Kernel panic at Ubuntu: IMA + Apparmor

[Eric W. Biederman]

Al Viro <viro at ZenIV.linux.org.uk> writes:

> On Fri, Apr 25, 2014 at 02:43:42PM -0700, Eric W. Biederman wrote:
>
>> ssize_t __vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
>> {
>> 	ssize_t ret;
>> 
>> 	if (!(file->f_mode & FMODE_READ))
>> 		return -EBADF;
>> 	if (!file->f_op->read && !file->f_op->aio_read)
>> 		return -EINVAL;
>> 	if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
>> 		return -EFAULT;
>> 
>> 	if (ret >= 0) {
>> 		count = ret;
>> 		if (file->f_op->read)
>> 			ret = file->f_op->read(file, buf, count, pos);
>> 		else
>> 			ret = do_sync_read(file, buf, count, pos);
>> 	}
>> 
>> 	return ret;
>> }
>
> ... which lacks the f_pos wraparound, etc. checks done by rw_verify_area().
> IOW, it's one more place to grep through while verifying that ->read()
> et.al. do not get called with such arguments.

Agreed it must be done more delicately than my sketch.  I am not
familiar with how much value such sanity checks add.  Especially when
the read is not coming from a potentially hostile userspace.

> fanotify probably could be skipped - ask the security circus crowd about
> that one, it's their bast^Wbaby.  

When the point is having a factor of read that skips the security circus
I think it makes sense to skip this too.  At least as a starting
position.

> add_rchar() and inc_syscr()... depends on
> whether you want those reads hidden from accounting.

I doubt it matters in practice, the code is cheap.

Still it feels wrong to account reads to a task that did not ask for
them.  It feels more correct to account that kind of read into a
different bucket.  Say the reads performed by the kernel for mysterious
kernel activities.

Eric