Commit f915a1cc754e3b82f238b64c69ad153e9be012c4
Exists in
ti-linux-3.15.y
and in
1 other branch
Merge branch 'master' of http://git.kernel.org/pub/scm/linux/kernel/git/torvalds…
…/linux into ti-linux-3.15.y * 'master' of http://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux: mm: rmap: fix use-after-free in __put_anon_vma Signed-off-by: Dan Murphy <DMurphy@ti.com>
Showing 1 changed file Inline Diff
mm/rmap.c
1 | /* | 1 | /* |
2 | * mm/rmap.c - physical to virtual reverse mappings | 2 | * mm/rmap.c - physical to virtual reverse mappings |
3 | * | 3 | * |
4 | * Copyright 2001, Rik van Riel <riel@conectiva.com.br> | 4 | * Copyright 2001, Rik van Riel <riel@conectiva.com.br> |
5 | * Released under the General Public License (GPL). | 5 | * Released under the General Public License (GPL). |
6 | * | 6 | * |
7 | * Simple, low overhead reverse mapping scheme. | 7 | * Simple, low overhead reverse mapping scheme. |
8 | * Please try to keep this thing as modular as possible. | 8 | * Please try to keep this thing as modular as possible. |
9 | * | 9 | * |
10 | * Provides methods for unmapping each kind of mapped page: | 10 | * Provides methods for unmapping each kind of mapped page: |
11 | * the anon methods track anonymous pages, and | 11 | * the anon methods track anonymous pages, and |
12 | * the file methods track pages belonging to an inode. | 12 | * the file methods track pages belonging to an inode. |
13 | * | 13 | * |
14 | * Original design by Rik van Riel <riel@conectiva.com.br> 2001 | 14 | * Original design by Rik van Riel <riel@conectiva.com.br> 2001 |
15 | * File methods by Dave McCracken <dmccr@us.ibm.com> 2003, 2004 | 15 | * File methods by Dave McCracken <dmccr@us.ibm.com> 2003, 2004 |
16 | * Anonymous methods by Andrea Arcangeli <andrea@suse.de> 2004 | 16 | * Anonymous methods by Andrea Arcangeli <andrea@suse.de> 2004 |
17 | * Contributions by Hugh Dickins 2003, 2004 | 17 | * Contributions by Hugh Dickins 2003, 2004 |
18 | */ | 18 | */ |
19 | 19 | ||
20 | /* | 20 | /* |
21 | * Lock ordering in mm: | 21 | * Lock ordering in mm: |
22 | * | 22 | * |
23 | * inode->i_mutex (while writing or truncating, not reading or faulting) | 23 | * inode->i_mutex (while writing or truncating, not reading or faulting) |
24 | * mm->mmap_sem | 24 | * mm->mmap_sem |
25 | * page->flags PG_locked (lock_page) | 25 | * page->flags PG_locked (lock_page) |
26 | * mapping->i_mmap_mutex | 26 | * mapping->i_mmap_mutex |
27 | * anon_vma->rwsem | 27 | * anon_vma->rwsem |
28 | * mm->page_table_lock or pte_lock | 28 | * mm->page_table_lock or pte_lock |
29 | * zone->lru_lock (in mark_page_accessed, isolate_lru_page) | 29 | * zone->lru_lock (in mark_page_accessed, isolate_lru_page) |
30 | * swap_lock (in swap_duplicate, swap_info_get) | 30 | * swap_lock (in swap_duplicate, swap_info_get) |
31 | * mmlist_lock (in mmput, drain_mmlist and others) | 31 | * mmlist_lock (in mmput, drain_mmlist and others) |
32 | * mapping->private_lock (in __set_page_dirty_buffers) | 32 | * mapping->private_lock (in __set_page_dirty_buffers) |
33 | * inode->i_lock (in set_page_dirty's __mark_inode_dirty) | 33 | * inode->i_lock (in set_page_dirty's __mark_inode_dirty) |
34 | * bdi.wb->list_lock (in set_page_dirty's __mark_inode_dirty) | 34 | * bdi.wb->list_lock (in set_page_dirty's __mark_inode_dirty) |
35 | * sb_lock (within inode_lock in fs/fs-writeback.c) | 35 | * sb_lock (within inode_lock in fs/fs-writeback.c) |
36 | * mapping->tree_lock (widely used, in set_page_dirty, | 36 | * mapping->tree_lock (widely used, in set_page_dirty, |
37 | * in arch-dependent flush_dcache_mmap_lock, | 37 | * in arch-dependent flush_dcache_mmap_lock, |
38 | * within bdi.wb->list_lock in __sync_single_inode) | 38 | * within bdi.wb->list_lock in __sync_single_inode) |
39 | * | 39 | * |
40 | * anon_vma->rwsem,mapping->i_mutex (memory_failure, collect_procs_anon) | 40 | * anon_vma->rwsem,mapping->i_mutex (memory_failure, collect_procs_anon) |
41 | * ->tasklist_lock | 41 | * ->tasklist_lock |
42 | * pte map lock | 42 | * pte map lock |
43 | */ | 43 | */ |
44 | 44 | ||
45 | #include <linux/mm.h> | 45 | #include <linux/mm.h> |
46 | #include <linux/pagemap.h> | 46 | #include <linux/pagemap.h> |
47 | #include <linux/swap.h> | 47 | #include <linux/swap.h> |
48 | #include <linux/swapops.h> | 48 | #include <linux/swapops.h> |
49 | #include <linux/slab.h> | 49 | #include <linux/slab.h> |
50 | #include <linux/init.h> | 50 | #include <linux/init.h> |
51 | #include <linux/ksm.h> | 51 | #include <linux/ksm.h> |
52 | #include <linux/rmap.h> | 52 | #include <linux/rmap.h> |
53 | #include <linux/rcupdate.h> | 53 | #include <linux/rcupdate.h> |
54 | #include <linux/export.h> | 54 | #include <linux/export.h> |
55 | #include <linux/memcontrol.h> | 55 | #include <linux/memcontrol.h> |
56 | #include <linux/mmu_notifier.h> | 56 | #include <linux/mmu_notifier.h> |
57 | #include <linux/migrate.h> | 57 | #include <linux/migrate.h> |
58 | #include <linux/hugetlb.h> | 58 | #include <linux/hugetlb.h> |
59 | #include <linux/backing-dev.h> | 59 | #include <linux/backing-dev.h> |
60 | 60 | ||
61 | #include <asm/tlbflush.h> | 61 | #include <asm/tlbflush.h> |
62 | 62 | ||
63 | #include "internal.h" | 63 | #include "internal.h" |
64 | 64 | ||
65 | static struct kmem_cache *anon_vma_cachep; | 65 | static struct kmem_cache *anon_vma_cachep; |
66 | static struct kmem_cache *anon_vma_chain_cachep; | 66 | static struct kmem_cache *anon_vma_chain_cachep; |
67 | 67 | ||
68 | static inline struct anon_vma *anon_vma_alloc(void) | 68 | static inline struct anon_vma *anon_vma_alloc(void) |
69 | { | 69 | { |
70 | struct anon_vma *anon_vma; | 70 | struct anon_vma *anon_vma; |
71 | 71 | ||
72 | anon_vma = kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL); | 72 | anon_vma = kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL); |
73 | if (anon_vma) { | 73 | if (anon_vma) { |
74 | atomic_set(&anon_vma->refcount, 1); | 74 | atomic_set(&anon_vma->refcount, 1); |
75 | /* | 75 | /* |
76 | * Initialise the anon_vma root to point to itself. If called | 76 | * Initialise the anon_vma root to point to itself. If called |
77 | * from fork, the root will be reset to the parents anon_vma. | 77 | * from fork, the root will be reset to the parents anon_vma. |
78 | */ | 78 | */ |
79 | anon_vma->root = anon_vma; | 79 | anon_vma->root = anon_vma; |
80 | } | 80 | } |
81 | 81 | ||
82 | return anon_vma; | 82 | return anon_vma; |
83 | } | 83 | } |
84 | 84 | ||
85 | static inline void anon_vma_free(struct anon_vma *anon_vma) | 85 | static inline void anon_vma_free(struct anon_vma *anon_vma) |
86 | { | 86 | { |
87 | VM_BUG_ON(atomic_read(&anon_vma->refcount)); | 87 | VM_BUG_ON(atomic_read(&anon_vma->refcount)); |
88 | 88 | ||
89 | /* | 89 | /* |
90 | * Synchronize against page_lock_anon_vma_read() such that | 90 | * Synchronize against page_lock_anon_vma_read() such that |
91 | * we can safely hold the lock without the anon_vma getting | 91 | * we can safely hold the lock without the anon_vma getting |
92 | * freed. | 92 | * freed. |
93 | * | 93 | * |
94 | * Relies on the full mb implied by the atomic_dec_and_test() from | 94 | * Relies on the full mb implied by the atomic_dec_and_test() from |
95 | * put_anon_vma() against the acquire barrier implied by | 95 | * put_anon_vma() against the acquire barrier implied by |
96 | * down_read_trylock() from page_lock_anon_vma_read(). This orders: | 96 | * down_read_trylock() from page_lock_anon_vma_read(). This orders: |
97 | * | 97 | * |
98 | * page_lock_anon_vma_read() VS put_anon_vma() | 98 | * page_lock_anon_vma_read() VS put_anon_vma() |
99 | * down_read_trylock() atomic_dec_and_test() | 99 | * down_read_trylock() atomic_dec_and_test() |
100 | * LOCK MB | 100 | * LOCK MB |
101 | * atomic_read() rwsem_is_locked() | 101 | * atomic_read() rwsem_is_locked() |
102 | * | 102 | * |
103 | * LOCK should suffice since the actual taking of the lock must | 103 | * LOCK should suffice since the actual taking of the lock must |
104 | * happen _before_ what follows. | 104 | * happen _before_ what follows. |
105 | */ | 105 | */ |
106 | if (rwsem_is_locked(&anon_vma->root->rwsem)) { | 106 | if (rwsem_is_locked(&anon_vma->root->rwsem)) { |
107 | anon_vma_lock_write(anon_vma); | 107 | anon_vma_lock_write(anon_vma); |
108 | anon_vma_unlock_write(anon_vma); | 108 | anon_vma_unlock_write(anon_vma); |
109 | } | 109 | } |
110 | 110 | ||
111 | kmem_cache_free(anon_vma_cachep, anon_vma); | 111 | kmem_cache_free(anon_vma_cachep, anon_vma); |
112 | } | 112 | } |
113 | 113 | ||
114 | static inline struct anon_vma_chain *anon_vma_chain_alloc(gfp_t gfp) | 114 | static inline struct anon_vma_chain *anon_vma_chain_alloc(gfp_t gfp) |
115 | { | 115 | { |
116 | return kmem_cache_alloc(anon_vma_chain_cachep, gfp); | 116 | return kmem_cache_alloc(anon_vma_chain_cachep, gfp); |
117 | } | 117 | } |
118 | 118 | ||
119 | static void anon_vma_chain_free(struct anon_vma_chain *anon_vma_chain) | 119 | static void anon_vma_chain_free(struct anon_vma_chain *anon_vma_chain) |
120 | { | 120 | { |
121 | kmem_cache_free(anon_vma_chain_cachep, anon_vma_chain); | 121 | kmem_cache_free(anon_vma_chain_cachep, anon_vma_chain); |
122 | } | 122 | } |
123 | 123 | ||
124 | static void anon_vma_chain_link(struct vm_area_struct *vma, | 124 | static void anon_vma_chain_link(struct vm_area_struct *vma, |
125 | struct anon_vma_chain *avc, | 125 | struct anon_vma_chain *avc, |
126 | struct anon_vma *anon_vma) | 126 | struct anon_vma *anon_vma) |
127 | { | 127 | { |
128 | avc->vma = vma; | 128 | avc->vma = vma; |
129 | avc->anon_vma = anon_vma; | 129 | avc->anon_vma = anon_vma; |
130 | list_add(&avc->same_vma, &vma->anon_vma_chain); | 130 | list_add(&avc->same_vma, &vma->anon_vma_chain); |
131 | anon_vma_interval_tree_insert(avc, &anon_vma->rb_root); | 131 | anon_vma_interval_tree_insert(avc, &anon_vma->rb_root); |
132 | } | 132 | } |
133 | 133 | ||
134 | /** | 134 | /** |
135 | * anon_vma_prepare - attach an anon_vma to a memory region | 135 | * anon_vma_prepare - attach an anon_vma to a memory region |
136 | * @vma: the memory region in question | 136 | * @vma: the memory region in question |
137 | * | 137 | * |
138 | * This makes sure the memory mapping described by 'vma' has | 138 | * This makes sure the memory mapping described by 'vma' has |
139 | * an 'anon_vma' attached to it, so that we can associate the | 139 | * an 'anon_vma' attached to it, so that we can associate the |
140 | * anonymous pages mapped into it with that anon_vma. | 140 | * anonymous pages mapped into it with that anon_vma. |
141 | * | 141 | * |
142 | * The common case will be that we already have one, but if | 142 | * The common case will be that we already have one, but if |
143 | * not we either need to find an adjacent mapping that we | 143 | * not we either need to find an adjacent mapping that we |
144 | * can re-use the anon_vma from (very common when the only | 144 | * can re-use the anon_vma from (very common when the only |
145 | * reason for splitting a vma has been mprotect()), or we | 145 | * reason for splitting a vma has been mprotect()), or we |
146 | * allocate a new one. | 146 | * allocate a new one. |
147 | * | 147 | * |
148 | * Anon-vma allocations are very subtle, because we may have | 148 | * Anon-vma allocations are very subtle, because we may have |
149 | * optimistically looked up an anon_vma in page_lock_anon_vma_read() | 149 | * optimistically looked up an anon_vma in page_lock_anon_vma_read() |
150 | * and that may actually touch the spinlock even in the newly | 150 | * and that may actually touch the spinlock even in the newly |
151 | * allocated vma (it depends on RCU to make sure that the | 151 | * allocated vma (it depends on RCU to make sure that the |
152 | * anon_vma isn't actually destroyed). | 152 | * anon_vma isn't actually destroyed). |
153 | * | 153 | * |
154 | * As a result, we need to do proper anon_vma locking even | 154 | * As a result, we need to do proper anon_vma locking even |
155 | * for the new allocation. At the same time, we do not want | 155 | * for the new allocation. At the same time, we do not want |
156 | * to do any locking for the common case of already having | 156 | * to do any locking for the common case of already having |
157 | * an anon_vma. | 157 | * an anon_vma. |
158 | * | 158 | * |
159 | * This must be called with the mmap_sem held for reading. | 159 | * This must be called with the mmap_sem held for reading. |
160 | */ | 160 | */ |
161 | int anon_vma_prepare(struct vm_area_struct *vma) | 161 | int anon_vma_prepare(struct vm_area_struct *vma) |
162 | { | 162 | { |
163 | struct anon_vma *anon_vma = vma->anon_vma; | 163 | struct anon_vma *anon_vma = vma->anon_vma; |
164 | struct anon_vma_chain *avc; | 164 | struct anon_vma_chain *avc; |
165 | 165 | ||
166 | might_sleep(); | 166 | might_sleep(); |
167 | if (unlikely(!anon_vma)) { | 167 | if (unlikely(!anon_vma)) { |
168 | struct mm_struct *mm = vma->vm_mm; | 168 | struct mm_struct *mm = vma->vm_mm; |
169 | struct anon_vma *allocated; | 169 | struct anon_vma *allocated; |
170 | 170 | ||
171 | avc = anon_vma_chain_alloc(GFP_KERNEL); | 171 | avc = anon_vma_chain_alloc(GFP_KERNEL); |
172 | if (!avc) | 172 | if (!avc) |
173 | goto out_enomem; | 173 | goto out_enomem; |
174 | 174 | ||
175 | anon_vma = find_mergeable_anon_vma(vma); | 175 | anon_vma = find_mergeable_anon_vma(vma); |
176 | allocated = NULL; | 176 | allocated = NULL; |
177 | if (!anon_vma) { | 177 | if (!anon_vma) { |
178 | anon_vma = anon_vma_alloc(); | 178 | anon_vma = anon_vma_alloc(); |
179 | if (unlikely(!anon_vma)) | 179 | if (unlikely(!anon_vma)) |
180 | goto out_enomem_free_avc; | 180 | goto out_enomem_free_avc; |
181 | allocated = anon_vma; | 181 | allocated = anon_vma; |
182 | } | 182 | } |
183 | 183 | ||
184 | anon_vma_lock_write(anon_vma); | 184 | anon_vma_lock_write(anon_vma); |
185 | /* page_table_lock to protect against threads */ | 185 | /* page_table_lock to protect against threads */ |
186 | spin_lock(&mm->page_table_lock); | 186 | spin_lock(&mm->page_table_lock); |
187 | if (likely(!vma->anon_vma)) { | 187 | if (likely(!vma->anon_vma)) { |
188 | vma->anon_vma = anon_vma; | 188 | vma->anon_vma = anon_vma; |
189 | anon_vma_chain_link(vma, avc, anon_vma); | 189 | anon_vma_chain_link(vma, avc, anon_vma); |
190 | allocated = NULL; | 190 | allocated = NULL; |
191 | avc = NULL; | 191 | avc = NULL; |
192 | } | 192 | } |
193 | spin_unlock(&mm->page_table_lock); | 193 | spin_unlock(&mm->page_table_lock); |
194 | anon_vma_unlock_write(anon_vma); | 194 | anon_vma_unlock_write(anon_vma); |
195 | 195 | ||
196 | if (unlikely(allocated)) | 196 | if (unlikely(allocated)) |
197 | put_anon_vma(allocated); | 197 | put_anon_vma(allocated); |
198 | if (unlikely(avc)) | 198 | if (unlikely(avc)) |
199 | anon_vma_chain_free(avc); | 199 | anon_vma_chain_free(avc); |
200 | } | 200 | } |
201 | return 0; | 201 | return 0; |
202 | 202 | ||
203 | out_enomem_free_avc: | 203 | out_enomem_free_avc: |
204 | anon_vma_chain_free(avc); | 204 | anon_vma_chain_free(avc); |
205 | out_enomem: | 205 | out_enomem: |
206 | return -ENOMEM; | 206 | return -ENOMEM; |
207 | } | 207 | } |
208 | 208 | ||
209 | /* | 209 | /* |
