Bounce-Back corner node

I just started LBM, I have a few questions about the bounce-back boundary conditions:

For example,
the D2Q9 lattice is like this:

7 3 6
4 1 2
8 5 9

Let’s assume node 1 is on the left-top corner, and node 2 is on the top boundary.

For node 2, after streaming, f1,f2,f4, f3,f7,f6 are known, only f8,f5,f9 are unknown, so we can apply the bounce back rules to calculate f8,f5,f9 like:
f8=f6; f5=f3; f9=f7;

But for node 1, I THINK after streaming, we only know f3, f4,f7, because those distribution functions are coming from the neighborhood nodes, but how can I get f6, f8, f2, f5, f9? if I apply bounce back rules, I might have:
f5=f3;f9=f7;f2=f4; how about f8 and f6?


What you will do with the unknown distribution functions will depend on what boundary conditions you want to impose on the left side.
For instance if you have periodic left-right, then you stream f6 from the right boundary of the domain.
If you have a wall on the left, you will want to bounce back f6 as well.
By the way, if you are using full way bounce-back, then most distributions only serve as a “temporary storage place”, and only f9 actually needs to be defined on your corner node, as the other distribution functions never go into the domain.