Doug / smarc-fsl-linux-kernel

Download as
Browse Code ยป

Commit a850ea30374ebed32a0724742601861853fde869

Authored by Naoya Horiguchi
Committed by Linus Torvalds
1 parent 56c9cfb13c
Exists in master and in 7 other branches imx_3.0.35_4.1.0, imx_3.10.17_1.0.1_ga, imx_3.10.53_1.1.0_ga, imx_3.14.28_1.0.0_ga, smarc-imx_3.10.53_1.1.0_ga, smarc-imx_3.14.28_1.0.0_ga, smarc-l5.0.0_1.0.0-ga

hugetlb, rmap: add BUG_ON(!PageLocked) in hugetlb_add_anon_rmap() 

Confirming page lock is held in hugetlb_add_anon_rmap() may be useful
to detect possible future problems.

Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Acked-by: Rik van Riel <riel@redhat.com>
Acked-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Showing 1 changed file with 2 additions and 0 deletions Inline Diff

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 * inode->i_alloc_sem (vmtruncate_range) 24 * inode->i_alloc_sem (vmtruncate_range)
25 * mm->mmap_sem 25 * mm->mmap_sem
26 * page->flags PG_locked (lock_page) 26 * page->flags PG_locked (lock_page)
27 * mapping->i_mmap_lock 27 * mapping->i_mmap_lock
28 * anon_vma->lock 28 * anon_vma->lock
29 * mm->page_table_lock or pte_lock 29 * mm->page_table_lock or pte_lock
30 * zone->lru_lock (in mark_page_accessed, isolate_lru_page) 30 * zone->lru_lock (in mark_page_accessed, isolate_lru_page)
31 * swap_lock (in swap_duplicate, swap_info_get) 31 * swap_lock (in swap_duplicate, swap_info_get)
32 * mmlist_lock (in mmput, drain_mmlist and others) 32 * mmlist_lock (in mmput, drain_mmlist and others)
33 * mapping->private_lock (in __set_page_dirty_buffers) 33 * mapping->private_lock (in __set_page_dirty_buffers)
34 * inode_lock (in set_page_dirty's __mark_inode_dirty) 34 * inode_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 inode_lock in __sync_single_inode) 38 * within inode_lock in __sync_single_inode)
39 * 39 *
40 * (code doesn't rely on that order so it could be switched around) 40 * (code doesn't rely on that order so it could be switched around)
41 * ->tasklist_lock 41 * ->tasklist_lock
42 * anon_vma->lock (memory_failure, collect_procs_anon) 42 * anon_vma->lock (memory_failure, collect_procs_anon)
43 * pte map lock 43 * pte map lock
44 */ 44 */
45 45
46 #include <linux/mm.h> 46 #include <linux/mm.h>
47 #include <linux/pagemap.h> 47 #include <linux/pagemap.h>
48 #include <linux/swap.h> 48 #include <linux/swap.h>
49 #include <linux/swapops.h> 49 #include <linux/swapops.h>
50 #include <linux/slab.h> 50 #include <linux/slab.h>
51 #include <linux/init.h> 51 #include <linux/init.h>
52 #include <linux/ksm.h> 52 #include <linux/ksm.h>
53 #include <linux/rmap.h> 53 #include <linux/rmap.h>
54 #include <linux/rcupdate.h> 54 #include <linux/rcupdate.h>
55 #include <linux/module.h> 55 #include <linux/module.h>
56 #include <linux/memcontrol.h> 56 #include <linux/memcontrol.h>
57 #include <linux/mmu_notifier.h> 57 #include <linux/mmu_notifier.h>
58 #include <linux/migrate.h> 58 #include <linux/migrate.h>
59 #include <linux/hugetlb.h> 59 #include <linux/hugetlb.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 return kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL); 70 return kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL);
71 } 71 }
72 72
73 void anon_vma_free(struct anon_vma *anon_vma) 73 void anon_vma_free(struct anon_vma *anon_vma)
74 { 74 {
75 kmem_cache_free(anon_vma_cachep, anon_vma); 75 kmem_cache_free(anon_vma_cachep, anon_vma);
76 } 76 }
77 77
78 static inline struct anon_vma_chain *anon_vma_chain_alloc(void) 78 static inline struct anon_vma_chain *anon_vma_chain_alloc(void)
79 { 79 {
80 return kmem_cache_alloc(anon_vma_chain_cachep, GFP_KERNEL); 80 return kmem_cache_alloc(anon_vma_chain_cachep, GFP_KERNEL);
81 } 81 }
82 82
83 void anon_vma_chain_free(struct anon_vma_chain *anon_vma_chain) 83 void anon_vma_chain_free(struct anon_vma_chain *anon_vma_chain)
84 { 84 {
85 kmem_cache_free(anon_vma_chain_cachep, anon_vma_chain); 85 kmem_cache_free(anon_vma_chain_cachep, anon_vma_chain);
86 } 86 }
87 87
88 /** 88 /**
89 * anon_vma_prepare - attach an anon_vma to a memory region 89 * anon_vma_prepare - attach an anon_vma to a memory region
90 * @vma: the memory region in question 90 * @vma: the memory region in question
91 * 91 *
92 * This makes sure the memory mapping described by 'vma' has 92 * This makes sure the memory mapping described by 'vma' has
93 * an 'anon_vma' attached to it, so that we can associate the 93 * an 'anon_vma' attached to it, so that we can associate the
94 * anonymous pages mapped into it with that anon_vma. 94 * anonymous pages mapped into it with that anon_vma.
95 * 95 *
96 * The common case will be that we already have one, but if 96 * The common case will be that we already have one, but if
97 * if not we either need to find an adjacent mapping that we 97 * if not we either need to find an adjacent mapping that we
98 * can re-use the anon_vma from (very common when the only 98 * can re-use the anon_vma from (very common when the only
99 * reason for splitting a vma has been mprotect()), or we 99 * reason for splitting a vma has been mprotect()), or we
100 * allocate a new one. 100 * allocate a new one.
101 * 101 *
102 * Anon-vma allocations are very subtle, because we may have 102 * Anon-vma allocations are very subtle, because we may have
103 * optimistically looked up an anon_vma in page_lock_anon_vma() 103 * optimistically looked up an anon_vma in page_lock_anon_vma()
104 * and that may actually touch the spinlock even in the newly 104 * and that may actually touch the spinlock even in the newly
105 * allocated vma (it depends on RCU to make sure that the 105 * allocated vma (it depends on RCU to make sure that the
106 * anon_vma isn't actually destroyed). 106 * anon_vma isn't actually destroyed).
107 * 107 *
108 * As a result, we need to do proper anon_vma locking even 108 * As a result, we need to do proper anon_vma locking even
109 * for the new allocation. At the same time, we do not want 109 * for the new allocation. At the same time, we do not want
110 * to do any locking for the common case of already having 110 * to do any locking for the common case of already having
111 * an anon_vma. 111 * an anon_vma.
112 * 112 *
113 * This must be called with the mmap_sem held for reading. 113 * This must be called with the mmap_sem held for reading.
114 */ 114 */
115 int anon_vma_prepare(struct vm_area_struct *vma) 115 int anon_vma_prepare(struct vm_area_struct *vma)
116 { 116 {
117 struct anon_vma *anon_vma = vma->anon_vma; 117 struct anon_vma *anon_vma = vma->anon_vma;
118 struct anon_vma_chain *avc; 118 struct anon_vma_chain *avc;
119 119
120 might_sleep(); 120 might_sleep();
121 if (unlikely(!anon_vma)) { 121 if (unlikely(!anon_vma)) {
122 struct mm_struct *mm = vma->vm_mm; 122 struct mm_struct *mm = vma->vm_mm;
123 struct anon_vma *allocated; 123 struct anon_vma *allocated;
124 124
125 avc = anon_vma_chain_alloc(); 125 avc = anon_vma_chain_alloc();
126 if (!avc) 126 if (!avc)
127 goto out_enomem; 127 goto out_enomem;
128 128
129 anon_vma = find_mergeable_anon_vma(vma); 129 anon_vma = find_mergeable_anon_vma(vma);
130 allocated = NULL; 130 allocated = NULL;
131 if (!anon_vma) { 131 if (!anon_vma) {
132 anon_vma = anon_vma_alloc(); 132 anon_vma = anon_vma_alloc();
133 if (unlikely(!anon_vma)) 133 if (unlikely(!anon_vma))
134 goto out_enomem_free_avc; 134 goto out_enomem_free_avc;
135 allocated = anon_vma; 135 allocated = anon_vma;
136 /* 136 /*
137 * This VMA had no anon_vma yet. This anon_vma is 137 * This VMA had no anon_vma yet. This anon_vma is
138 * the root of any anon_vma tree that might form. 138 * the root of any anon_vma tree that might form.
139 */ 139 */
140 anon_vma->root = anon_vma; 140 anon_vma->root = anon_vma;
141 } 141 }
142 142
143 anon_vma_lock(anon_vma); 143 anon_vma_lock(anon_vma);
144 /* page_table_lock to protect against threads */ 144 /* page_table_lock to protect against threads */
145 spin_lock(&mm->page_table_lock); 145 spin_lock(&mm->page_table_lock);
146 if (likely(!vma->anon_vma)) { 146 if (likely(!vma->anon_vma)) {
147 vma->anon_vma = anon_vma; 147 vma->anon_vma = anon_vma;
148 avc->anon_vma = anon_vma; 148 avc->anon_vma = anon_vma;
149 avc->vma = vma; 149 avc->vma = vma;
150 list_add(&avc->same_vma, &vma->anon_vma_chain); 150 list_add(&avc->same_vma, &vma->anon_vma_chain);
151 list_add_tail(&avc->same_anon_vma, &anon_vma->head); 151 list_add_tail(&avc->same_anon_vma, &anon_vma->head);
152 allocated = NULL; 152 allocated = NULL;
153 avc = NULL; 153 avc = NULL;
154 } 154 }
155 spin_unlock(&mm->page_table_lock); 155 spin_unlock(&mm->page_table_lock);
156 anon_vma_unlock(anon_vma); 156 anon_vma_unlock(anon_vma);
157 157
158 if (unlikely(allocated)) 158 if (unlikely(allocated))
159 anon_vma_free(allocated); 159 anon_vma_free(allocated);
160 if (unlikely(avc)) 160 if (unlikely(avc))
161 anon_vma_chain_free(avc); 161 anon_vma_chain_free(avc);
162 } 162 }
163 return 0; 163 return 0;
164 164
165 out_enomem_free_avc: 165 out_enomem_free_avc:
166 anon_vma_chain_free(avc); 166 anon_vma_chain_free(avc);
167 out_enomem: 167 out_enomem:
168 return -ENOMEM; 168 return -ENOMEM;
169 } 169 }
170 170
171 static void anon_vma_chain_link(struct vm_area_struct *vma, 171 static void anon_vma_chain_link(struct vm_area_struct *vma,
172 struct anon_vma_chain *avc, 172 struct anon_vma_chain *avc,
173 struct anon_vma *anon_vma) 173 struct anon_vma *anon_vma)
174 { 174 {
175 avc->vma = vma; 175 avc->vma = vma;
176 avc->anon_vma = anon_vma; 176 avc->anon_vma = anon_vma;
177 list_add(&avc->same_vma, &vma->anon_vma_chain); 177 list_add(&avc->same_vma, &vma->anon_vma_chain);
178 178
179 anon_vma_lock(anon_vma); 179 anon_vma_lock(anon_vma);
180 list_add_tail(&avc->same_anon_vma, &anon_vma->head); 180 list_add_tail(&avc->same_anon_vma, &anon_vma->head);
181 anon_vma_unlock(anon_vma); 181 anon_vma_unlock(anon_vma);
182 } 182 }
183 183
184 /* 184 /*
185 * Attach the anon_vmas from src to dst. 185 * Attach the anon_vmas from src to dst.
186 * Returns 0 on success, -ENOMEM on failure. 186 * Returns 0 on success, -ENOMEM on failure.
187 */ 187 */
188 int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src) 188 int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src)
189 { 189 {
190 struct anon_vma_chain *avc, *pavc; 190 struct anon_vma_chain *avc, *pavc;
191 191
192 list_for_each_entry_reverse(pavc, &src->anon_vma_chain, same_vma) { 192 list_for_each_entry_reverse(pavc, &src->anon_vma_chain, same_vma) {
193 avc = anon_vma_chain_alloc(); 193 avc = anon_vma_chain_alloc();
194 if (!avc) 194 if (!avc)
195 goto enomem_failure; 195 goto enomem_failure;
196 anon_vma_chain_link(dst, avc, pavc->anon_vma); 196 anon_vma_chain_link(dst, avc, pavc->anon_vma);
197 } 197 }
198 return 0; 198 return 0;
199 199
200 enomem_failure: 200 enomem_failure:
201 unlink_anon_vmas(dst); 201 unlink_anon_vmas(dst);
202 return -ENOMEM; 202 return -ENOMEM;
203 } 203 }
204 204
205 /* 205 /*
206 * Attach vma to its own anon_vma, as well as to the anon_vmas that 206 * Attach vma to its own anon_vma, as well as to the anon_vmas that
207 * the corresponding VMA in the parent process is attached to. 207 * the corresponding VMA in the parent process is attached to.
208 * Returns 0 on success, non-zero on failure. 208 * Returns 0 on success, non-zero on failure.
209 */ 209 */
210 int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma) 210 int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma)
211 { 211 {
212 struct anon_vma_chain *avc; 212 struct anon_vma_chain *avc;
213 struct anon_vma *anon_vma; 213 struct anon_vma *anon_vma;
214 214
215 /* Don't bother if the parent process has no anon_vma here. */ 215 /* Don't bother if the parent process has no anon_vma here. */
216 if (!pvma->anon_vma) 216 if (!pvma->anon_vma)
217 return 0; 217 return 0;
218 218
219 /* 219 /*
220 * First, attach the new VMA to the parent VMA's anon_vmas, 220 * First, attach the new VMA to the parent VMA's anon_vmas,
221 * so rmap can find non-COWed pages in child processes. 221 * so rmap can find non-COWed pages in child processes.
