On Sun, Dec 08, 2019 at 07:49:07PM +0000, Al Viro wrote:
> On Mon, Nov 25, 2019 at 01:53:42PM +0100, Vladis Dronov wrote:
> > In a case when a chardev file (like /dev/ptp0) is open but an underlying
> > device is removed, closing this file leads to a use-after-free. This
> > reproduces easily in a KVM virtual machine:
> >
> > # cat openptp0.c
> > int main() { ... fp = fopen("/dev/ptp0", "r"); ... sleep(10); }
>
> > static void __fput(struct file *file)
> > { ...
> > if (file->f_op->release)
> > file->f_op->release(inode, file); <<< cdev is kfree'd here
>
> > if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
> > !(mode & FMODE_PATH))) {
> > cdev_put(inode->i_cdev); <<< cdev fields are accessed here
> >
> > because of:
> >
> > __fput()
> > posix_clock_release()
> > kref_put(&clk->kref, delete_clock) <<< the last reference
> > delete_clock()
> > delete_ptp_clock()
> > kfree(ptp) <<< cdev is embedded in ptp
> > cdev_put
> > module_put(p->owner) <<< *p is kfree'd
> >
> > The fix is to call cdev_put() before file->f_op->release(). This fix the
> > class of bugs when a chardev device is removed when its file is open, for
> > example:
>
> And what's to prevent rmmod coming and freeing ->release code right as you
> are executing it?
FWIW, the bug here seems to be that the lifetime rules of cdev are fucked -
if it can get freed while its ->kobj is still alive, we have something
very wrong there. IOW, you have ptp lifetime controlled by *TWO*
refcounts - that of clk and that of of cdev->kobj. That's doesn't work.
Replace that kfree() with dropping a kobject reference, perhaps, so
that freeing would've been done by its release callback?
