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