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