210 | * This is a useful helper function for locking the anon_vma root as | 210 | * This is a useful helper function for locking the anon_vma root as |
211 | * we traverse the vma->anon_vma_chain, looping over anon_vma's that | 211 | * we traverse the vma->anon_vma_chain, looping over anon_vma's that |
212 | * have the same vma. | 212 | * have the same vma. |
213 | * | 213 | * |
214 | * Such anon_vma's should have the same root, so you'd expect to see | 214 | * Such anon_vma's should have the same root, so you'd expect to see |
215 | * just a single mutex_lock for the whole traversal. | 215 | * just a single mutex_lock for the whole traversal. |
216 | */ | 216 | */ |
217 | static inline struct anon_vma *lock_anon_vma_root(struct anon_vma *root, struct anon_vma *anon_vma) | 217 | static inline struct anon_vma *lock_anon_vma_root(struct anon_vma *root, struct anon_vma *anon_vma) |
218 | { | 218 | { |
219 | struct anon_vma *new_root = anon_vma->root; | 219 | struct anon_vma *new_root = anon_vma->root; |
220 | if (new_root != root) { | 220 | if (new_root != root) { |
221 | if (WARN_ON_ONCE(root)) | 221 | if (WARN_ON_ONCE(root)) |
222 | up_write(&root->rwsem); | 222 | up_write(&root->rwsem); |
223 | root = new_root; | 223 | root = new_root; |
224 | down_write(&root->rwsem); | 224 | down_write(&root->rwsem); |
225 | } | 225 | } |
226 | return root; | 226 | return root; |
227 | } | 227 | } |
228 | 228 | ||
229 | static inline void unlock_anon_vma_root(struct anon_vma *root) | 229 | static inline void unlock_anon_vma_root(struct anon_vma *root) |
230 | { | 230 | { |
231 | if (root) | 231 | if (root) |
232 | up_write(&root->rwsem); | 232 | up_write(&root->rwsem); |
233 | } | 233 | } |
234 | 234 | ||
235 | /* | 235 | /* |
236 | * Attach the anon_vmas from src to dst. | 236 | * Attach the anon_vmas from src to dst. |
237 | * Returns 0 on success, -ENOMEM on failure. | 237 | * Returns 0 on success, -ENOMEM on failure. |
238 | */ | 238 | */ |
239 | int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src) | 239 | int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src) |
240 | { | 240 | { |
241 | struct anon_vma_chain *avc, *pavc; | 241 | struct anon_vma_chain *avc, *pavc; |
242 | struct anon_vma *root = NULL; | 242 | struct anon_vma *root = NULL; |
243 | 243 | ||
244 | list_for_each_entry_reverse(pavc, &src->anon_vma_chain, same_vma) { | 244 | list_for_each_entry_reverse(pavc, &src->anon_vma_chain, same_vma) { |
245 | struct anon_vma *anon_vma; | 245 | struct anon_vma *anon_vma; |
246 | 246 | ||
247 | avc = anon_vma_chain_alloc(GFP_NOWAIT | __GFP_NOWARN); | 247 | avc = anon_vma_chain_alloc(GFP_NOWAIT | __GFP_NOWARN); |
248 | if (unlikely(!avc)) { | 248 | if (unlikely(!avc)) { |
249 | unlock_anon_vma_root(root); | 249 | unlock_anon_vma_root(root); |
250 | root = NULL; | 250 | root = NULL; |
251 | avc = anon_vma_chain_alloc(GFP_KERNEL); | 251 | avc = anon_vma_chain_alloc(GFP_KERNEL); |
252 | if (!avc) | 252 | if (!avc) |
253 | goto enomem_failure; | 253 | goto enomem_failure; |
254 | } | 254 | } |
255 | anon_vma = pavc->anon_vma; | 255 | anon_vma = pavc->anon_vma; |
256 | root = lock_anon_vma_root(root, anon_vma); | 256 | root = lock_anon_vma_root(root, anon_vma); |
257 | anon_vma_chain_link(dst, avc, anon_vma); | 257 | anon_vma_chain_link(dst, avc, anon_vma); |
258 | } | 258 | } |
259 | unlock_anon_vma_root(root); | 259 | unlock_anon_vma_root(root); |
260 | return 0; | 260 | return 0; |
261 | 261 | ||
262 | enomem_failure: | 262 | enomem_failure: |
263 | unlink_anon_vmas(dst); | 263 | unlink_anon_vmas(dst); |
264 | return -ENOMEM; | 264 | return -ENOMEM; |
265 | } | 265 | } |
266 | 266 | ||
267 | /* | 267 | /* |
268 | * Attach vma to its own anon_vma, as well as to the anon_vmas that | 268 | * Attach vma to its own anon_vma, as well as to the anon_vmas that |
269 | * the corresponding VMA in the parent process is attached to. | 269 | * the corresponding VMA in the parent process is attached to. |
270 | * Returns 0 on success, non-zero on failure. | 270 | * Returns 0 on success, non-zero on failure. |
271 | */ | 271 | */ |
272 | int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma) | 272 | int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma) |
273 | { | 273 | { |
274 | struct anon_vma_chain *avc; | 274 | struct anon_vma_chain *avc; |
275 | struct anon_vma *anon_vma; | 275 | struct anon_vma *anon_vma; |
276 | 276 | ||
277 | /* Don't bother if the parent process has no anon_vma here. */ | 277 | /* Don't bother if the parent process has no anon_vma here. */ |
278 | if (!pvma->anon_vma) | 278 | if (!pvma->anon_vma) |
279 | return 0; | 279 | return 0; |
280 | 280 | ||
281 | /* | 281 | /* |
282 | * First, attach the new VMA to the parent VMA's anon_vmas, | 282 | * First, attach the new VMA to the parent VMA's anon_vmas, |
283 | * so rmap can find non-COWed pages in child processes. | 283 | * so rmap can find non-COWed pages in child processes. |
284 | */ | 284 | */ |
285 | if (anon_vma_clone(vma, pvma)) | 285 | if (anon_vma_clone(vma, pvma)) |
286 | return -ENOMEM; | 286 | return -ENOMEM; |
287 | 287 | ||
288 | /* Then add our own anon_vma. */ | 288 | /* Then add our own anon_vma. */ |
289 | anon_vma = anon_vma_alloc(); | 289 | anon_vma = anon_vma_alloc(); |
290 | if (!anon_vma) | 290 | if (!anon_vma) |
291 | goto out_error; | 291 | goto out_error; |
292 | avc = anon_vma_chain_alloc(GFP_KERNEL); | 292 | avc = anon_vma_chain_alloc(GFP_KERNEL); |
293 | if (!avc) | 293 | if (!avc) |
294 | goto out_error_free_anon_vma; | 294 | goto out_error_free_anon_vma; |
295 | 295 | ||
296 | /* | 296 | /* |
297 | * The root anon_vma's spinlock is the lock actually used when we | 297 | * The root anon_vma's spinlock is the lock actually used when we |
298 | * lock any of the anon_vmas in this anon_vma tree. | 298 | * lock any of the anon_vmas in this anon_vma tree. |
299 | */ | 299 | */ |
300 | anon_vma->root = pvma->anon_vma->root; | 300 | anon_vma->root = pvma->anon_vma->root; |
301 | /* | 301 | /* |
302 | * With refcounts, an anon_vma can stay around longer than the | 302 | * With refcounts, an anon_vma can stay around longer than the |
303 | * process it belongs to. The root anon_vma needs to be pinned until | 303 | * process it belongs to. The root anon_vma needs to be pinned until |
304 | * this anon_vma is freed, because the lock lives in the root. | 304 | * this anon_vma is freed, because the lock lives in the root. |
305 | */ | 305 | */ |
306 | get_anon_vma(anon_vma->root); | 306 | get_anon_vma(anon_vma->root); |
307 | /* Mark this anon_vma as the one where our new (COWed) pages go. */ | 307 | /* Mark this anon_vma as the one where our new (COWed) pages go. */ |
308 | vma->anon_vma = anon_vma; | 308 | vma->anon_vma = anon_vma; |
309 | anon_vma_lock_write(anon_vma); | 309 | anon_vma_lock_write(anon_vma); |
310 | anon_vma_chain_link(vma, avc, anon_vma); | 310 | anon_vma_chain_link(vma, avc, anon_vma); |
311 | anon_vma_unlock_write(anon_vma); | 311 | anon_vma_unlock_write(anon_vma); |
312 | 312 | ||
313 | return 0; | 313 | return 0; |
314 | 314 | ||
315 | out_error_free_anon_vma: | 315 | out_error_free_anon_vma: |
316 | put_anon_vma(anon_vma); | 316 | put_anon_vma(anon_vma); |
317 | out_error: | 317 | out_error: |
318 | unlink_anon_vmas(vma); | 318 | unlink_anon_vmas(vma); |
319 | return -ENOMEM; | 319 | return -ENOMEM; |
320 | } | 320 | } |
321 | 321 | ||
322 | void unlink_anon_vmas(struct vm_area_struct *vma) | 322 | void unlink_anon_vmas(struct vm_area_struct *vma) |
323 | { | 323 | { |
324 | struct anon_vma_chain *avc, *next; | 324 | struct anon_vma_chain *avc, *next; |
325 | struct anon_vma *root = NULL; | 325 | struct anon_vma *root = NULL; |
326 | 326 | ||
327 | /* | 327 | /* |
328 | * Unlink each anon_vma chained to the VMA. This list is ordered | 328 | * Unlink each anon_vma chained to the VMA. This list is ordered |
329 | * from newest to oldest, ensuring the root anon_vma gets freed last. | 329 | * from newest to oldest, ensuring the root anon_vma gets freed last. |
330 | */ | 330 | */ |
331 | list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) { | 331 | list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) { |
332 | struct anon_vma *anon_vma = avc->anon_vma; | 332 | struct anon_vma *anon_vma = avc->anon_vma; |
333 | 333 | ||
334 | root = lock_anon_vma_root(root, anon_vma); | 334 | root = lock_anon_vma_root(root, anon_vma); |
335 | anon_vma_interval_tree_remove(avc, &anon_vma->rb_root); | 335 | anon_vma_interval_tree_remove(avc, &anon_vma->rb_root); |
336 | 336 | ||
337 | /* | 337 | /* |
338 | * Leave empty anon_vmas on the list - we'll need | 338 | * Leave empty anon_vmas on the list - we'll need |
339 | * to free them outside the lock. | 339 | * to free them outside the lock. |
340 | */ | 340 | */ |
341 | if (RB_EMPTY_ROOT(&anon_vma->rb_root)) | 341 | if (RB_EMPTY_ROOT(&anon_vma->rb_root)) |
342 | continue; | 342 | continue; |
343 | 343 | ||
344 | list_del(&avc->same_vma); | 344 | list_del(&avc->same_vma); |
345 | anon_vma_chain_free(avc); | 345 | anon_vma_chain_free(avc); |
346 | } | 346 | } |
347 | unlock_anon_vma_root(root); | 347 | unlock_anon_vma_root(root); |
348 | 348 | ||
349 | /* | 349 | /* |
350 | * Iterate the list once more, it now only contains empty and unlinked | 350 | * Iterate the list once more, it now only contains empty and unlinked |
351 | * anon_vmas, destroy them. Could not do before due to __put_anon_vma() | 351 | * anon_vmas, destroy them. Could not do before due to __put_anon_vma() |
352 | * needing to write-acquire the anon_vma->root->rwsem. | 352 | * needing to write-acquire the anon_vma->root->rwsem. |
353 | */ | 353 | */ |
354 | list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) { | 354 | list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) { |
355 | struct anon_vma *anon_vma = avc->anon_vma; | 355 | struct anon_vma *anon_vma = avc->anon_vma; |
356 | 356 | ||
357 | put_anon_vma(anon_vma); | 357 | put_anon_vma(anon_vma); |
358 | 358 | ||
359 | list_del(&avc->same_vma); | 359 | list_del(&avc->same_vma); |
360 | anon_vma_chain_free(avc); | 360 | anon_vma_chain_free(avc); |
361 | } | 361 | } |
362 | } | 362 | } |
363 | 363 | ||
364 | static void anon_vma_ctor(void *data) | 364 | static void anon_vma_ctor(void *data) |
365 | { | 365 | { |
366 | struct anon_vma *anon_vma = data; | 366 | struct anon_vma *anon_vma = data; |
367 | 367 | ||
368 | init_rwsem(&anon_vma->rwsem); | 368 | init_rwsem(&anon_vma->rwsem); |
369 | atomic_set(&anon_vma->refcount, 0); | 369 | atomic_set(&anon_vma->refcount, 0); |
370 | anon_vma->rb_root = RB_ROOT; | 370 | anon_vma->rb_root = RB_ROOT; |
371 | } | 371 | } |
372 | 372 | ||
373 | void __init anon_vma_init(void) | 373 | void __init anon_vma_init(void) |
374 | { | 374 | { |
375 | anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma), | 375 | anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma), |
376 | 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor); | 376 | 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor); |
377 | anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain, SLAB_PANIC); | 377 | anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain, SLAB_PANIC); |
378 | } | 378 | } |
379 | 379 | ||
380 | /* | 380 | /* |
381 | * Getting a lock on a stable anon_vma from a page off the LRU is tricky! | 381 | * Getting a lock on a stable anon_vma from a page off the LRU is tricky! |
382 | * | 382 | * |
383 | * Since there is no serialization what so ever against page_remove_rmap() | 383 | * Since there is no serialization what so ever against page_remove_rmap() |
384 | * the best this function can do is return a locked anon_vma that might | 384 | * the best this function can do is return a locked anon_vma that might |
385 | * have been relevant to this page. | 385 | * have been relevant to this page. |
386 | * | 386 | * |
387 | * The page might have been remapped to a different anon_vma or the anon_vma | 387 | * The page might have been remapped to a different anon_vma or the anon_vma |
388 | * returned may already be freed (and even reused). | 388 | * returned may already be freed (and even reused). |
389 | * | 389 | * |
390 | * In case it was remapped to a different anon_vma, the new anon_vma will be a | 390 | * In case it was remapped to a different anon_vma, the new anon_vma will be a |
391 | * child of the old anon_vma, and the anon_vma lifetime rules will therefore | 391 | * child of the old anon_vma, and the anon_vma lifetime rules will therefore |
392 | * ensure that any anon_vma obtained from the page will still be valid for as | 392 | * ensure that any anon_vma obtained from the page will still be valid for as |
393 | * long as we observe page_mapped() [ hence all those page_mapped() tests ]. | 393 | * long as we observe page_mapped() [ hence all those page_mapped() tests ]. |
394 | * | 394 | * |
395 | * All users of this function must be very careful when walking the anon_vma | 395 | * All users of this function must be very careful when walking the anon_vma |
396 | * chain and verify that the page in question is indeed mapped in it | 396 | * chain and verify that the page in question is indeed mapped in it |
397 | * [ something equivalent to page_mapped_in_vma() ]. | 397 | * [ something equivalent to page_mapped_in_vma() ]. |
398 | * | 398 | * |
399 | * Since anon_vma's slab is DESTROY_BY_RCU and we know from page_remove_rmap() | 399 | * Since anon_vma's slab is DESTROY_BY_RCU and we know from page_remove_rmap() |
400 | * that the anon_vma pointer from page->mapping is valid if there is a | 400 | * that the anon_vma pointer from page->mapping is valid if there is a |
401 | * mapcount, we can dereference the anon_vma after observing those. | 401 | * mapcount, we can dereference the anon_vma after observing those. |
402 | */ | 402 | */ |
403 | struct anon_vma *page_get_anon_vma(struct page *page) | 403 | struct anon_vma *page_get_anon_vma(struct page *page) |
404 | { | 404 | { |
405 | struct anon_vma *anon_vma = NULL; | 405 | struct anon_vma *anon_vma = NULL; |
406 | unsigned long anon_mapping; | 406 | unsigned long anon_mapping; |
407 | 407 | ||
408 | rcu_read_lock(); | 408 | rcu_read_lock(); |
409 | anon_mapping = (unsigned long) ACCESS_ONCE(page->mapping); | 409 | anon_mapping = (unsigned long) ACCESS_ONCE(page->mapping); |
410 | if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON) | 410 | if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON) |
411 | goto out; | 411 | goto out; |
412 | if (!page_mapped(page)) | 412 | if (!page_mapped(page)) |
413 | goto out; | 413 | goto out; |
414 | 414 | ||
415 | anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON); | 415 | anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON); |
416 | if (!atomic_inc_not_zero(&anon_vma->refcount)) { | 416 | if (!atomic_inc_not_zero(&anon_vma->refcount)) { |
417 | anon_vma = NULL; | 417 | anon_vma = NULL; |
418 | goto out; | 418 | goto out; |
419 | } | 419 | } |
420 | 420 | ||
421 | /* | 421 | /* |
422 | * If this page is still mapped, then its anon_vma cannot have been | 422 | * If this page is still mapped, then its anon_vma cannot have been |
423 | * freed. But if it has been unmapped, we have no security against the | 423 | * freed. But if it has been unmapped, we have no security against the |
