328
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -07001329 /*
1330 * It would be tidy to reset the PageAnon mapping here,
1331 * but that might overwrite a racing page_add_anon_rmap
1332 * which increments mapcount after us but sets mapping
1333 * before us: so leave the reset to free_hot_cold_page,
1334 * and remember that it's only reliable while mapped.
1335 * Leaving it set also helps swapoff to reinstate ptes
1336 * faster for those pages still in swapcache.
1337 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338}
1339
1340/*
Joonsoo Kim52629502014-01-21 15:49:50 -08001341 * @arg: enum ttu_flags will be passed to this argument
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 */
Kirill A. Shutemovac769502014-06-04 16:08:17 -07001343static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
Joonsoo Kim52629502014-01-21 15:49:50 -08001344 unsigned long address, void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345{
1346 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 pte_t *pte;
1348 pte_t pteval;
Hugh Dickinsc0718802005-10-29 18:16:31 -07001349 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 int ret = SWAP_AGAIN;
Joonsoo Kim52629502014-01-21 15:49:50 -08001351 enum ttu_flags flags = (enum ttu_flags)arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352
Hugh Dickinsb87537d9e2015-11-05 18:49:33 -08001353 /* munlock has nothing to gain from examining un-locked vmas */
1354 if ((flags & TTU_MUNLOCK) && !(vma->vm_flags & VM_LOCKED))
1355 goto out;
1356
Nick Piggin479db0b2008-08-20 14:09:18 -07001357 pte = page_check_address(page, mm, address, &ptl, 0);
Hugh Dickinsc0718802005-10-29 18:16:31 -07001358 if (!pte)
Nikita Danilov81b40822005-05-01 08:58:36 -07001359 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360
1361 /*
1362 * If the page is mlock()d, we cannot swap it out.
1363 * If it's recently referenced (perhaps page_referenced
1364 * skipped over this mm) then we should reactivate it.
1365 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001366 if (!(flags & TTU_IGNORE_MLOCK)) {
Hugh Dickinsb87537d9e2015-11-05 18:49:33 -08001367 if (vma->vm_flags & VM_LOCKED) {
1368 /* Holding pte lock, we do *not* need mmap_sem here */
1369 mlock_vma_page(page);
1370 ret = SWAP_MLOCK;
1371 goto out_unmap;
1372 }
Konstantin Khlebnikovdaa5ba72014-06-04 16:10:52 -07001373 if (flags & TTU_MUNLOCK)
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001374 goto out_unmap;
Andi Kleen14fa31b2009-09-16 11:50:10 +02001375 }
1376 if (!(flags & TTU_IGNORE_ACCESS)) {
Nick Pigginb291f002008-10-18 20:26:44 -07001377 if (ptep_clear_flush_young_notify(vma, address, pte)) {
1378 ret = SWAP_FAIL;
1379 goto out_unmap;
1380 }
1381 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 /* Nuke the page table entry. */
1384 flush_cache_page(vma, address, page_to_pfn(page));
Mel Gorman72b252a2015-09-04 15:47:32 -07001385 if (should_defer_flush(mm, flags)) {
1386 /*
1387 * We clear the PTE but do not flush so potentially a remote
1388 * CPU could still be writing to the page. If the entry was
1389 * previously clean then the architecture must guarantee that
1390 * a clear->dirty transition on a cached TLB entry is written
1391 * through and traps if the PTE is unmapped.
1392 */
1393 pteval = ptep_get_and_clear(mm, address, pte);
1394
Mel Gormand950c942015-09-04 15:47:35 -07001395 set_tlb_ubc_flush_pending(mm, page, pte_dirty(pteval));
Mel Gorman72b252a2015-09-04 15:47:32 -07001396 } else {
1397 pteval = ptep_clear_flush(vma, address, pte);
1398 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399
1400 /* Move the dirty bit to the physical page now the pte is gone. */
1401 if (pte_dirty(pteval))
1402 set_page_dirty(page);
1403
Hugh Dickins365e9c872005-10-29 18:16:18 -07001404 /* Update high watermark before we lower rss */
1405 update_hiwater_rss(mm);
1406
Andi Kleen888b9f72009-09-16 11:50:11 +02001407 if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) {
Naoya Horiguchi5d317b22015-11-05 18:47:14 -08001408 if (PageHuge(page)) {
1409 hugetlb_count_sub(1 << compound_order(page), mm);
1410 } else {
Jerome Marchandeca56ff2016-01-14 15:19:26 -08001411 dec_mm_counter(mm, mm_counter(page));
Naoya Horiguchi5f24ae52012-12-12 13:52:30 -08001412 }
Andi Kleen888b9f72009-09-16 11:50:11 +02001413 set_pte_at(mm, address, pte,
Naoya Horiguchi5f24ae52012-12-12 13:52:30 -08001414 swp_entry_to_pte(make_hwpoison_entry(page)));
Konstantin Weitz45961722013-04-17 13:59:32 +02001415 } else if (pte_unused(pteval)) {
1416 /*
1417 * The guest indicated that the page content is of no
1418 * interest anymore. Simply discard the pte, vmscan
1419 * will take care of the rest.
