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/capability.h>
   9 #include <linux/mman.h>
  10 #include <linux/mm.h>
  11 #include <linux/mempolicy.h>
  12 #include <linux/syscalls.h>
  13 #include <linux/vs_memory.h>
  14
  15
  16 static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
  17         unsigned long start, unsigned long end, unsigned int newflags)
  18 {
  19         struct mm_struct * mm = vma->vm_mm;
  20         pgoff_t pgoff;
  21         int pages;
  22         int ret = 0;
  23
  24         if (newflags == vma->vm_flags) {
  25                 *prev = vma;
  26                 goto out;
  27         }
  28
  29         pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
  30         *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma,
  31                           vma->vm_file, pgoff, vma_policy(vma));
  32         if (*prev) {
  33                 vma = *prev;
  34                 goto success;
  35         }
  36
  37         *prev = vma;
  38
  39         if (start != vma->vm_start) {
  40                 ret = split_vma(mm, vma, start, 1);
  41                 if (ret)
  42                         goto out;
  43         }
  44
  45         if (end != vma->vm_end) {
  46                 ret = split_vma(mm, vma, end, 0);
  47                 if (ret)
  48                         goto out;
  49         }
  50
  51 success:
  52         /*
  53          * vm_flags is protected by the mmap_sem held in write mode.
  54          * It's okay if try_to_unmap_one unmaps a page just after we
  55          * set VM_LOCKED, make_pages_present below will bring it back.
  56          */
  57         vma->vm_flags = newflags;
  58
  59         /*
  60          * Keep track of amount of locked VM.
  61          */
  62         pages = (end - start) >> PAGE_SHIFT;
  63         if (newflags & VM_LOCKED) {
  64                 pages = -pages;
  65                 if (!(newflags & VM_IO))
  66                         ret = make_pages_present(start, end);
  67         }
  68
  69         vx_vmlocked_sub(mm, pages);
  70 out:
  71         if (ret == -ENOMEM)
  72                 ret = -EAGAIN;
  73         return ret;
  74 }
  75
  76 static int do_mlock(unsigned long start, size_t len, int on)
  77 {
  78         unsigned long nstart, end, tmp;
  79         struct vm_area_struct * vma, * prev;
  80         int error;
  81
  82         len = PAGE_ALIGN(len);
  83         end = start + len;
  84         if (end < start)
  85                 return -EINVAL;
  86         if (end == start)
  87                 return 0;
  88         vma = find_vma_prev(current->mm, start, &prev);
  89         if (!vma || vma->vm_start > start)
  90                 return -ENOMEM;
  91
  92         if (start > vma->vm_start)
  93                 prev = vma;
  94
  95         for (nstart = start ; ; ) {
  96                 unsigned int newflags;
  97
  98                 /* Here we know that  vma->vm_start <= nstart < vma->vm_end. */
  99
 100                 newflags = vma->vm_flags | VM_LOCKED;
 101                 if (!on)
 102                         newflags &= ~VM_LOCKED;
 103
 104                 tmp = vma->vm_end;
 105                 if (tmp > end)
 106                         tmp = end;
 107                 error = mlock_fixup(vma, &prev, nstart, tmp, newflags);
 108                 if (error)
 109                         break;
 110                 nstart = tmp;
 111                 if (nstart < prev->vm_end)
 112                         nstart = prev->vm_end;
 113                 if (nstart >= end)
 114                         break;
 115
 116                 vma = prev->vm_next;
 117                 if (!vma || vma->vm_start != nstart) {
 118                         error = -ENOMEM;
 119                         break;
 120                 }
 121         }
 122         return error;
 123 }
 124
 125 asmlinkage long sys_mlock(unsigned long start, size_t len)
 126 {
 127         unsigned long locked, grow;
 128         unsigned long lock_limit;
 129         int error = -ENOMEM;
 130
 131         if (!can_do_mlock())
 132                 return -EPERM;
 133
