In e-learning Cisco (section 9.11), we have: Redistribution of core IGP into aggregation IGPs and vice-versa is in principle not performed as it would defeat the purpose of the UMPLS scaling benefits. The only exception is the redistribution of the RR (ABR) loopbacks from core IGP into aggregation IGPs because the loopback prefix of the RRs must be known in the aggregation IGP also, so that BGP on the PE router can peer with the loopback of the RR. Other prefixes are never redistributed between the core and aggregation layers.
Route Reflectors are configured with the next-hop-self all feature to allow for RRs to change next hop to their own loopback IP even for the reflected prefixes. LDP operates normally inside the IGP islands, however ***BGP labeled unicast layer*** is responsible for the inter network segment label exchange.
https://www.cisco.com/c/en/us/support/docs/multiprotocol-label-switching-mpls/mpls/118846- config-mpls-00.html