424 | * anon_vma structure being freed and reused (for another anon_vma: | 424 | * anon_vma structure being freed and reused (for another anon_vma: |
425 | * SLAB_DESTROY_BY_RCU guarantees that - so the atomic_inc_not_zero() | 425 | * SLAB_DESTROY_BY_RCU guarantees that - so the atomic_inc_not_zero() |
426 | * above cannot corrupt). | 426 | * above cannot corrupt). |
427 | */ | 427 | */ |
428 | if (!page_mapped(page)) { | 428 | if (!page_mapped(page)) { |
429 | put_anon_vma(anon_vma); | 429 | put_anon_vma(anon_vma); |
430 | anon_vma = NULL; | 430 | anon_vma = NULL; |
431 | } | 431 | } |
432 | out: | 432 | out: |
433 | rcu_read_unlock(); | 433 | rcu_read_unlock(); |
434 | 434 | ||
435 | return anon_vma; | 435 | return anon_vma; |
436 | } | 436 | } |
437 | 437 | ||
438 | /* | 438 | /* |
439 | * Similar to page_get_anon_vma() except it locks the anon_vma. | 439 | * Similar to page_get_anon_vma() except it locks the anon_vma. |
440 | * | 440 | * |
441 | * Its a little more complex as it tries to keep the fast path to a single | 441 | * Its a little more complex as it tries to keep the fast path to a single |
442 | * atomic op -- the trylock. If we fail the trylock, we fall back to getting a | 442 | * atomic op -- the trylock. If we fail the trylock, we fall back to getting a |
443 | * reference like with page_get_anon_vma() and then block on the mutex. | 443 | * reference like with page_get_anon_vma() and then block on the mutex. |
444 | */ | 444 | */ |
445 | struct anon_vma *page_lock_anon_vma_read(struct page *page) | 445 | struct anon_vma *page_lock_anon_vma_read(struct page *page) |
446 | { | 446 | { |
447 | struct anon_vma *anon_vma = NULL; | 447 | struct anon_vma *anon_vma = NULL; |
448 | struct anon_vma *root_anon_vma; | 448 | struct anon_vma *root_anon_vma; |
449 | unsigned long anon_mapping; | 449 | unsigned long anon_mapping; |
450 | 450 | ||
451 | rcu_read_lock(); | 451 | rcu_read_lock(); |
452 | anon_mapping = (unsigned long) ACCESS_ONCE(page->mapping); | 452 | anon_mapping = (unsigned long) ACCESS_ONCE(page->mapping); |
453 | if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON) | 453 | if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON) |
454 | goto out; | 454 | goto out; |
455 | if (!page_mapped(page)) | 455 | if (!page_mapped(page)) |
456 | goto out; | 456 | goto out; |
457 | 457 | ||
458 | anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON); | 458 | anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON); |
459 | root_anon_vma = ACCESS_ONCE(anon_vma->root); | 459 | root_anon_vma = ACCESS_ONCE(anon_vma->root); |
460 | if (down_read_trylock(&root_anon_vma->rwsem)) { | 460 | if (down_read_trylock(&root_anon_vma->rwsem)) { |
461 | /* | 461 | /* |
462 | * If the page is still mapped, then this anon_vma is still | 462 | * If the page is still mapped, then this anon_vma is still |
463 | * its anon_vma, and holding the mutex ensures that it will | 463 | * its anon_vma, and holding the mutex ensures that it will |
464 | * not go away, see anon_vma_free(). | 464 | * not go away, see anon_vma_free(). |
465 | */ | 465 | */ |
466 | if (!page_mapped(page)) { | 466 | if (!page_mapped(page)) { |
467 | up_read(&root_anon_vma->rwsem); | 467 | up_read(&root_anon_vma->rwsem); |
468 | anon_vma = NULL; | 468 | anon_vma = NULL; |
469 | } | 469 | } |
470 | goto out; | 470 | goto out; |
471 | } | 471 | } |
472 | 472 | ||
473 | /* trylock failed, we got to sleep */ | 473 | /* trylock failed, we got to sleep */ |
474 | if (!atomic_inc_not_zero(&anon_vma->refcount)) { | 474 | if (!atomic_inc_not_zero(&anon_vma->refcount)) { |
475 | anon_vma = NULL; | 475 | anon_vma = NULL; |
476 | goto out; | 476 | goto out; |
477 | } | 477 | } |
478 | 478 | ||
479 | if (!page_mapped(page)) { | 479 | if (!page_mapped(page)) { |
480 | put_anon_vma(anon_vma); | 480 | put_anon_vma(anon_vma); |
481 | anon_vma = NULL; | 481 | anon_vma = NULL; |
482 | goto out; | 482 | goto out; |
483 | } | 483 | } |
484 | 484 | ||
485 | /* we pinned the anon_vma, its safe to sleep */ | 485 | /* we pinned the anon_vma, its safe to sleep */ |
486 | rcu_read_unlock(); | 486 | rcu_read_unlock(); |
487 | anon_vma_lock_read(anon_vma); | 487 | anon_vma_lock_read(anon_vma); |
488 | 488 | ||
489 | if (atomic_dec_and_test(&anon_vma->refcount)) { | 489 | if (atomic_dec_and_test(&anon_vma->refcount)) { |
490 | /* | 490 | /* |
491 | * Oops, we held the last refcount, release the lock | 491 | * Oops, we held the last refcount, release the lock |
492 | * and bail -- can't simply use put_anon_vma() because | 492 | * and bail -- can't simply use put_anon_vma() because |
493 | * we'll deadlock on the anon_vma_lock_write() recursion. | 493 | * we'll deadlock on the anon_vma_lock_write() recursion. |
494 | */ | 494 | */ |
495 | anon_vma_unlock_read(anon_vma); | 495 | anon_vma_unlock_read(anon_vma); |
496 | __put_anon_vma(anon_vma); | 496 | __put_anon_vma(anon_vma); |
497 | anon_vma = NULL; | 497 | anon_vma = NULL; |
498 | } | 498 | } |
499 | 499 | ||
500 | return anon_vma; | 500 | return anon_vma; |
501 | 501 | ||
502 | out: | 502 | out: |
503 | rcu_read_unlock(); | 503 | rcu_read_unlock(); |
504 | return anon_vma; | 504 | return anon_vma; |
505 | } | 505 | } |
506 | 506 | ||
507 | void page_unlock_anon_vma_read(struct anon_vma *anon_vma) | 507 | void page_unlock_anon_vma_read(struct anon_vma *anon_vma) |
508 | { | 508 | { |
509 | anon_vma_unlock_read(anon_vma); | 509 | anon_vma_unlock_read(anon_vma); |
510 | } | 510 | } |
511 | 511 | ||
512 | /* | 512 | /* |
513 | * At what user virtual address is page expected in @vma? | 513 | * At what user virtual address is page expected in @vma? |
514 | */ | 514 | */ |
515 | static inline unsigned long | 515 | static inline unsigned long |
516 | __vma_address(struct page *page, struct vm_area_struct *vma) | 516 | __vma_address(struct page *page, struct vm_area_struct *vma) |
517 | { | 517 | { |
518 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | 518 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); |
519 | 519 | ||
520 | if (unlikely(is_vm_hugetlb_page(vma))) | 520 | if (unlikely(is_vm_hugetlb_page(vma))) |
521 | pgoff = page->index << huge_page_order(page_hstate(page)); | 521 | pgoff = page->index << huge_page_order(page_hstate(page)); |
522 | 522 | ||
523 | return vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); | 523 | return vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); |
524 | } | 524 | } |
525 | 525 | ||
526 | inline unsigned long | 526 | inline unsigned long |
527 | vma_address(struct page *page, struct vm_area_struct *vma) | 527 | vma_address(struct page *page, struct vm_area_struct *vma) |
528 | { | 528 | { |
529 | unsigned long address = __vma_address(page, vma); | 529 | unsigned long address = __vma_address(page, vma); |
530 | 530 | ||
531 | /* page should be within @vma mapping range */ | 531 | /* page should be within @vma mapping range */ |
532 | VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); | 532 | VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); |
533 | 533 | ||
534 | return address; | 534 | return address; |
535 | } | 535 | } |
536 | 536 | ||
537 | /* | 537 | /* |
538 | * At what user virtual address is page expected in vma? | 538 | * At what user virtual address is page expected in vma? |
539 | * Caller should check the page is actually part of the vma. | 539 | * Caller should check the page is actually part of the vma. |
540 | */ | 540 | */ |
541 | unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma) | 541 | unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma) |
542 | { | 542 | { |
543 | unsigned long address; | 543 | unsigned long address; |
544 | if (PageAnon(page)) { | 544 | if (PageAnon(page)) { |
545 | struct anon_vma *page__anon_vma = page_anon_vma(page); | 545 | struct anon_vma *page__anon_vma = page_anon_vma(page); |
546 | /* | 546 | /* |
547 | * Note: swapoff's unuse_vma() is more efficient with this | 547 | * Note: swapoff's unuse_vma() is more efficient with this |
548 | * check, and needs it to match anon_vma when KSM is active. | 548 | * check, and needs it to match anon_vma when KSM is active. |
549 | */ | 549 | */ |
550 | if (!vma->anon_vma || !page__anon_vma || | 550 | if (!vma->anon_vma || !page__anon_vma || |
551 | vma->anon_vma->root != page__anon_vma->root) | 551 | vma->anon_vma->root != page__anon_vma->root) |
552 | return -EFAULT; | 552 | return -EFAULT; |
553 | } else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) { | 553 | } else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) { |
554 | if (!vma->vm_file || | 554 | if (!vma->vm_file || |
555 | vma->vm_file->f_mapping != page->mapping) | 555 | vma->vm_file->f_mapping != page->mapping) |
556 | return -EFAULT; | 556 | return -EFAULT; |
557 | } else | 557 | } else |
558 | return -EFAULT; | 558 | return -EFAULT; |
559 | address = __vma_address(page, vma); | 559 | address = __vma_address(page, vma); |
560 | if (unlikely(address < vma->vm_start || address >= vma->vm_end)) | 560 | if (unlikely(address < vma->vm_start || address >= vma->vm_end)) |
561 | return -EFAULT; | 561 | return -EFAULT; |
562 | return address; | 562 | return address; |
563 | } | 563 | } |
564 | 564 | ||
565 | pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address) | 565 | pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address) |
566 | { | 566 | { |
567 | pgd_t *pgd; | 567 | pgd_t *pgd; |
568 | pud_t *pud; | 568 | pud_t *pud; |
569 | pmd_t *pmd = NULL; | 569 | pmd_t *pmd = NULL; |
570 | 570 | ||
571 | pgd = pgd_offset(mm, address); | 571 | pgd = pgd_offset(mm, address); |
572 | if (!pgd_present(*pgd)) | 572 | if (!pgd_present(*pgd)) |
573 | goto out; | 573 | goto out; |
574 | 574 | ||
575 | pud = pud_offset(pgd, address); | 575 | pud = pud_offset(pgd, address); |
576 | if (!pud_present(*pud)) | 576 | if (!pud_present(*pud)) |
577 | goto out; | 577 | goto out; |
578 | 578 | ||
579 | pmd = pmd_offset(pud, address); | 579 | pmd = pmd_offset(pud, address); |
580 | if (!pmd_present(*pmd)) | 580 | if (!pmd_present(*pmd)) |
581 | pmd = NULL; | 581 | pmd = NULL; |
582 | out: | 582 | out: |
583 | return pmd; | 583 | return pmd; |
584 | } | 584 | } |
585 | 585 | ||
586 | /* | 586 | /* |
587 | * Check that @page is mapped at @address into @mm. | 587 | * Check that @page is mapped at @address into @mm. |
588 | * | 588 | * |
589 | * If @sync is false, page_check_address may perform a racy check to avoid | 589 | * If @sync is false, page_check_address may perform a racy check to avoid |
590 | * the page table lock when the pte is not present (helpful when reclaiming | 590 | * the page table lock when the pte is not present (helpful when reclaiming |
591 | * highly shared pages). | 591 | * highly shared pages). |
592 | * | 592 | * |
593 | * On success returns with pte mapped and locked. | 593 | * On success returns with pte mapped and locked. |
594 | */ | 594 | */ |
595 | pte_t *__page_check_address(struct page *page, struct mm_struct *mm, | 595 | pte_t *__page_check_address(struct page *page, struct mm_struct *mm, |
596 | unsigned long address, spinlock_t **ptlp, int sync) | 596 | unsigned long address, spinlock_t **ptlp, int sync) |
597 | { | 597 | { |
598 | pmd_t *pmd; | 598 | pmd_t *pmd; |
599 | pte_t *pte; | 599 | pte_t *pte; |
600 | spinlock_t *ptl; | 600 | spinlock_t *ptl; |
601 | 601 | ||
602 | if (unlikely(PageHuge(page))) { | 602 | if (unlikely(PageHuge(page))) { |
603 | /* when pud is not present, pte will be NULL */ | 603 | /* when pud is not present, pte will be NULL */ |
604 | pte = huge_pte_offset(mm, address); | 604 | pte = huge_pte_offset(mm, address); |
605 | if (!pte) | 605 | if (!pte) |
606 | return NULL; | 606 | return NULL; |
607 | 607 | ||
608 | ptl = huge_pte_lockptr(page_hstate(page), mm, pte); | 608 | ptl = huge_pte_lockptr(page_hstate(page), mm, pte); |
609 | goto check; | 609 | goto check; |
610 | } | 610 | } |
611 | 611 | ||
612 | pmd = mm_find_pmd(mm, address); | 612 | pmd = mm_find_pmd(mm, address); |
613 | if (!pmd) | 613 | if (!pmd) |
614 | return NULL; | 614 | return NULL; |
615 | 615 | ||
616 | if (pmd_trans_huge(*pmd)) | 616 | if (pmd_trans_huge(*pmd)) |
617 | return NULL; | 617 | return NULL; |
618 | 618 | ||
619 | pte = pte_offset_map(pmd, address); | 619 | pte = pte_offset_map(pmd, address); |
620 | /* Make a quick check before getting the lock */ | 620 | /* Make a quick check before getting the lock */ |
621 | if (!sync && !pte_present(*pte)) { | 621 | if (!sync && !pte_present(*pte)) { |
622 | pte_unmap(pte); | 622 | pte_unmap(pte); |
623 | return NULL; | 623 | return NULL; |
624 | } | 624 | } |
625 | 625 | ||
626 | ptl = pte_lockptr(mm, pmd); | 626 | ptl = pte_lockptr(mm, pmd); |
627 | check: | 627 | check: |
628 | spin_lock(ptl); | 628 | spin_lock(ptl); |
629 | if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) { | 629 | if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) { |
630 | *ptlp = ptl; | 630 | *ptlp = ptl; |
631 | return pte; | 631 | return pte; |
632 | } | 632 | } |
633 | pte_unmap_unlock(pte, ptl); | 633 | pte_unmap_unlock(pte, ptl); |
634 | return NULL; | 634 | return NULL; |
635 | } | 635 | } |
636 | 636 | ||
637 | /** | 637 | /** |
638 | * page_mapped_in_vma - check whether a page is really mapped in a VMA | 638 | * page_mapped_in_vma - check whether a page is really mapped in a VMA |
639 | * @page: the page to test | 639 | * @page: the page to test |
640 | * @vma: the VMA to test | 640 | * @vma: the VMA to test |
641 | * | 641 | * |
642 | * Returns 1 if the page is mapped into the page tables of the VMA, 0 | 642 | * Returns 1 if the page is mapped into the page tables of the VMA, 0 |
643 | * if the page is not mapped into the page tables of this VMA. Only | 643 | * if the page is not mapped into the page tables of this VMA. Only |
644 | * valid for normal file or anonymous VMAs. | 644 | * valid for normal file or anonymous VMAs. |
645 | */ | 645 | */ |
646 | int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma) | 646 | int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma) |
647 | { | 647 | { |
648 | unsigned long address; | 648 | unsigned long address; |
649 | pte_t *pte; | 649 | pte_t *pte; |
650 | spinlock_t *ptl; | 650 | spinlock_t *ptl; |
651 | 651 | ||
652 | address = __vma_address(page, vma); | 652 | address = __vma_address(page, vma); |
653 | if (unlikely(address < vma->vm_start || address >= vma->vm_end)) | 653 | if (unlikely(address < vma->vm_start || address >= vma->vm_end)) |
654 | return 0; | 654 | return 0; |
655 | pte = page_check_address(page, vma->vm_mm, address, &ptl, 1); | 655 | pte = page_check_address(page, vma->vm_mm, address, &ptl, 1); |
656 | if (!pte) /* the page is not in this mm */ | 656 | if (!pte) /* the page is not in this mm */ |
657 | return 0; | 657 | return 0; |
658 | pte_unmap_unlock(pte, ptl); | 658 | pte_unmap_unlock(pte, ptl); |
659 | 659 | ||
660 | return 1; | 660 | return 1; |
661 | } | 661 | } |
662 | 662 | ||
663 | struct page_referenced_arg { | 663 | struct page_referenced_arg { |
664 | int mapcount; | 664 | int mapcount; |
665 | int referenced; | 665 | int referenced; |
666 | unsigned long vm_flags; | 666 | unsigned long vm_flags; |