222 */ 222 */
223 if (anon_vma_clone(vma, pvma)) 223 if (anon_vma_clone(vma, pvma))
224 return -ENOMEM; 224 return -ENOMEM;
225 225
226 /* Then add our own anon_vma. */ 226 /* Then add our own anon_vma. */
227 anon_vma = anon_vma_alloc(); 227 anon_vma = anon_vma_alloc();
228 if (!anon_vma) 228 if (!anon_vma)
229 goto out_error; 229 goto out_error;
230 avc = anon_vma_chain_alloc(); 230 avc = anon_vma_chain_alloc();
231 if (!avc) 231 if (!avc)
232 goto out_error_free_anon_vma; 232 goto out_error_free_anon_vma;
233 233
234 /* 234 /*
235 * The root anon_vma's spinlock is the lock actually used when we 235 * The root anon_vma's spinlock is the lock actually used when we
236 * lock any of the anon_vmas in this anon_vma tree. 236 * lock any of the anon_vmas in this anon_vma tree.
237 */ 237 */
238 anon_vma->root = pvma->anon_vma->root; 238 anon_vma->root = pvma->anon_vma->root;
239 /* 239 /*
240 * With KSM refcounts, an anon_vma can stay around longer than the 240 * With KSM refcounts, an anon_vma can stay around longer than the
241 * process it belongs to. The root anon_vma needs to be pinned 241 * process it belongs to. The root anon_vma needs to be pinned
242 * until this anon_vma is freed, because the lock lives in the root. 242 * until this anon_vma is freed, because the lock lives in the root.
243 */ 243 */
244 get_anon_vma(anon_vma->root); 244 get_anon_vma(anon_vma->root);
245 /* Mark this anon_vma as the one where our new (COWed) pages go. */ 245 /* Mark this anon_vma as the one where our new (COWed) pages go. */
246 vma->anon_vma = anon_vma; 246 vma->anon_vma = anon_vma;
247 anon_vma_chain_link(vma, avc, anon_vma); 247 anon_vma_chain_link(vma, avc, anon_vma);
248 248
249 return 0; 249 return 0;
250 250
251 out_error_free_anon_vma: 251 out_error_free_anon_vma:
252 anon_vma_free(anon_vma); 252 anon_vma_free(anon_vma);
253 out_error: 253 out_error:
254 unlink_anon_vmas(vma); 254 unlink_anon_vmas(vma);
255 return -ENOMEM; 255 return -ENOMEM;
256 } 256 }
257 257
258 static void anon_vma_unlink(struct anon_vma_chain *anon_vma_chain) 258 static void anon_vma_unlink(struct anon_vma_chain *anon_vma_chain)
259 { 259 {
260 struct anon_vma *anon_vma = anon_vma_chain->anon_vma; 260 struct anon_vma *anon_vma = anon_vma_chain->anon_vma;
261 int empty; 261 int empty;
262 262
263 /* If anon_vma_fork fails, we can get an empty anon_vma_chain. */ 263 /* If anon_vma_fork fails, we can get an empty anon_vma_chain. */
264 if (!anon_vma) 264 if (!anon_vma)
265 return; 265 return;
266 266
267 anon_vma_lock(anon_vma); 267 anon_vma_lock(anon_vma);
268 list_del(&anon_vma_chain->same_anon_vma); 268 list_del(&anon_vma_chain->same_anon_vma);
269 269
270 /* We must garbage collect the anon_vma if it's empty */ 270 /* We must garbage collect the anon_vma if it's empty */
271 empty = list_empty(&anon_vma->head) && !anonvma_external_refcount(anon_vma); 271 empty = list_empty(&anon_vma->head) && !anonvma_external_refcount(anon_vma);
272 anon_vma_unlock(anon_vma); 272 anon_vma_unlock(anon_vma);
273 273
274 if (empty) { 274 if (empty) {
275 /* We no longer need the root anon_vma */ 275 /* We no longer need the root anon_vma */
276 if (anon_vma->root != anon_vma) 276 if (anon_vma->root != anon_vma)
277 drop_anon_vma(anon_vma->root); 277 drop_anon_vma(anon_vma->root);
278 anon_vma_free(anon_vma); 278 anon_vma_free(anon_vma);
279 } 279 }
280 } 280 }
281 281
282 void unlink_anon_vmas(struct vm_area_struct *vma) 282 void unlink_anon_vmas(struct vm_area_struct *vma)
283 { 283 {
284 struct anon_vma_chain *avc, *next; 284 struct anon_vma_chain *avc, *next;
285 285
286 /* 286 /*
287 * Unlink each anon_vma chained to the VMA. This list is ordered 287 * Unlink each anon_vma chained to the VMA. This list is ordered
288 * from newest to oldest, ensuring the root anon_vma gets freed last. 288 * from newest to oldest, ensuring the root anon_vma gets freed last.
289 */ 289 */
290 list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) { 290 list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) {
291 anon_vma_unlink(avc); 291 anon_vma_unlink(avc);
292 list_del(&avc->same_vma); 292 list_del(&avc->same_vma);
293 anon_vma_chain_free(avc); 293 anon_vma_chain_free(avc);
294 } 294 }
295 } 295 }
296 296
297 static void anon_vma_ctor(void *data) 297 static void anon_vma_ctor(void *data)
298 { 298 {
299 struct anon_vma *anon_vma = data; 299 struct anon_vma *anon_vma = data;
300 300
301 spin_lock_init(&anon_vma->lock); 301 spin_lock_init(&anon_vma->lock);
302 anonvma_external_refcount_init(anon_vma); 302 anonvma_external_refcount_init(anon_vma);
303 INIT_LIST_HEAD(&anon_vma->head); 303 INIT_LIST_HEAD(&anon_vma->head);
304 } 304 }
305 305
306 void __init anon_vma_init(void) 306 void __init anon_vma_init(void)
307 { 307 {
308 anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma), 308 anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma),
309 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor); 309 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor);
310 anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain, SLAB_PANIC); 310 anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain, SLAB_PANIC);
311 } 311 }
312 312
313 /* 313 /*
314 * Getting a lock on a stable anon_vma from a page off the LRU is 314 * Getting a lock on a stable anon_vma from a page off the LRU is
315 * tricky: page_lock_anon_vma rely on RCU to guard against the races. 315 * tricky: page_lock_anon_vma rely on RCU to guard against the races.
316 */ 316 */
317 struct anon_vma *page_lock_anon_vma(struct page *page) 317 struct anon_vma *page_lock_anon_vma(struct page *page)
318 { 318 {
319 struct anon_vma *anon_vma, *root_anon_vma; 319 struct anon_vma *anon_vma, *root_anon_vma;
320 unsigned long anon_mapping; 320 unsigned long anon_mapping;
321 321
322 rcu_read_lock(); 322 rcu_read_lock();
323 anon_mapping = (unsigned long) ACCESS_ONCE(page->mapping); 323 anon_mapping = (unsigned long) ACCESS_ONCE(page->mapping);
324 if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON) 324 if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
325 goto out; 325 goto out;
326 if (!page_mapped(page)) 326 if (!page_mapped(page))
327 goto out; 327 goto out;
328 328
329 anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON); 329 anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
330 root_anon_vma = ACCESS_ONCE(anon_vma->root); 330 root_anon_vma = ACCESS_ONCE(anon_vma->root);
331 spin_lock(&root_anon_vma->lock); 331 spin_lock(&root_anon_vma->lock);
332 332
333 /* 333 /*
334 * If this page is still mapped, then its anon_vma cannot have been 334 * If this page is still mapped, then its anon_vma cannot have been
335 * freed. But if it has been unmapped, we have no security against 335 * freed. But if it has been unmapped, we have no security against
336 * the anon_vma structure being freed and reused (for another anon_vma: 336 * the anon_vma structure being freed and reused (for another anon_vma:
337 * SLAB_DESTROY_BY_RCU guarantees that - so the spin_lock above cannot 337 * SLAB_DESTROY_BY_RCU guarantees that - so the spin_lock above cannot
338 * corrupt): with anon_vma_prepare() or anon_vma_fork() redirecting 338 * corrupt): with anon_vma_prepare() or anon_vma_fork() redirecting
339 * anon_vma->root before page_unlock_anon_vma() is called to unlock. 339 * anon_vma->root before page_unlock_anon_vma() is called to unlock.
340 */ 340 */
341 if (page_mapped(page)) 341 if (page_mapped(page))
342 return anon_vma; 342 return anon_vma;
343 343
344 spin_unlock(&root_anon_vma->lock); 344 spin_unlock(&root_anon_vma->lock);
345 out: 345 out:
346 rcu_read_unlock(); 346 rcu_read_unlock();
347 return NULL; 347 return NULL;
348 } 348 }
349 349
350 void page_unlock_anon_vma(struct anon_vma *anon_vma) 350 void page_unlock_anon_vma(struct anon_vma *anon_vma)
351 { 351 {
352 anon_vma_unlock(anon_vma); 352 anon_vma_unlock(anon_vma);
353 rcu_read_unlock(); 353 rcu_read_unlock();
354 } 354 }
355 355
356 /* 356 /*
357 * At what user virtual address is page expected in @vma? 357 * At what user virtual address is page expected in @vma?
358 * Returns virtual address or -EFAULT if page's index/offset is not 358 * Returns virtual address or -EFAULT if page's index/offset is not
359 * within the range mapped the @vma. 359 * within the range mapped the @vma.
360 */ 360 */
361 static inline unsigned long 361 static inline unsigned long
362 vma_address(struct page *page, struct vm_area_struct *vma) 362 vma_address(struct page *page, struct vm_area_struct *vma)
363 { 363 {
364 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); 364 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
365 unsigned long address; 365 unsigned long address;
366 366
367 if (unlikely(is_vm_hugetlb_page(vma))) 367 if (unlikely(is_vm_hugetlb_page(vma)))
368 pgoff = page->index << huge_page_order(page_hstate(page)); 368 pgoff = page->index << huge_page_order(page_hstate(page));
369 address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); 369 address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
370 if (unlikely(address < vma->vm_start || address >= vma->vm_end)) { 370 if (unlikely(address < vma->vm_start || address >= vma->vm_end)) {
371 /* page should be within @vma mapping range */ 371 /* page should be within @vma mapping range */
372 return -EFAULT; 372 return -EFAULT;
373 } 373 }
374 return address; 374 return address;
375 } 375 }
376 376
377 /* 377 /*
378 * At what user virtual address is page expected in vma? 378 * At what user virtual address is page expected in vma?
379 * Caller should check the page is actually part of the vma. 379 * Caller should check the page is actually part of the vma.
380 */ 380 */
381 unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma) 381 unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma)
382 { 382 {
383 if (PageAnon(page)) { 383 if (PageAnon(page)) {
384 if (vma->anon_vma->root != page_anon_vma(page)->root) 384 if (vma->anon_vma->root != page_anon_vma(page)->root)
385 return -EFAULT; 385 return -EFAULT;
386 } else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) { 386 } else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) {
387 if (!vma->vm_file || 387 if (!vma->vm_file ||
388 vma->vm_file->f_mapping != page->mapping) 388 vma->vm_file->f_mapping != page->mapping)
389 return -EFAULT; 389 return -EFAULT;
390 } else 390 } else
391 return -EFAULT; 391 return -EFAULT;
392 return vma_address(page, vma); 392 return vma_address(page, vma);
393 } 393 }
394 394
395 /* 395 /*
396 * Check that @page is mapped at @address into @mm. 396 * Check that @page is mapped at @address into @mm.
397 * 397 *
398 * If @sync is false, page_check_address may perform a racy check to avoid 398 * If @sync is false, page_check_address may perform a racy check to avoid
399 * the page table lock when the pte is not present (helpful when reclaiming 399 * the page table lock when the pte is not present (helpful when reclaiming
400 * highly shared pages). 400 * highly shared pages).
401 * 401 *
402 * On success returns with pte mapped and locked. 402 * On success returns with pte mapped and locked.
403 */ 403 */
404 pte_t *page_check_address(struct page *page, struct mm_struct *mm, 404 pte_t *page_check_address(struct page *page, struct mm_struct *mm,
405 unsigned long address, spinlock_t **ptlp, int sync) 405 unsigned long address, spinlock_t **ptlp, int sync)
406 { 406 {
407 pgd_t *pgd; 407 pgd_t *pgd;
408 pud_t *pud; 408 pud_t *pud;
409 pmd_t *pmd; 409 pmd_t *pmd;
410 pte_t *pte; 410 pte_t *pte;
411 spinlock_t *ptl; 411 spinlock_t *ptl;
412 412
413 if (unlikely(PageHuge(page))) { 413 if (unlikely(PageHuge(page))) {
414 pte = huge_pte_offset(mm, address); 414 pte = huge_pte_offset(mm, address);
415 ptl = &mm->page_table_lock; 415 ptl = &mm->page_table_lock;
416 goto check; 416 goto check;
417 } 417 }
418 418
419 pgd = pgd_offset(mm, address); 419 pgd = pgd_offset(mm, address);
420 if (!pgd_present(*pgd)) 420 if (!pgd_present(*pgd))
421 return NULL; 421 return NULL;
422 422
423 pud = pud_offset(pgd, address); 423 pud = pud_offset(pgd, address);
424 if (!pud_present(*pud)) 424 if (!pud_present(*pud))
425 return NULL; 425 return NULL;
426 426
427 pmd = pmd_offset(pud, address); 427 pmd = pmd_offset(pud, address);
428 if (!pmd_present(*pmd)) 428 if (!pmd_present(*pmd))
429 return NULL; 429 return NULL;
430 430
431 pte = pte_offset_map(pmd, address); 431 pte = pte_offset_map(pmd, address);
432 /* Make a quick check before getting the lock */ 432 /* Make a quick check before getting the lock */
433 if (!sync && !pte_present(*pte)) { 433 if (!sync && !pte_present(*pte)) {
434 pte_unmap(pte); 434 pte_unmap(pte);
435 return NULL; 435 return NULL;
436 } 436 }
437 437
438 ptl = pte_lockptr(mm, pmd); 438 ptl = pte_lockptr(mm, pmd);
439 check: 439 check:
440 spin_lock(ptl); 440 spin_lock(ptl);
441 if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) { 441 if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) {
442 *ptlp = ptl; 442 *ptlp = ptl;
443 return pte; 443 return pte;
444 } 444 }
445 pte_unmap_unlock(pte, ptl); 445 pte_unmap_unlock(pte, ptl);
446 return NULL; 446 return NULL;
447 } 447 }
448 448
449 /** 449 /**
450 * page_mapped_in_vma - check whether a page is really mapped in a VMA 450 * page_mapped_in_vma - check whether a page is really mapped in a VMA
451 * @page: the page to test 451 * @page: the page to test
452 * @vma: the VMA to test 452 * @vma: the VMA to test
453 * 453 *
454 * Returns 1 if the page is mapped into the page tables of the VMA, 0 454 * Returns 1 if the page is mapped into the page tables of the VMA, 0
455 * if the page is not mapped into the page tables of this VMA. Only 455 * if the page is not mapped into the page tables of this VMA. Only
456 * valid for normal file or anonymous VMAs. 456 * valid for normal file or anonymous VMAs.
