I've got a very simple VPLS/MPLS/OSPF setup following the examples from the Internet.
Everything from "Client Router" to "Computer 2" works without any issues.
In fact, "Computer 1" can start communication with "Computer 2". Then, what's the problem?
"Computer 2" cannot start communication with "Computer 1".

Attempts made from "Computer 1" to "Computer 2" work. Connection attempts initiated from "Computer 2" to "Computer 1" fail.

Going deep on the attempt I figured out where the problem is.
Original packet from "Computer 2" reaches "Computer 1". But the response packet from "Computer 1" only goes as far as reaching CE1.
It fails to reach PE. I suppose CE1, for some reason, can't MPLS'lize it.
The funny thing is, packets starting from "Computer 1" have no problem reaching "Computer 2" and its response also have no problem making its way to "Computer 1"

=== Detail ===
Computer 1 @ Site A = 192.168.2.2/24

Client Router @ Site A
interface facing computer 1= 192.168.2.1/24
interface facing CE1 = 172.17.0.53/30

CE1 @ Site A
interface facing Client Router = 172.17.0.54/30
interface facing PE = x.x.x.1/30

PE
interface facing CE1 = x.x.x.2/30
interface facing CE2 = y.y.y.2/30

CE2 @ Site B
interface facing PE = y.y.y.1/30
interface facing Computer 2 = 192.168.3.1/24

Computer 2 @ Site B = 192.168.3.2/24

"Client Router" has OSPF configured. Its routing is sent to CE1, PE, and CE2 dynamically.
CE2's local network (192.168.3.0/24) appears at "Client Router" dynamically as it should.
No problem with OSPF.

My general understanding of the network is that if "Computer 1" can start communication with "Computer 2" then "Computer 2" can surely start communication with "Computer 1". However, with mpls/vpls involvement, it says otherwise.

Can someone help me find the missing piece of the puzzle?

Okay. Something else I found it.
On CE1, response packets were getting DROPed by the FORWARD chain as its state was determined as INVALID.
Why would it be INVALID?
Should I not check the state?