1420 */
Jerome Marchandeca56ff2016-01-14 15:19:26 -08001421 dec_mm_counter(mm, mm_counter(page));
Hugh Dickins470f1192015-11-05 18:49:59 -08001422 } else if (IS_ENABLED(CONFIG_MIGRATION) && (flags & TTU_MIGRATION)) {
1423 swp_entry_t entry;
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001424 pte_t swp_pte;
Hugh Dickins470f1192015-11-05 18:49:59 -08001425 /*
1426 * Store the pfn of the page in a special migration
1427 * pte. do_swap_page() will wait until the migration
1428 * pte is removed and then restart fault handling.
1429 */
1430 entry = make_migration_entry(page, pte_write(pteval));
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07001431 swp_pte = swp_entry_to_pte(entry);
1432 if (pte_soft_dirty(pteval))
1433 swp_pte = pte_swp_mksoft_dirty(swp_pte);
1434 set_pte_at(mm, address, pte, swp_pte);
Hugh Dickins470f1192015-11-05 18:49:59 -08001435 } else if (PageAnon(page)) {
1436 swp_entry_t entry = { .val = page_private(page) };
1437 pte_t swp_pte;
1438 /*
1439 * Store the swap location in the pte.
1440 * See handle_pte_fault() ...
1441 */
1442 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
1443 if (swap_duplicate(entry) < 0) {
1444 set_pte_at(mm, address, pte, pteval);
1445 ret = SWAP_FAIL;
1446 goto out_unmap;
1447 }
1448 if (list_empty(&mm->mmlist)) {
1449 spin_lock(&mmlist_lock);
1450 if (list_empty(&mm->mmlist))
1451 list_add(&mm->mmlist, &init_mm.mmlist);
1452 spin_unlock(&mmlist_lock);
1453 }
1454 dec_mm_counter(mm, MM_ANONPAGES);
1455 inc_mm_counter(mm, MM_SWAPENTS);
1456 swp_pte = swp_entry_to_pte(entry);
1457 if (pte_soft_dirty(pteval))
1458 swp_pte = pte_swp_mksoft_dirty(swp_pte);
1459 set_pte_at(mm, address, pte, swp_pte);
Christoph Lameter04e62a22006-06-23 02:03:38 -07001460 } else
Jerome Marchandeca56ff2016-01-14 15:19:26 -08001461 dec_mm_counter(mm, mm_counter_file(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08001463 page_remove_rmap(page, PageHuge(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 page_cache_release(page);
1465
1466out_unmap:
Hugh Dickinsc0718802005-10-29 18:16:31 -07001467 pte_unmap_unlock(pte, ptl);
Hugh Dickinsb87537d9e2015-11-05 18:49:33 -08001468 if (ret != SWAP_FAIL && ret != SWAP_MLOCK && !(flags & TTU_MUNLOCK))
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001469 mmu_notifier_invalidate_page(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470out:
1471 return ret;
1472}
1473
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001474bool is_vma_temporary_stack(struct vm_area_struct *vma)
Mel Gormana8bef8ff2010-05-24 14:32:24 -07001475{
1476 int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP);
1477
1478 if (!maybe_stack)
1479 return false;
1480
1481 if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) ==
1482 VM_STACK_INCOMPLETE_SETUP)
1483 return true;
1484
1485 return false;
1486}
1487
Joonsoo Kim52629502014-01-21 15:49:50 -08001488static bool invalid_migration_vma(struct vm_area_struct *vma, void *arg)
1489{
1490 return is_vma_temporary_stack(vma);
1491}
1492
Joonsoo Kim52629502014-01-21 15:49:50 -08001493static int page_not_mapped(struct page *page)
1494{
1495 return !page_mapped(page);
1496};
1497
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498/**
1499 * try_to_unmap - try to remove all page table mappings to a page
1500 * @page: the page to get unmapped
Andi Kleen14fa31b2009-09-16 11:50:10 +02001501 * @flags: action and flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 *
1503 * Tries to remove all the page table entries which are mapping this
1504 * page, used in the pageout path. Caller must hold the page lock.
