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