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