1505 * Return values are:
1506 *
1507 * SWAP_SUCCESS - we succeeded in removing all mappings
1508 * SWAP_AGAIN - we missed a mapping, try again later
1509 * SWAP_FAIL - the page is unswappable
Nick Pigginb291f002008-10-18 20:26:44 -07001510 * SWAP_MLOCK - page is mlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001512int try_to_unmap(struct page *page, enum ttu_flags flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513{
1514 int ret;
Joonsoo Kim52629502014-01-21 15:49:50 -08001515 struct rmap_walk_control rwc = {
1516 .rmap_one = try_to_unmap_one,
1517 .arg = (void *)flags,
1518 .done = page_not_mapped,
Joonsoo Kim52629502014-01-21 15:49:50 -08001519 .anon_lock = page_lock_anon_vma_read,
1520 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521
Sasha Levin309381fea2014-01-23 15:52:54 -08001522 VM_BUG_ON_PAGE(!PageHuge(page) && PageTransHuge(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523
Joonsoo Kim52629502014-01-21 15:49:50 -08001524 /*
1525 * During exec, a temporary VMA is setup and later moved.
1526 * The VMA is moved under the anon_vma lock but not the
1527 * page tables leading to a race where migration cannot
1528 * find the migration ptes. Rather than increasing the
1529 * locking requirements of exec(), migration skips
1530 * temporary VMAs until after exec() completes.
1531 */
Konstantin Khlebnikovdaa5ba72014-06-04 16:10:52 -07001532 if ((flags & TTU_MIGRATION) && !PageKsm(page) && PageAnon(page))
Joonsoo Kim52629502014-01-21 15:49:50 -08001533 rwc.invalid_vma = invalid_migration_vma;
1534
1535 ret = rmap_walk(page, &rwc);
1536
Nick Pigginb291f002008-10-18 20:26:44 -07001537 if (ret != SWAP_MLOCK && !page_mapped(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 ret = SWAP_SUCCESS;
1539 return ret;
1540}
Nikita Danilov81b40822005-05-01 08:58:36 -07001541
Nick Pigginb291f002008-10-18 20:26:44 -07001542/**
1543 * try_to_munlock - try to munlock a page
1544 * @page: the page to be munlocked
1545 *
1546 * Called from munlock code. Checks all of the VMAs mapping the page
1547 * to make sure nobody else has this page mlocked. The page will be
1548 * returned with PG_mlocked cleared if no other vmas have it mlocked.
1549 *
1550 * Return values are:
1551 *
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001552 * SWAP_AGAIN - no vma is holding page mlocked, or,
Nick Pigginb291f002008-10-18 20:26:44 -07001553 * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem
Hugh Dickins5ad64682009-12-14 17:59:24 -08001554 * SWAP_FAIL - page cannot be located at present
Nick Pigginb291f002008-10-18 20:26:44 -07001555 * SWAP_MLOCK - page is now mlocked.
1556 */
1557int try_to_munlock(struct page *page)
1558{
Joonsoo Kime8351ac2014-01-21 15:49:52 -08001559 int ret;
1560 struct rmap_walk_control rwc = {
1561 .rmap_one = try_to_unmap_one,
1562 .arg = (void *)TTU_MUNLOCK,
1563 .done = page_not_mapped,
Joonsoo Kime8351ac2014-01-21 15:49:52 -08001564 .anon_lock = page_lock_anon_vma_read,
1565
1566 };
1567
Sasha Levin309381fea2014-01-23 15:52:54 -08001568 VM_BUG_ON_PAGE(!PageLocked(page) || PageLRU(page), page);
Nick Pigginb291f002008-10-18 20:26:44 -07001569
Joonsoo Kime8351ac2014-01-21 15:49:52 -08001570 ret = rmap_walk(page, &rwc);
1571 return ret;
Nick Pigginb291f002008-10-18 20:26:44 -07001572}
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001573
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001574void __put_anon_vma(struct anon_vma *anon_vma)
Rik van Riel76545062010-08-09 17:18:41 -07001575{
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001576 struct anon_vma *root = anon_vma->root;
Rik van Riel76545062010-08-09 17:18:41 -07001577
Andrey Ryabinin624483f2014-06-06 19:09:30 +04001578 anon_vma_free(anon_vma);
Peter Zijlstra01d8b202011-03-22 16:32:49 -07001579 if (root != anon_vma && atomic_dec_and_test(&root->refcount))
1580 anon_vma_free(root);
Rik van Riel76545062010-08-09 17:18:41 -07001581}
Rik van Riel76545062010-08-09 17:18:41 -07001582
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001583static struct anon_vma *rmap_walk_anon_lock(struct page *page,
1584 struct rmap_walk_control *rwc)
Joonsoo Kimfaecd8d2014-01-21 15:49:46 -08001585{
1586 struct anon_vma *anon_vma;
1587
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001588 if (rwc->anon_lock)
1589 return rwc->anon_lock(page);
1590
Joonsoo Kimfaecd8d2014-01-21 15:49:46 -08001591 /*
1592 * Note: remove_migration_ptes() cannot use page_lock_anon_vma_read()
