Re: JunOS/FRR/Nokia et al BGP critical issue

Mailman List Mirror (Read Only)
1

The blog was updated. Correct link:

https://blog.benjojo.co.uk/post/bgp-path-attributes-grave-error-handling

The attribute was not malformed.

This is the hex dump of the attribute: “E0 1C 00”

It is described here.

https://www.rfc-editor.org/rfc/rfc6790#section-5.2

This attribute is deprecated, but that does not prevent routers from originating it or passing it on.

2

Fair update. To be clear, though, the main point of the article stands, and is maybe even strengthened by the update. A corrupted attribute def can cause the behavior (personal experience speaking here with a different attribute) and vendors should adopt RFC7606 and not be absolutely awful at responding to vulnerability reporting.

3

IOS-XR passes on the attribute by default.

Some other routers incorrectly claim it to be malformed and reset the BGP session.

IOS-XR has a configuration to discard an attribute, so it will not pass it on.

It will pass the route with all its other attributes.

Here is an example configuration:

router bgp {asn}

attribute-filter group block_elc

attribute 28 discard

!

neighbor {ip address}

update in filtering

attribute-filter group block_elc

!

!

!

More info:

https://www.cisco.com/c/en/us/td/docs/routers/asr9000/software/routing/command/reference/b-routing-cr-asr9000/bgp-commands.html#wp3145726977

https://www.cisco.com/c/en/us/td/docs/routers/asr9000/software/asr9k-r7-8/routing/configuration/guide/b-routing-cg-asr9000-78x/implementing-bgp.html#concept_77EE033C2F0C4BDDB8423C25FA71E3F9

4

vendors should adopt RFC7606

Yes

and not be absolutely awful at responding to vulnerability reporting.

  1. This isn’t exactly new. It’s been possible to do this since the original days of BGP.
  2. Probably not wise to assume that’s accurate just because he thinks that is true.
5

You may treat-as-withdraw instead of discard.

However, this attribute does not affect routing.

It only affects whether a sender of packets to the route will add the entropy

label or not to the MPLS header, if such an MPLS header is added.

Therefore, it is safe to discard the attribute.

6

vendors should adopt RFC7606

Yes

and not be absolutely awful at responding to vulnerability reporting.

  1. This isn’t exactly new. It’s been possible to do this since the original days of BGP.

Literally the first thing that came into my mind was the as-set issue from 2 decades ago where some vendors passed it, others dropped sessions…