binder: Don't modify VMA bounds in ->mmap handler

binder_mmap() tries to prevent the creation of overly big binder mappings
by silently truncating the size of the VMA to 4MiB. However, this violates
the API contract of mmap(). If userspace attempts to create a large binder
VMA, and later attempts to unmap that VMA, it will call munmap() on a range
beyond the end of the VMA, which may have been allocated to another VMA in
the meantime. This can lead to userspace memory corruption.

The following sequence of calls leads to a segfault without this commit:

int main(void) {
  int binder_fd = open("/dev/binder", O_RDWR);
  if (binder_fd == -1) err(1, "open binder");
  void *binder_mapping = mmap(NULL, 0x800000UL, PROT_READ, MAP_SHARED,
                              binder_fd, 0);
  if (binder_mapping == MAP_FAILED) err(1, "mmap binder");
  void *data_mapping = mmap(NULL, 0x400000UL, PROT_READ|PROT_WRITE,
                            MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
  if (data_mapping == MAP_FAILED) err(1, "mmap data");
  munmap(binder_mapping, 0x800000UL);
  *(char*)data_mapping = 1;
  return 0;
}

Cc: stable@vger.kernel.org
Signed-off-by: Jann Horn <jannh@google.com>
Acked-by: Todd Kjos <tkjos@google.com>
Acked-by: Christian Brauner <christian.brauner@ubuntu.com>
Link: https://lore.kernel.org/r/20191016150119.154756-1-jannh@google.com


Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>