1593 * because that depends on page_mapped(); but not all its usages
1594 * are holding mmap_sem. Users without mmap_sem are required to
1595 * take a reference count to prevent the anon_vma disappearing
1596 */
1597 anon_vma = page_anon_vma(page);
1598 if (!anon_vma)
1599 return NULL;
1600
1601 anon_vma_lock_read(anon_vma);
1602 return anon_vma;
1603}
1604
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001605/*
Joonsoo Kime8351ac2014-01-21 15:49:52 -08001606 * rmap_walk_anon - do something to anonymous page using the object-based
1607 * rmap method
1608 * @page: the page to be handled
1609 * @rwc: control variable according to each walk type
1610 *
1611 * Find all the mappings of a page using the mapping pointer and the vma chains
1612 * contained in the anon_vma struct it points to.
1613 *
1614 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1615 * where the page was found will be held for write. So, we won't recheck
1616 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1617 * LOCKED.
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001618 */
Joonsoo Kim051ac832014-01-21 15:49:48 -08001619static int rmap_walk_anon(struct page *page, struct rmap_walk_control *rwc)
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001620{
1621 struct anon_vma *anon_vma;
Davidlohr Buesob258d862014-12-12 16:55:04 -08001622 pgoff_t pgoff;
Rik van Riel5beb4932010-03-05 13:42:07 -08001623 struct anon_vma_chain *avc;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001624 int ret = SWAP_AGAIN;
1625
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001626 anon_vma = rmap_walk_anon_lock(page, rwc);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001627 if (!anon_vma)
1628 return ret;
Joonsoo Kimfaecd8d2014-01-21 15:49:46 -08001629
Davidlohr Buesob258d862014-12-12 16:55:04 -08001630 pgoff = page_to_pgoff(page);
Michel Lespinassebf181b92012-10-08 16:31:39 -07001631 anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) {
Rik van Riel5beb4932010-03-05 13:42:07 -08001632 struct vm_area_struct *vma = avc->vma;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001633 unsigned long address = vma_address(page, vma);
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001634
Andrea Arcangeliad126952015-11-05 18:49:07 -08001635 cond_resched();
1636
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001637 if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg))
1638 continue;
1639
Joonsoo Kim051ac832014-01-21 15:49:48 -08001640 ret = rwc->rmap_one(page, vma, address, rwc->arg);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001641 if (ret != SWAP_AGAIN)
1642 break;
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001643 if (rwc->done && rwc->done(page))
1644 break;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001645 }
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +00001646 anon_vma_unlock_read(anon_vma);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001647 return ret;
1648}
1649
Joonsoo Kime8351ac2014-01-21 15:49:52 -08001650/*
1651 * rmap_walk_file - do something to file page using the object-based rmap method
1652 * @page: the page to be handled
1653 * @rwc: control variable according to each walk type
1654 *
1655 * Find all the mappings of a page using the mapping pointer and the vma chains
1656 * contained in the address_space struct it points to.
1657 *
1658 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1659 * where the page was found will be held for write. So, we won't recheck
1660 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1661 * LOCKED.
1662 */
Joonsoo Kim051ac832014-01-21 15:49:48 -08001663static int rmap_walk_file(struct page *page, struct rmap_walk_control *rwc)
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001664{
1665 struct address_space *mapping = page->mapping;
Davidlohr Buesob258d862014-12-12 16:55:04 -08001666 pgoff_t pgoff;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001667 struct vm_area_struct *vma;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001668 int ret = SWAP_AGAIN;
1669
Joonsoo Kim9f326242014-01-21 15:49:53 -08001670 /*
1671 * The page lock not only makes sure that page->mapping cannot
1672 * suddenly be NULLified by truncation, it makes sure that the
1673 * structure at mapping cannot be freed and reused yet,
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -08001674 * so we can safely take mapping->i_mmap_rwsem.