457 */ 457 */
458 int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma) 458 int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma)
459 { 459 {
460 unsigned long address; 460 unsigned long address;
461 pte_t *pte; 461 pte_t *pte;
462 spinlock_t *ptl; 462 spinlock_t *ptl;
463 463
464 address = vma_address(page, vma); 464 address = vma_address(page, vma);
465 if (address == -EFAULT) /* out of vma range */ 465 if (address == -EFAULT) /* out of vma range */
466 return 0; 466 return 0;
467 pte = page_check_address(page, vma->vm_mm, address, &ptl, 1); 467 pte = page_check_address(page, vma->vm_mm, address, &ptl, 1);
468 if (!pte) /* the page is not in this mm */ 468 if (!pte) /* the page is not in this mm */
469 return 0; 469 return 0;
470 pte_unmap_unlock(pte, ptl); 470 pte_unmap_unlock(pte, ptl);
471 471
472 return 1; 472 return 1;
473 } 473 }
474 474
475 /* 475 /*
476 * Subfunctions of page_referenced: page_referenced_one called 476 * Subfunctions of page_referenced: page_referenced_one called
477 * repeatedly from either page_referenced_anon or page_referenced_file. 477 * repeatedly from either page_referenced_anon or page_referenced_file.
478 */ 478 */
479 int page_referenced_one(struct page *page, struct vm_area_struct *vma, 479 int page_referenced_one(struct page *page, struct vm_area_struct *vma,
480 unsigned long address, unsigned int *mapcount, 480 unsigned long address, unsigned int *mapcount,
481 unsigned long *vm_flags) 481 unsigned long *vm_flags)
482 { 482 {
483 struct mm_struct *mm = vma->vm_mm; 483 struct mm_struct *mm = vma->vm_mm;
484 pte_t *pte; 484 pte_t *pte;
485 spinlock_t *ptl; 485 spinlock_t *ptl;
486 int referenced = 0; 486 int referenced = 0;
487 487
488 pte = page_check_address(page, mm, address, &ptl, 0); 488 pte = page_check_address(page, mm, address, &ptl, 0);
489 if (!pte) 489 if (!pte)
490 goto out; 490 goto out;
491 491
492 /* 492 /*
493 * Don't want to elevate referenced for mlocked page that gets this far, 493 * Don't want to elevate referenced for mlocked page that gets this far,
494 * in order that it progresses to try_to_unmap and is moved to the 494 * in order that it progresses to try_to_unmap and is moved to the
495 * unevictable list. 495 * unevictable list.
496 */ 496 */
497 if (vma->vm_flags & VM_LOCKED) { 497 if (vma->vm_flags & VM_LOCKED) {
498 *mapcount = 1; /* break early from loop */ 498 *mapcount = 1; /* break early from loop */
499 *vm_flags |= VM_LOCKED; 499 *vm_flags |= VM_LOCKED;
500 goto out_unmap; 500 goto out_unmap;
501 } 501 }
502 502
503 if (ptep_clear_flush_young_notify(vma, address, pte)) { 503 if (ptep_clear_flush_young_notify(vma, address, pte)) {
504 /* 504 /*
505 * Don't treat a reference through a sequentially read 505 * Don't treat a reference through a sequentially read
506 * mapping as such. If the page has been used in 506 * mapping as such. If the page has been used in
507 * another mapping, we will catch it; if this other 507 * another mapping, we will catch it; if this other
508 * mapping is already gone, the unmap path will have 508 * mapping is already gone, the unmap path will have
509 * set PG_referenced or activated the page. 509 * set PG_referenced or activated the page.
510 */ 510 */
511 if (likely(!VM_SequentialReadHint(vma))) 511 if (likely(!VM_SequentialReadHint(vma)))
512 referenced++; 512 referenced++;
513 } 513 }
514 514
515 /* Pretend the page is referenced if the task has the 515 /* Pretend the page is referenced if the task has the
516 swap token and is in the middle of a page fault. */ 516 swap token and is in the middle of a page fault. */
517 if (mm != current->mm && has_swap_token(mm) && 517 if (mm != current->mm && has_swap_token(mm) &&
518 rwsem_is_locked(&mm->mmap_sem)) 518 rwsem_is_locked(&mm->mmap_sem))
519 referenced++; 519 referenced++;
520 520
521 out_unmap: 521 out_unmap:
522 (*mapcount)--; 522 (*mapcount)--;
523 pte_unmap_unlock(pte, ptl); 523 pte_unmap_unlock(pte, ptl);
524 524
525 if (referenced) 525 if (referenced)
526 *vm_flags |= vma->vm_flags; 526 *vm_flags |= vma->vm_flags;
527 out: 527 out:
528 return referenced; 528 return referenced;
529 } 529 }
530 530
531 static int page_referenced_anon(struct page *page, 531 static int page_referenced_anon(struct page *page,
532 struct mem_cgroup *mem_cont, 532 struct mem_cgroup *mem_cont,
533 unsigned long *vm_flags) 533 unsigned long *vm_flags)
534 { 534 {
535 unsigned int mapcount; 535 unsigned int mapcount;
536 struct anon_vma *anon_vma; 536 struct anon_vma *anon_vma;
537 struct anon_vma_chain *avc; 537 struct anon_vma_chain *avc;
538 int referenced = 0; 538 int referenced = 0;
539 539
540 anon_vma = page_lock_anon_vma(page); 540 anon_vma = page_lock_anon_vma(page);
541 if (!anon_vma) 541 if (!anon_vma)
542 return referenced; 542 return referenced;
543 543
544 mapcount = page_mapcount(page); 544 mapcount = page_mapcount(page);
545 list_for_each_entry(avc, &anon_vma->head, same_anon_vma) { 545 list_for_each_entry(avc, &anon_vma->head, same_anon_vma) {
546 struct vm_area_struct *vma = avc->vma; 546 struct vm_area_struct *vma = avc->vma;
547 unsigned long address = vma_address(page, vma); 547 unsigned long address = vma_address(page, vma);
548 if (address == -EFAULT) 548 if (address == -EFAULT)
549 continue; 549 continue;
550 /* 550 /*
551 * If we are reclaiming on behalf of a cgroup, skip 551 * If we are reclaiming on behalf of a cgroup, skip
552 * counting on behalf of references from different 552 * counting on behalf of references from different
553 * cgroups 553 * cgroups
554 */ 554 */
555 if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont)) 555 if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont))
556 continue; 556 continue;
557 referenced += page_referenced_one(page, vma, address, 557 referenced += page_referenced_one(page, vma, address,
558 &mapcount, vm_flags); 558 &mapcount, vm_flags);
559 if (!mapcount) 559 if (!mapcount)
560 break; 560 break;
561 } 561 }
562 562
563 page_unlock_anon_vma(anon_vma); 563 page_unlock_anon_vma(anon_vma);
564 return referenced; 564 return referenced;
565 } 565 }
566 566
567 /** 567 /**
568 * page_referenced_file - referenced check for object-based rmap 568 * page_referenced_file - referenced check for object-based rmap
569 * @page: the page we're checking references on. 569 * @page: the page we're checking references on.
570 * @mem_cont: target memory controller 570 * @mem_cont: target memory controller
571 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page 571 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
572 * 572 *
573 * For an object-based mapped page, find all the places it is mapped and 573 * For an object-based mapped page, find all the places it is mapped and
574 * check/clear the referenced flag. This is done by following the page->mapping 574 * check/clear the referenced flag. This is done by following the page->mapping
575 * pointer, then walking the chain of vmas it holds. It returns the number 575 * pointer, then walking the chain of vmas it holds. It returns the number
576 * of references it found. 576 * of references it found.
577 * 577 *
578 * This function is only called from page_referenced for object-based pages. 578 * This function is only called from page_referenced for object-based pages.
579 */ 579 */
580 static int page_referenced_file(struct page *page, 580 static int page_referenced_file(struct page *page,
581 struct mem_cgroup *mem_cont, 581 struct mem_cgroup *mem_cont,
582 unsigned long *vm_flags) 582 unsigned long *vm_flags)
583 { 583 {
584 unsigned int mapcount; 584 unsigned int mapcount;
585 struct address_space *mapping = page->mapping; 585 struct address_space *mapping = page->mapping;
586 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); 586 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
587 struct vm_area_struct *vma; 587 struct vm_area_struct *vma;
588 struct prio_tree_iter iter; 588 struct prio_tree_iter iter;
589 int referenced = 0; 589 int referenced = 0;
590 590
591 /* 591 /*
592 * The caller's checks on page->mapping and !PageAnon have made 592 * The caller's checks on page->mapping and !PageAnon have made
593 * sure that this is a file page: the check for page->mapping 593 * sure that this is a file page: the check for page->mapping
594 * excludes the case just before it gets set on an anon page. 594 * excludes the case just before it gets set on an anon page.
595 */ 595 */
596 BUG_ON(PageAnon(page)); 596 BUG_ON(PageAnon(page));
597 597
598 /* 598 /*
599 * The page lock not only makes sure that page->mapping cannot 599 * The page lock not only makes sure that page->mapping cannot
600 * suddenly be NULLified by truncation, it makes sure that the 600 * suddenly be NULLified by truncation, it makes sure that the
601 * structure at mapping cannot be freed and reused yet, 601 * structure at mapping cannot be freed and reused yet,
602 * so we can safely take mapping->i_mmap_lock. 602 * so we can safely take mapping->i_mmap_lock.
603 */ 603 */
604 BUG_ON(!PageLocked(page)); 604 BUG_ON(!PageLocked(page));
605 605
606 spin_lock(&mapping->i_mmap_lock); 606 spin_lock(&mapping->i_mmap_lock);
607 607
608 /* 608 /*
609 * i_mmap_lock does not stabilize mapcount at all, but mapcount 609 * i_mmap_lock does not stabilize mapcount at all, but mapcount
610 * is more likely to be accurate if we note it after spinning. 610 * is more likely to be accurate if we note it after spinning.
611 */ 611 */
612 mapcount = page_mapcount(page); 612 mapcount = page_mapcount(page);
613 613
614 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { 614 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
615 unsigned long address = vma_address(page, vma); 615 unsigned long address = vma_address(page, vma);
616 if (address == -EFAULT) 616 if (address == -EFAULT)
617 continue; 617 continue;
618 /* 618 /*
619 * If we are reclaiming on behalf of a cgroup, skip 619 * If we are reclaiming on behalf of a cgroup, skip
620 * counting on behalf of references from different 620 * counting on behalf of references from different
621 * cgroups 621 * cgroups
622 */ 622 */
623 if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont)) 623 if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont))
624 continue; 624 continue;
625 referenced += page_referenced_one(page, vma, address, 625 referenced += page_referenced_one(page, vma, address,
626 &mapcount, vm_flags); 626 &mapcount, vm_flags);
627 if (!mapcount) 627 if (!mapcount)
628 break; 628 break;
629 } 629 }
630 630
631 spin_unlock(&mapping->i_mmap_lock); 631 spin_unlock(&mapping->i_mmap_lock);
632 return referenced; 632 return referenced;
633 } 633 }
634 634
635 /** 635 /**
636 * page_referenced - test if the page was referenced 636 * page_referenced - test if the page was referenced
637 * @page: the page to test 637 * @page: the page to test
638 * @is_locked: caller holds lock on the page 638 * @is_locked: caller holds lock on the page
639 * @mem_cont: target memory controller 639 * @mem_cont: target memory controller
640 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page 640 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
641 * 641 *
642 * Quick test_and_clear_referenced for all mappings to a page, 642 * Quick test_and_clear_referenced for all mappings to a page,
643 * returns the number of ptes which referenced the page. 643 * returns the number of ptes which referenced the page.
