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