blob: 7b2656055d6a96a6b940574e4d23f0a4fc7dd6bc [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/mlock.c
3 *
4 * (C) Copyright 1995 Linus Torvalds
5 * (C) Copyright 2002 Christoph Hellwig
6 */
7
Randy.Dunlapc59ede72006-01-11 12:17:46 -08008#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -07009#include <linux/mman.h>
10#include <linux/mm.h>
11#include <linux/mempolicy.h>
12#include <linux/syscalls.h>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040013#include <linux/sched.h>
14#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040016int can_do_mlock(void)
17{
18 if (capable(CAP_IPC_LOCK))
19 return 1;
20 if (current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur != 0)
21 return 1;
22 return 0;
23}
24EXPORT_SYMBOL(can_do_mlock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070025
26static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
27 unsigned long start, unsigned long end, unsigned int newflags)
28{
29 struct mm_struct * mm = vma->vm_mm;
30 pgoff_t pgoff;
31 int pages;
32 int ret = 0;
33
34 if (newflags == vma->vm_flags) {
35 *prev = vma;
36 goto out;
37 }
38
39 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
40 *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma,
41 vma->vm_file, pgoff, vma_policy(vma));
42 if (*prev) {
43 vma = *prev;
44 goto success;
45 }
46
47 *prev = vma;
48
49 if (start != vma->vm_start) {
50 ret = split_vma(mm, vma, start, 1);
51 if (ret)
52 goto out;
53 }
54
55 if (end != vma->vm_end) {
56 ret = split_vma(mm, vma, end, 0);
57 if (ret)
58 goto out;
59 }
60
61success:
62 /*
63 * vm_flags is protected by the mmap_sem held in write mode.
64 * It's okay if try_to_unmap_one unmaps a page just after we
65 * set VM_LOCKED, make_pages_present below will bring it back.
66 */
67 vma->vm_flags = newflags;
68
69 /*
70 * Keep track of amount of locked VM.
71 */
72 pages = (end - start) >> PAGE_SHIFT;
73 if (newflags & VM_LOCKED) {
74 pages = -pages;
75 if (!(newflags & VM_IO))
76 ret = make_pages_present(start, end);
77 }
78
Rik Bobbaersa3eea482006-12-06 20:32:25 -080079 mm->locked_vm -= pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -070080out:
81 if (ret == -ENOMEM)
82 ret = -EAGAIN;
83 return ret;
84}
85
86static int do_mlock(unsigned long start, size_t len, int on)
87{
88 unsigned long nstart, end, tmp;
89 struct vm_area_struct * vma, * prev;
90 int error;
91
92 len = PAGE_ALIGN(len);
93 end = start + len;
94 if (end < start)
95 return -EINVAL;
96 if (end == start)
97 return 0;
98 vma = find_vma_prev(current->mm, start, &prev);
99 if (!vma || vma->vm_start > start)
100 return -ENOMEM;
101
102 if (start > vma->vm_start)
103 prev = vma;
104
105 for (nstart = start ; ; ) {
106 unsigned int newflags;
107
108 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
109
110 newflags = vma->vm_flags | VM_LOCKED;
111 if (!on)
112 newflags &= ~VM_LOCKED;
113
114 tmp = vma->vm_end;
115 if (tmp > end)
116 tmp = end;
117 error = mlock_fixup(vma, &prev, nstart, tmp, newflags);
118 if (error)
119 break;
120 nstart = tmp;
121 if (nstart < prev->vm_end)
122 nstart = prev->vm_end;
123 if (nstart >= end)
124 break;
125
126 vma = prev->vm_next;
127 if (!vma || vma->vm_start != nstart) {
128 error = -ENOMEM;
129 break;
130 }
131 }
132 return error;
133}
134
135asmlinkage long sys_mlock(unsigned long start, size_t len)
136{
137 unsigned long locked;
138 unsigned long lock_limit;
139 int error = -ENOMEM;
140
141 if (!can_do_mlock())
142 return -EPERM;
143
144 down_write(&current->mm->mmap_sem);
145 len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
146 start &= PAGE_MASK;
147
148 locked = len >> PAGE_SHIFT;
149 locked += current->mm->locked_vm;
150
151 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
152 lock_limit >>= PAGE_SHIFT;
153
154 /* check against resource limits */
155 if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
156 error = do_mlock(start, len, 1);
157 up_write(&current->mm->mmap_sem);
158 return error;
159}
160
161asmlinkage long sys_munlock(unsigned long start, size_t len)
162{
163 int ret;
164
165 down_write(&current->mm->mmap_sem);
166 len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
167 start &= PAGE_MASK;
168 ret = do_mlock(start, len, 0);
169 up_write(&current->mm->mmap_sem);
170 return ret;
171}
172
173static int do_mlockall(int flags)
174{
175 struct vm_area_struct * vma, * prev = NULL;
176 unsigned int def_flags = 0;
177
178 if (flags & MCL_FUTURE)
179 def_flags = VM_LOCKED;
180 current->mm->def_flags = def_flags;
181 if (flags == MCL_FUTURE)
182 goto out;
183
184 for (vma = current->mm->mmap; vma ; vma = prev->vm_next) {
185 unsigned int newflags;
186
187 newflags = vma->vm_flags | VM_LOCKED;
188 if (!(flags & MCL_CURRENT))
189 newflags &= ~VM_LOCKED;
190
191 /* Ignore errors */
192 mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags);
193 }
194out:
195 return 0;
196}
197
198asmlinkage long sys_mlockall(int flags)
199{
200 unsigned long lock_limit;
201 int ret = -EINVAL;
202
203 if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE)))
204 goto out;
205
206 ret = -EPERM;
207 if (!can_do_mlock())
208 goto out;
209
210 down_write(&current->mm->mmap_sem);
211
212 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
213 lock_limit >>= PAGE_SHIFT;
214
215 ret = -ENOMEM;
216 if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
217 capable(CAP_IPC_LOCK))
218 ret = do_mlockall(flags);
219 up_write(&current->mm->mmap_sem);
220out:
221 return ret;
222}
223
224asmlinkage long sys_munlockall(void)
225{
226 int ret;
227
228 down_write(&current->mm->mmap_sem);
229 ret = do_mlockall(0);
230 up_write(&current->mm->mmap_sem);
231 return ret;
232}
233
234/*
235 * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
236 * shm segments) get accounted against the user_struct instead.
237 */
238static DEFINE_SPINLOCK(shmlock_user_lock);
239
240int user_shm_lock(size_t size, struct user_struct *user)
241{
242 unsigned long lock_limit, locked;
243 int allowed = 0;
244
245 locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
246 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
Herbert van den Bergh5ed44a42007-07-15 23:38:25 -0700247 if (lock_limit == RLIM_INFINITY)
248 allowed = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 lock_limit >>= PAGE_SHIFT;
250 spin_lock(&shmlock_user_lock);
Herbert van den Bergh5ed44a42007-07-15 23:38:25 -0700251 if (!allowed &&
252 locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 goto out;
254 get_uid(user);
255 user->locked_shm += locked;
256 allowed = 1;
257out:
258 spin_unlock(&shmlock_user_lock);
259 return allowed;
260}
261
262void user_shm_unlock(size_t size, struct user_struct *user)
263{
264 spin_lock(&shmlock_user_lock);
265 user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
266 spin_unlock(&shmlock_user_lock);
267 free_uid(user);
268}