644 */ 644 */
645 int page_referenced(struct page *page, 645 int page_referenced(struct page *page,
646 int is_locked, 646 int is_locked,
647 struct mem_cgroup *mem_cont, 647 struct mem_cgroup *mem_cont,
648 unsigned long *vm_flags) 648 unsigned long *vm_flags)
649 { 649 {
650 int referenced = 0; 650 int referenced = 0;
651 int we_locked = 0; 651 int we_locked = 0;
652 652
653 *vm_flags = 0; 653 *vm_flags = 0;
654 if (page_mapped(page) && page_rmapping(page)) { 654 if (page_mapped(page) && page_rmapping(page)) {
655 if (!is_locked && (!PageAnon(page) || PageKsm(page))) { 655 if (!is_locked && (!PageAnon(page) || PageKsm(page))) {
656 we_locked = trylock_page(page); 656 we_locked = trylock_page(page);
657 if (!we_locked) { 657 if (!we_locked) {
658 referenced++; 658 referenced++;
659 goto out; 659 goto out;
660 } 660 }
661 } 661 }
662 if (unlikely(PageKsm(page))) 662 if (unlikely(PageKsm(page)))
663 referenced += page_referenced_ksm(page, mem_cont, 663 referenced += page_referenced_ksm(page, mem_cont,
664 vm_flags); 664 vm_flags);
665 else if (PageAnon(page)) 665 else if (PageAnon(page))
666 referenced += page_referenced_anon(page, mem_cont, 666 referenced += page_referenced_anon(page, mem_cont,
667 vm_flags); 667 vm_flags);
668 else if (page->mapping) 668 else if (page->mapping)
669 referenced += page_referenced_file(page, mem_cont, 669 referenced += page_referenced_file(page, mem_cont,
670 vm_flags); 670 vm_flags);
671 if (we_locked) 671 if (we_locked)
672 unlock_page(page); 672 unlock_page(page);
673 } 673 }
674 out: 674 out:
675 if (page_test_and_clear_young(page)) 675 if (page_test_and_clear_young(page))
676 referenced++; 676 referenced++;
677 677
678 return referenced; 678 return referenced;
679 } 679 }
680 680
681 static int page_mkclean_one(struct page *page, struct vm_area_struct *vma, 681 static int page_mkclean_one(struct page *page, struct vm_area_struct *vma,
682 unsigned long address) 682 unsigned long address)
683 { 683 {
684 struct mm_struct *mm = vma->vm_mm; 684 struct mm_struct *mm = vma->vm_mm;
685 pte_t *pte; 685 pte_t *pte;
686 spinlock_t *ptl; 686 spinlock_t *ptl;
687 int ret = 0; 687 int ret = 0;
688 688
689 pte = page_check_address(page, mm, address, &ptl, 1); 689 pte = page_check_address(page, mm, address, &ptl, 1);
690 if (!pte) 690 if (!pte)
691 goto out; 691 goto out;
692 692
693 if (pte_dirty(*pte) || pte_write(*pte)) { 693 if (pte_dirty(*pte) || pte_write(*pte)) {
694 pte_t entry; 694 pte_t entry;
695 695
696 flush_cache_page(vma, address, pte_pfn(*pte)); 696 flush_cache_page(vma, address, pte_pfn(*pte));
697 entry = ptep_clear_flush_notify(vma, address, pte); 697 entry = ptep_clear_flush_notify(vma, address, pte);
698 entry = pte_wrprotect(entry); 698 entry = pte_wrprotect(entry);
699 entry = pte_mkclean(entry); 699 entry = pte_mkclean(entry);
700 set_pte_at(mm, address, pte, entry); 700 set_pte_at(mm, address, pte, entry);
701 ret = 1; 701 ret = 1;
702 } 702 }
703 703
704 pte_unmap_unlock(pte, ptl); 704 pte_unmap_unlock(pte, ptl);
705 out: 705 out:
706 return ret; 706 return ret;
707 } 707 }
708 708
709 static int page_mkclean_file(struct address_space *mapping, struct page *page) 709 static int page_mkclean_file(struct address_space *mapping, struct page *page)
710 { 710 {
711 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); 711 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
712 struct vm_area_struct *vma; 712 struct vm_area_struct *vma;
713 struct prio_tree_iter iter; 713 struct prio_tree_iter iter;
714 int ret = 0; 714 int ret = 0;
715 715
716 BUG_ON(PageAnon(page)); 716 BUG_ON(PageAnon(page));
717 717
718 spin_lock(&mapping->i_mmap_lock); 718 spin_lock(&mapping->i_mmap_lock);
719 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { 719 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
720 if (vma->vm_flags & VM_SHARED) { 720 if (vma->vm_flags & VM_SHARED) {
721 unsigned long address = vma_address(page, vma); 721 unsigned long address = vma_address(page, vma);
722 if (address == -EFAULT) 722 if (address == -EFAULT)
723 continue; 723 continue;
724 ret += page_mkclean_one(page, vma, address); 724 ret += page_mkclean_one(page, vma, address);
725 } 725 }
726 } 726 }
727 spin_unlock(&mapping->i_mmap_lock); 727 spin_unlock(&mapping->i_mmap_lock);
728 return ret; 728 return ret;
729 } 729 }
730 730
731 int page_mkclean(struct page *page) 731 int page_mkclean(struct page *page)
732 { 732 {
733 int ret = 0; 733 int ret = 0;
734 734
735 BUG_ON(!PageLocked(page)); 735 BUG_ON(!PageLocked(page));
736 736
737 if (page_mapped(page)) { 737 if (page_mapped(page)) {
738 struct address_space *mapping = page_mapping(page); 738 struct address_space *mapping = page_mapping(page);
739 if (mapping) { 739 if (mapping) {
740 ret = page_mkclean_file(mapping, page); 740 ret = page_mkclean_file(mapping, page);
741 if (page_test_dirty(page)) { 741 if (page_test_dirty(page)) {
742 page_clear_dirty(page); 742 page_clear_dirty(page);
743 ret = 1; 743 ret = 1;
744 } 744 }
745 } 745 }
746 } 746 }
747 747
748 return ret; 748 return ret;
749 } 749 }
750 EXPORT_SYMBOL_GPL(page_mkclean); 750 EXPORT_SYMBOL_GPL(page_mkclean);
751 751
752 /** 752 /**
753 * page_move_anon_rmap - move a page to our anon_vma 753 * page_move_anon_rmap - move a page to our anon_vma
754 * @page: the page to move to our anon_vma 754 * @page: the page to move to our anon_vma
755 * @vma: the vma the page belongs to 755 * @vma: the vma the page belongs to
756 * @address: the user virtual address mapped 756 * @address: the user virtual address mapped
757 * 757 *
758 * When a page belongs exclusively to one process after a COW event, 758 * When a page belongs exclusively to one process after a COW event,
759 * that page can be moved into the anon_vma that belongs to just that 759 * that page can be moved into the anon_vma that belongs to just that
760 * process, so the rmap code will not search the parent or sibling 760 * process, so the rmap code will not search the parent or sibling
761 * processes. 761 * processes.
762 */ 762 */
763 void page_move_anon_rmap(struct page *page, 763 void page_move_anon_rmap(struct page *page,
764 struct vm_area_struct *vma, unsigned long address) 764 struct vm_area_struct *vma, unsigned long address)
765 { 765 {
766 struct anon_vma *anon_vma = vma->anon_vma; 766 struct anon_vma *anon_vma = vma->anon_vma;
767 767
768 VM_BUG_ON(!PageLocked(page)); 768 VM_BUG_ON(!PageLocked(page));
769 VM_BUG_ON(!anon_vma); 769 VM_BUG_ON(!anon_vma);
770 VM_BUG_ON(page->index != linear_page_index(vma, address)); 770 VM_BUG_ON(page->index != linear_page_index(vma, address));
771 771
772 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; 772 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
773 page->mapping = (struct address_space *) anon_vma; 773 page->mapping = (struct address_space *) anon_vma;
774 } 774 }
775 775
776 /** 776 /**
777 * __page_set_anon_rmap - setup new anonymous rmap 777 * __page_set_anon_rmap - setup new anonymous rmap
778 * @page: the page to add the mapping to 778 * @page: the page to add the mapping to
779 * @vma: the vm area in which the mapping is added 779 * @vma: the vm area in which the mapping is added
780 * @address: the user virtual address mapped 780 * @address: the user virtual address mapped
781 * @exclusive: the page is exclusively owned by the current process 781 * @exclusive: the page is exclusively owned by the current process
782 */ 782 */
783 static void __page_set_anon_rmap(struct page *page, 783 static void __page_set_anon_rmap(struct page *page,
784 struct vm_area_struct *vma, unsigned long address, int exclusive) 784 struct vm_area_struct *vma, unsigned long address, int exclusive)
785 { 785 {
786 struct anon_vma *anon_vma = vma->anon_vma; 786 struct anon_vma *anon_vma = vma->anon_vma;
787 787
788 BUG_ON(!anon_vma); 788 BUG_ON(!anon_vma);
789 789
790 /* 790 /*
791 * If the page isn't exclusively mapped into this vma, 791 * If the page isn't exclusively mapped into this vma,
792 * we must use the _oldest_ possible anon_vma for the 792 * we must use the _oldest_ possible anon_vma for the
793 * page mapping! 793 * page mapping!
794 */ 794 */
795 if (!exclusive) { 795 if (!exclusive) {
796 if (PageAnon(page)) 796 if (PageAnon(page))
797 return; 797 return;
798 anon_vma = anon_vma->root; 798 anon_vma = anon_vma->root;
799 } else { 799 } else {
800 /* 800 /*
801 * In this case, swapped-out-but-not-discarded swap-cache 801 * In this case, swapped-out-but-not-discarded swap-cache
802 * is remapped. So, no need to update page->mapping here. 802 * is remapped. So, no need to update page->mapping here.
803 * We convice anon_vma poitned by page->mapping is not obsolete 803 * We convice anon_vma poitned by page->mapping is not obsolete
804 * because vma->anon_vma is necessary to be a family of it. 804 * because vma->anon_vma is necessary to be a family of it.
805 */ 805 */
806 if (PageAnon(page)) 806 if (PageAnon(page))
807 return; 807 return;
808 } 808 }
809 809
810 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; 810 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
811 page->mapping = (struct address_space *) anon_vma; 811 page->mapping = (struct address_space *) anon_vma;
812 page->index = linear_page_index(vma, address); 812 page->index = linear_page_index(vma, address);
813 } 813 }
814 814
815 /** 815 /**
816 * __page_check_anon_rmap - sanity check anonymous rmap addition 816 * __page_check_anon_rmap - sanity check anonymous rmap addition
817 * @page: the page to add the mapping to 817 * @page: the page to add the mapping to
818 * @vma: the vm area in which the mapping is added 818 * @vma: the vm area in which the mapping is added
819 * @address: the user virtual address mapped 819 * @address: the user virtual address mapped
820 */ 820 */
821 static void __page_check_anon_rmap(struct page *page, 821 static void __page_check_anon_rmap(struct page *page,
822 struct vm_area_struct *vma, unsigned long address) 822 struct vm_area_struct *vma, unsigned long address)
823 { 823 {
824 #ifdef CONFIG_DEBUG_VM 824 #ifdef CONFIG_DEBUG_VM
825 /* 825 /*
826 * The page's anon-rmap details (mapping and index) are guaranteed to 826 * The page's anon-rmap details (mapping and index) are guaranteed to
827 * be set up correctly at this point. 827 * be set up correctly at this point.
828 * 828 *
829 * We have exclusion against page_add_anon_rmap because the caller 829 * We have exclusion against page_add_anon_rmap because the caller
830 * always holds the page locked, except if called from page_dup_rmap, 830 * always holds the page locked, except if called from page_dup_rmap,
831 * in which case the page is already known to be setup. 831 * in which case the page is already known to be setup.
832 * 832 *
833 * We have exclusion against page_add_new_anon_rmap because those pages 833 * We have exclusion against page_add_new_anon_rmap because those pages
834 * are initially only visible via the pagetables, and the pte is locked 834 * are initially only visible via the pagetables, and the pte is locked
835 * over the call to page_add_new_anon_rmap. 835 * over the call to page_add_new_anon_rmap.
836 */ 836 */
837 BUG_ON(page_anon_vma(page)->root != vma->anon_vma->root); 837 BUG_ON(page_anon_vma(page)->root != vma->anon_vma->root);
838 BUG_ON(page->index != linear_page_index(vma, address)); 838 BUG_ON(page->index != linear_page_index(vma, address));
839 #endif 839 #endif
840 } 840 }
841 841
842 /** 842 /**
843 * page_add_anon_rmap - add pte mapping to an anonymous page 843 * page_add_anon_rmap - add pte mapping to an anonymous page
844 * @page: the page to add the mapping to 844 * @page: the page to add the mapping to
845 * @vma: the vm area in which the mapping is added 845 * @vma: the vm area in which the mapping is added
846 * @address: the user virtual address mapped 846 * @address: the user virtual address mapped
847 * 847 *
848 * The caller needs to hold the pte lock, and the page must be locked in 848 * The caller needs to hold the pte lock, and the page must be locked in
849 * the anon_vma case: to serialize mapping,index checking after setting, 849 * the anon_vma case: to serialize mapping,index checking after setting,
850 * and to ensure that PageAnon is not being upgraded racily to PageKsm 850 * and to ensure that PageAnon is not being upgraded racily to PageKsm
851 * (but PageKsm is never downgraded to PageAnon). 851 * (but PageKsm is never downgraded to PageAnon).
852 */ 852 */
853 void page_add_anon_rmap(struct page *page, 853 void page_add_anon_rmap(struct page *page,
854 struct vm_area_struct *vma, unsigned long address) 854 struct vm_area_struct *vma, unsigned long address)
855 { 855 {
856 do_page_add_anon_rmap(page, vma, address, 0); 856 do_page_add_anon_rmap(page, vma, address, 0);
857 } 857 }
858 858
859 /* 859 /*
860 * Special version of the above for do_swap_page, which often runs 860 * Special version of the above for do_swap_page, which often runs
861 * into pages that are exclusively owned by the current process. 861 * into pages that are exclusively owned by the current process.
862 * Everybody else should continue to use page_add_anon_rmap above. 862 * Everybody else should continue to use page_add_anon_rmap above.
