Re: [RFC] netlink: limit recursion depth in policy validation

[Pablo Neira Ayuso <pablo@netfilter.org>]

On Fri, Apr 26, 2019 at 07:03:10PM +0200, Johannes Berg wrote:
> On Fri, 2019-04-26 at 18:57 +0200, Pablo Neira Ayuso wrote:
> > 
> > > +/*
> > > + * Nested policies might refer back to the original
> > > + * policy in some cases, and userspace could try to
> > > + * abuse that and recurse by nesting in the right
> > > + * ways. Limit recursion to avoid this problem.
> > > + */
> > > +#define MAX_POLICY_RECURSION_DEPTH	10
> > 
> > In your policy description approach, you iterate over the policy
> > structures. How do you deal with this recursions from there?
> 
> Well, check out the code :-)
> 
> It doesn't actually recurse. What it does is build a list of policies
> that are reachable from the root policy and each policy in the list. So
> basically, there we do:
> 
> list = [root policy]
> list_len = 1
> i = 0
> 
> walk_policy(policy)
> {
>    for_each_policy_entry(entry, policy) {
>       nested = nested_policy_or_null(entry);
>       if (nested) {
>          list[i] = nested;
>          list_len += 1
>       }
>    }
> }
> 
> while (i < list_len) {
>     walk_policy(list[i]);
>     i++;
> }
> 
> Then, we walk the list again:
> 
> for (i = 0; i < list_len; i++) {
>     for_each_policy_entry(entry, list[i]) {
>        send_entry_to_userspace(i, entry); // mark it as occurring in policy i
>     }
> }
> 
> 
> This basically flattens the whole thing.
> 
> Obviously, the walking may allocate some memory, and the last loop to
> send it out isn't actually a loop like that because it's a netlink dump
> with each entry being in a separate netlink message, but that's the gist
> of it.

I see, following this approach, I can just remove the duplicated code
in my netlink description stuff by using the list of policy
structures.