Convert InfoBond::ArpValidate into an enum.

[Gilnaa]

Closes #80

[codecov]

Codecov Report

Attention: Patch coverage is 31.57895% with 26 lines in your changes are missing coverage. Please review.

Project coverage is 63.82%. Comparing base (3e16f9c) to head (ae7a60a).
Report is 5 commits behind head on main.

❗ Current head ae7a60a differs from pull request most recent head 236fbe9

Please upload reports for the commit 236fbe9 to get more accurate results.

Files	Patch %	Lines
src/link/link_info/bond.rs	31.57%	26 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #110      +/-   ##
==========================================
- Coverage   63.97%   63.82%   -0.16%     
==========================================
  Files         140      140              
  Lines        8617     8653      +36     
==========================================
+ Hits         5513     5523      +10     
- Misses       3104     3130      +26     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[Gilnaa]

Also, it seems that these PRs steadily lower the code-coverage.
What meaningful tests can I add?
Use nlmon to record ip link add bond0 type bond mode blablablah and make sure we produce an equivalent output?

Is there place for tests that open an actual socket and try to create rtnl objects; is this problematic with CI running us in a non-privileged context?

[cathay4t]

Also, it seems that these PRs steadily lower the code-coverage. What meaningful tests can I add? Use nlmon to record ip link add bond0 type bond mode blablablah and make sure we produce an equivalent output?

I have document on how to capture netlink package and convert it into Vec at:

https://github.com/rust-netlink/netlink-packet-route/blob/main/README.md

You may setup up a bond with arp_validate value changed to non-default one. Then capture the netlink package kernel replied to the request of ip -d link show bond0. The unit test should parse this Vec into structs/enums, and also emit() should generate identical Vec<u8> as kernel did.

Is there place for tests that open an actual socket and try to create rtnl objects; is this problematic with CI running us in a non-privileged context?

For this crate, there is no real netlink communication to kernel. Because this crate only handle parsing and emitting netlink package itself, no socket communication code included. All tests in this crate are unit test, no privilege required.

[little-dude]

@Gilnaa I just added a test in my own MR, in case you want to see how it looks: #98

I used wireshark to get the hex dump of a captured netlink packet, and isolate the different packet parts. It's very tedious, and you don't have to go into so much detail though.

[Gilnaa]

Thank you both!

Unfortunately life got in the way;
will hopefully get back to it eventually

[Gilnaa]

Unfortunately life got in the way; will hopefully get back to it eventually

Life got out of the way!

I've added a test as instructed; dump is taken of tcpdump, with the last few fields removed for brevity, and the length patched in the header.

---

I did notice that serializing stats64 will result in a slightly different payload (I think the length field was different, probably counted padding or something).
Unrelated to this change, though.

EDIT: Yep, the kernel-sent message claimed that the length of Stats64 is 196, and omitted the last 4 fields in stats64.rs (rx_compressed, tx_compressed, rx_nohandler, rx_otherhost_dropped).
Probably due to the link type, nothing to worry about.

[cathay4t]

I have amend the patch:

• Rebased to latest main branch.
• Changed enum ArpValidate to enum BondArpValidate.