667 | struct mem_cgroup *memcg; | 667 | struct mem_cgroup *memcg; |
668 | }; | 668 | }; |
669 | /* | 669 | /* |
670 | * arg: page_referenced_arg will be passed | 670 | * arg: page_referenced_arg will be passed |
671 | */ | 671 | */ |
672 | int page_referenced_one(struct page *page, struct vm_area_struct *vma, | 672 | int page_referenced_one(struct page *page, struct vm_area_struct *vma, |
673 | unsigned long address, void *arg) | 673 | unsigned long address, void *arg) |
674 | { | 674 | { |
675 | struct mm_struct *mm = vma->vm_mm; | 675 | struct mm_struct *mm = vma->vm_mm; |
676 | spinlock_t *ptl; | 676 | spinlock_t *ptl; |
677 | int referenced = 0; | 677 | int referenced = 0; |
678 | struct page_referenced_arg *pra = arg; | 678 | struct page_referenced_arg *pra = arg; |
679 | 679 | ||
680 | if (unlikely(PageTransHuge(page))) { | 680 | if (unlikely(PageTransHuge(page))) { |
681 | pmd_t *pmd; | 681 | pmd_t *pmd; |
682 | 682 | ||
683 | /* | 683 | /* |
684 | * rmap might return false positives; we must filter | 684 | * rmap might return false positives; we must filter |
685 | * these out using page_check_address_pmd(). | 685 | * these out using page_check_address_pmd(). |
686 | */ | 686 | */ |
687 | pmd = page_check_address_pmd(page, mm, address, | 687 | pmd = page_check_address_pmd(page, mm, address, |
688 | PAGE_CHECK_ADDRESS_PMD_FLAG, &ptl); | 688 | PAGE_CHECK_ADDRESS_PMD_FLAG, &ptl); |
689 | if (!pmd) | 689 | if (!pmd) |
690 | return SWAP_AGAIN; | 690 | return SWAP_AGAIN; |
691 | 691 | ||
692 | if (vma->vm_flags & VM_LOCKED) { | 692 | if (vma->vm_flags & VM_LOCKED) { |
693 | spin_unlock(ptl); | 693 | spin_unlock(ptl); |
694 | pra->vm_flags |= VM_LOCKED; | 694 | pra->vm_flags |= VM_LOCKED; |
695 | return SWAP_FAIL; /* To break the loop */ | 695 | return SWAP_FAIL; /* To break the loop */ |
696 | } | 696 | } |
697 | 697 | ||
698 | /* go ahead even if the pmd is pmd_trans_splitting() */ | 698 | /* go ahead even if the pmd is pmd_trans_splitting() */ |
699 | if (pmdp_clear_flush_young_notify(vma, address, pmd)) | 699 | if (pmdp_clear_flush_young_notify(vma, address, pmd)) |
700 | referenced++; | 700 | referenced++; |
701 | spin_unlock(ptl); | 701 | spin_unlock(ptl); |
702 | } else { | 702 | } else { |
703 | pte_t *pte; | 703 | pte_t *pte; |
704 | 704 | ||
705 | /* | 705 | /* |
706 | * rmap might return false positives; we must filter | 706 | * rmap might return false positives; we must filter |
707 | * these out using page_check_address(). | 707 | * these out using page_check_address(). |
708 | */ | 708 | */ |
709 | pte = page_check_address(page, mm, address, &ptl, 0); | 709 | pte = page_check_address(page, mm, address, &ptl, 0); |
710 | if (!pte) | 710 | if (!pte) |
711 | return SWAP_AGAIN; | 711 | return SWAP_AGAIN; |
712 | 712 | ||
713 | if (vma->vm_flags & VM_LOCKED) { | 713 | if (vma->vm_flags & VM_LOCKED) { |
714 | pte_unmap_unlock(pte, ptl); | 714 | pte_unmap_unlock(pte, ptl); |
715 | pra->vm_flags |= VM_LOCKED; | 715 | pra->vm_flags |= VM_LOCKED; |
716 | return SWAP_FAIL; /* To break the loop */ | 716 | return SWAP_FAIL; /* To break the loop */ |
717 | } | 717 | } |
718 | 718 | ||
719 | if (ptep_clear_flush_young_notify(vma, address, pte)) { | 719 | if (ptep_clear_flush_young_notify(vma, address, pte)) { |
720 | /* | 720 | /* |
721 | * Don't treat a reference through a sequentially read | 721 | * Don't treat a reference through a sequentially read |
722 | * mapping as such. If the page has been used in | 722 | * mapping as such. If the page has been used in |
723 | * another mapping, we will catch it; if this other | 723 | * another mapping, we will catch it; if this other |
724 | * mapping is already gone, the unmap path will have | 724 | * mapping is already gone, the unmap path will have |
725 | * set PG_referenced or activated the page. | 725 | * set PG_referenced or activated the page. |
726 | */ | 726 | */ |
727 | if (likely(!(vma->vm_flags & VM_SEQ_READ))) | 727 | if (likely(!(vma->vm_flags & VM_SEQ_READ))) |
728 | referenced++; | 728 | referenced++; |
729 | } | 729 | } |
730 | pte_unmap_unlock(pte, ptl); | 730 | pte_unmap_unlock(pte, ptl); |
731 | } | 731 | } |
732 | 732 | ||
733 | if (referenced) { | 733 | if (referenced) { |
734 | pra->referenced++; | 734 | pra->referenced++; |
735 | pra->vm_flags |= vma->vm_flags; | 735 | pra->vm_flags |= vma->vm_flags; |
736 | } | 736 | } |
737 | 737 | ||
738 | pra->mapcount--; | 738 | pra->mapcount--; |
739 | if (!pra->mapcount) | 739 | if (!pra->mapcount) |
740 | return SWAP_SUCCESS; /* To break the loop */ | 740 | return SWAP_SUCCESS; /* To break the loop */ |
741 | 741 | ||
742 | return SWAP_AGAIN; | 742 | return SWAP_AGAIN; |
743 | } | 743 | } |
744 | 744 | ||
745 | static bool invalid_page_referenced_vma(struct vm_area_struct *vma, void *arg) | 745 | static bool invalid_page_referenced_vma(struct vm_area_struct *vma, void *arg) |
746 | { | 746 | { |
747 | struct page_referenced_arg *pra = arg; | 747 | struct page_referenced_arg *pra = arg; |
748 | struct mem_cgroup *memcg = pra->memcg; | 748 | struct mem_cgroup *memcg = pra->memcg; |
749 | 749 | ||
750 | if (!mm_match_cgroup(vma->vm_mm, memcg)) | 750 | if (!mm_match_cgroup(vma->vm_mm, memcg)) |
751 | return true; | 751 | return true; |
752 | 752 | ||
753 | return false; | 753 | return false; |
754 | } | 754 | } |
755 | 755 | ||
756 | /** | 756 | /** |
757 | * page_referenced - test if the page was referenced | 757 | * page_referenced - test if the page was referenced |
758 | * @page: the page to test | 758 | * @page: the page to test |
759 | * @is_locked: caller holds lock on the page | 759 | * @is_locked: caller holds lock on the page |
760 | * @memcg: target memory cgroup | 760 | * @memcg: target memory cgroup |
761 | * @vm_flags: collect encountered vma->vm_flags who actually referenced the page | 761 | * @vm_flags: collect encountered vma->vm_flags who actually referenced the page |
762 | * | 762 | * |
763 | * Quick test_and_clear_referenced for all mappings to a page, | 763 | * Quick test_and_clear_referenced for all mappings to a page, |
764 | * returns the number of ptes which referenced the page. | 764 | * returns the number of ptes which referenced the page. |
765 | */ | 765 | */ |
766 | int page_referenced(struct page *page, | 766 | int page_referenced(struct page *page, |
767 | int is_locked, | 767 | int is_locked, |
768 | struct mem_cgroup *memcg, | 768 | struct mem_cgroup *memcg, |
769 | unsigned long *vm_flags) | 769 | unsigned long *vm_flags) |
770 | { | 770 | { |
771 | int ret; | 771 | int ret; |
772 | int we_locked = 0; | 772 | int we_locked = 0; |
773 | struct page_referenced_arg pra = { | 773 | struct page_referenced_arg pra = { |
774 | .mapcount = page_mapcount(page), | 774 | .mapcount = page_mapcount(page), |
775 | .memcg = memcg, | 775 | .memcg = memcg, |
776 | }; | 776 | }; |
777 | struct rmap_walk_control rwc = { | 777 | struct rmap_walk_control rwc = { |
778 | .rmap_one = page_referenced_one, | 778 | .rmap_one = page_referenced_one, |
779 | .arg = (void *)&pra, | 779 | .arg = (void *)&pra, |
780 | .anon_lock = page_lock_anon_vma_read, | 780 | .anon_lock = page_lock_anon_vma_read, |
781 | }; | 781 | }; |
782 | 782 | ||
783 | *vm_flags = 0; | 783 | *vm_flags = 0; |
784 | if (!page_mapped(page)) | 784 | if (!page_mapped(page)) |
785 | return 0; | 785 | return 0; |
786 | 786 | ||
787 | if (!page_rmapping(page)) | 787 | if (!page_rmapping(page)) |
788 | return 0; | 788 | return 0; |
789 | 789 | ||
790 | if (!is_locked && (!PageAnon(page) || PageKsm(page))) { | 790 | if (!is_locked && (!PageAnon(page) || PageKsm(page))) { |
791 | we_locked = trylock_page(page); | 791 | we_locked = trylock_page(page); |
792 | if (!we_locked) | 792 | if (!we_locked) |
793 | return 1; | 793 | return 1; |
794 | } | 794 | } |
795 | 795 | ||
796 | /* | 796 | /* |
797 | * If we are reclaiming on behalf of a cgroup, skip | 797 | * If we are reclaiming on behalf of a cgroup, skip |
798 | * counting on behalf of references from different | 798 | * counting on behalf of references from different |
799 | * cgroups | 799 | * cgroups |
800 | */ | 800 | */ |
801 | if (memcg) { | 801 | if (memcg) { |
802 | rwc.invalid_vma = invalid_page_referenced_vma; | 802 | rwc.invalid_vma = invalid_page_referenced_vma; |
803 | } | 803 | } |
804 | 804 | ||
805 | ret = rmap_walk(page, &rwc); | 805 | ret = rmap_walk(page, &rwc); |
806 | *vm_flags = pra.vm_flags; | 806 | *vm_flags = pra.vm_flags; |
807 | 807 | ||
808 | if (we_locked) | 808 | if (we_locked) |
809 | unlock_page(page); | 809 | unlock_page(page); |
810 | 810 | ||
811 | return pra.referenced; | 811 | return pra.referenced; |
812 | } | 812 | } |
813 | 813 | ||
814 | static int page_mkclean_one(struct page *page, struct vm_area_struct *vma, | 814 | static int page_mkclean_one(struct page *page, struct vm_area_struct *vma, |
815 | unsigned long address, void *arg) | 815 | unsigned long address, void *arg) |
816 | { | 816 | { |
817 | struct mm_struct *mm = vma->vm_mm; | 817 | struct mm_struct *mm = vma->vm_mm; |
818 | pte_t *pte; | 818 | pte_t *pte; |
819 | spinlock_t *ptl; | 819 | spinlock_t *ptl; |
820 | int ret = 0; | 820 | int ret = 0; |
821 | int *cleaned = arg; | 821 | int *cleaned = arg; |
822 | 822 | ||
823 | pte = page_check_address(page, mm, address, &ptl, 1); | 823 | pte = page_check_address(page, mm, address, &ptl, 1); |
824 | if (!pte) | 824 | if (!pte) |
825 | goto out; | 825 | goto out; |
826 | 826 | ||
827 | if (pte_dirty(*pte) || pte_write(*pte)) { | 827 | if (pte_dirty(*pte) || pte_write(*pte)) { |
828 | pte_t entry; | 828 | pte_t entry; |
829 | 829 | ||
830 | flush_cache_page(vma, address, pte_pfn(*pte)); | 830 | flush_cache_page(vma, address, pte_pfn(*pte)); |
831 | entry = ptep_clear_flush(vma, address, pte); | 831 | entry = ptep_clear_flush(vma, address, pte); |
832 | entry = pte_wrprotect(entry); | 832 | entry = pte_wrprotect(entry); |
833 | entry = pte_mkclean(entry); | 833 | entry = pte_mkclean(entry); |
834 | set_pte_at(mm, address, pte, entry); | 834 | set_pte_at(mm, address, pte, entry); |
835 | ret = 1; | 835 | ret = 1; |
836 | } | 836 | } |
837 | 837 | ||
838 | pte_unmap_unlock(pte, ptl); | 838 | pte_unmap_unlock(pte, ptl); |
839 | 839 | ||
840 | if (ret) { | 840 | if (ret) { |
841 | mmu_notifier_invalidate_page(mm, address); | 841 | mmu_notifier_invalidate_page(mm, address); |
842 | (*cleaned)++; | 842 | (*cleaned)++; |
843 | } | 843 | } |
844 | out: | 844 | out: |
845 | return SWAP_AGAIN; | 845 | return SWAP_AGAIN; |
846 | } | 846 | } |
847 | 847 | ||
848 | static bool invalid_mkclean_vma(struct vm_area_struct *vma, void *arg) | 848 | static bool invalid_mkclean_vma(struct vm_area_struct *vma, void *arg) |
849 | { | 849 | { |
850 | if (vma->vm_flags & VM_SHARED) | 850 | if (vma->vm_flags & VM_SHARED) |
851 | return false; | 851 | return false; |
852 | 852 | ||
853 | return true; | 853 | return true; |
854 | } | 854 | } |
855 | 855 | ||
856 | int page_mkclean(struct page *page) | 856 | int page_mkclean(struct page *page) |
857 | { | 857 | { |
858 | int cleaned = 0; | 858 | int cleaned = 0; |
859 | struct address_space *mapping; | 859 | struct address_space *mapping; |
860 | struct rmap_walk_control rwc = { | 860 | struct rmap_walk_control rwc = { |
861 | .arg = (void *)&cleaned, | 861 | .arg = (void *)&cleaned, |
862 | .rmap_one = page_mkclean_one, | 862 | .rmap_one = page_mkclean_one, |
863 | .invalid_vma = invalid_mkclean_vma, | 863 | .invalid_vma = invalid_mkclean_vma, |
864 | }; | 864 | }; |
865 | 865 | ||
866 | BUG_ON(!PageLocked(page)); | 866 | BUG_ON(!PageLocked(page)); |
867 | 867 | ||
868 | if (!page_mapped(page)) | 868 | if (!page_mapped(page)) |
869 | return 0; | 869 | return 0; |
870 | 870 | ||
871 | mapping = page_mapping(page); | 871 | mapping = page_mapping(page); |
872 | if (!mapping) | 872 | if (!mapping) |
873 | return 0; | 873 | return 0; |
874 | 874 | ||
875 | rmap_walk(page, &rwc); | 875 | rmap_walk(page, &rwc); |
876 | 876 | ||
877 | return cleaned; | 877 | return cleaned; |
878 | } | 878 | } |
879 | EXPORT_SYMBOL_GPL(page_mkclean); | 879 | EXPORT_SYMBOL_GPL(page_mkclean); |
880 | 880 | ||
881 | /** | 881 | /** |
882 | * page_move_anon_rmap - move a page to our anon_vma | 882 | * page_move_anon_rmap - move a page to our anon_vma |
883 | * @page: the page to move to our anon_vma | 883 | * @page: the page to move to our anon_vma |
884 | * @vma: the vma the page belongs to | 884 | * @vma: the vma the page belongs to |
885 | * @address: the user virtual address mapped | 885 | * @address: the user virtual address mapped |
886 | * | 886 | * |
887 | * When a page belongs exclusively to one process after a COW event, | 887 | * When a page belongs exclusively to one process after a COW event, |
888 | * that page can be moved into the anon_vma that belongs to just that | 888 | * that page can be moved into the anon_vma that belongs to just that |
889 | * process, so the rmap code will not search the parent or sibling | 889 | * process, so the rmap code will not search the parent or sibling |
890 | * processes. | 890 | * processes. |
891 | */ | 891 | */ |
892 | void page_move_anon_rmap(struct page *page, | 892 | void page_move_anon_rmap(struct page *page, |
893 | struct vm_area_struct *vma, unsigned long address) | 893 | struct vm_area_struct *vma, unsigned long address) |
894 | { | 894 | { |
895 | struct anon_vma *anon_vma = vma->anon_vma; | 895 | struct anon_vma *anon_vma = vma->anon_vma; |
896 | 896 | ||
897 | VM_BUG_ON_PAGE(!PageLocked(page), page); | 897 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
898 | VM_BUG_ON(!anon_vma); | 898 | VM_BUG_ON(!anon_vma); |
899 | VM_BUG_ON_PAGE(page->index != linear_page_index(vma, address), page); | 899 | VM_BUG_ON_PAGE(page->index != linear_page_index(vma, address), page); |
900 | 900 | ||
901 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; | 901 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; |
902 | page->mapping = (struct address_space *) anon_vma; | 902 | page->mapping = (struct address_space *) anon_vma; |
903 | } | 903 | } |
904 | 904 | ||
905 | /** | 905 | /** |
906 | * __page_set_anon_rmap - set up new anonymous rmap | 906 | * __page_set_anon_rmap - set up new anonymous rmap |
907 | * @page: Page to add to rmap | 907 | * @page: Page to add to rmap |
908 | * @vma: VM area to add page to. | 908 | * @vma: VM area to add page to. |
909 | * @address: User virtual address of the mapping | 909 | * @address: User virtual address of the mapping |
910 | * @exclusive: the page is exclusively owned by the current process | 910 | * @exclusive: the page is exclusively owned by the current process |
911 | */ | 911 | */ |
912 | static void __page_set_anon_rmap(struct page *page, | 912 | static void __page_set_anon_rmap(struct page *page, |
913 | struct vm_area_struct *vma, unsigned long address, int exclusive) | 913 | struct vm_area_struct *vma, unsigned long address, int exclusive) |
914 | { | 914 | { |
915 | struct anon_vma *anon_vma = vma->anon_vma; | 915 | struct anon_vma *anon_vma = vma->anon_vma; |
916 | 916 | ||
917 | BUG_ON(!anon_vma); | 917 | BUG_ON(!anon_vma); |
918 | 918 | ||
919 | if (PageAnon(page)) | 919 | if (PageAnon(page)) |
920 | return; | 920 | return; |
