Hello Sinbad,
in cisco routers MPLS implementation is based on CEF.
So the first step for each label switch router (LSR) is to have CEF enabled and to build the CEF related tables including the FIB ( Forwarding Information Base).
the FIB provides forwarding information for each IP prefix that is known in the IP routing table including rewriting information to encapsulate the packet.
The second building component is a classless IGP like OSPF, ISIS but also EIGRP or RIPv2 could be used.
The third component in the signalling plane is an MPLS label distribution protocol like LDP or RSVP-TE.
Let's make the case of LDP. Each LSR builds LDP sessions ( TCP based on port 646) with all adjacent routers and advertises the local bindings for each IPv4 prefix that is known from IGP protocol in use.
Each node allocates MPLS labels to all known IPv4 prefixes (local labels) and can receive remote bindings that are the labels to be used when sending traffic to neighbor router.
The MPLS label space is unique and once a label is picked it is not avaiable for other uses.
Labels have local meaning in each node.
Different nodes will assign different label values to the same FEC IPv4 prefix.
The duty of LDP is to advertise these label to IP prefix mappings that are called bindings.
LDP does not use split horizon or other routing protocol concepts so the bindings are exchanged bidirectionally with all neighbors regardless of best route to destination.
all the MPLS bindings are stored in the LIB label information base database.
By checking the IGP best route an MPLS binding using the same outgoing interface/neighbor for a given IPv4 destination is picked from the LIB and placed in the LFIB ( label forwarding information base).
So the LFIB contains the installed MPLS entries used for forwarding.
If an entry exists in the LFIB it is used instead of the corresponding entry in the FIB.
LDP in this way can populate the LFIB with entries for all IGP routes.
For a given destination each router on the IGP best path makes its own choice of label and publishes its label upstream (opposite direction of the destination). The ordered list of chosen labels make the LSP for the destination.
For BGP routes things are little different: BGP routes use recursion to use forwarding the BGP next-hop.
This recursion is used also in the MPLS forwarding so the BGP next-hop is usually the loopback address of a remote edge router (PE node) that is published in the IGP.
With the introduction of BGP the IGP is only requested to provide infrastructure connectivity, that is to advertise the loopback addresses.
MPLS traffic engineering uses RSVP-TE and relies on OSPF or ISIS only with their TE extensions.
Hope to help
Giuseppe
