[PATCH] memory hotplug prep: break out zone initialization
[linux-2.6] / mm / mlock.c
1 /*
2  *      linux/mm/mlock.c
3  *
4  *  (C) Copyright 1995 Linus Torvalds
5  *  (C) Copyright 2002 Christoph Hellwig
6  */
7
8 #include <linux/mman.h>
9 #include <linux/mm.h>
10 #include <linux/mempolicy.h>
11 #include <linux/syscalls.h>
12
13
14 static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
15         unsigned long start, unsigned long end, unsigned int newflags)
16 {
17         struct mm_struct * mm = vma->vm_mm;
18         pgoff_t pgoff;
19         int pages;
20         int ret = 0;
21
22         if (newflags == vma->vm_flags) {
23                 *prev = vma;
24                 goto out;
25         }
26
27         pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
28         *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma,
29                           vma->vm_file, pgoff, vma_policy(vma));
30         if (*prev) {
31                 vma = *prev;
32                 goto success;
33         }
34
35         *prev = vma;
36
37         if (start != vma->vm_start) {
38                 ret = split_vma(mm, vma, start, 1);
39                 if (ret)
40                         goto out;
41         }
42
43         if (end != vma->vm_end) {
44                 ret = split_vma(mm, vma, end, 0);
45                 if (ret)
46                         goto out;
47         }
48
49 success:
50         /*
51          * vm_flags is protected by the mmap_sem held in write mode.
52          * It's okay if try_to_unmap_one unmaps a page just after we
53          * set VM_LOCKED, make_pages_present below will bring it back.
54          */
55         vma->vm_flags = newflags;
56
57         /*
58          * Keep track of amount of locked VM.
59          */
60         pages = (end - start) >> PAGE_SHIFT;
61         if (newflags & VM_LOCKED) {
62                 pages = -pages;
63                 if (!(newflags & VM_IO))
64                         ret = make_pages_present(start, end);
65         }
66
67         vma->vm_mm->locked_vm -= pages;
68 out:
69         if (ret == -ENOMEM)
70                 ret = -EAGAIN;
71         return ret;
72 }
73
74 static int do_mlock(unsigned long start, size_t len, int on)
75 {
76         unsigned long nstart, end, tmp;
77         struct vm_area_struct * vma, * prev;
78         int error;
79
80         len = PAGE_ALIGN(len);
81         end = start + len;
82         if (end < start)
83                 return -EINVAL;
84         if (end == start)
85                 return 0;
86         vma = find_vma_prev(current->mm, start, &prev);
87         if (!vma || vma->vm_start > start)
88                 return -ENOMEM;
89
90         if (start > vma->vm_start)
91                 prev = vma;
92
93         for (nstart = start ; ; ) {
94                 unsigned int newflags;
95
96                 /* Here we know that  vma->vm_start <= nstart < vma->vm_end. */
97
98                 newflags = vma->vm_flags | VM_LOCKED;
99                 if (!on)
100                         newflags &= ~VM_LOCKED;
101
102                 tmp = vma->vm_end;
103                 if (tmp > end)
104                         tmp = end;
105                 error = mlock_fixup(vma, &prev, nstart, tmp, newflags);
106                 if (error)
107                         break;
108                 nstart = tmp;
109                 if (nstart < prev->vm_end)
110                         nstart = prev->vm_end;
111                 if (nstart >= end)
112                         break;
113
114                 vma = prev->vm_next;
115                 if (!vma || vma->vm_start != nstart) {
116                         error = -ENOMEM;
117                         break;
118                 }
119         }
120         return error;
121 }
122
123 asmlinkage long sys_mlock(unsigned long start, size_t len)
124 {
125         unsigned long locked;
126         unsigned long lock_limit;
127         int error = -ENOMEM;
128
129         if (!can_do_mlock())
130                 return -EPERM;
131
132         down_write(&current->mm->mmap_sem);
133         len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
134         start &= PAGE_MASK;
135
136         locked = len >> PAGE_SHIFT;
137         locked += current->mm->locked_vm;
138
139         lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
140         lock_limit >>= PAGE_SHIFT;
141
142         /* check against resource limits */
143         if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
144                 error = do_mlock(start, len, 1);
145         up_write(&current->mm->mmap_sem);
146         return error;
147 }
148
149 asmlinkage long sys_munlock(unsigned long start, size_t len)
150 {
151         int ret;
152
153         down_write(&current->mm->mmap_sem);
154         len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
155         start &= PAGE_MASK;
156         ret = do_mlock(start, len, 0);
157         up_write(&current->mm->mmap_sem);
158         return ret;
159 }
160
161 static int do_mlockall(int flags)
162 {
163         struct vm_area_struct * vma, * prev = NULL;
164         unsigned int def_flags = 0;
165
166         if (flags & MCL_FUTURE)
167                 def_flags = VM_LOCKED;
168         current->mm->def_flags = def_flags;
169         if (flags == MCL_FUTURE)
170                 goto out;
171
172         for (vma = current->mm->mmap; vma ; vma = prev->vm_next) {
173                 unsigned int newflags;
174
175                 newflags = vma->vm_flags | VM_LOCKED;
176                 if (!(flags & MCL_CURRENT))
177                         newflags &= ~VM_LOCKED;
178
179                 /* Ignore errors */
180                 mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags);
181         }
182 out:
183         return 0;
184 }
185
186 asmlinkage long sys_mlockall(int flags)
187 {
188         unsigned long lock_limit;
189         int ret = -EINVAL;
190
191         if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE)))
192                 goto out;
193
194         ret = -EPERM;
195         if (!can_do_mlock())
196                 goto out;
197
198         down_write(&current->mm->mmap_sem);
199
200         lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
201         lock_limit >>= PAGE_SHIFT;
202
203         ret = -ENOMEM;
204         if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
205             capable(CAP_IPC_LOCK))
206                 ret = do_mlockall(flags);
207         up_write(&current->mm->mmap_sem);
208 out:
209         return ret;
210 }
211
212 asmlinkage long sys_munlockall(void)
213 {
214         int ret;
215
216         down_write(&current->mm->mmap_sem);
217         ret = do_mlockall(0);
218         up_write(&current->mm->mmap_sem);
219         return ret;
220 }
221
222 /*
223  * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
224  * shm segments) get accounted against the user_struct instead.
225  */
226 static DEFINE_SPINLOCK(shmlock_user_lock);
227
228 int user_shm_lock(size_t size, struct user_struct *user)
229 {
230         unsigned long lock_limit, locked;
231         int allowed = 0;
232
233         locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
234         lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
235         lock_limit >>= PAGE_SHIFT;
236         spin_lock(&shmlock_user_lock);
237         if (locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK))
238                 goto out;
239         get_uid(user);
240         user->locked_shm += locked;
241         allowed = 1;
242 out:
243         spin_unlock(&shmlock_user_lock);
244         return allowed;
245 }
246
247 void user_shm_unlock(size_t size, struct user_struct *user)
248 {
249         spin_lock(&shmlock_user_lock);
250         user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
251         spin_unlock(&shmlock_user_lock);
252         free_uid(user);
253 }