863 */ 863 */
864 void do_page_add_anon_rmap(struct page *page, 864 void do_page_add_anon_rmap(struct page *page,
865 struct vm_area_struct *vma, unsigned long address, int exclusive) 865 struct vm_area_struct *vma, unsigned long address, int exclusive)
866 { 866 {
867 int first = atomic_inc_and_test(&page->_mapcount); 867 int first = atomic_inc_and_test(&page->_mapcount);
868 if (first) 868 if (first)
869 __inc_zone_page_state(page, NR_ANON_PAGES); 869 __inc_zone_page_state(page, NR_ANON_PAGES);
870 if (unlikely(PageKsm(page))) 870 if (unlikely(PageKsm(page)))
871 return; 871 return;
872 872
873 VM_BUG_ON(!PageLocked(page)); 873 VM_BUG_ON(!PageLocked(page));
874 VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); 874 VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end);
875 if (first) 875 if (first)
876 __page_set_anon_rmap(page, vma, address, exclusive); 876 __page_set_anon_rmap(page, vma, address, exclusive);
877 else 877 else
878 __page_check_anon_rmap(page, vma, address); 878 __page_check_anon_rmap(page, vma, address);
879 } 879 }
880 880
881 /** 881 /**
882 * page_add_new_anon_rmap - add pte mapping to a new anonymous page 882 * page_add_new_anon_rmap - add pte mapping to a new anonymous page
883 * @page: the page to add the mapping to 883 * @page: the page to add the mapping to
884 * @vma: the vm area in which the mapping is added 884 * @vma: the vm area in which the mapping is added
885 * @address: the user virtual address mapped 885 * @address: the user virtual address mapped
886 * 886 *
887 * Same as page_add_anon_rmap but must only be called on *new* pages. 887 * Same as page_add_anon_rmap but must only be called on *new* pages.
888 * This means the inc-and-test can be bypassed. 888 * This means the inc-and-test can be bypassed.
889 * Page does not have to be locked. 889 * Page does not have to be locked.
890 */ 890 */
891 void page_add_new_anon_rmap(struct page *page, 891 void page_add_new_anon_rmap(struct page *page,
892 struct vm_area_struct *vma, unsigned long address) 892 struct vm_area_struct *vma, unsigned long address)
893 { 893 {
894 VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); 894 VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end);
895 SetPageSwapBacked(page); 895 SetPageSwapBacked(page);
896 atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */ 896 atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */
897 __inc_zone_page_state(page, NR_ANON_PAGES); 897 __inc_zone_page_state(page, NR_ANON_PAGES);
898 __page_set_anon_rmap(page, vma, address, 1); 898 __page_set_anon_rmap(page, vma, address, 1);
899 if (page_evictable(page, vma)) 899 if (page_evictable(page, vma))
900 lru_cache_add_lru(page, LRU_ACTIVE_ANON); 900 lru_cache_add_lru(page, LRU_ACTIVE_ANON);
901 else 901 else
902 add_page_to_unevictable_list(page); 902 add_page_to_unevictable_list(page);
903 } 903 }
904 904
905 /** 905 /**
906 * page_add_file_rmap - add pte mapping to a file page 906 * page_add_file_rmap - add pte mapping to a file page
907 * @page: the page to add the mapping to 907 * @page: the page to add the mapping to
908 * 908 *
909 * The caller needs to hold the pte lock. 909 * The caller needs to hold the pte lock.
910 */ 910 */
911 void page_add_file_rmap(struct page *page) 911 void page_add_file_rmap(struct page *page)
912 { 912 {
913 if (atomic_inc_and_test(&page->_mapcount)) { 913 if (atomic_inc_and_test(&page->_mapcount)) {
914 __inc_zone_page_state(page, NR_FILE_MAPPED); 914 __inc_zone_page_state(page, NR_FILE_MAPPED);
915 mem_cgroup_update_file_mapped(page, 1); 915 mem_cgroup_update_file_mapped(page, 1);
916 } 916 }
917 } 917 }
918 918
919 /** 919 /**
920 * page_remove_rmap - take down pte mapping from a page 920 * page_remove_rmap - take down pte mapping from a page
921 * @page: page to remove mapping from 921 * @page: page to remove mapping from
922 * 922 *
923 * The caller needs to hold the pte lock. 923 * The caller needs to hold the pte lock.
924 */ 924 */
925 void page_remove_rmap(struct page *page) 925 void page_remove_rmap(struct page *page)
926 { 926 {
927 /* page still mapped by someone else? */ 927 /* page still mapped by someone else? */
928 if (!atomic_add_negative(-1, &page->_mapcount)) 928 if (!atomic_add_negative(-1, &page->_mapcount))
929 return; 929 return;
930 930
931 /* 931 /*
932 * Now that the last pte has gone, s390 must transfer dirty 932 * Now that the last pte has gone, s390 must transfer dirty
933 * flag from storage key to struct page. We can usually skip 933 * flag from storage key to struct page. We can usually skip
934 * this if the page is anon, so about to be freed; but perhaps 934 * this if the page is anon, so about to be freed; but perhaps
935 * not if it's in swapcache - there might be another pte slot 935 * not if it's in swapcache - there might be another pte slot
936 * containing the swap entry, but page not yet written to swap. 936 * containing the swap entry, but page not yet written to swap.
937 */ 937 */
938 if ((!PageAnon(page) || PageSwapCache(page)) && page_test_dirty(page)) { 938 if ((!PageAnon(page) || PageSwapCache(page)) && page_test_dirty(page)) {
939 page_clear_dirty(page); 939 page_clear_dirty(page);
940 set_page_dirty(page); 940 set_page_dirty(page);
941 } 941 }
942 /* 942 /*
943 * Hugepages are not counted in NR_ANON_PAGES nor NR_FILE_MAPPED 943 * Hugepages are not counted in NR_ANON_PAGES nor NR_FILE_MAPPED
944 * and not charged by memcg for now. 944 * and not charged by memcg for now.
945 */ 945 */
946 if (unlikely(PageHuge(page))) 946 if (unlikely(PageHuge(page)))
947 return; 947 return;
948 if (PageAnon(page)) { 948 if (PageAnon(page)) {
949 mem_cgroup_uncharge_page(page); 949 mem_cgroup_uncharge_page(page);
950 __dec_zone_page_state(page, NR_ANON_PAGES); 950 __dec_zone_page_state(page, NR_ANON_PAGES);
951 } else { 951 } else {
952 __dec_zone_page_state(page, NR_FILE_MAPPED); 952 __dec_zone_page_state(page, NR_FILE_MAPPED);
953 mem_cgroup_update_file_mapped(page, -1); 953 mem_cgroup_update_file_mapped(page, -1);
954 } 954 }
955 /* 955 /*
956 * It would be tidy to reset the PageAnon mapping here, 956 * It would be tidy to reset the PageAnon mapping here,
957 * but that might overwrite a racing page_add_anon_rmap 957 * but that might overwrite a racing page_add_anon_rmap
958 * which increments mapcount after us but sets mapping 958 * which increments mapcount after us but sets mapping
959 * before us: so leave the reset to free_hot_cold_page, 959 * before us: so leave the reset to free_hot_cold_page,
960 * and remember that it's only reliable while mapped. 960 * and remember that it's only reliable while mapped.
961 * Leaving it set also helps swapoff to reinstate ptes 961 * Leaving it set also helps swapoff to reinstate ptes
962 * faster for those pages still in swapcache. 962 * faster for those pages still in swapcache.
963 */ 963 */
964 } 964 }
965 965
966 /* 966 /*
967 * Subfunctions of try_to_unmap: try_to_unmap_one called 967 * Subfunctions of try_to_unmap: try_to_unmap_one called
968 * repeatedly from either try_to_unmap_anon or try_to_unmap_file. 968 * repeatedly from either try_to_unmap_anon or try_to_unmap_file.
969 */ 969 */
970 int try_to_unmap_one(struct page *page, struct vm_area_struct *vma, 970 int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
971 unsigned long address, enum ttu_flags flags) 971 unsigned long address, enum ttu_flags flags)
972 { 972 {
973 struct mm_struct *mm = vma->vm_mm; 973 struct mm_struct *mm = vma->vm_mm;
974 pte_t *pte; 974 pte_t *pte;
975 pte_t pteval; 975 pte_t pteval;
976 spinlock_t *ptl; 976 spinlock_t *ptl;
977 int ret = SWAP_AGAIN; 977 int ret = SWAP_AGAIN;
978 978
979 pte = page_check_address(page, mm, address, &ptl, 0); 979 pte = page_check_address(page, mm, address, &ptl, 0);
980 if (!pte) 980 if (!pte)
981 goto out; 981 goto out;
982 982
983 /* 983 /*
984 * If the page is mlock()d, we cannot swap it out. 984 * If the page is mlock()d, we cannot swap it out.
985 * If it's recently referenced (perhaps page_referenced 985 * If it's recently referenced (perhaps page_referenced
986 * skipped over this mm) then we should reactivate it. 986 * skipped over this mm) then we should reactivate it.
987 */ 987 */
988 if (!(flags & TTU_IGNORE_MLOCK)) { 988 if (!(flags & TTU_IGNORE_MLOCK)) {
989 if (vma->vm_flags & VM_LOCKED) 989 if (vma->vm_flags & VM_LOCKED)
990 goto out_mlock; 990 goto out_mlock;
991 991
992 if (TTU_ACTION(flags) == TTU_MUNLOCK) 992 if (TTU_ACTION(flags) == TTU_MUNLOCK)
993 goto out_unmap; 993 goto out_unmap;
994 } 994 }
995 if (!(flags & TTU_IGNORE_ACCESS)) { 995 if (!(flags & TTU_IGNORE_ACCESS)) {
996 if (ptep_clear_flush_young_notify(vma, address, pte)) { 996 if (ptep_clear_flush_young_notify(vma, address, pte)) {
997 ret = SWAP_FAIL; 997 ret = SWAP_FAIL;
998 goto out_unmap; 998 goto out_unmap;
999 } 999 }
1000 } 1000 }
1001 1001
1002 /* Nuke the page table entry. */ 1002 /* Nuke the page table entry. */
1003 flush_cache_page(vma, address, page_to_pfn(page)); 1003 flush_cache_page(vma, address, page_to_pfn(page));
1004 pteval = ptep_clear_flush_notify(vma, address, pte); 1004 pteval = ptep_clear_flush_notify(vma, address, pte);
1005 1005
1006 /* Move the dirty bit to the physical page now the pte is gone. */ 1006 /* Move the dirty bit to the physical page now the pte is gone. */
1007 if (pte_dirty(pteval)) 1007 if (pte_dirty(pteval))
1008 set_page_dirty(page); 1008 set_page_dirty(page);
1009 1009
1010 /* Update high watermark before we lower rss */ 1010 /* Update high watermark before we lower rss */
1011 update_hiwater_rss(mm); 1011 update_hiwater_rss(mm);
1012 1012
1013 if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) { 1013 if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) {
1014 if (PageAnon(page)) 1014 if (PageAnon(page))
1015 dec_mm_counter(mm, MM_ANONPAGES); 1015 dec_mm_counter(mm, MM_ANONPAGES);
1016 else 1016 else
1017 dec_mm_counter(mm, MM_FILEPAGES); 1017 dec_mm_counter(mm, MM_FILEPAGES);
1018 set_pte_at(mm, address, pte, 1018 set_pte_at(mm, address, pte,
1019 swp_entry_to_pte(make_hwpoison_entry(page))); 1019 swp_entry_to_pte(make_hwpoison_entry(page)));
1020 } else if (PageAnon(page)) { 1020 } else if (PageAnon(page)) {
1021 swp_entry_t entry = { .val = page_private(page) }; 1021 swp_entry_t entry = { .val = page_private(page) };
1022 1022
1023 if (PageSwapCache(page)) { 1023 if (PageSwapCache(page)) {
1024 /* 1024 /*
1025 * Store the swap location in the pte. 1025 * Store the swap location in the pte.
1026 * See handle_pte_fault() ... 1026 * See handle_pte_fault() ...
1027 */ 1027 */
1028 if (swap_duplicate(entry) < 0) { 1028 if (swap_duplicate(entry) < 0) {
1029 set_pte_at(mm, address, pte, pteval); 1029 set_pte_at(mm, address, pte, pteval);
1030 ret = SWAP_FAIL; 1030 ret = SWAP_FAIL;
1031 goto out_unmap; 1031 goto out_unmap;
1032 } 1032 }
1033 if (list_empty(&mm->mmlist)) { 1033 if (list_empty(&mm->mmlist)) {
1034 spin_lock(&mmlist_lock); 1034 spin_lock(&mmlist_lock);
1035 if (list_empty(&mm->mmlist)) 1035 if (list_empty(&mm->mmlist))
1036 list_add(&mm->mmlist, &init_mm.mmlist); 1036 list_add(&mm->mmlist, &init_mm.mmlist);
1037 spin_unlock(&mmlist_lock); 1037 spin_unlock(&mmlist_lock);
1038 } 1038 }
1039 dec_mm_counter(mm, MM_ANONPAGES); 1039 dec_mm_counter(mm, MM_ANONPAGES);
1040 inc_mm_counter(mm, MM_SWAPENTS); 1040 inc_mm_counter(mm, MM_SWAPENTS);
1041 } else if (PAGE_MIGRATION) { 1041 } else if (PAGE_MIGRATION) {
1042 /* 1042 /*
1043 * Store the pfn of the page in a special migration 1043 * Store the pfn of the page in a special migration
1044 * pte. do_swap_page() will wait until the migration 1044 * pte. do_swap_page() will wait until the migration
1045 * pte is removed and then restart fault handling. 1045 * pte is removed and then restart fault handling.