Joonsoo Kim9f326242014-01-21 15:49:53 -08001675 */
Sasha Levin81d1b092014-10-09 15:28:10 -07001676 VM_BUG_ON_PAGE(!PageLocked(page), page);
Joonsoo Kim9f326242014-01-21 15:49:53 -08001677
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001678 if (!mapping)
1679 return ret;
Davidlohr Bueso3dec0ba2014-12-12 16:54:27 -08001680
Davidlohr Buesob258d862014-12-12 16:55:04 -08001681 pgoff = page_to_pgoff(page);
Davidlohr Bueso3dec0ba2014-12-12 16:54:27 -08001682 i_mmap_lock_read(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07001683 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001684 unsigned long address = vma_address(page, vma);
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001685
Andrea Arcangeliad126952015-11-05 18:49:07 -08001686 cond_resched();
1687
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001688 if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg))
1689 continue;
1690
Joonsoo Kim051ac832014-01-21 15:49:48 -08001691 ret = rwc->rmap_one(page, vma, address, rwc->arg);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001692 if (ret != SWAP_AGAIN)
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001693 goto done;
1694 if (rwc->done && rwc->done(page))
1695 goto done;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001696 }
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001697
Joonsoo Kim0dd1c7b2014-01-21 15:49:49 -08001698done:
Davidlohr Bueso3dec0ba2014-12-12 16:54:27 -08001699 i_mmap_unlock_read(mapping);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001700 return ret;
1701}
1702
Joonsoo Kim051ac832014-01-21 15:49:48 -08001703int rmap_walk(struct page *page, struct rmap_walk_control *rwc)
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001704{
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001705 if (unlikely(PageKsm(page)))
Joonsoo Kim051ac832014-01-21 15:49:48 -08001706 return rmap_walk_ksm(page, rwc);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001707 else if (PageAnon(page))
Joonsoo Kim051ac832014-01-21 15:49:48 -08001708 return rmap_walk_anon(page, rwc);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001709 else
Joonsoo Kim051ac832014-01-21 15:49:48 -08001710 return rmap_walk_file(page, rwc);
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001711}
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001712
Naoya Horiguchie3390f62010-06-15 13:18:13 +09001713#ifdef CONFIG_HUGETLB_PAGE
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001714/*
1715 * The following three functions are for anonymous (private mapped) hugepages.
1716 * Unlike common anonymous pages, anonymous hugepages have no accounting code
1717 * and no lru code, because we handle hugepages differently from common pages.
1718 */
1719static void __hugepage_set_anon_rmap(struct page *page,
1720 struct vm_area_struct *vma, unsigned long address, int exclusive)
1721{
1722 struct anon_vma *anon_vma = vma->anon_vma;
Naoya Horiguchi433abed2010-09-10 13:23:03 +09001723
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001724 BUG_ON(!anon_vma);
Naoya Horiguchi433abed2010-09-10 13:23:03 +09001725
1726 if (PageAnon(page))
1727 return;
1728 if (!exclusive)
1729 anon_vma = anon_vma->root;
1730
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001731 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
1732 page->mapping = (struct address_space *) anon_vma;
1733 page->index = linear_page_index(vma, address);
1734}
1735
1736void hugepage_add_anon_rmap(struct page *page,
1737 struct vm_area_struct *vma, unsigned long address)
1738{
1739 struct anon_vma *anon_vma = vma->anon_vma;
1740 int first;
Naoya Horiguchia850ea32010-09-10 13:23:06 +09001741
1742 BUG_ON(!PageLocked(page));
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001743 BUG_ON(!anon_vma);
Hugh Dickins5dbe0af2011-05-28 13:17:04 -07001744 /* address might be in next vma when migration races vma_adjust */
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08001745 first = atomic_inc_and_test(compound_mapcount_ptr(page));
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001746 if (first)
1747 __hugepage_set_anon_rmap(page, vma, address, 0);
1748}
1749
1750void hugepage_add_new_anon_rmap(struct page *page,
1751 struct vm_area_struct *vma, unsigned long address)
1752{
1753 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08001754 atomic_set(compound_mapcount_ptr(page), 0);
Naoya Horiguchi0fe6e202010-05-28 09:29:16 +09001755 __hugepage_set_anon_rmap(page, vma, address, 1);
1756}
Naoya Horiguchie3390f62010-06-15 13:18:13 +09001757#endif /* CONFIG_HUGETLB_PAGE */