 134         down_write(&current->mm->mmap_sem);
 135         len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
 136         start &= PAGE_MASK;
 137
 138         grow = len >> PAGE_SHIFT;
 139         if (!vx_vmlocked_avail(current->mm, grow))
 140                 goto out;
 141         locked = current->mm->locked_vm + grow;
 142
 143         lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
 144         lock_limit >>= PAGE_SHIFT;
 145
 146         /* check against resource limits */
 147         if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
 148                 error = do_mlock(start, len, 1);
 149 out:
 150         up_write(&current->mm->mmap_sem);
 151         return error;
 152 }
 153
 154 asmlinkage long sys_munlock(unsigned long start, size_t len)
 155 {
 156         int ret;
 157
 158         down_write(&current->mm->mmap_sem);
 159         len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
 160         start &= PAGE_MASK;
 161         ret = do_mlock(start, len, 0);
 162         up_write(&current->mm->mmap_sem);
 163         return ret;
 164 }
 165
 166 static int do_mlockall(int flags)
 167 {
 168         struct vm_area_struct * vma, * prev = NULL;
 169         unsigned int def_flags = 0;
 170
 171         if (flags & MCL_FUTURE)
 172                 def_flags = VM_LOCKED;
 173         current->mm->def_flags = def_flags;
 174         if (flags == MCL_FUTURE)
 175                 goto out;
 176
 177         for (vma = current->mm->mmap; vma ; vma = prev->vm_next) {
 178                 unsigned int newflags;
 179
 180                 newflags = vma->vm_flags | VM_LOCKED;
 181                 if (!(flags & MCL_CURRENT))
 182                         newflags &= ~VM_LOCKED;
 183
 184                 /* Ignore errors */
 185                 mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags);
 186         }
 187 out:
 188         return 0;
 189 }
 190
 191 asmlinkage long sys_mlockall(int flags)
 192 {
 193         unsigned long lock_limit;
 194         int ret = -EINVAL;
 195
 196         if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE)))
 197                 goto out;
 198
 199         ret = -EPERM;
 200         if (!can_do_mlock())
 201                 goto out;
 202
 203         down_write(&current->mm->mmap_sem);
 204
 205         lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
 206         lock_limit >>= PAGE_SHIFT;
 207
 208         ret = -ENOMEM;
 209         if (!vx_vmlocked_avail(current->mm, current->mm->total_vm))
 210                 goto out;
 211         if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
 212             capable(CAP_IPC_LOCK))
 213                 ret = do_mlockall(flags);
 214         up_write(&current->mm->mmap_sem);
 215 out:
 216         return ret;
 217 }
 218
 219 asmlinkage long sys_munlockall(void)
 220 {
 221         int ret;
 222
 223         down_write(&current->mm->mmap_sem);
 224         ret = do_mlockall(0);
 225         up_write(&current->mm->mmap_sem);
 226         return ret;
 227 }
 228
 229 /*
 230  * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
 231  * shm segments) get accounted against the user_struct instead.
 232  */
 233 static DEFINE_SPINLOCK(shmlock_user_lock);
 234
 235 int user_shm_lock(size_t size, struct user_struct *user)
 236 {
 237         unsigned long lock_limit, locked;
 238         int allowed = 0;
 239
 240         locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
 241         lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
 242         lock_limit >>= PAGE_SHIFT;
 243         spin_lock(&shmlock_user_lock);
 244         if (locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK))
 245                 goto out;
 246         get_uid(user);
 247         user->locked_shm += locked;
 248         allowed = 1;
 249 out:
 250         spin_unlock(&shmlock_user_lock);
 251         return allowed;
 252 }
 253
 254 void user_shm_unlock(size_t size, struct user_struct *user)
 255 {
 256         spin_lock(&shmlock_user_lock);
 257         user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
 258         spin_unlock(&shmlock_user_lock);
 259         free_uid(user);
 260 }