921 | 921 | ||
922 | /* | 922 | /* |
923 | * If the page isn't exclusively mapped into this vma, | 923 | * If the page isn't exclusively mapped into this vma, |
924 | * we must use the _oldest_ possible anon_vma for the | 924 | * we must use the _oldest_ possible anon_vma for the |
925 | * page mapping! | 925 | * page mapping! |
926 | */ | 926 | */ |
927 | if (!exclusive) | 927 | if (!exclusive) |
928 | anon_vma = anon_vma->root; | 928 | anon_vma = anon_vma->root; |
929 | 929 | ||
930 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; | 930 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; |
931 | page->mapping = (struct address_space *) anon_vma; | 931 | page->mapping = (struct address_space *) anon_vma; |
932 | page->index = linear_page_index(vma, address); | 932 | page->index = linear_page_index(vma, address); |
933 | } | 933 | } |
934 | 934 | ||
935 | /** | 935 | /** |
936 | * __page_check_anon_rmap - sanity check anonymous rmap addition | 936 | * __page_check_anon_rmap - sanity check anonymous rmap addition |
937 | * @page: the page to add the mapping to | 937 | * @page: the page to add the mapping to |
938 | * @vma: the vm area in which the mapping is added | 938 | * @vma: the vm area in which the mapping is added |
939 | * @address: the user virtual address mapped | 939 | * @address: the user virtual address mapped |
940 | */ | 940 | */ |
941 | static void __page_check_anon_rmap(struct page *page, | 941 | static void __page_check_anon_rmap(struct page *page, |
942 | struct vm_area_struct *vma, unsigned long address) | 942 | struct vm_area_struct *vma, unsigned long address) |
943 | { | 943 | { |
944 | #ifdef CONFIG_DEBUG_VM | 944 | #ifdef CONFIG_DEBUG_VM |
945 | /* | 945 | /* |
946 | * The page's anon-rmap details (mapping and index) are guaranteed to | 946 | * The page's anon-rmap details (mapping and index) are guaranteed to |
947 | * be set up correctly at this point. | 947 | * be set up correctly at this point. |
948 | * | 948 | * |
949 | * We have exclusion against page_add_anon_rmap because the caller | 949 | * We have exclusion against page_add_anon_rmap because the caller |
950 | * always holds the page locked, except if called from page_dup_rmap, | 950 | * always holds the page locked, except if called from page_dup_rmap, |
951 | * in which case the page is already known to be setup. | 951 | * in which case the page is already known to be setup. |
952 | * | 952 | * |
953 | * We have exclusion against page_add_new_anon_rmap because those pages | 953 | * We have exclusion against page_add_new_anon_rmap because those pages |
954 | * are initially only visible via the pagetables, and the pte is locked | 954 | * are initially only visible via the pagetables, and the pte is locked |
955 | * over the call to page_add_new_anon_rmap. | 955 | * over the call to page_add_new_anon_rmap. |
956 | */ | 956 | */ |
957 | BUG_ON(page_anon_vma(page)->root != vma->anon_vma->root); | 957 | BUG_ON(page_anon_vma(page)->root != vma->anon_vma->root); |
958 | BUG_ON(page->index != linear_page_index(vma, address)); | 958 | BUG_ON(page->index != linear_page_index(vma, address)); |
959 | #endif | 959 | #endif |
960 | } | 960 | } |
961 | 961 | ||
962 | /** | 962 | /** |
963 | * page_add_anon_rmap - add pte mapping to an anonymous page | 963 | * page_add_anon_rmap - add pte mapping to an anonymous page |
964 | * @page: the page to add the mapping to | 964 | * @page: the page to add the mapping to |
965 | * @vma: the vm area in which the mapping is added | 965 | * @vma: the vm area in which the mapping is added |
966 | * @address: the user virtual address mapped | 966 | * @address: the user virtual address mapped |
967 | * | 967 | * |
968 | * The caller needs to hold the pte lock, and the page must be locked in | 968 | * The caller needs to hold the pte lock, and the page must be locked in |
969 | * the anon_vma case: to serialize mapping,index checking after setting, | 969 | * the anon_vma case: to serialize mapping,index checking after setting, |
970 | * and to ensure that PageAnon is not being upgraded racily to PageKsm | 970 | * and to ensure that PageAnon is not being upgraded racily to PageKsm |
971 | * (but PageKsm is never downgraded to PageAnon). | 971 | * (but PageKsm is never downgraded to PageAnon). |
972 | */ | 972 | */ |
973 | void page_add_anon_rmap(struct page *page, | 973 | void page_add_anon_rmap(struct page *page, |
974 | struct vm_area_struct *vma, unsigned long address) | 974 | struct vm_area_struct *vma, unsigned long address) |
975 | { | 975 | { |
976 | do_page_add_anon_rmap(page, vma, address, 0); | 976 | do_page_add_anon_rmap(page, vma, address, 0); |
977 | } | 977 | } |
978 | 978 | ||
979 | /* | 979 | /* |
980 | * Special version of the above for do_swap_page, which often runs | 980 | * Special version of the above for do_swap_page, which often runs |
981 | * into pages that are exclusively owned by the current process. | 981 | * into pages that are exclusively owned by the current process. |
982 | * Everybody else should continue to use page_add_anon_rmap above. | 982 | * Everybody else should continue to use page_add_anon_rmap above. |
983 | */ | 983 | */ |
984 | void do_page_add_anon_rmap(struct page *page, | 984 | void do_page_add_anon_rmap(struct page *page, |
985 | struct vm_area_struct *vma, unsigned long address, int exclusive) | 985 | struct vm_area_struct *vma, unsigned long address, int exclusive) |
986 | { | 986 | { |
987 | int first = atomic_inc_and_test(&page->_mapcount); | 987 | int first = atomic_inc_and_test(&page->_mapcount); |
988 | if (first) { | 988 | if (first) { |
989 | if (PageTransHuge(page)) | 989 | if (PageTransHuge(page)) |
990 | __inc_zone_page_state(page, | 990 | __inc_zone_page_state(page, |
991 | NR_ANON_TRANSPARENT_HUGEPAGES); | 991 | NR_ANON_TRANSPARENT_HUGEPAGES); |
992 | __mod_zone_page_state(page_zone(page), NR_ANON_PAGES, | 992 | __mod_zone_page_state(page_zone(page), NR_ANON_PAGES, |
993 | hpage_nr_pages(page)); | 993 | hpage_nr_pages(page)); |
994 | } | 994 | } |
995 | if (unlikely(PageKsm(page))) | 995 | if (unlikely(PageKsm(page))) |
996 | return; | 996 | return; |
997 | 997 | ||
998 | VM_BUG_ON_PAGE(!PageLocked(page), page); | 998 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
999 | /* address might be in next vma when migration races vma_adjust */ | 999 | /* address might be in next vma when migration races vma_adjust */ |
1000 | if (first) | 1000 | if (first) |
1001 | __page_set_anon_rmap(page, vma, address, exclusive); | 1001 | __page_set_anon_rmap(page, vma, address, exclusive); |
1002 | else | 1002 | else |
1003 | __page_check_anon_rmap(page, vma, address); | 1003 | __page_check_anon_rmap(page, vma, address); |
1004 | } | 1004 | } |
1005 | 1005 | ||
1006 | /** | 1006 | /** |
1007 | * page_add_new_anon_rmap - add pte mapping to a new anonymous page | 1007 | * page_add_new_anon_rmap - add pte mapping to a new anonymous page |
1008 | * @page: the page to add the mapping to | 1008 | * @page: the page to add the mapping to |
1009 | * @vma: the vm area in which the mapping is added | 1009 | * @vma: the vm area in which the mapping is added |
1010 | * @address: the user virtual address mapped | 1010 | * @address: the user virtual address mapped |
1011 | * | 1011 | * |
1012 | * Same as page_add_anon_rmap but must only be called on *new* pages. | 1012 | * Same as page_add_anon_rmap but must only be called on *new* pages. |
1013 | * This means the inc-and-test can be bypassed. | 1013 | * This means the inc-and-test can be bypassed. |
1014 | * Page does not have to be locked. | 1014 | * Page does not have to be locked. |
1015 | */ | 1015 | */ |
1016 | void page_add_new_anon_rmap(struct page *page, | 1016 | void page_add_new_anon_rmap(struct page *page, |
1017 | struct vm_area_struct *vma, unsigned long address) | 1017 | struct vm_area_struct *vma, unsigned long address) |
1018 | { | 1018 | { |
1019 | VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); | 1019 | VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); |
1020 | SetPageSwapBacked(page); | 1020 | SetPageSwapBacked(page); |
1021 | atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */ | 1021 | atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */ |
1022 | if (PageTransHuge(page)) | 1022 | if (PageTransHuge(page)) |
1023 | __inc_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES); | 1023 | __inc_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES); |
1024 | __mod_zone_page_state(page_zone(page), NR_ANON_PAGES, | 1024 | __mod_zone_page_state(page_zone(page), NR_ANON_PAGES, |
1025 | hpage_nr_pages(page)); | 1025 | hpage_nr_pages(page)); |
1026 | __page_set_anon_rmap(page, vma, address, 1); | 1026 | __page_set_anon_rmap(page, vma, address, 1); |
1027 | if (!mlocked_vma_newpage(vma, page)) { | 1027 | if (!mlocked_vma_newpage(vma, page)) { |
1028 | SetPageActive(page); | 1028 | SetPageActive(page); |
1029 | lru_cache_add(page); | 1029 | lru_cache_add(page); |
1030 | } else | 1030 | } else |
1031 | add_page_to_unevictable_list(page); | 1031 | add_page_to_unevictable_list(page); |
1032 | } | 1032 | } |
1033 | 1033 | ||
1034 | /** | 1034 | /** |
1035 | * page_add_file_rmap - add pte mapping to a file page | 1035 | * page_add_file_rmap - add pte mapping to a file page |
1036 | * @page: the page to add the mapping to | 1036 | * @page: the page to add the mapping to |
1037 | * | 1037 | * |
1038 | * The caller needs to hold the pte lock. | 1038 | * The caller needs to hold the pte lock. |
1039 | */ | 1039 | */ |
1040 | void page_add_file_rmap(struct page *page) | 1040 | void page_add_file_rmap(struct page *page) |
1041 | { | 1041 | { |
1042 | bool locked; | 1042 | bool locked; |
1043 | unsigned long flags; | 1043 | unsigned long flags; |
1044 | 1044 | ||
1045 | mem_cgroup_begin_update_page_stat(page, &locked, &flags); | 1045 | mem_cgroup_begin_update_page_stat(page, &locked, &flags); |
1046 | if (atomic_inc_and_test(&page->_mapcount)) { | 1046 | if (atomic_inc_and_test(&page->_mapcount)) { |
1047 | __inc_zone_page_state(page, NR_FILE_MAPPED); | 1047 | __inc_zone_page_state(page, NR_FILE_MAPPED); |
1048 | mem_cgroup_inc_page_stat(page, MEM_CGROUP_STAT_FILE_MAPPED); | 1048 | mem_cgroup_inc_page_stat(page, MEM_CGROUP_STAT_FILE_MAPPED); |
1049 | } | 1049 | } |
1050 | mem_cgroup_end_update_page_stat(page, &locked, &flags); | 1050 | mem_cgroup_end_update_page_stat(page, &locked, &flags); |
1051 | } | 1051 | } |
1052 | 1052 | ||
1053 | /** | 1053 | /** |
1054 | * page_remove_rmap - take down pte mapping from a page | 1054 | * page_remove_rmap - take down pte mapping from a page |
1055 | * @page: page to remove mapping from | 1055 | * @page: page to remove mapping from |
1056 | * | 1056 | * |
1057 | * The caller needs to hold the pte lock. | 1057 | * The caller needs to hold the pte lock. |
1058 | */ | 1058 | */ |
1059 | void page_remove_rmap(struct page *page) | 1059 | void page_remove_rmap(struct page *page) |
1060 | { | 1060 | { |
1061 | bool anon = PageAnon(page); | 1061 | bool anon = PageAnon(page); |
1062 | bool locked; | 1062 | bool locked; |
1063 | unsigned long flags; | 1063 | unsigned long flags; |
1064 | 1064 | ||
1065 | /* | 1065 | /* |
1066 | * The anon case has no mem_cgroup page_stat to update; but may | 1066 | * The anon case has no mem_cgroup page_stat to update; but may |
1067 | * uncharge_page() below, where the lock ordering can deadlock if | 1067 | * uncharge_page() below, where the lock ordering can deadlock if |
1068 | * we hold the lock against page_stat move: so avoid it on anon. | 1068 | * we hold the lock against page_stat move: so avoid it on anon. |
1069 | */ | 1069 | */ |
1070 | if (!anon) | 1070 | if (!anon) |
1071 | mem_cgroup_begin_update_page_stat(page, &locked, &flags); | 1071 | mem_cgroup_begin_update_page_stat(page, &locked, &flags); |
1072 | 1072 | ||
1073 | /* page still mapped by someone else? */ | 1073 | /* page still mapped by someone else? */ |
1074 | if (!atomic_add_negative(-1, &page->_mapcount)) | 1074 | if (!atomic_add_negative(-1, &page->_mapcount)) |
1075 | goto out; | 1075 | goto out; |
1076 | 1076 | ||
1077 | /* | 1077 | /* |
1078 | * Hugepages are not counted in NR_ANON_PAGES nor NR_FILE_MAPPED | 1078 | * Hugepages are not counted in NR_ANON_PAGES nor NR_FILE_MAPPED |
1079 | * and not charged by memcg for now. | 1079 | * and not charged by memcg for now. |
1080 | */ | 1080 | */ |
1081 | if (unlikely(PageHuge(page))) | 1081 | if (unlikely(PageHuge(page))) |
1082 | goto out; | 1082 | goto out; |
1083 | if (anon) { | 1083 | if (anon) { |
1084 | mem_cgroup_uncharge_page(page); | 1084 | mem_cgroup_uncharge_page(page); |
1085 | if (PageTransHuge(page)) | 1085 | if (PageTransHuge(page)) |
1086 | __dec_zone_page_state(page, | 1086 | __dec_zone_page_state(page, |
1087 | NR_ANON_TRANSPARENT_HUGEPAGES); | 1087 | NR_ANON_TRANSPARENT_HUGEPAGES); |
1088 | __mod_zone_page_state(page_zone(page), NR_ANON_PAGES, | 1088 | __mod_zone_page_state(page_zone(page), NR_ANON_PAGES, |
1089 | -hpage_nr_pages(page)); | 1089 | -hpage_nr_pages(page)); |
1090 | } else { | 1090 | } else { |
1091 | __dec_zone_page_state(page, NR_FILE_MAPPED); | 1091 | __dec_zone_page_state(page, NR_FILE_MAPPED); |
1092 | mem_cgroup_dec_page_stat(page, MEM_CGROUP_STAT_FILE_MAPPED); | 1092 | mem_cgroup_dec_page_stat(page, MEM_CGROUP_STAT_FILE_MAPPED); |
1093 | mem_cgroup_end_update_page_stat(page, &locked, &flags); | 1093 | mem_cgroup_end_update_page_stat(page, &locked, &flags); |
1094 | } | 1094 | } |
1095 | if (unlikely(PageMlocked(page))) | 1095 | if (unlikely(PageMlocked(page))) |
1096 | clear_page_mlock(page); | 1096 | clear_page_mlock(page); |
1097 | /* | 1097 | /* |
1098 | * It would be tidy to reset the PageAnon mapping here, | 1098 | * It would be tidy to reset the PageAnon mapping here, |
1099 | * but that might overwrite a racing page_add_anon_rmap | 1099 | * but that might overwrite a racing page_add_anon_rmap |
1100 | * which increments mapcount after us but sets mapping | 1100 | * which increments mapcount after us but sets mapping |
1101 | * before us: so leave the reset to free_hot_cold_page, | 1101 | * before us: so leave the reset to free_hot_cold_page, |
1102 | * and remember that it's only reliable while mapped. | 1102 | * and remember that it's only reliable while mapped. |
1103 | * Leaving it set also helps swapoff to reinstate ptes | 1103 | * Leaving it set also helps swapoff to reinstate ptes |
1104 | * faster for those pages still in swapcache. | 1104 | * faster for those pages still in swapcache. |
1105 | */ | 1105 | */ |
1106 | return; | 1106 | return; |
1107 | out: | 1107 | out: |
1108 | if (!anon) | 1108 | if (!anon) |
1109 | mem_cgroup_end_update_page_stat(page, &locked, &flags); | 1109 | mem_cgroup_end_update_page_stat(page, &locked, &flags); |
1110 | } | 1110 | } |
1111 | 1111 | ||
1112 | /* | 1112 | /* |
1113 | * @arg: enum ttu_flags will be passed to this argument | 1113 | * @arg: enum ttu_flags will be passed to this argument |
1114 | */ | 1114 | */ |
1115 | int try_to_unmap_one(struct page *page, struct vm_area_struct *vma, | 1115 | int try_to_unmap_one(struct page *page, struct vm_area_struct *vma, |
1116 | unsigned long address, void *arg) | 1116 | unsigned long address, void *arg) |