1046 */ 1046 */
1047 BUG_ON(TTU_ACTION(flags) != TTU_MIGRATION); 1047 BUG_ON(TTU_ACTION(flags) != TTU_MIGRATION);
1048 entry = make_migration_entry(page, pte_write(pteval)); 1048 entry = make_migration_entry(page, pte_write(pteval));
1049 } 1049 }
1050 set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); 1050 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
1051 BUG_ON(pte_file(*pte)); 1051 BUG_ON(pte_file(*pte));
1052 } else if (PAGE_MIGRATION && (TTU_ACTION(flags) == TTU_MIGRATION)) { 1052 } else if (PAGE_MIGRATION && (TTU_ACTION(flags) == TTU_MIGRATION)) {
1053 /* Establish migration entry for a file page */ 1053 /* Establish migration entry for a file page */
1054 swp_entry_t entry; 1054 swp_entry_t entry;
1055 entry = make_migration_entry(page, pte_write(pteval)); 1055 entry = make_migration_entry(page, pte_write(pteval));
1056 set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); 1056 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
1057 } else 1057 } else
1058 dec_mm_counter(mm, MM_FILEPAGES); 1058 dec_mm_counter(mm, MM_FILEPAGES);
1059 1059
1060 page_remove_rmap(page); 1060 page_remove_rmap(page);
1061 page_cache_release(page); 1061 page_cache_release(page);
1062 1062
1063 out_unmap: 1063 out_unmap:
1064 pte_unmap_unlock(pte, ptl); 1064 pte_unmap_unlock(pte, ptl);
1065 out: 1065 out:
1066 return ret; 1066 return ret;
1067 1067
1068 out_mlock: 1068 out_mlock:
1069 pte_unmap_unlock(pte, ptl); 1069 pte_unmap_unlock(pte, ptl);
1070 1070
1071 1071
1072 /* 1072 /*
1073 * We need mmap_sem locking, Otherwise VM_LOCKED check makes 1073 * We need mmap_sem locking, Otherwise VM_LOCKED check makes
1074 * unstable result and race. Plus, We can't wait here because 1074 * unstable result and race. Plus, We can't wait here because
1075 * we now hold anon_vma->lock or mapping->i_mmap_lock. 1075 * we now hold anon_vma->lock or mapping->i_mmap_lock.
1076 * if trylock failed, the page remain in evictable lru and later 1076 * if trylock failed, the page remain in evictable lru and later
1077 * vmscan could retry to move the page to unevictable lru if the 1077 * vmscan could retry to move the page to unevictable lru if the
1078 * page is actually mlocked. 1078 * page is actually mlocked.
1079 */ 1079 */
1080 if (down_read_trylock(&vma->vm_mm->mmap_sem)) { 1080 if (down_read_trylock(&vma->vm_mm->mmap_sem)) {
1081 if (vma->vm_flags & VM_LOCKED) { 1081 if (vma->vm_flags & VM_LOCKED) {
1082 mlock_vma_page(page); 1082 mlock_vma_page(page);
1083 ret = SWAP_MLOCK; 1083 ret = SWAP_MLOCK;
1084 } 1084 }
1085 up_read(&vma->vm_mm->mmap_sem); 1085 up_read(&vma->vm_mm->mmap_sem);
1086 } 1086 }
1087 return ret; 1087 return ret;
1088 } 1088 }
1089 1089
1090 /* 1090 /*
1091 * objrmap doesn't work for nonlinear VMAs because the assumption that 1091 * objrmap doesn't work for nonlinear VMAs because the assumption that
1092 * offset-into-file correlates with offset-into-virtual-addresses does not hold. 1092 * offset-into-file correlates with offset-into-virtual-addresses does not hold.
1093 * Consequently, given a particular page and its ->index, we cannot locate the 1093 * Consequently, given a particular page and its ->index, we cannot locate the
1094 * ptes which are mapping that page without an exhaustive linear search. 1094 * ptes which are mapping that page without an exhaustive linear search.
1095 * 1095 *
1096 * So what this code does is a mini "virtual scan" of each nonlinear VMA which 1096 * So what this code does is a mini "virtual scan" of each nonlinear VMA which
1097 * maps the file to which the target page belongs. The ->vm_private_data field 1097 * maps the file to which the target page belongs. The ->vm_private_data field
1098 * holds the current cursor into that scan. Successive searches will circulate 1098 * holds the current cursor into that scan. Successive searches will circulate
1099 * around the vma's virtual address space. 1099 * around the vma's virtual address space.
1100 * 1100 *
1101 * So as more replacement pressure is applied to the pages in a nonlinear VMA, 1101 * So as more replacement pressure is applied to the pages in a nonlinear VMA,
1102 * more scanning pressure is placed against them as well. Eventually pages 1102 * more scanning pressure is placed against them as well. Eventually pages
1103 * will become fully unmapped and are eligible for eviction. 1103 * will become fully unmapped and are eligible for eviction.
1104 * 1104 *
1105 * For very sparsely populated VMAs this is a little inefficient - chances are 1105 * For very sparsely populated VMAs this is a little inefficient - chances are
1106 * there there won't be many ptes located within the scan cluster. In this case 1106 * there there won't be many ptes located within the scan cluster. In this case
1107 * maybe we could scan further - to the end of the pte page, perhaps. 1107 * maybe we could scan further - to the end of the pte page, perhaps.
1108 * 1108 *
1109 * Mlocked pages: check VM_LOCKED under mmap_sem held for read, if we can 1109 * Mlocked pages: check VM_LOCKED under mmap_sem held for read, if we can
1110 * acquire it without blocking. If vma locked, mlock the pages in the cluster, 1110 * acquire it without blocking. If vma locked, mlock the pages in the cluster,
1111 * rather than unmapping them. If we encounter the "check_page" that vmscan is 1111 * rather than unmapping them. If we encounter the "check_page" that vmscan is
1112 * trying to unmap, return SWAP_MLOCK, else default SWAP_AGAIN. 1112 * trying to unmap, return SWAP_MLOCK, else default SWAP_AGAIN.
1113 */ 1113 */
1114 #define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE) 1114 #define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE)
1115 #define CLUSTER_MASK (~(CLUSTER_SIZE - 1)) 1115 #define CLUSTER_MASK (~(CLUSTER_SIZE - 1))
1116 1116
1117 static int try_to_unmap_cluster(unsigned long cursor, unsigned int *mapcount, 1117 static int try_to_unmap_cluster(unsigned long cursor, unsigned int *mapcount,
1118 struct vm_area_struct *vma, struct page *check_page) 1118 struct vm_area_struct *vma, struct page *check_page)
1119 { 1119 {
1120 struct mm_struct *mm = vma->vm_mm; 1120 struct mm_struct *mm = vma->vm_mm;
1121 pgd_t *pgd; 1121 pgd_t *pgd;
1122 pud_t *pud; 1122 pud_t *pud;
1123 pmd_t *pmd; 1123 pmd_t *pmd;
1124 pte_t *pte; 1124 pte_t *pte;
1125 pte_t pteval; 1125 pte_t pteval;
1126 spinlock_t *ptl; 1126 spinlock_t *ptl;
1127 struct page *page; 1127 struct page *page;
1128 unsigned long address; 1128 unsigned long address;
1129 unsigned long end; 1129 unsigned long end;
1130 int ret = SWAP_AGAIN; 1130 int ret = SWAP_AGAIN;
1131 int locked_vma = 0; 1131 int locked_vma = 0;
1132 1132
1133 address = (vma->vm_start + cursor) & CLUSTER_MASK; 1133 address = (vma->vm_start + cursor) & CLUSTER_MASK;
1134 end = address + CLUSTER_SIZE; 1134 end = address + CLUSTER_SIZE;
1135 if (address < vma->vm_start) 1135 if (address < vma->vm_start)
1136 address = vma->vm_start; 1136 address = vma->vm_start;
1137 if (end > vma->vm_end) 1137 if (end > vma->vm_end)
1138 end = vma->vm_end; 1138 end = vma->vm_end;
1139 1139
1140 pgd = pgd_offset(mm, address); 1140 pgd = pgd_offset(mm, address);
1141 if (!pgd_present(*pgd)) 1141 if (!pgd_present(*pgd))
1142 return ret; 1142 return ret;
1143 1143
1144 pud = pud_offset(pgd, address); 1144 pud = pud_offset(pgd, address);
1145 if (!pud_present(*pud)) 1145 if (!pud_present(*pud))
1146 return ret; 1146 return ret;
1147 1147
1148 pmd = pmd_offset(pud, address); 1148 pmd = pmd_offset(pud, address);
1149 if (!pmd_present(*pmd)) 1149 if (!pmd_present(*pmd))
1150 return ret; 1150 return ret;
1151 1151
1152 /* 1152 /*
1153 * If we can acquire the mmap_sem for read, and vma is VM_LOCKED, 1153 * If we can acquire the mmap_sem for read, and vma is VM_LOCKED,
1154 * keep the sem while scanning the cluster for mlocking pages. 1154 * keep the sem while scanning the cluster for mlocking pages.
1155 */ 1155 */
1156 if (down_read_trylock(&vma->vm_mm->mmap_sem)) { 1156 if (down_read_trylock(&vma->vm_mm->mmap_sem)) {
1157 locked_vma = (vma->vm_flags & VM_LOCKED); 1157 locked_vma = (vma->vm_flags & VM_LOCKED);
1158 if (!locked_vma) 1158 if (!locked_vma)
1159 up_read(&vma->vm_mm->mmap_sem); /* don't need it */ 1159 up_read(&vma->vm_mm->mmap_sem); /* don't need it */
1160 } 1160 }
1161 1161
1162 pte = pte_offset_map_lock(mm, pmd, address, &ptl); 1162 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
1163 1163
1164 /* Update high watermark before we lower rss */ 1164 /* Update high watermark before we lower rss */
1165 update_hiwater_rss(mm); 1165 update_hiwater_rss(mm);
1166 1166
1167 for (; address < end; pte++, address += PAGE_SIZE) { 1167 for (; address < end; pte++, address += PAGE_SIZE) {
1168 if (!pte_present(*pte)) 1168 if (!pte_present(*pte))
1169 continue; 1169 continue;
1170 page = vm_normal_page(vma, address, *pte); 1170 page = vm_normal_page(vma, address, *pte);
1171 BUG_ON(!page || PageAnon(page)); 1171 BUG_ON(!page || PageAnon(page));
1172 1172
1173 if (locked_vma) { 1173 if (locked_vma) {
1174 mlock_vma_page(page); /* no-op if already mlocked */ 1174 mlock_vma_page(page); /* no-op if already mlocked */
1175 if (page == check_page) 1175 if (page == check_page)
1176 ret = SWAP_MLOCK; 1176 ret = SWAP_MLOCK;
1177 continue; /* don't unmap */ 1177 continue; /* don't unmap */
1178 } 1178 }
1179 1179
1180 if (ptep_clear_flush_young_notify(vma, address, pte)) 1180 if (ptep_clear_flush_young_notify(vma, address, pte))
1181 continue; 1181 continue;
1182 1182
1183 /* Nuke the page table entry. */ 1183 /* Nuke the page table entry. */
1184 flush_cache_page(vma, address, pte_pfn(*pte)); 1184 flush_cache_page(vma, address, pte_pfn(*pte));
1185 pteval = ptep_clear_flush_notify(vma, address, pte); 1185 pteval = ptep_clear_flush_notify(vma, address, pte);
1186 1186
1187 /* If nonlinear, store the file page offset in the pte. */ 1187 /* If nonlinear, store the file page offset in the pte. */
1188 if (page->index != linear_page_index(vma, address)) 1188 if (page->index != linear_page_index(vma, address))
1189 set_pte_at(mm, address, pte, pgoff_to_pte(page->index)); 1189 set_pte_at(mm, address, pte, pgoff_to_pte(page->index));
1190 1190
1191 /* Move the dirty bit to the physical page now the pte is gone. */ 1191 /* Move the dirty bit to the physical page now the pte is gone. */
1192 if (pte_dirty(pteval)) 1192 if (pte_dirty(pteval))
1193 set_page_dirty(page); 1193 set_page_dirty(page);
1194 1194
1195 page_remove_rmap(page); 1195 page_remove_rmap(page);
1196 page_cache_release(page); 1196 page_cache_release(page);
1197 dec_mm_counter(mm, MM_FILEPAGES); 1197 dec_mm_counter(mm, MM_FILEPAGES);
1198 (*mapcount)--; 1198 (*mapcount)--;
1199 } 1199 }
1200 pte_unmap_unlock(pte - 1, ptl); 1200 pte_unmap_unlock(pte - 1, ptl);
1201 if (locked_vma) 1201 if (locked_vma)
1202 up_read(&vma->vm_mm->mmap_sem); 1202 up_read(&vma->vm_mm->mmap_sem);
1203 return ret; 1203 return ret;
1204 } 1204 }
1205 1205
1206 static bool is_vma_temporary_stack(struct vm_area_struct *vma) 1206 static bool is_vma_temporary_stack(struct vm_area_struct *vma)
1207 { 1207 {
1208 int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP); 1208 int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP);
1209 1209
1210 if (!maybe_stack) 1210 if (!maybe_stack)
1211 return false; 1211 return false;
1212 1212
1213 if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) == 1213 if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) ==
1214 VM_STACK_INCOMPLETE_SETUP) 1214 VM_STACK_INCOMPLETE_SETUP)
1215 return true; 1215 return true;
1216 1216
1217 return false; 1217 return false;
1218 } 1218 }
1219 1219
1220 /** 1220 /**
1221 * try_to_unmap_anon - unmap or unlock anonymous page using the object-based 1221 * try_to_unmap_anon - unmap or unlock anonymous page using the object-based
1222 * rmap method 1222 * rmap method
1223 * @page: the page to unmap/unlock 1223 * @page: the page to unmap/unlock
1224 * @flags: action and flags 1224 * @flags: action and flags
1225 * 1225 *
1226 * Find all the mappings of a page using the mapping pointer and the vma chains 1226 * Find all the mappings of a page using the mapping pointer and the vma chains
1227 * contained in the anon_vma struct it points to. 1227 * contained in the anon_vma struct it points to.
1228 * 1228 *
1229 * This function is only called from try_to_unmap/try_to_munlock for 1229 * This function is only called from try_to_unmap/try_to_munlock for
1230 * anonymous pages. 1230 * anonymous pages.
1231 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma 1231 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1232 * where the page was found will be held for write. So, we won't recheck 1232 * where the page was found will be held for write. So, we won't recheck
1233 * vm_flags for that VMA. That should be OK, because that vma shouldn't be 1233 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1234 * 'LOCKED. 1234 * 'LOCKED.
