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 #include <linux/vs_memory.h>
  13
  14
  15 static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
  16         unsigned long start, unsigned long end, unsigned int newflags)
  17 {
  18         struct mm_struct * mm = vma->vm_mm;
  19         pgoff_t pgoff;
  20         int pages;
  21         int ret = 0;
  22
  23         if (newflags == vma->vm_flags) {
  24                 *prev = vma;
  25                 goto out;
  26         }
  27
  28         pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
  29         *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma,
  30                           vma->vm_file, pgoff, vma_policy(vma));
  31         if (*prev) {
  32                 vma = *prev;
  33                 goto success;
  34         }
  35
  36         *prev = vma;
  37
  38         if (start != vma->vm_start) {
  39                 ret = split_vma(mm, vma, start, 1);
  40                 if (ret)
  41                         goto out;
  42         }
  43
  44         if (end != vma->vm_end) {
  45                 ret = split_vma(mm, vma, end, 0);
  46                 if (ret)
  47                         goto out;
  48         }
  49
  50 success:
  51         /*
  52          * vm_flags is protected by the mmap_sem held in write mode.
  53          * It's okay if try_to_unmap_one unmaps a page just after we
  54          * set VM_LOCKED, make_pages_present below will bring it back.
  55          */
  56         vma->vm_flags = newflags;
  57
  58         /*
  59          * Keep track of amount of locked VM.
  60          */
  61         pages = (end - start) >> PAGE_SHIFT;
  62         if (newflags & VM_LOCKED) {
  63                 pages = -pages;
  64                 if (!(newflags & VM_IO))
  65                         ret = make_pages_present(start, end);
  66         }
  67
  68         vx_vmlocked_sub(vma->vm_mm, pages);
  69 out:
  70         if (ret == -ENOMEM)
  71                 ret = -EAGAIN;
  72         return ret;
  73 }
  74
  75 static int do_mlock(unsigned long start, size_t len, int on)
  76 {
  77         unsigned long nstart, end, tmp;
  78         struct vm_area_struct * vma, * prev;
  79         int error;
  80
  81         len = PAGE_ALIGN(len);
  82         end = start + len;
  83         if (end < start)
  84                 return -EINVAL;
  85         if (end == start)
  86                 return 0;
  87         vma = find_vma_prev(current->mm, start, &prev);
  88         if (!vma || vma->vm_start > start)
  89                 return -ENOMEM;
  90
  91         if (start > vma->vm_start)
  92                 prev = vma;
  93
  94         for (nstart = start ; ; ) {
  95                 unsigned int newflags;
  96
  97                 /* Here we know that  vma->vm_start <= nstart < vma->vm_end. */
  98
  99                 newflags = vma->vm_flags | VM_LOCKED;
 100                 if (!on)
 101                         newflags &= ~VM_LOCKED;
 102
 103                 tmp = vma->vm_end;
 104                 if (tmp > end)
 105                         tmp = end;
 106                 error = mlock_fixup(vma, &prev, nstart, tmp, newflags);
 107                 if (error)
 108                         break;
 109                 nstart = tmp;
 110                 if (nstart < prev->vm_end)
 111                         nstart = prev->vm_end;
 112                 if (nstart >= end)
 113                         break;
 114
 115                 vma = prev->vm_next;
 116                 if (!vma || vma->vm_start != nstart) {
 117                         error = -ENOMEM;
 118                         break;
 119                 }
 120         }
 121         return error;
 122 }
 123
 124 asmlinkage long sys_mlock(unsigned long start, size_t len)
 125 {
 126         unsigned long locked, grow;
 127         unsigned long lock_limit;
 128         int error = -ENOMEM;
 129
 130         if (!can_do_mlock())
 131                 return -EPERM;
 132
 133         down_write(&current->mm->mmap_sem);
 134         len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
 135         start &= PAGE_MASK;
 136
 137         grow = len >> PAGE_SHIFT;
 138         if (!vx_vmlocked_avail(current->mm, grow))
 139                 goto out;
 140         locked = current->mm->locked_vm + grow;
 141
 142         lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
 143         lock_limit >>= PAGE_SHIFT;
 144
 145         /* check against resource limits */
 146         if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
 147                 error = do_mlock(start, len, 1);
 148 out:
 149         up_write(&current->mm->mmap_sem);
 150         return error;
 151 }
 152
 153 asmlinkage long sys_munlock(unsigned long start, size_t len)
 154 {
 155         int ret;
 156
 157         down_write(&current->mm->mmap_sem);
 158         len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
 159         start &= PAGE_MASK;
 160         ret = do_mlock(start, len, 0);
 161         up_write(&current->mm->mmap_sem);
 162         return ret;
 163 }
 164
 165 static int do_mlockall(int flags)
 166 {
 167         struct vm_area_struct * vma, * prev = NULL;
 168         unsigned int def_flags = 0;
 169
 170         if (flags & MCL_FUTURE)
 171                 def_flags = VM_LOCKED;
 172         current->mm->def_flags = def_flags;
 173         if (flags == MCL_FUTURE)
 174                 goto out;
 175
 176         for (vma = current->mm->mmap; vma ; vma = prev->vm_next) {
 177                 unsigned int newflags;
 178
 179                 newflags = vma->vm_flags | VM_LOCKED;
 180                 if (!(flags & MCL_CURRENT))
 181                         newflags &= ~VM_LOCKED;
 182
 183                 /* Ignore errors */
 184                 mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags);
 185         }
 186 out:
 187         return 0;
 188 }
 189
 190 asmlinkage long sys_mlockall(int flags)
 191 {
 192         unsigned long lock_limit;
 193         int ret = -EINVAL;
 194
 195         if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE)))
 196                 goto out;
 197
 198         ret = -EPERM;
 199         if (!can_do_mlock())
 200                 goto out;
 201
 202         down_write(&current->mm->mmap_sem);
 203
 204         lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
 205         lock_limit >>= PAGE_SHIFT;
 206
 207         ret = -ENOMEM;
 208         if (!vx_vmlocked_avail(current->mm, current->mm->total_vm))
 209                 goto out;
 210         if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
 211             capable(CAP_IPC_LOCK))
 212                 ret = do_mlockall(flags);
 213         up_write(&current->mm->mmap_sem);
 214 out:
 215         return ret;
 216 }
 217
 218 asmlinkage long sys_munlockall(void)
 219 {
 220         int ret;
 221
 222         down_write(&current->mm->mmap_sem);
 223         ret = do_mlockall(0);
 224         up_write(&current->mm->mmap_sem);
 225         return ret;
 226 }
 227
 228 /*
 229  * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
 230  * shm segments) get accounted against the user_struct instead.
 231  */
 232 static DEFINE_SPINLOCK(shmlock_user_lock);
 233
 234 int user_shm_lock(size_t size, struct user_struct *user)
 235 {
 236         unsigned long lock_limit, locked;
 237         int allowed = 0;
 238
 239         locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
 240         lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
 241         lock_limit >>= PAGE_SHIFT;
 242         spin_lock(&shmlock_user_lock);
 243         if (locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK))
 244                 goto out;
 245         get_uid(user);
 246         user->locked_shm += locked;
 247         allowed = 1;
 248 out:
 249         spin_unlock(&shmlock_user_lock);
 250         return allowed;
 251 }
 252
 253 void user_shm_unlock(size_t size, struct user_struct *user)
 254 {
 255         spin_lock(&shmlock_user_lock);
 256         user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
 257         spin_unlock(&shmlock_user_lock);
 258         free_uid(user);
 259 }