1117 | { | 1117 | { |
1118 | struct mm_struct *mm = vma->vm_mm; | 1118 | struct mm_struct *mm = vma->vm_mm; |
1119 | pte_t *pte; | 1119 | pte_t *pte; |
1120 | pte_t pteval; | 1120 | pte_t pteval; |
1121 | spinlock_t *ptl; | 1121 | spinlock_t *ptl; |
1122 | int ret = SWAP_AGAIN; | 1122 | int ret = SWAP_AGAIN; |
1123 | enum ttu_flags flags = (enum ttu_flags)arg; | 1123 | enum ttu_flags flags = (enum ttu_flags)arg; |
1124 | 1124 | ||
1125 | pte = page_check_address(page, mm, address, &ptl, 0); | 1125 | pte = page_check_address(page, mm, address, &ptl, 0); |
1126 | if (!pte) | 1126 | if (!pte) |
1127 | goto out; | 1127 | goto out; |
1128 | 1128 | ||
1129 | /* | 1129 | /* |
1130 | * If the page is mlock()d, we cannot swap it out. | 1130 | * If the page is mlock()d, we cannot swap it out. |
1131 | * If it's recently referenced (perhaps page_referenced | 1131 | * If it's recently referenced (perhaps page_referenced |
1132 | * skipped over this mm) then we should reactivate it. | 1132 | * skipped over this mm) then we should reactivate it. |
1133 | */ | 1133 | */ |
1134 | if (!(flags & TTU_IGNORE_MLOCK)) { | 1134 | if (!(flags & TTU_IGNORE_MLOCK)) { |
1135 | if (vma->vm_flags & VM_LOCKED) | 1135 | if (vma->vm_flags & VM_LOCKED) |
1136 | goto out_mlock; | 1136 | goto out_mlock; |
1137 | 1137 | ||
1138 | if (TTU_ACTION(flags) == TTU_MUNLOCK) | 1138 | if (TTU_ACTION(flags) == TTU_MUNLOCK) |
1139 | goto out_unmap; | 1139 | goto out_unmap; |
1140 | } | 1140 | } |
1141 | if (!(flags & TTU_IGNORE_ACCESS)) { | 1141 | if (!(flags & TTU_IGNORE_ACCESS)) { |
1142 | if (ptep_clear_flush_young_notify(vma, address, pte)) { | 1142 | if (ptep_clear_flush_young_notify(vma, address, pte)) { |
1143 | ret = SWAP_FAIL; | 1143 | ret = SWAP_FAIL; |
1144 | goto out_unmap; | 1144 | goto out_unmap; |
1145 | } | 1145 | } |
1146 | } | 1146 | } |
1147 | 1147 | ||
1148 | /* Nuke the page table entry. */ | 1148 | /* Nuke the page table entry. */ |
1149 | flush_cache_page(vma, address, page_to_pfn(page)); | 1149 | flush_cache_page(vma, address, page_to_pfn(page)); |
1150 | pteval = ptep_clear_flush(vma, address, pte); | 1150 | pteval = ptep_clear_flush(vma, address, pte); |
1151 | 1151 | ||
1152 | /* Move the dirty bit to the physical page now the pte is gone. */ | 1152 | /* Move the dirty bit to the physical page now the pte is gone. */ |
1153 | if (pte_dirty(pteval)) | 1153 | if (pte_dirty(pteval)) |
1154 | set_page_dirty(page); | 1154 | set_page_dirty(page); |
1155 | 1155 | ||
1156 | /* Update high watermark before we lower rss */ | 1156 | /* Update high watermark before we lower rss */ |
1157 | update_hiwater_rss(mm); | 1157 | update_hiwater_rss(mm); |
1158 | 1158 | ||
1159 | if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) { | 1159 | if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) { |
1160 | if (!PageHuge(page)) { | 1160 | if (!PageHuge(page)) { |
1161 | if (PageAnon(page)) | 1161 | if (PageAnon(page)) |
1162 | dec_mm_counter(mm, MM_ANONPAGES); | 1162 | dec_mm_counter(mm, MM_ANONPAGES); |
1163 | else | 1163 | else |
1164 | dec_mm_counter(mm, MM_FILEPAGES); | 1164 | dec_mm_counter(mm, MM_FILEPAGES); |
1165 | } | 1165 | } |
1166 | set_pte_at(mm, address, pte, | 1166 | set_pte_at(mm, address, pte, |
1167 | swp_entry_to_pte(make_hwpoison_entry(page))); | 1167 | swp_entry_to_pte(make_hwpoison_entry(page))); |
1168 | } else if (pte_unused(pteval)) { | 1168 | } else if (pte_unused(pteval)) { |
1169 | /* | 1169 | /* |
1170 | * The guest indicated that the page content is of no | 1170 | * The guest indicated that the page content is of no |
1171 | * interest anymore. Simply discard the pte, vmscan | 1171 | * interest anymore. Simply discard the pte, vmscan |
1172 | * will take care of the rest. | 1172 | * will take care of the rest. |
1173 | */ | 1173 | */ |
1174 | if (PageAnon(page)) | 1174 | if (PageAnon(page)) |
1175 | dec_mm_counter(mm, MM_ANONPAGES); | 1175 | dec_mm_counter(mm, MM_ANONPAGES); |
1176 | else | 1176 | else |
1177 | dec_mm_counter(mm, MM_FILEPAGES); | 1177 | dec_mm_counter(mm, MM_FILEPAGES); |
1178 | } else if (PageAnon(page)) { | 1178 | } else if (PageAnon(page)) { |
1179 | swp_entry_t entry = { .val = page_private(page) }; | 1179 | swp_entry_t entry = { .val = page_private(page) }; |
1180 | pte_t swp_pte; | 1180 | pte_t swp_pte; |
1181 | 1181 | ||
1182 | if (PageSwapCache(page)) { | 1182 | if (PageSwapCache(page)) { |
1183 | /* | 1183 | /* |
1184 | * Store the swap location in the pte. | 1184 | * Store the swap location in the pte. |
1185 | * See handle_pte_fault() ... | 1185 | * See handle_pte_fault() ... |
1186 | */ | 1186 | */ |
1187 | if (swap_duplicate(entry) < 0) { | 1187 | if (swap_duplicate(entry) < 0) { |
1188 | set_pte_at(mm, address, pte, pteval); | 1188 | set_pte_at(mm, address, pte, pteval); |
1189 | ret = SWAP_FAIL; | 1189 | ret = SWAP_FAIL; |
1190 | goto out_unmap; | 1190 | goto out_unmap; |
1191 | } | 1191 | } |
1192 | if (list_empty(&mm->mmlist)) { | 1192 | if (list_empty(&mm->mmlist)) { |
1193 | spin_lock(&mmlist_lock); | 1193 | spin_lock(&mmlist_lock); |
1194 | if (list_empty(&mm->mmlist)) | 1194 | if (list_empty(&mm->mmlist)) |
1195 | list_add(&mm->mmlist, &init_mm.mmlist); | 1195 | list_add(&mm->mmlist, &init_mm.mmlist); |
1196 | spin_unlock(&mmlist_lock); | 1196 | spin_unlock(&mmlist_lock); |
1197 | } | 1197 | } |
1198 | dec_mm_counter(mm, MM_ANONPAGES); | 1198 | dec_mm_counter(mm, MM_ANONPAGES); |
1199 | inc_mm_counter(mm, MM_SWAPENTS); | 1199 | inc_mm_counter(mm, MM_SWAPENTS); |
1200 | } else if (IS_ENABLED(CONFIG_MIGRATION)) { | 1200 | } else if (IS_ENABLED(CONFIG_MIGRATION)) { |
1201 | /* | 1201 | /* |
1202 | * Store the pfn of the page in a special migration | 1202 | * Store the pfn of the page in a special migration |
1203 | * pte. do_swap_page() will wait until the migration | 1203 | * pte. do_swap_page() will wait until the migration |
1204 | * pte is removed and then restart fault handling. | 1204 | * pte is removed and then restart fault handling. |
1205 | */ | 1205 | */ |
1206 | BUG_ON(TTU_ACTION(flags) != TTU_MIGRATION); | 1206 | BUG_ON(TTU_ACTION(flags) != TTU_MIGRATION); |
1207 | entry = make_migration_entry(page, pte_write(pteval)); | 1207 | entry = make_migration_entry(page, pte_write(pteval)); |
1208 | } | 1208 | } |
1209 | swp_pte = swp_entry_to_pte(entry); | 1209 | swp_pte = swp_entry_to_pte(entry); |
1210 | if (pte_soft_dirty(pteval)) | 1210 | if (pte_soft_dirty(pteval)) |
1211 | swp_pte = pte_swp_mksoft_dirty(swp_pte); | 1211 | swp_pte = pte_swp_mksoft_dirty(swp_pte); |
1212 | set_pte_at(mm, address, pte, swp_pte); | 1212 | set_pte_at(mm, address, pte, swp_pte); |
1213 | BUG_ON(pte_file(*pte)); | 1213 | BUG_ON(pte_file(*pte)); |
1214 | } else if (IS_ENABLED(CONFIG_MIGRATION) && | 1214 | } else if (IS_ENABLED(CONFIG_MIGRATION) && |
1215 | (TTU_ACTION(flags) == TTU_MIGRATION)) { | 1215 | (TTU_ACTION(flags) == TTU_MIGRATION)) { |
1216 | /* Establish migration entry for a file page */ | 1216 | /* Establish migration entry for a file page */ |
1217 | swp_entry_t entry; | 1217 | swp_entry_t entry; |
1218 | entry = make_migration_entry(page, pte_write(pteval)); | 1218 | entry = make_migration_entry(page, pte_write(pteval)); |
1219 | set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); | 1219 | set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); |
1220 | } else | 1220 | } else |
1221 | dec_mm_counter(mm, MM_FILEPAGES); | 1221 | dec_mm_counter(mm, MM_FILEPAGES); |
1222 | 1222 | ||
1223 | page_remove_rmap(page); | 1223 | page_remove_rmap(page); |
1224 | page_cache_release(page); | 1224 | page_cache_release(page); |
1225 | 1225 | ||
1226 | out_unmap: | 1226 | out_unmap: |
1227 | pte_unmap_unlock(pte, ptl); | 1227 | pte_unmap_unlock(pte, ptl); |
1228 | if (ret != SWAP_FAIL) | 1228 | if (ret != SWAP_FAIL) |
1229 | mmu_notifier_invalidate_page(mm, address); | 1229 | mmu_notifier_invalidate_page(mm, address); |
1230 | out: | 1230 | out: |
1231 | return ret; | 1231 | return ret; |
1232 | 1232 | ||
1233 | out_mlock: | 1233 | out_mlock: |
1234 | pte_unmap_unlock(pte, ptl); | 1234 | pte_unmap_unlock(pte, ptl); |
1235 | 1235 | ||
1236 | 1236 | ||
1237 | /* | 1237 | /* |
1238 | * We need mmap_sem locking, Otherwise VM_LOCKED check makes | 1238 | * We need mmap_sem locking, Otherwise VM_LOCKED check makes |
1239 | * unstable result and race. Plus, We can't wait here because | 1239 | * unstable result and race. Plus, We can't wait here because |
1240 | * we now hold anon_vma->rwsem or mapping->i_mmap_mutex. | 1240 | * we now hold anon_vma->rwsem or mapping->i_mmap_mutex. |
1241 | * if trylock failed, the page remain in evictable lru and later | 1241 | * if trylock failed, the page remain in evictable lru and later |
1242 | * vmscan could retry to move the page to unevictable lru if the | 1242 | * vmscan could retry to move the page to unevictable lru if the |
1243 | * page is actually mlocked. | 1243 | * page is actually mlocked. |
1244 | */ | 1244 | */ |
1245 | if (down_read_trylock(&vma->vm_mm->mmap_sem)) { | 1245 | if (down_read_trylock(&vma->vm_mm->mmap_sem)) { |
1246 | if (vma->vm_flags & VM_LOCKED) { | 1246 | if (vma->vm_flags & VM_LOCKED) { |
1247 | mlock_vma_page(page); | 1247 | mlock_vma_page(page); |
1248 | ret = SWAP_MLOCK; | 1248 | ret = SWAP_MLOCK; |
1249 | } | 1249 | } |
1250 | up_read(&vma->vm_mm->mmap_sem); | 1250 | up_read(&vma->vm_mm->mmap_sem); |
1251 | } | 1251 | } |
1252 | return ret; | 1252 | return ret; |
1253 | } | 1253 | } |
1254 | 1254 | ||
1255 | /* | 1255 | /* |
1256 | * objrmap doesn't work for nonlinear VMAs because the assumption that | 1256 | * objrmap doesn't work for nonlinear VMAs because the assumption that |
1257 | * offset-into-file correlates with offset-into-virtual-addresses does not hold. | 1257 | * offset-into-file correlates with offset-into-virtual-addresses does not hold. |
1258 | * Consequently, given a particular page and its ->index, we cannot locate the | 1258 | * Consequently, given a particular page and its ->index, we cannot locate the |
1259 | * ptes which are mapping that page without an exhaustive linear search. | 1259 | * ptes which are mapping that page without an exhaustive linear search. |
1260 | * | 1260 | * |
1261 | * So what this code does is a mini "virtual scan" of each nonlinear VMA which | 1261 | * So what this code does is a mini "virtual scan" of each nonlinear VMA which |
1262 | * maps the file to which the target page belongs. The ->vm_private_data field | 1262 | * maps the file to which the target page belongs. The ->vm_private_data field |
1263 | * holds the current cursor into that scan. Successive searches will circulate | 1263 | * holds the current cursor into that scan. Successive searches will circulate |
1264 | * around the vma's virtual address space. | 1264 | * around the vma's virtual address space. |
1265 | * | 1265 | * |
1266 | * So as more replacement pressure is applied to the pages in a nonlinear VMA, | 1266 | * So as more replacement pressure is applied to the pages in a nonlinear VMA, |
1267 | * more scanning pressure is placed against them as well. Eventually pages | 1267 | * more scanning pressure is placed against them as well. Eventually pages |
1268 | * will become fully unmapped and are eligible for eviction. | 1268 | * will become fully unmapped and are eligible for eviction. |
1269 | * | 1269 | * |
1270 | * For very sparsely populated VMAs this is a little inefficient - chances are | 1270 | * For very sparsely populated VMAs this is a little inefficient - chances are |
1271 | * there there won't be many ptes located within the scan cluster. In this case | 1271 | * there there won't be many ptes located within the scan cluster. In this case |
1272 | * maybe we could scan further - to the end of the pte page, perhaps. | 1272 | * maybe we could scan further - to the end of the pte page, perhaps. |
1273 | * | 1273 | * |
1274 | * Mlocked pages: check VM_LOCKED under mmap_sem held for read, if we can | 1274 | * Mlocked pages: check VM_LOCKED under mmap_sem held for read, if we can |
1275 | * acquire it without blocking. If vma locked, mlock the pages in the cluster, | 1275 | * acquire it without blocking. If vma locked, mlock the pages in the cluster, |
1276 | * rather than unmapping them. If we encounter the "check_page" that vmscan is | 1276 | * rather than unmapping them. If we encounter the "check_page" that vmscan is |
1277 | * trying to unmap, return SWAP_MLOCK, else default SWAP_AGAIN. | 1277 | * trying to unmap, return SWAP_MLOCK, else default SWAP_AGAIN. |
1278 | */ | 1278 | */ |
1279 | #define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE) | 1279 | #define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE) |
1280 | #define CLUSTER_MASK (~(CLUSTER_SIZE - 1)) | 1280 | #define CLUSTER_MASK (~(CLUSTER_SIZE - 1)) |
1281 | 1281 | ||
1282 | static int try_to_unmap_cluster(unsigned long cursor, unsigned int *mapcount, | 1282 | static int try_to_unmap_cluster(unsigned long cursor, unsigned int *mapcount, |
1283 | struct vm_area_struct *vma, struct page *check_page) | 1283 | struct vm_area_struct *vma, struct page *check_page) |
1284 | { | 1284 | { |
1285 | struct mm_struct *mm = vma->vm_mm; | 1285 | struct mm_struct *mm = vma->vm_mm; |
1286 | pmd_t *pmd; | 1286 | pmd_t *pmd; |
1287 | pte_t *pte; | 1287 | pte_t *pte; |
1288 | pte_t pteval; | 1288 | pte_t pteval; |
1289 | spinlock_t *ptl; | 1289 | spinlock_t *ptl; |
1290 | struct page *page; | 1290 | struct page *page; |
1291 | unsigned long address; | 1291 | unsigned long address; |
1292 | unsigned long mmun_start; /* For mmu_notifiers */ | 1292 | unsigned long mmun_start; /* For mmu_notifiers */ |
1293 | unsigned long mmun_end; /* For mmu_notifiers */ | 1293 | unsigned long mmun_end; /* For mmu_notifiers */ |
1294 | unsigned long end; | 1294 | unsigned long end; |
1295 | int ret = SWAP_AGAIN; | 1295 | int ret = SWAP_AGAIN; |
1296 | int locked_vma = 0; | 1296 | int locked_vma = 0; |
1297 | 1297 | ||
1298 | address = (vma->vm_start + cursor) & CLUSTER_MASK; | 1298 | address = (vma->vm_start + cursor) & CLUSTER_MASK; |
1299 | end = address + CLUSTER_SIZE; | 1299 | end = address + CLUSTER_SIZE; |
1300 | if (address < vma->vm_start) | 1300 | if (address < vma->vm_start) |
1301 | address = vma->vm_start; | 1301 | address = vma->vm_start; |
1302 | if (end > vma->vm_end) | 1302 | if (end > vma->vm_end) |
1303 | end = vma->vm_end; | 1303 | end = vma->vm_end; |
1304 | 1304 | ||
1305 | pmd = mm_find_pmd(mm, address); | 1305 | pmd = mm_find_pmd(mm, address); |
1306 | if (!pmd) | 1306 | if (!pmd) |
1307 | return ret; | 1307 | return ret; |
1308 | 1308 | ||
1309 | mmun_start = address; | 1309 | mmun_start = address; |
1310 | mmun_end = end; | 1310 | mmun_end = end; |
1311 | mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); | 1311 | mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); |
1312 | 1312 | ||
1313 | /* | 1313 | /* |
1314 | * If we can acquire the mmap_sem for read, and vma is VM_LOCKED, | 1314 | * If we can acquire the mmap_sem for read, and vma is VM_LOCKED, |
1315 | * keep the sem while scanning the cluster for mlocking pages. | 1315 | * keep the sem while scanning the cluster for mlocking pages. |
1316 | */ | 1316 | */ |
