Hello Americo,
Great questions, and deep thinking about OSPF! Respect!
There are two very important things about LSA-3 to remember.
First, they are not flooded across ABRs even if it appears that way. Rather, ABRs originate and re-originate those LSA-3. As an example, imagine that you have an OSPF network that contains three areas: Area 1, Area 0, Area 2. For some network X in area 1, the ABR between areas 1 and 0 will originate an LSA-3 and flood it into area 0. The other ABR between areas 0 and 2 will receive the LSA-3, but it won't flood it directly into area 2. Instead, after computing its own internal routing table for OSPF routes, it will originate its own LSA-3 on behalf of the network X, increase the cost accordingly to account for the distance across area 0, and only then flood it into area 2. The LSA-3 flooded into area 2 is not the one that was initially originated by the ABR between areas 1 and 0 - it's a reoriginated LSA-3 created by the ABR between areas 0 and 2. Note that because the ABR re-originates the LSA-3 instead of just flooding it, it can also decide to not to originate it if it has reasons to do so.
And this is the gist of the second fact to notice: An OSPF router won't originate a LSA-3 for a network X back into an area whose link-state dabase led to the discovery of the network X in the first place, or that holds the next hops for the network X. This is mandated by RFC2328 Section 12.4.3:
To determine which routes to advertise into an attached Area
A, each routing table entry is processed as follows.
Remember that each routing table entry describes a set of
equal-cost best paths to a particular destination:
[...]
o Else, if the area associated with this set of paths is
the Area A itself, do not generate a summary-LSA for the
route.[17]
o Else, if the next hops associated with this set of paths
belong to Area A itself, do not generate a summary-LSA
for the route.[18] This is the logical equivalent of a
Distance Vector protocol's split horizon logic.
[...]
It is an extract from a larger set of rules on how to generate LSA-3 for routes in the internal OSPF's routing table. The first bullet in the extract above prevents an ABR from injecting an LSA-3 into the same area whose own link-state database (through LSA-1 and LSA-2 in case of the "home" area of network X, or LSA-3 in case of area 0 or other non-backbone areas) led to creating the routing table entry for the network X in the first place. The second bullet prevents an ABR from injecting an LSA-3 into another area that holds the next hop toward the network X.
Needs a little head-wrapping around the stuff : ) Please feel free to ask further!
Best regards,
Peter
