On 31/01/17 00:17, Eric W. Biederman wrote:

Robert Shearman [off-list ref] writes:

quoted

It is sometimes desirable to present an MPLS transport network as a
single hop to traffic transiting it because it prevents confusion when
diagnosing failures. An example of where confusion can be generated is
when addresses used in the provider network overlap with addresses in
the overlay network and the addresses get exposed through ICMP errors
generated as packets transit the provider network.

Therefore, provide the ability to control whether the TTL value from
an MPLS packet is propagated to an IPv4/IPv6 packet when the last
label is popped through the addition of a new per-namespace sysctl:
"net.mpls.ip_ttl_propagate" which defaults to enabled.

Use the same sysctl to control whether the TTL is propagated from IP
packets into the MPLS header. If the TTL isn't propagated then a
default TTL value is used which can be configured via a new sysctl:
"net.mpls.default_ttl".

Instead of having a global sysctl can we please have a different way
to configure the ingress/egress?

My general memory is that this makes sense for a slightly different
tunnel type.   Making it a per mpls tunnel property instead of global
property feels like it should be much more maintainable.

RFC 3443 that David Ahern referenced does indeed infer that this should 
be a per-LSP property. However, it says:

   We also note here that signaling the LSP type (Pipe, Short Pipe or
   Uniform Model) is out of the scope of this document, and that is also
   not addressed in the current versions of the label distribution
   protocols, e.g. LDP [MPLS-LDP] and RSVP-TE [MPLS-RSVP].  Currently,
   the LSP type is configured by the network operator manually by means
   of either a command line or network management interface.

AIUI, the situation of label distribution protocols not signaling this 
property hasn't changed from when this RFC has written, which limits the 
usefulness of a per-LSP property, and perhaps also indicates a lack of 
desire from users of this.

Do you still feel it's worth implementing on a per-LSP basis? If so, any 
opinion on how it should be done for the pop case? Either a new per-path 
RTA attribute can be added, e.g. RTA_TTL_PROPAGATE, or a new rtnh flag 
could be added, e.g. RTNH_F_TTL_PROPAGATE.

Similarly with the related behavior of what to do if the mpls ttl is
exhausted during the trip through the tunnel.  Drop or dig through the
packet and send an ICMP error message at the ip layer.

That's an interesting suggestion, but I don't think it will be useful 
when carrying another LSP over the LSP in question, since the LSR will 
have no idea what the label is being used for (i.e. the payload). If 
there is only one label in the packet then the router should know what 
the payload is of the label and since this is implicitly IPv4 or IPv6 at 
the moment (since those are the only types of traffic for which the 
labels can be used) then surely the ICMP should always be generated in 
that case?

Thanks,
Rob