1317 | if (down_read_trylock(&vma->vm_mm->mmap_sem)) { | 1317 | if (down_read_trylock(&vma->vm_mm->mmap_sem)) { |
1318 | locked_vma = (vma->vm_flags & VM_LOCKED); | 1318 | locked_vma = (vma->vm_flags & VM_LOCKED); |
1319 | if (!locked_vma) | 1319 | if (!locked_vma) |
1320 | up_read(&vma->vm_mm->mmap_sem); /* don't need it */ | 1320 | up_read(&vma->vm_mm->mmap_sem); /* don't need it */ |
1321 | } | 1321 | } |
1322 | 1322 | ||
1323 | pte = pte_offset_map_lock(mm, pmd, address, &ptl); | 1323 | pte = pte_offset_map_lock(mm, pmd, address, &ptl); |
1324 | 1324 | ||
1325 | /* Update high watermark before we lower rss */ | 1325 | /* Update high watermark before we lower rss */ |
1326 | update_hiwater_rss(mm); | 1326 | update_hiwater_rss(mm); |
1327 | 1327 | ||
1328 | for (; address < end; pte++, address += PAGE_SIZE) { | 1328 | for (; address < end; pte++, address += PAGE_SIZE) { |
1329 | if (!pte_present(*pte)) | 1329 | if (!pte_present(*pte)) |
1330 | continue; | 1330 | continue; |
1331 | page = vm_normal_page(vma, address, *pte); | 1331 | page = vm_normal_page(vma, address, *pte); |
1332 | BUG_ON(!page || PageAnon(page)); | 1332 | BUG_ON(!page || PageAnon(page)); |
1333 | 1333 | ||
1334 | if (locked_vma) { | 1334 | if (locked_vma) { |
1335 | if (page == check_page) { | 1335 | if (page == check_page) { |
1336 | /* we know we have check_page locked */ | 1336 | /* we know we have check_page locked */ |
1337 | mlock_vma_page(page); | 1337 | mlock_vma_page(page); |
1338 | ret = SWAP_MLOCK; | 1338 | ret = SWAP_MLOCK; |
1339 | } else if (trylock_page(page)) { | 1339 | } else if (trylock_page(page)) { |
1340 | /* | 1340 | /* |
1341 | * If we can lock the page, perform mlock. | 1341 | * If we can lock the page, perform mlock. |
1342 | * Otherwise leave the page alone, it will be | 1342 | * Otherwise leave the page alone, it will be |
1343 | * eventually encountered again later. | 1343 | * eventually encountered again later. |
1344 | */ | 1344 | */ |
1345 | mlock_vma_page(page); | 1345 | mlock_vma_page(page); |
1346 | unlock_page(page); | 1346 | unlock_page(page); |
1347 | } | 1347 | } |
1348 | continue; /* don't unmap */ | 1348 | continue; /* don't unmap */ |
1349 | } | 1349 | } |
1350 | 1350 | ||
1351 | if (ptep_clear_flush_young_notify(vma, address, pte)) | 1351 | if (ptep_clear_flush_young_notify(vma, address, pte)) |
1352 | continue; | 1352 | continue; |
1353 | 1353 | ||
1354 | /* Nuke the page table entry. */ | 1354 | /* Nuke the page table entry. */ |
1355 | flush_cache_page(vma, address, pte_pfn(*pte)); | 1355 | flush_cache_page(vma, address, pte_pfn(*pte)); |
1356 | pteval = ptep_clear_flush(vma, address, pte); | 1356 | pteval = ptep_clear_flush(vma, address, pte); |
1357 | 1357 | ||
1358 | /* If nonlinear, store the file page offset in the pte. */ | 1358 | /* If nonlinear, store the file page offset in the pte. */ |
1359 | if (page->index != linear_page_index(vma, address)) { | 1359 | if (page->index != linear_page_index(vma, address)) { |
1360 | pte_t ptfile = pgoff_to_pte(page->index); | 1360 | pte_t ptfile = pgoff_to_pte(page->index); |
1361 | if (pte_soft_dirty(pteval)) | 1361 | if (pte_soft_dirty(pteval)) |
1362 | pte_file_mksoft_dirty(ptfile); | 1362 | pte_file_mksoft_dirty(ptfile); |
1363 | set_pte_at(mm, address, pte, ptfile); | 1363 | set_pte_at(mm, address, pte, ptfile); |
1364 | } | 1364 | } |
1365 | 1365 | ||
1366 | /* Move the dirty bit to the physical page now the pte is gone. */ | 1366 | /* Move the dirty bit to the physical page now the pte is gone. */ |
1367 | if (pte_dirty(pteval)) | 1367 | if (pte_dirty(pteval)) |
1368 | set_page_dirty(page); | 1368 | set_page_dirty(page); |
1369 | 1369 | ||
1370 | page_remove_rmap(page); | 1370 | page_remove_rmap(page); |
1371 | page_cache_release(page); | 1371 | page_cache_release(page); |
1372 | dec_mm_counter(mm, MM_FILEPAGES); | 1372 | dec_mm_counter(mm, MM_FILEPAGES); |
1373 | (*mapcount)--; | 1373 | (*mapcount)--; |
1374 | } | 1374 | } |
1375 | pte_unmap_unlock(pte - 1, ptl); | 1375 | pte_unmap_unlock(pte - 1, ptl); |
1376 | mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); | 1376 | mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); |
1377 | if (locked_vma) | 1377 | if (locked_vma) |
1378 | up_read(&vma->vm_mm->mmap_sem); | 1378 | up_read(&vma->vm_mm->mmap_sem); |
1379 | return ret; | 1379 | return ret; |
1380 | } | 1380 | } |
1381 | 1381 | ||
1382 | static int try_to_unmap_nonlinear(struct page *page, | 1382 | static int try_to_unmap_nonlinear(struct page *page, |
1383 | struct address_space *mapping, void *arg) | 1383 | struct address_space *mapping, void *arg) |
1384 | { | 1384 | { |
1385 | struct vm_area_struct *vma; | 1385 | struct vm_area_struct *vma; |
1386 | int ret = SWAP_AGAIN; | 1386 | int ret = SWAP_AGAIN; |
1387 | unsigned long cursor; | 1387 | unsigned long cursor; |
1388 | unsigned long max_nl_cursor = 0; | 1388 | unsigned long max_nl_cursor = 0; |
1389 | unsigned long max_nl_size = 0; | 1389 | unsigned long max_nl_size = 0; |
1390 | unsigned int mapcount; | 1390 | unsigned int mapcount; |
1391 | 1391 | ||
1392 | list_for_each_entry(vma, | 1392 | list_for_each_entry(vma, |
1393 | &mapping->i_mmap_nonlinear, shared.nonlinear) { | 1393 | &mapping->i_mmap_nonlinear, shared.nonlinear) { |
1394 | 1394 | ||
1395 | cursor = (unsigned long) vma->vm_private_data; | 1395 | cursor = (unsigned long) vma->vm_private_data; |
1396 | if (cursor > max_nl_cursor) | 1396 | if (cursor > max_nl_cursor) |
1397 | max_nl_cursor = cursor; | 1397 | max_nl_cursor = cursor; |
1398 | cursor = vma->vm_end - vma->vm_start; | 1398 | cursor = vma->vm_end - vma->vm_start; |
1399 | if (cursor > max_nl_size) | 1399 | if (cursor > max_nl_size) |
1400 | max_nl_size = cursor; | 1400 | max_nl_size = cursor; |
1401 | } | 1401 | } |
1402 | 1402 | ||
1403 | if (max_nl_size == 0) { /* all nonlinears locked or reserved ? */ | 1403 | if (max_nl_size == 0) { /* all nonlinears locked or reserved ? */ |
1404 | return SWAP_FAIL; | 1404 | return SWAP_FAIL; |
1405 | } | 1405 | } |
1406 | 1406 | ||
1407 | /* | 1407 | /* |
1408 | * We don't try to search for this page in the nonlinear vmas, | 1408 | * We don't try to search for this page in the nonlinear vmas, |
1409 | * and page_referenced wouldn't have found it anyway. Instead | 1409 | * and page_referenced wouldn't have found it anyway. Instead |
1410 | * just walk the nonlinear vmas trying to age and unmap some. | 1410 | * just walk the nonlinear vmas trying to age and unmap some. |
1411 | * The mapcount of the page we came in with is irrelevant, | 1411 | * The mapcount of the page we came in with is irrelevant, |
1412 | * but even so use it as a guide to how hard we should try? | 1412 | * but even so use it as a guide to how hard we should try? |
1413 | */ | 1413 | */ |
1414 | mapcount = page_mapcount(page); | 1414 | mapcount = page_mapcount(page); |
1415 | if (!mapcount) | 1415 | if (!mapcount) |
1416 | return ret; | 1416 | return ret; |
1417 | 1417 | ||
1418 | cond_resched(); | 1418 | cond_resched(); |
1419 | 1419 | ||
1420 | max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK; | 1420 | max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK; |
1421 | if (max_nl_cursor == 0) | 1421 | if (max_nl_cursor == 0) |
1422 | max_nl_cursor = CLUSTER_SIZE; | 1422 | max_nl_cursor = CLUSTER_SIZE; |
1423 | 1423 | ||
1424 | do { | 1424 | do { |
1425 | list_for_each_entry(vma, | 1425 | list_for_each_entry(vma, |
1426 | &mapping->i_mmap_nonlinear, shared.nonlinear) { | 1426 | &mapping->i_mmap_nonlinear, shared.nonlinear) { |
1427 | 1427 | ||
1428 | cursor = (unsigned long) vma->vm_private_data; | 1428 | cursor = (unsigned long) vma->vm_private_data; |
1429 | while (cursor < max_nl_cursor && | 1429 | while (cursor < max_nl_cursor && |
1430 | cursor < vma->vm_end - vma->vm_start) { | 1430 | cursor < vma->vm_end - vma->vm_start) { |
1431 | if (try_to_unmap_cluster(cursor, &mapcount, | 1431 | if (try_to_unmap_cluster(cursor, &mapcount, |
1432 | vma, page) == SWAP_MLOCK) | 1432 | vma, page) == SWAP_MLOCK) |
1433 | ret = SWAP_MLOCK; | 1433 | ret = SWAP_MLOCK; |
1434 | cursor += CLUSTER_SIZE; | 1434 | cursor += CLUSTER_SIZE; |
1435 | vma->vm_private_data = (void *) cursor; | 1435 | vma->vm_private_data = (void *) cursor; |
1436 | if ((int)mapcount <= 0) | 1436 | if ((int)mapcount <= 0) |
1437 | return ret; | 1437 | return ret; |
1438 | } | 1438 | } |
1439 | vma->vm_private_data = (void *) max_nl_cursor; | 1439 | vma->vm_private_data = (void *) max_nl_cursor; |
1440 | } | 1440 | } |
1441 | cond_resched(); | 1441 | cond_resched(); |
1442 | max_nl_cursor += CLUSTER_SIZE; | 1442 | max_nl_cursor += CLUSTER_SIZE; |
1443 | } while (max_nl_cursor <= max_nl_size); | 1443 | } while (max_nl_cursor <= max_nl_size); |
1444 | 1444 | ||
1445 | /* | 1445 | /* |
1446 | * Don't loop forever (perhaps all the remaining pages are | 1446 | * Don't loop forever (perhaps all the remaining pages are |
1447 | * in locked vmas). Reset cursor on all unreserved nonlinear | 1447 | * in locked vmas). Reset cursor on all unreserved nonlinear |
1448 | * vmas, now forgetting on which ones it had fallen behind. | 1448 | * vmas, now forgetting on which ones it had fallen behind. |
1449 | */ | 1449 | */ |
1450 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.nonlinear) | 1450 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.nonlinear) |
1451 | vma->vm_private_data = NULL; | 1451 | vma->vm_private_data = NULL; |
1452 | 1452 | ||
1453 | return ret; | 1453 | return ret; |
1454 | } | 1454 | } |
1455 | 1455 | ||
1456 | bool is_vma_temporary_stack(struct vm_area_struct *vma) | 1456 | bool is_vma_temporary_stack(struct vm_area_struct *vma) |
1457 | { | 1457 | { |
1458 | int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP); | 1458 | int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP); |
1459 | 1459 | ||
1460 | if (!maybe_stack) | 1460 | if (!maybe_stack) |
1461 | return false; | 1461 | return false; |
1462 | 1462 | ||
1463 | if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) == | 1463 | if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) == |
1464 | VM_STACK_INCOMPLETE_SETUP) | 1464 | VM_STACK_INCOMPLETE_SETUP) |
1465 | return true; | 1465 | return true; |
1466 | 1466 | ||
1467 | return false; | 1467 | return false; |
1468 | } | 1468 | } |
1469 | 1469 | ||
1470 | static bool invalid_migration_vma(struct vm_area_struct *vma, void *arg) | 1470 | static bool invalid_migration_vma(struct vm_area_struct *vma, void *arg) |
1471 | { | 1471 | { |
1472 | return is_vma_temporary_stack(vma); | 1472 | return is_vma_temporary_stack(vma); |
1473 | } | 1473 | } |
1474 | 1474 | ||
1475 | static int page_not_mapped(struct page *page) | 1475 | static int page_not_mapped(struct page *page) |
1476 | { | 1476 | { |
1477 | return !page_mapped(page); | 1477 | return !page_mapped(page); |
1478 | }; | 1478 | }; |
1479 | 1479 | ||
1480 | /** | 1480 | /** |
1481 | * try_to_unmap - try to remove all page table mappings to a page | 1481 | * try_to_unmap - try to remove all page table mappings to a page |
1482 | * @page: the page to get unmapped | 1482 | * @page: the page to get unmapped |
1483 | * @flags: action and flags | 1483 | * @flags: action and flags |
1484 | * | 1484 | * |
1485 | * Tries to remove all the page table entries which are mapping this | 1485 | * Tries to remove all the page table entries which are mapping this |
1486 | * page, used in the pageout path. Caller must hold the page lock. | 1486 | * page, used in the pageout path. Caller must hold the page lock. |
1487 | * Return values are: | 1487 | * Return values are: |
1488 | * | 1488 | * |
1489 | * SWAP_SUCCESS - we succeeded in removing all mappings | 1489 | * SWAP_SUCCESS - we succeeded in removing all mappings |
1490 | * SWAP_AGAIN - we missed a mapping, try again later | 1490 | * SWAP_AGAIN - we missed a mapping, try again later |
1491 | * SWAP_FAIL - the page is unswappable | 1491 | * SWAP_FAIL - the page is unswappable |
1492 | * SWAP_MLOCK - page is mlocked. | 1492 | * SWAP_MLOCK - page is mlocked. |
1493 | */ | 1493 | */ |
1494 | int try_to_unmap(struct page *page, enum ttu_flags flags) | 1494 | int try_to_unmap(struct page *page, enum ttu_flags flags) |
1495 | { | 1495 | { |
1496 | int ret; | 1496 | int ret; |
1497 | struct rmap_walk_control rwc = { | 1497 | struct rmap_walk_control rwc = { |
1498 | .rmap_one = try_to_unmap_one, | 1498 | .rmap_one = try_to_unmap_one, |
1499 | .arg = (void *)flags, | 1499 | .arg = (void *)flags, |
1500 | .done = page_not_mapped, | 1500 | .done = page_not_mapped, |
1501 | .file_nonlinear = try_to_unmap_nonlinear, | 1501 | .file_nonlinear = try_to_unmap_nonlinear, |
1502 | .anon_lock = page_lock_anon_vma_read, | 1502 | .anon_lock = page_lock_anon_vma_read, |
1503 | }; | 1503 | }; |
1504 | 1504 | ||
1505 | VM_BUG_ON_PAGE(!PageHuge(page) && PageTransHuge(page), page); | 1505 | VM_BUG_ON_PAGE(!PageHuge(page) && PageTransHuge(page), page); |
1506 | 1506 | ||
1507 | /* | 1507 | /* |
1508 | * During exec, a temporary VMA is setup and later moved. | 1508 | * During exec, a temporary VMA is setup and later moved. |
1509 | * The VMA is moved under the anon_vma lock but not the | 1509 | * The VMA is moved under the anon_vma lock but not the |
1510 | * page tables leading to a race where migration cannot | 1510 | * page tables leading to a race where migration cannot |
1511 | * find the migration ptes. Rather than increasing the | 1511 | * find the migration ptes. Rather than increasing the |
1512 | * locking requirements of exec(), migration skips | 1512 | * locking requirements of exec(), migration skips |
1513 | * temporary VMAs until after exec() completes. | 1513 | * temporary VMAs until after exec() completes. |
1514 | */ | 1514 | */ |
1515 | if (flags & TTU_MIGRATION && !PageKsm(page) && PageAnon(page)) | 1515 | if (flags & TTU_MIGRATION && !PageKsm(page) && PageAnon(page)) |
1516 | rwc.invalid_vma = invalid_migration_vma; | 1516 | rwc.invalid_vma = invalid_migration_vma; |
1517 | 1517 | ||
1518 | ret = rmap_walk(page, &rwc); | 1518 | ret = rmap_walk(page, &rwc); |
1519 | 1519 | ||
1520 | if (ret != SWAP_MLOCK && !page_mapped(page)) | 1520 | if (ret != SWAP_MLOCK && !page_mapped(page)) |
1521 | ret = SWAP_SUCCESS; | 1521 | ret = SWAP_SUCCESS; |
1522 | return ret; | 1522 | return ret; |
1523 | } | 1523 | } |
1524 | 1524 | ||
1525 | /** | 1525 | /** |
1526 | * try_to_munlock - try to munlock a page | 1526 | * try_to_munlock - try to munlock a page |
1527 | * @page: the page to be munlocked | 1527 | * @page: the page to be munlocked |
1528 | * | 1528 | * |
1529 | * Called from munlock code. Checks all of the VMAs mapping the page | 1529 | * Called from munlock code. Checks all of the VMAs mapping the page |
1530 | * to make sure nobody else has this page mlocked. The page will be | 1530 | * to make sure nobody else has this page mlocked. The page will be |
1531 | * returned with PG_mlocked cleared if no other vmas have it mlocked. | 1531 | * returned with PG_mlocked cleared if no other vmas have it mlocked. |
1532 | * | 1532 | * |
1533 | * Return values are: | 1533 | * Return values are: |
1534 | * | 1534 | * |
1535 | * SWAP_AGAIN - no vma is holding page mlocked, or, | 1535 | * SWAP_AGAIN - no vma is holding page mlocked, or, |