1235 */ 1235 */
1236 static int try_to_unmap_anon(struct page *page, enum ttu_flags flags) 1236 static int try_to_unmap_anon(struct page *page, enum ttu_flags flags)
1237 { 1237 {
1238 struct anon_vma *anon_vma; 1238 struct anon_vma *anon_vma;
1239 struct anon_vma_chain *avc; 1239 struct anon_vma_chain *avc;
1240 int ret = SWAP_AGAIN; 1240 int ret = SWAP_AGAIN;
1241 1241
1242 anon_vma = page_lock_anon_vma(page); 1242 anon_vma = page_lock_anon_vma(page);
1243 if (!anon_vma) 1243 if (!anon_vma)
1244 return ret; 1244 return ret;
1245 1245
1246 list_for_each_entry(avc, &anon_vma->head, same_anon_vma) { 1246 list_for_each_entry(avc, &anon_vma->head, same_anon_vma) {
1247 struct vm_area_struct *vma = avc->vma; 1247 struct vm_area_struct *vma = avc->vma;
1248 unsigned long address; 1248 unsigned long address;
1249 1249
1250 /* 1250 /*
1251 * During exec, a temporary VMA is setup and later moved. 1251 * During exec, a temporary VMA is setup and later moved.
1252 * The VMA is moved under the anon_vma lock but not the 1252 * The VMA is moved under the anon_vma lock but not the
1253 * page tables leading to a race where migration cannot 1253 * page tables leading to a race where migration cannot
1254 * find the migration ptes. Rather than increasing the 1254 * find the migration ptes. Rather than increasing the
1255 * locking requirements of exec(), migration skips 1255 * locking requirements of exec(), migration skips
1256 * temporary VMAs until after exec() completes. 1256 * temporary VMAs until after exec() completes.
1257 */ 1257 */
1258 if (PAGE_MIGRATION && (flags & TTU_MIGRATION) && 1258 if (PAGE_MIGRATION && (flags & TTU_MIGRATION) &&
1259 is_vma_temporary_stack(vma)) 1259 is_vma_temporary_stack(vma))
1260 continue; 1260 continue;
1261 1261
1262 address = vma_address(page, vma); 1262 address = vma_address(page, vma);
1263 if (address == -EFAULT) 1263 if (address == -EFAULT)
1264 continue; 1264 continue;
1265 ret = try_to_unmap_one(page, vma, address, flags); 1265 ret = try_to_unmap_one(page, vma, address, flags);
1266 if (ret != SWAP_AGAIN || !page_mapped(page)) 1266 if (ret != SWAP_AGAIN || !page_mapped(page))
1267 break; 1267 break;
1268 } 1268 }
1269 1269
1270 page_unlock_anon_vma(anon_vma); 1270 page_unlock_anon_vma(anon_vma);
1271 return ret; 1271 return ret;
1272 } 1272 }
1273 1273
1274 /** 1274 /**
1275 * try_to_unmap_file - unmap/unlock file page using the object-based rmap method 1275 * try_to_unmap_file - unmap/unlock file page using the object-based rmap method
1276 * @page: the page to unmap/unlock 1276 * @page: the page to unmap/unlock
1277 * @flags: action and flags 1277 * @flags: action and flags
1278 * 1278 *
1279 * Find all the mappings of a page using the mapping pointer and the vma chains 1279 * Find all the mappings of a page using the mapping pointer and the vma chains
1280 * contained in the address_space struct it points to. 1280 * contained in the address_space struct it points to.
1281 * 1281 *
1282 * This function is only called from try_to_unmap/try_to_munlock for 1282 * This function is only called from try_to_unmap/try_to_munlock for
1283 * object-based pages. 1283 * object-based pages.
1284 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma 1284 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1285 * where the page was found will be held for write. So, we won't recheck 1285 * where the page was found will be held for write. So, we won't recheck
1286 * vm_flags for that VMA. That should be OK, because that vma shouldn't be 1286 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1287 * 'LOCKED. 1287 * 'LOCKED.
1288 */ 1288 */
1289 static int try_to_unmap_file(struct page *page, enum ttu_flags flags) 1289 static int try_to_unmap_file(struct page *page, enum ttu_flags flags)
1290 { 1290 {
1291 struct address_space *mapping = page->mapping; 1291 struct address_space *mapping = page->mapping;
1292 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); 1292 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
1293 struct vm_area_struct *vma; 1293 struct vm_area_struct *vma;
1294 struct prio_tree_iter iter; 1294 struct prio_tree_iter iter;
1295 int ret = SWAP_AGAIN; 1295 int ret = SWAP_AGAIN;
1296 unsigned long cursor; 1296 unsigned long cursor;
1297 unsigned long max_nl_cursor = 0; 1297 unsigned long max_nl_cursor = 0;
1298 unsigned long max_nl_size = 0; 1298 unsigned long max_nl_size = 0;
1299 unsigned int mapcount; 1299 unsigned int mapcount;
1300 1300
1301 spin_lock(&mapping->i_mmap_lock); 1301 spin_lock(&mapping->i_mmap_lock);
1302 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { 1302 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
1303 unsigned long address = vma_address(page, vma); 1303 unsigned long address = vma_address(page, vma);
1304 if (address == -EFAULT) 1304 if (address == -EFAULT)
1305 continue; 1305 continue;
1306 ret = try_to_unmap_one(page, vma, address, flags); 1306 ret = try_to_unmap_one(page, vma, address, flags);
1307 if (ret != SWAP_AGAIN || !page_mapped(page)) 1307 if (ret != SWAP_AGAIN || !page_mapped(page))
1308 goto out; 1308 goto out;
1309 } 1309 }
1310 1310
1311 if (list_empty(&mapping->i_mmap_nonlinear)) 1311 if (list_empty(&mapping->i_mmap_nonlinear))
1312 goto out; 1312 goto out;
1313 1313
1314 /* 1314 /*
1315 * We don't bother to try to find the munlocked page in nonlinears. 1315 * We don't bother to try to find the munlocked page in nonlinears.
1316 * It's costly. Instead, later, page reclaim logic may call 1316 * It's costly. Instead, later, page reclaim logic may call
1317 * try_to_unmap(TTU_MUNLOCK) and recover PG_mlocked lazily. 1317 * try_to_unmap(TTU_MUNLOCK) and recover PG_mlocked lazily.
1318 */ 1318 */
1319 if (TTU_ACTION(flags) == TTU_MUNLOCK) 1319 if (TTU_ACTION(flags) == TTU_MUNLOCK)
1320 goto out; 1320 goto out;
1321 1321
1322 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, 1322 list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
1323 shared.vm_set.list) { 1323 shared.vm_set.list) {
1324 cursor = (unsigned long) vma->vm_private_data; 1324 cursor = (unsigned long) vma->vm_private_data;
1325 if (cursor > max_nl_cursor) 1325 if (cursor > max_nl_cursor)
1326 max_nl_cursor = cursor; 1326 max_nl_cursor = cursor;
1327 cursor = vma->vm_end - vma->vm_start; 1327 cursor = vma->vm_end - vma->vm_start;
1328 if (cursor > max_nl_size) 1328 if (cursor > max_nl_size)
1329 max_nl_size = cursor; 1329 max_nl_size = cursor;
1330 } 1330 }
1331 1331
1332 if (max_nl_size == 0) { /* all nonlinears locked or reserved ? */ 1332 if (max_nl_size == 0) { /* all nonlinears locked or reserved ? */
1333 ret = SWAP_FAIL; 1333 ret = SWAP_FAIL;
1334 goto out; 1334 goto out;
1335 } 1335 }
1336 1336
1337 /* 1337 /*
1338 * We don't try to search for this page in the nonlinear vmas, 1338 * We don't try to search for this page in the nonlinear vmas,
1339 * and page_referenced wouldn't have found it anyway. Instead 1339 * and page_referenced wouldn't have found it anyway. Instead
1340 * just walk the nonlinear vmas trying to age and unmap some. 1340 * just walk the nonlinear vmas trying to age and unmap some.
1341 * The mapcount of the page we came in with is irrelevant, 1341 * The mapcount of the page we came in with is irrelevant,
1342 * but even so use it as a guide to how hard we should try? 1342 * but even so use it as a guide to how hard we should try?
1343 */ 1343 */
1344 mapcount = page_mapcount(page); 1344 mapcount = page_mapcount(page);
1345 if (!mapcount) 1345 if (!mapcount)
1346 goto out; 1346 goto out;
1347 cond_resched_lock(&mapping->i_mmap_lock); 1347 cond_resched_lock(&mapping->i_mmap_lock);
1348 1348
1349 max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK; 1349 max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK;
1350 if (max_nl_cursor == 0) 1350 if (max_nl_cursor == 0)
1351 max_nl_cursor = CLUSTER_SIZE; 1351 max_nl_cursor = CLUSTER_SIZE;
1352 1352
1353 do { 1353 do {
1354 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, 1354 list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
1355 shared.vm_set.list) { 1355 shared.vm_set.list) {
1356 cursor = (unsigned long) vma->vm_private_data; 1356 cursor = (unsigned long) vma->vm_private_data;
1357 while ( cursor < max_nl_cursor && 1357 while ( cursor < max_nl_cursor &&
1358 cursor < vma->vm_end - vma->vm_start) { 1358 cursor < vma->vm_end - vma->vm_start) {
1359 if (try_to_unmap_cluster(cursor, &mapcount, 1359 if (try_to_unmap_cluster(cursor, &mapcount,
1360 vma, page) == SWAP_MLOCK) 1360 vma, page) == SWAP_MLOCK)
1361 ret = SWAP_MLOCK; 1361 ret = SWAP_MLOCK;
1362 cursor += CLUSTER_SIZE; 1362 cursor += CLUSTER_SIZE;
1363 vma->vm_private_data = (void *) cursor; 1363 vma->vm_private_data = (void *) cursor;
1364 if ((int)mapcount <= 0) 1364 if ((int)mapcount <= 0)
1365 goto out; 1365 goto out;
1366 } 1366 }
1367 vma->vm_private_data = (void *) max_nl_cursor; 1367 vma->vm_private_data = (void *) max_nl_cursor;
1368 } 1368 }
1369 cond_resched_lock(&mapping->i_mmap_lock); 1369 cond_resched_lock(&mapping->i_mmap_lock);
1370 max_nl_cursor += CLUSTER_SIZE; 1370 max_nl_cursor += CLUSTER_SIZE;
1371 } while (max_nl_cursor <= max_nl_size); 1371 } while (max_nl_cursor <= max_nl_size);
1372 1372
1373 /* 1373 /*
1374 * Don't loop forever (perhaps all the remaining pages are 1374 * Don't loop forever (perhaps all the remaining pages are
1375 * in locked vmas). Reset cursor on all unreserved nonlinear 1375 * in locked vmas). Reset cursor on all unreserved nonlinear
1376 * vmas, now forgetting on which ones it had fallen behind. 1376 * vmas, now forgetting on which ones it had fallen behind.
1377 */ 1377 */
1378 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list) 1378 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
1379 vma->vm_private_data = NULL; 1379 vma->vm_private_data = NULL;
1380 out: 1380 out:
1381 spin_unlock(&mapping->i_mmap_lock); 1381 spin_unlock(&mapping->i_mmap_lock);
1382 return ret; 1382 return ret;
1383 } 1383 }
1384 1384
1385 /** 1385 /**
1386 * try_to_unmap - try to remove all page table mappings to a page 1386 * try_to_unmap - try to remove all page table mappings to a page
1387 * @page: the page to get unmapped 1387 * @page: the page to get unmapped
1388 * @flags: action and flags 1388 * @flags: action and flags
1389 * 1389 *
1390 * Tries to remove all the page table entries which are mapping this 1390 * Tries to remove all the page table entries which are mapping this
1391 * page, used in the pageout path. Caller must hold the page lock. 1391 * page, used in the pageout path. Caller must hold the page lock.
1392 * Return values are: 1392 * Return values are:
1393 * 1393 *
1394 * SWAP_SUCCESS - we succeeded in removing all mappings 1394 * SWAP_SUCCESS - we succeeded in removing all mappings
1395 * SWAP_AGAIN - we missed a mapping, try again later 1395 * SWAP_AGAIN - we missed a mapping, try again later
1396 * SWAP_FAIL - the page is unswappable 1396 * SWAP_FAIL - the page is unswappable
1397 * SWAP_MLOCK - page is mlocked. 1397 * SWAP_MLOCK - page is mlocked.
1398 */ 1398 */
1399 int try_to_unmap(struct page *page, enum ttu_flags flags) 1399 int try_to_unmap(struct page *page, enum ttu_flags flags)
1400 { 1400 {
1401 int ret; 1401 int ret;
1402 1402
1403 BUG_ON(!PageLocked(page)); 1403 BUG_ON(!PageLocked(page));
1404 1404
1405 if (unlikely(PageKsm(page))) 1405 if (unlikely(PageKsm(page)))
1406 ret = try_to_unmap_ksm(page, flags); 1406 ret = try_to_unmap_ksm(page, flags);
1407 else if (PageAnon(page)) 1407 else if (PageAnon(page))
1408 ret = try_to_unmap_anon(page, flags); 1408 ret = try_to_unmap_anon(page, flags);
1409 else 1409 else
1410 ret = try_to_unmap_file(page, flags); 1410 ret = try_to_unmap_file(page, flags);
1411 if (ret != SWAP_MLOCK && !page_mapped(page)) 1411 if (ret != SWAP_MLOCK && !page_mapped(page))
1412 ret = SWAP_SUCCESS; 1412 ret = SWAP_SUCCESS;
1413 return ret; 1413 return ret;
1414 } 1414 }
1415 1415
1416 /** 1416 /**
1417 * try_to_munlock - try to munlock a page 1417 * try_to_munlock - try to munlock a page
1418 * @page: the page to be munlocked 1418 * @page: the page to be munlocked
1419 * 1419 *
1420 * Called from munlock code. Checks all of the VMAs mapping the page 1420 * Called from munlock code. Checks all of the VMAs mapping the page
1421 * to make sure nobody else has this page mlocked. The page will be 1421 * to make sure nobody else has this page mlocked. The page will be
1422 * returned with PG_mlocked cleared if no other vmas have it mlocked. 1422 * returned with PG_mlocked cleared if no other vmas have it mlocked.
