I think there's a misunderstanding of the sentence from RFC4364: "This
procedure requires that there be a label switched path leading from a
packet's ingress PE to its egress PE."
If a VPN packet is traveling the following path:
PE1--->P1--->ASBR1--->ASBR2--->P2--->PE2
where the ASBRs are supporting option B, then there is necessarily an LSP
consisting of the following sequence of routers: <PE1,ASBR1,ASBR2,PE2>.
What makes this sequence of routers an LSP is that they all operate on the
"VPN label". (See section 3.15 of RFC 3031 for the precise definition of
"Label Switched Path". Note that "LSP" is defined relative to the position
of a particular label in the stack of a particular packet.) PE1 pushes on
the label, ASBR1 swaps it with a label assigned by ASBR2, ASBR2 swaps it
with a label assigned by PE2, PE2 pops it. We could call this the "VPN
LSP", because it follows the path of distribution of a VPN-IP route.
The cited text is intended to distinguish option B from option A; in option
A, the sequence <PE1,ASBR1,ASBR2,PE2> is not an LSP, because ASBR2 does not
distribute labels to ASBR1.
> which requires two SPs pre-provisioning process that may not apply to
> here.
The LSP exists by virtue of the distribution of the labeled VPN-IP routes.
Note that, although <PE1, ASBR1, ASBR2, PE2> is an LSP according to the
definition in RFC 3031, each router in the sequence only has to know which
router is next in the sequence; the ingress router does not need to know how
to reach to the egress router. Also, there is no requirement that an LSP is
used to carry packets from PE1 to ASBR1, or from ASBR2 to PE2. Either or
both of those "hops" could be via a GRE tunnel, or any other method of
transport.
It's true that in a typical deployment, there is a "top-level" LSP
<PE1,P1,ASBR1> and another "top-level" LSP <ASBR2,P2,PE2>. Neither of these
is the VPN LSP; rather, each is used to carry packets from one member of the
VPN LSP to the next member in sequence. But option B is independent of how
packets get between adjacent members of the VPN LSP.
> WAN may use MPLS LSP tunnel and DC may use IP based tunnel.
That is compatible with RFC4364 option B, because the LSP
<PE1,ASBR1,ASBR2,PE2> still exists.
> This will be an issue when use IP GRE tunnels.
It's only an issue if you want to have a single GRE tunnel that goes from
PE1 to PE2. If that is what you want, then you should probably be using
option C. Option B does not tell the ingress PE who the egress PE is.
(Well, not unless RFC 6513/6514 is also in use ...)
> Option C request set up two LSP tunnel, one from ingress PE to egress PE,
> second from ingress PE to ASBR to solve the issue in IGP.
In option C, the "VPN LSP" would be <PE1,PE2>. If one wants to use MPLS to
move the packets from PE1 to PE2, one needs a "top-level" LSP like
<PE1,P1,ASBR1,ASBR2,P2,PE2>; but this is not the "VPN LSP". I think you
interpreted the text from RFC 4364 as requiring this top-level LSP for
option B, but it is only intended to require the VPN LSP
<PE1,ASBR1,ASBR2,PE2>.
I'm not sure whether this makes things any clearer or not ...
