dp_rte_flow refactoring

[PlagueCZ]

I was looking for places where return value handling is still not refactored, so I entered the rte_flow subdir.

Most of the changes is just moving stuff around and creating wrappers. The overall architecture of each flow creation is till the same, i.e. all steps are still repeated for similar flows to be easier to read.

I tried to create small commits due to the fact that I cannot properly test on a cluster, but I do finally have two real machines connected. These commits should make it easier to review and (hopefully not needed) bisect on errors.

I would rather squash them at the end as the history does not really need that many I think.

I put this as a draft so we can discuss the changes first and also address a few leftover problems/misunderstandings by me.

Maybe just take a look first, test if this is totally broken and then we can sync using this draft PR as a visual guide.

[byteocean]

@PlagueCZ The changes look good to me. As we communicated over the unused variable issue, you could cherry-pick removal of this variable at your convenience. LB test has some an issue, which however is not caused by this refactoring, which I will further investigate.

[PlagueCZ]

I cherry-picked your commit, thank you for clearing that up.

I also changed the variable declaration sections as discussed.

Please take a look at the three TODOs I mentioned in the code now, as I would like to clear them up before providing a final PR state.

[PlagueCZ]

@guvenc I created many commits to make the review/test easier, but should I squash them in the end (after some tests by Tao) as not to pollute the history?

[guvenc]

@PlagueCZ Yes, please squash them when the PR is mature to be merged and thanks for making the review easier with smaller commits.

[PlagueCZ]

@byteocean I had to rebase onto fix/flow_table_flush for this because that PR created unified freeup function for flows which is something needed for unrolling the hairpin flow if the subsequent normal one fails.

[guvenc]

@PlagueCZ
I think this is now mature enough to become a PR ?
Or is it better to merge this one after #363 ? As I see that it is rebased on #363

[PlagueCZ]

@guvenc yes, I had to rebase to make this work (and prevent future merge issues), so #363 needs to go first.

From my side this seems finished, but I can only test VM-VM offloading and a small subset of host-host offloading configurations.

[byteocean]

@PlagueCZ I tested the code after rebasing, especially on NAT and LB scenarios. The tests I usually use can actually work.

[PlagueCZ]

I've created an archive/dp_rte_flow_refactor as a copy of the original commit log and squashed this assigned branch.