1423 * 1423 *
1424 * Return values are: 1424 * Return values are:
1425 * 1425 *
1426 * SWAP_AGAIN - no vma is holding page mlocked, or, 1426 * SWAP_AGAIN - no vma is holding page mlocked, or,
1427 * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem 1427 * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem
1428 * SWAP_FAIL - page cannot be located at present 1428 * SWAP_FAIL - page cannot be located at present
1429 * SWAP_MLOCK - page is now mlocked. 1429 * SWAP_MLOCK - page is now mlocked.
1430 */ 1430 */
1431 int try_to_munlock(struct page *page) 1431 int try_to_munlock(struct page *page)
1432 { 1432 {
1433 VM_BUG_ON(!PageLocked(page) || PageLRU(page)); 1433 VM_BUG_ON(!PageLocked(page) || PageLRU(page));
1434 1434
1435 if (unlikely(PageKsm(page))) 1435 if (unlikely(PageKsm(page)))
1436 return try_to_unmap_ksm(page, TTU_MUNLOCK); 1436 return try_to_unmap_ksm(page, TTU_MUNLOCK);
1437 else if (PageAnon(page)) 1437 else if (PageAnon(page))
1438 return try_to_unmap_anon(page, TTU_MUNLOCK); 1438 return try_to_unmap_anon(page, TTU_MUNLOCK);
1439 else 1439 else
1440 return try_to_unmap_file(page, TTU_MUNLOCK); 1440 return try_to_unmap_file(page, TTU_MUNLOCK);
1441 } 1441 }
1442 1442
1443 #if defined(CONFIG_KSM) || defined(CONFIG_MIGRATION) 1443 #if defined(CONFIG_KSM) || defined(CONFIG_MIGRATION)
1444 /* 1444 /*
1445 * Drop an anon_vma refcount, freeing the anon_vma and anon_vma->root 1445 * Drop an anon_vma refcount, freeing the anon_vma and anon_vma->root
1446 * if necessary. Be careful to do all the tests under the lock. Once 1446 * if necessary. Be careful to do all the tests under the lock. Once
1447 * we know we are the last user, nobody else can get a reference and we 1447 * we know we are the last user, nobody else can get a reference and we
1448 * can do the freeing without the lock. 1448 * can do the freeing without the lock.
1449 */ 1449 */
1450 void drop_anon_vma(struct anon_vma *anon_vma) 1450 void drop_anon_vma(struct anon_vma *anon_vma)
1451 { 1451 {
1452 BUG_ON(atomic_read(&anon_vma->external_refcount) <= 0); 1452 BUG_ON(atomic_read(&anon_vma->external_refcount) <= 0);
1453 if (atomic_dec_and_lock(&anon_vma->external_refcount, &anon_vma->root->lock)) { 1453 if (atomic_dec_and_lock(&anon_vma->external_refcount, &anon_vma->root->lock)) {
1454 struct anon_vma *root = anon_vma->root; 1454 struct anon_vma *root = anon_vma->root;
1455 int empty = list_empty(&anon_vma->head); 1455 int empty = list_empty(&anon_vma->head);
1456 int last_root_user = 0; 1456 int last_root_user = 0;
1457 int root_empty = 0; 1457 int root_empty = 0;
1458 1458
1459 /* 1459 /*
1460 * The refcount on a non-root anon_vma got dropped. Drop 1460 * The refcount on a non-root anon_vma got dropped. Drop
1461 * the refcount on the root and check if we need to free it. 1461 * the refcount on the root and check if we need to free it.
1462 */ 1462 */
1463 if (empty && anon_vma != root) { 1463 if (empty && anon_vma != root) {
1464 BUG_ON(atomic_read(&root->external_refcount) <= 0); 1464 BUG_ON(atomic_read(&root->external_refcount) <= 0);
1465 last_root_user = atomic_dec_and_test(&root->external_refcount); 1465 last_root_user = atomic_dec_and_test(&root->external_refcount);
1466 root_empty = list_empty(&root->head); 1466 root_empty = list_empty(&root->head);
1467 } 1467 }
1468 anon_vma_unlock(anon_vma); 1468 anon_vma_unlock(anon_vma);
1469 1469
1470 if (empty) { 1470 if (empty) {
1471 anon_vma_free(anon_vma); 1471 anon_vma_free(anon_vma);
1472 if (root_empty && last_root_user) 1472 if (root_empty && last_root_user)
1473 anon_vma_free(root); 1473 anon_vma_free(root);
1474 } 1474 }
1475 } 1475 }
1476 } 1476 }
1477 #endif 1477 #endif
1478 1478
1479 #ifdef CONFIG_MIGRATION 1479 #ifdef CONFIG_MIGRATION
1480 /* 1480 /*
1481 * rmap_walk() and its helpers rmap_walk_anon() and rmap_walk_file(): 1481 * rmap_walk() and its helpers rmap_walk_anon() and rmap_walk_file():
1482 * Called by migrate.c to remove migration ptes, but might be used more later. 1482 * Called by migrate.c to remove migration ptes, but might be used more later.
1483 */ 1483 */
1484 static int rmap_walk_anon(struct page *page, int (*rmap_one)(struct page *, 1484 static int rmap_walk_anon(struct page *page, int (*rmap_one)(struct page *,
1485 struct vm_area_struct *, unsigned long, void *), void *arg) 1485 struct vm_area_struct *, unsigned long, void *), void *arg)
1486 { 1486 {
1487 struct anon_vma *anon_vma; 1487 struct anon_vma *anon_vma;
1488 struct anon_vma_chain *avc; 1488 struct anon_vma_chain *avc;
1489 int ret = SWAP_AGAIN; 1489 int ret = SWAP_AGAIN;
1490 1490
1491 /* 1491 /*
1492 * Note: remove_migration_ptes() cannot use page_lock_anon_vma() 1492 * Note: remove_migration_ptes() cannot use page_lock_anon_vma()
1493 * because that depends on page_mapped(); but not all its usages 1493 * because that depends on page_mapped(); but not all its usages
1494 * are holding mmap_sem. Users without mmap_sem are required to 1494 * are holding mmap_sem. Users without mmap_sem are required to
1495 * take a reference count to prevent the anon_vma disappearing 1495 * take a reference count to prevent the anon_vma disappearing
1496 */ 1496 */
1497 anon_vma = page_anon_vma(page); 1497 anon_vma = page_anon_vma(page);
1498 if (!anon_vma) 1498 if (!anon_vma)
1499 return ret; 1499 return ret;
1500 anon_vma_lock(anon_vma); 1500 anon_vma_lock(anon_vma);
1501 list_for_each_entry(avc, &anon_vma->head, same_anon_vma) { 1501 list_for_each_entry(avc, &anon_vma->head, same_anon_vma) {
1502 struct vm_area_struct *vma = avc->vma; 1502 struct vm_area_struct *vma = avc->vma;
1503 unsigned long address = vma_address(page, vma); 1503 unsigned long address = vma_address(page, vma);
1504 if (address == -EFAULT) 1504 if (address == -EFAULT)
1505 continue; 1505 continue;
1506 ret = rmap_one(page, vma, address, arg); 1506 ret = rmap_one(page, vma, address, arg);
1507 if (ret != SWAP_AGAIN) 1507 if (ret != SWAP_AGAIN)
1508 break; 1508 break;
1509 } 1509 }
1510 anon_vma_unlock(anon_vma); 1510 anon_vma_unlock(anon_vma);
1511 return ret; 1511 return ret;
1512 } 1512 }
1513 1513
1514 static int rmap_walk_file(struct page *page, int (*rmap_one)(struct page *, 1514 static int rmap_walk_file(struct page *page, int (*rmap_one)(struct page *,
1515 struct vm_area_struct *, unsigned long, void *), void *arg) 1515 struct vm_area_struct *, unsigned long, void *), void *arg)
1516 { 1516 {
1517 struct address_space *mapping = page->mapping; 1517 struct address_space *mapping = page->mapping;
1518 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); 1518 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
1519 struct vm_area_struct *vma; 1519 struct vm_area_struct *vma;
1520 struct prio_tree_iter iter; 1520 struct prio_tree_iter iter;
1521 int ret = SWAP_AGAIN; 1521 int ret = SWAP_AGAIN;
1522 1522
1523 if (!mapping) 1523 if (!mapping)
1524 return ret; 1524 return ret;
1525 spin_lock(&mapping->i_mmap_lock); 1525 spin_lock(&mapping->i_mmap_lock);
1526 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { 1526 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
1527 unsigned long address = vma_address(page, vma); 1527 unsigned long address = vma_address(page, vma);
1528 if (address == -EFAULT) 1528 if (address == -EFAULT)
1529 continue; 1529 continue;
1530 ret = rmap_one(page, vma, address, arg); 1530 ret = rmap_one(page, vma, address, arg);
1531 if (ret != SWAP_AGAIN) 1531 if (ret != SWAP_AGAIN)
1532 break; 1532 break;
1533 } 1533 }
1534 /* 1534 /*
1535 * No nonlinear handling: being always shared, nonlinear vmas 1535 * No nonlinear handling: being always shared, nonlinear vmas
1536 * never contain migration ptes. Decide what to do about this 1536 * never contain migration ptes. Decide what to do about this
1537 * limitation to linear when we need rmap_walk() on nonlinear. 1537 * limitation to linear when we need rmap_walk() on nonlinear.
1538 */ 1538 */
1539 spin_unlock(&mapping->i_mmap_lock); 1539 spin_unlock(&mapping->i_mmap_lock);
1540 return ret; 1540 return ret;
1541 } 1541 }
1542 1542
1543 int rmap_walk(struct page *page, int (*rmap_one)(struct page *, 1543 int rmap_walk(struct page *page, int (*rmap_one)(struct page *,
1544 struct vm_area_struct *, unsigned long, void *), void *arg) 1544 struct vm_area_struct *, unsigned long, void *), void *arg)
1545 { 1545 {
1546 VM_BUG_ON(!PageLocked(page)); 1546 VM_BUG_ON(!PageLocked(page));
1547 1547
1548 if (unlikely(PageKsm(page))) 1548 if (unlikely(PageKsm(page)))
1549 return rmap_walk_ksm(page, rmap_one, arg); 1549 return rmap_walk_ksm(page, rmap_one, arg);
1550 else if (PageAnon(page)) 1550 else if (PageAnon(page))
1551 return rmap_walk_anon(page, rmap_one, arg); 1551 return rmap_walk_anon(page, rmap_one, arg);
1552 else 1552 else
1553 return rmap_walk_file(page, rmap_one, arg); 1553 return rmap_walk_file(page, rmap_one, arg);
1554 } 1554 }
1555 #endif /* CONFIG_MIGRATION */ 1555 #endif /* CONFIG_MIGRATION */
1556 1556
1557 #ifdef CONFIG_HUGETLB_PAGE 1557 #ifdef CONFIG_HUGETLB_PAGE
1558 /* 1558 /*
1559 * The following three functions are for anonymous (private mapped) hugepages. 1559 * The following three functions are for anonymous (private mapped) hugepages.
1560 * Unlike common anonymous pages, anonymous hugepages have no accounting code 1560 * Unlike common anonymous pages, anonymous hugepages have no accounting code
1561 * and no lru code, because we handle hugepages differently from common pages. 1561 * and no lru code, because we handle hugepages differently from common pages.
1562 */ 1562 */
1563 static void __hugepage_set_anon_rmap(struct page *page, 1563 static void __hugepage_set_anon_rmap(struct page *page,
1564 struct vm_area_struct *vma, unsigned long address, int exclusive) 1564 struct vm_area_struct *vma, unsigned long address, int exclusive)
1565 { 1565 {
1566 struct anon_vma *anon_vma = vma->anon_vma; 1566 struct anon_vma *anon_vma = vma->anon_vma;
1567 1567
1568 BUG_ON(!anon_vma); 1568 BUG_ON(!anon_vma);
1569 1569
1570 if (PageAnon(page)) 1570 if (PageAnon(page))
1571 return; 1571 return;
1572 if (!exclusive) 1572 if (!exclusive)
1573 anon_vma = anon_vma->root; 1573 anon_vma = anon_vma->root;
1574 1574
1575 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; 1575 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
1576 page->mapping = (struct address_space *) anon_vma; 1576 page->mapping = (struct address_space *) anon_vma;
1577 page->index = linear_page_index(vma, address); 1577 page->index = linear_page_index(vma, address);
1578 } 1578 }
1579 1579
1580 void hugepage_add_anon_rmap(struct page *page, 1580 void hugepage_add_anon_rmap(struct page *page,
1581 struct vm_area_struct *vma, unsigned long address) 1581 struct vm_area_struct *vma, unsigned long address)
1582 { 1582 {
1583 struct anon_vma *anon_vma = vma->anon_vma; 1583 struct anon_vma *anon_vma = vma->anon_vma;
1584 int first; 1584 int first;
1585
1586 BUG_ON(!PageLocked(page));
1585 BUG_ON(!anon_vma); 1587 BUG_ON(!anon_vma);
1586 BUG_ON(address < vma->vm_start || address >= vma->vm_end); 1588 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
1587 first = atomic_inc_and_test(&page->_mapcount); 1589 first = atomic_inc_and_test(&page->_mapcount);
1588 if (first) 1590 if (first)
1589 __hugepage_set_anon_rmap(page, vma, address, 0); 1591 __hugepage_set_anon_rmap(page, vma, address, 0);
1590 } 1592 }
1591 1593
1592 void hugepage_add_new_anon_rmap(struct page *page, 1594 void hugepage_add_new_anon_rmap(struct page *page,
1593 struct vm_area_struct *vma, unsigned long address) 1595 struct vm_area_struct *vma, unsigned long address)
1594 { 1596 {
1595 BUG_ON(address < vma->vm_start || address >= vma->vm_end); 1597 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
1596 atomic_set(&page->_mapcount, 0); 1598 atomic_set(&page->_mapcount, 0);
1597 __hugepage_set_anon_rmap(page, vma, address, 1); 1599 __hugepage_set_anon_rmap(page, vma, address, 1);
1598 } 1600 }
1599 #endif /* CONFIG_HUGETLB_PAGE */ 1601 #endif /* CONFIG_HUGETLB_PAGE */
1600 1602