1536 | * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem | 1536 | * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem |
1537 | * SWAP_FAIL - page cannot be located at present | 1537 | * SWAP_FAIL - page cannot be located at present |
1538 | * SWAP_MLOCK - page is now mlocked. | 1538 | * SWAP_MLOCK - page is now mlocked. |
1539 | */ | 1539 | */ |
1540 | int try_to_munlock(struct page *page) | 1540 | int try_to_munlock(struct page *page) |
1541 | { | 1541 | { |
1542 | int ret; | 1542 | int ret; |
1543 | struct rmap_walk_control rwc = { | 1543 | struct rmap_walk_control rwc = { |
1544 | .rmap_one = try_to_unmap_one, | 1544 | .rmap_one = try_to_unmap_one, |
1545 | .arg = (void *)TTU_MUNLOCK, | 1545 | .arg = (void *)TTU_MUNLOCK, |
1546 | .done = page_not_mapped, | 1546 | .done = page_not_mapped, |
1547 | /* | 1547 | /* |
1548 | * We don't bother to try to find the munlocked page in | 1548 | * We don't bother to try to find the munlocked page in |
1549 | * nonlinears. It's costly. Instead, later, page reclaim logic | 1549 | * nonlinears. It's costly. Instead, later, page reclaim logic |
1550 | * may call try_to_unmap() and recover PG_mlocked lazily. | 1550 | * may call try_to_unmap() and recover PG_mlocked lazily. |
1551 | */ | 1551 | */ |
1552 | .file_nonlinear = NULL, | 1552 | .file_nonlinear = NULL, |
1553 | .anon_lock = page_lock_anon_vma_read, | 1553 | .anon_lock = page_lock_anon_vma_read, |
1554 | 1554 | ||
1555 | }; | 1555 | }; |
1556 | 1556 | ||
1557 | VM_BUG_ON_PAGE(!PageLocked(page) || PageLRU(page), page); | 1557 | VM_BUG_ON_PAGE(!PageLocked(page) || PageLRU(page), page); |
1558 | 1558 | ||
1559 | ret = rmap_walk(page, &rwc); | 1559 | ret = rmap_walk(page, &rwc); |
1560 | return ret; | 1560 | return ret; |
1561 | } | 1561 | } |
1562 | 1562 | ||
1563 | void __put_anon_vma(struct anon_vma *anon_vma) | 1563 | void __put_anon_vma(struct anon_vma *anon_vma) |
1564 | { | 1564 | { |
1565 | struct anon_vma *root = anon_vma->root; | 1565 | struct anon_vma *root = anon_vma->root; |
1566 | 1566 | ||
1567 | anon_vma_free(anon_vma); | ||
1567 | if (root != anon_vma && atomic_dec_and_test(&root->refcount)) | 1568 | if (root != anon_vma && atomic_dec_and_test(&root->refcount)) |
1568 | anon_vma_free(root); | 1569 | anon_vma_free(root); |
1569 | |||
1570 | anon_vma_free(anon_vma); | ||
1571 | } | 1570 | } |
1572 | 1571 | ||
1573 | static struct anon_vma *rmap_walk_anon_lock(struct page *page, | 1572 | static struct anon_vma *rmap_walk_anon_lock(struct page *page, |
1574 | struct rmap_walk_control *rwc) | 1573 | struct rmap_walk_control *rwc) |
1575 | { | 1574 | { |
1576 | struct anon_vma *anon_vma; | 1575 | struct anon_vma *anon_vma; |
1577 | 1576 | ||
1578 | if (rwc->anon_lock) | 1577 | if (rwc->anon_lock) |
1579 | return rwc->anon_lock(page); | 1578 | return rwc->anon_lock(page); |
1580 | 1579 | ||
1581 | /* | 1580 | /* |
1582 | * Note: remove_migration_ptes() cannot use page_lock_anon_vma_read() | 1581 | * Note: remove_migration_ptes() cannot use page_lock_anon_vma_read() |
1583 | * because that depends on page_mapped(); but not all its usages | 1582 | * because that depends on page_mapped(); but not all its usages |
1584 | * are holding mmap_sem. Users without mmap_sem are required to | 1583 | * are holding mmap_sem. Users without mmap_sem are required to |
1585 | * take a reference count to prevent the anon_vma disappearing | 1584 | * take a reference count to prevent the anon_vma disappearing |
1586 | */ | 1585 | */ |
1587 | anon_vma = page_anon_vma(page); | 1586 | anon_vma = page_anon_vma(page); |
1588 | if (!anon_vma) | 1587 | if (!anon_vma) |
1589 | return NULL; | 1588 | return NULL; |
1590 | 1589 | ||
1591 | anon_vma_lock_read(anon_vma); | 1590 | anon_vma_lock_read(anon_vma); |
1592 | return anon_vma; | 1591 | return anon_vma; |
1593 | } | 1592 | } |
1594 | 1593 | ||
1595 | /* | 1594 | /* |
1596 | * rmap_walk_anon - do something to anonymous page using the object-based | 1595 | * rmap_walk_anon - do something to anonymous page using the object-based |
1597 | * rmap method | 1596 | * rmap method |
1598 | * @page: the page to be handled | 1597 | * @page: the page to be handled |
1599 | * @rwc: control variable according to each walk type | 1598 | * @rwc: control variable according to each walk type |
1600 | * | 1599 | * |
1601 | * Find all the mappings of a page using the mapping pointer and the vma chains | 1600 | * Find all the mappings of a page using the mapping pointer and the vma chains |
1602 | * contained in the anon_vma struct it points to. | 1601 | * contained in the anon_vma struct it points to. |
1603 | * | 1602 | * |
1604 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma | 1603 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma |
1605 | * where the page was found will be held for write. So, we won't recheck | 1604 | * where the page was found will be held for write. So, we won't recheck |
1606 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be | 1605 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be |
1607 | * LOCKED. | 1606 | * LOCKED. |
1608 | */ | 1607 | */ |
1609 | static int rmap_walk_anon(struct page *page, struct rmap_walk_control *rwc) | 1608 | static int rmap_walk_anon(struct page *page, struct rmap_walk_control *rwc) |
1610 | { | 1609 | { |
1611 | struct anon_vma *anon_vma; | 1610 | struct anon_vma *anon_vma; |
1612 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | 1611 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); |
1613 | struct anon_vma_chain *avc; | 1612 | struct anon_vma_chain *avc; |
1614 | int ret = SWAP_AGAIN; | 1613 | int ret = SWAP_AGAIN; |
1615 | 1614 | ||
1616 | anon_vma = rmap_walk_anon_lock(page, rwc); | 1615 | anon_vma = rmap_walk_anon_lock(page, rwc); |
1617 | if (!anon_vma) | 1616 | if (!anon_vma) |
1618 | return ret; | 1617 | return ret; |
1619 | 1618 | ||
1620 | anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) { | 1619 | anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) { |
1621 | struct vm_area_struct *vma = avc->vma; | 1620 | struct vm_area_struct *vma = avc->vma; |
1622 | unsigned long address = vma_address(page, vma); | 1621 | unsigned long address = vma_address(page, vma); |
1623 | 1622 | ||
1624 | if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg)) | 1623 | if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg)) |
1625 | continue; | 1624 | continue; |
1626 | 1625 | ||
1627 | ret = rwc->rmap_one(page, vma, address, rwc->arg); | 1626 | ret = rwc->rmap_one(page, vma, address, rwc->arg); |
1628 | if (ret != SWAP_AGAIN) | 1627 | if (ret != SWAP_AGAIN) |
1629 | break; | 1628 | break; |
1630 | if (rwc->done && rwc->done(page)) | 1629 | if (rwc->done && rwc->done(page)) |
1631 | break; | 1630 | break; |
1632 | } | 1631 | } |
1633 | anon_vma_unlock_read(anon_vma); | 1632 | anon_vma_unlock_read(anon_vma); |
1634 | return ret; | 1633 | return ret; |
1635 | } | 1634 | } |
1636 | 1635 | ||
1637 | /* | 1636 | /* |
1638 | * rmap_walk_file - do something to file page using the object-based rmap method | 1637 | * rmap_walk_file - do something to file page using the object-based rmap method |
1639 | * @page: the page to be handled | 1638 | * @page: the page to be handled |
1640 | * @rwc: control variable according to each walk type | 1639 | * @rwc: control variable according to each walk type |
1641 | * | 1640 | * |
1642 | * Find all the mappings of a page using the mapping pointer and the vma chains | 1641 | * Find all the mappings of a page using the mapping pointer and the vma chains |
1643 | * contained in the address_space struct it points to. | 1642 | * contained in the address_space struct it points to. |
1644 | * | 1643 | * |
1645 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma | 1644 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma |
1646 | * where the page was found will be held for write. So, we won't recheck | 1645 | * where the page was found will be held for write. So, we won't recheck |
1647 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be | 1646 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be |
1648 | * LOCKED. | 1647 | * LOCKED. |
1649 | */ | 1648 | */ |
1650 | static int rmap_walk_file(struct page *page, struct rmap_walk_control *rwc) | 1649 | static int rmap_walk_file(struct page *page, struct rmap_walk_control *rwc) |
1651 | { | 1650 | { |
1652 | struct address_space *mapping = page->mapping; | 1651 | struct address_space *mapping = page->mapping; |
1653 | pgoff_t pgoff = page->index << compound_order(page); | 1652 | pgoff_t pgoff = page->index << compound_order(page); |
1654 | struct vm_area_struct *vma; | 1653 | struct vm_area_struct *vma; |
1655 | int ret = SWAP_AGAIN; | 1654 | int ret = SWAP_AGAIN; |
1656 | 1655 | ||
1657 | /* | 1656 | /* |
1658 | * The page lock not only makes sure that page->mapping cannot | 1657 | * The page lock not only makes sure that page->mapping cannot |
1659 | * suddenly be NULLified by truncation, it makes sure that the | 1658 | * suddenly be NULLified by truncation, it makes sure that the |
1660 | * structure at mapping cannot be freed and reused yet, | 1659 | * structure at mapping cannot be freed and reused yet, |
1661 | * so we can safely take mapping->i_mmap_mutex. | 1660 | * so we can safely take mapping->i_mmap_mutex. |
1662 | */ | 1661 | */ |
1663 | VM_BUG_ON(!PageLocked(page)); | 1662 | VM_BUG_ON(!PageLocked(page)); |
1664 | 1663 | ||
1665 | if (!mapping) | 1664 | if (!mapping) |
1666 | return ret; | 1665 | return ret; |
1667 | mutex_lock(&mapping->i_mmap_mutex); | 1666 | mutex_lock(&mapping->i_mmap_mutex); |
1668 | vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) { | 1667 | vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) { |
1669 | unsigned long address = vma_address(page, vma); | 1668 | unsigned long address = vma_address(page, vma); |
1670 | 1669 | ||
1671 | if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg)) | 1670 | if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg)) |
1672 | continue; | 1671 | continue; |
1673 | 1672 | ||
1674 | ret = rwc->rmap_one(page, vma, address, rwc->arg); | 1673 | ret = rwc->rmap_one(page, vma, address, rwc->arg); |
1675 | if (ret != SWAP_AGAIN) | 1674 | if (ret != SWAP_AGAIN) |
1676 | goto done; | 1675 | goto done; |
1677 | if (rwc->done && rwc->done(page)) | 1676 | if (rwc->done && rwc->done(page)) |
1678 | goto done; | 1677 | goto done; |
1679 | } | 1678 | } |
1680 | 1679 | ||
1681 | if (!rwc->file_nonlinear) | 1680 | if (!rwc->file_nonlinear) |
1682 | goto done; | 1681 | goto done; |
1683 | 1682 | ||
1684 | if (list_empty(&mapping->i_mmap_nonlinear)) | 1683 | if (list_empty(&mapping->i_mmap_nonlinear)) |
1685 | goto done; | 1684 | goto done; |
1686 | 1685 | ||
1687 | ret = rwc->file_nonlinear(page, mapping, rwc->arg); | 1686 | ret = rwc->file_nonlinear(page, mapping, rwc->arg); |
1688 | 1687 | ||
1689 | done: | 1688 | done: |
1690 | mutex_unlock(&mapping->i_mmap_mutex); | 1689 | mutex_unlock(&mapping->i_mmap_mutex); |
1691 | return ret; | 1690 | return ret; |
1692 | } | 1691 | } |
1693 | 1692 | ||
1694 | int rmap_walk(struct page *page, struct rmap_walk_control *rwc) | 1693 | int rmap_walk(struct page *page, struct rmap_walk_control *rwc) |
1695 | { | 1694 | { |
1696 | if (unlikely(PageKsm(page))) | 1695 | if (unlikely(PageKsm(page))) |
1697 | return rmap_walk_ksm(page, rwc); | 1696 | return rmap_walk_ksm(page, rwc); |
1698 | else if (PageAnon(page)) | 1697 | else if (PageAnon(page)) |
1699 | return rmap_walk_anon(page, rwc); | 1698 | return rmap_walk_anon(page, rwc); |
1700 | else | 1699 | else |
1701 | return rmap_walk_file(page, rwc); | 1700 | return rmap_walk_file(page, rwc); |
1702 | } | 1701 | } |
1703 | 1702 | ||
1704 | #ifdef CONFIG_HUGETLB_PAGE | 1703 | #ifdef CONFIG_HUGETLB_PAGE |
1705 | /* | 1704 | /* |
1706 | * The following three functions are for anonymous (private mapped) hugepages. | 1705 | * The following three functions are for anonymous (private mapped) hugepages. |
1707 | * Unlike common anonymous pages, anonymous hugepages have no accounting code | 1706 | * Unlike common anonymous pages, anonymous hugepages have no accounting code |
1708 | * and no lru code, because we handle hugepages differently from common pages. | 1707 | * and no lru code, because we handle hugepages differently from common pages. |
1709 | */ | 1708 | */ |
1710 | static void __hugepage_set_anon_rmap(struct page *page, | 1709 | static void __hugepage_set_anon_rmap(struct page *page, |
1711 | struct vm_area_struct *vma, unsigned long address, int exclusive) | 1710 | struct vm_area_struct *vma, unsigned long address, int exclusive) |
1712 | { | 1711 | { |
1713 | struct anon_vma *anon_vma = vma->anon_vma; | 1712 | struct anon_vma *anon_vma = vma->anon_vma; |
1714 | 1713 | ||
1715 | BUG_ON(!anon_vma); | 1714 | BUG_ON(!anon_vma); |
1716 | 1715 | ||
1717 | if (PageAnon(page)) | 1716 | if (PageAnon(page)) |
1718 | return; | 1717 | return; |
1719 | if (!exclusive) | 1718 | if (!exclusive) |
1720 | anon_vma = anon_vma->root; | 1719 | anon_vma = anon_vma->root; |
1721 | 1720 | ||
1722 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; | 1721 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; |
1723 | page->mapping = (struct address_space *) anon_vma; | 1722 | page->mapping = (struct address_space *) anon_vma; |
1724 | page->index = linear_page_index(vma, address); | 1723 | page->index = linear_page_index(vma, address); |
1725 | } | 1724 | } |
1726 | 1725 | ||
1727 | void hugepage_add_anon_rmap(struct page *page, | 1726 | void hugepage_add_anon_rmap(struct page *page, |
1728 | struct vm_area_struct *vma, unsigned long address) | 1727 | struct vm_area_struct *vma, unsigned long address) |
1729 | { | 1728 | { |
1730 | struct anon_vma *anon_vma = vma->anon_vma; | 1729 | struct anon_vma *anon_vma = vma->anon_vma; |
1731 | int first; | 1730 | int first; |
1732 | 1731 | ||
1733 | BUG_ON(!PageLocked(page)); | 1732 | BUG_ON(!PageLocked(page)); |
1734 | BUG_ON(!anon_vma); | 1733 | BUG_ON(!anon_vma); |
1735 | /* address might be in next vma when migration races vma_adjust */ | 1734 | /* address might be in next vma when migration races vma_adjust */ |
1736 | first = atomic_inc_and_test(&page->_mapcount); | 1735 | first = atomic_inc_and_test(&page->_mapcount); |
1737 | if (first) | 1736 | if (first) |
1738 | __hugepage_set_anon_rmap(page, vma, address, 0); | 1737 | __hugepage_set_anon_rmap(page, vma, address, 0); |
1739 | } | 1738 | } |
1740 | 1739 | ||
1741 | void hugepage_add_new_anon_rmap(struct page *page, | 1740 | void hugepage_add_new_anon_rmap(struct page *page, |
1742 | struct vm_area_struct *vma, unsigned long address) | 1741 | struct vm_area_struct *vma, unsigned long address) |
1743 | { | 1742 | { |
1744 | BUG_ON(address < vma->vm_start || address >= vma->vm_end); | 1743 | BUG_ON(address < vma->vm_start || address >= vma->vm_end); |
1745 | atomic_set(&page->_mapcount, 0); | 1744 | atomic_set(&page->_mapcount, 0); |
1746 | __hugepage_set_anon_rmap(page, vma, address, 1); | 1745 | __hugepage_set_anon_rmap(page, vma, address, 1); |
1747 | } | 1746 | } |
1748 | #endif /* CONFIG_HUGETLB_PAGE */ | 1747 | #endif /* CONFIG_HUGETLB_PAGE */ |