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