Add SocketDiagTCPInfo with some constants and structs to get tcp_info…

[RealFatCat]

… via sock_diag.

Probably, needs some restructurisation.

[randomizedcoder]

G'day @RealFatCat , This is a really great PR.

• How did you actually test this?
• What are your thoughts in terms of keeping github.com/vishvananda/netlink in line with the kernel struct? Some kind of kernel version detection, to map to the correct struct? e.g. Latest iproute2 has even more good stuff in the struct:
  https://github.com/shemminger/iproute2/blob/afa588490b7e87c5adfb05d5163074e20b6ff14a/misc/ss.c#L807

[RealFatCat]

Yeah, found the problem that in different kernels we have different size of tcp_info struct.
Moreover, some fields can be added in the middle of structure.
For example, in 4.19 there is tcpi_delivery_rate_app_limited right after scales https://elixir.bootlin.com/linux/v4.19/source/include/uapi/linux/tcp.h#L184
But in v4.4 we don't have it: https://elixir.bootlin.com/linux/v4.4/source/include/uapi/linux/tcp.h#L159
I didn't count on such "inserts" in the kernel.

Any advises how to handle this?

UPD: Probably, need to determine version of current kernel and read that byte or not.

[khasanovbi]

Yeap, understand problem now.
I'm think, we could determine version or use size of b.
Maybe we could write some private structs for different versions (looks like blame should help https://github.com/torvalds/linux/blame/29d9f30d4ce6c7a38745a54a8cddface10013490/include/uapi/linux/tcp.h) and converters to our unified struct.
We can use struct embedding to don't duplicate same fields between different linux versions.

[RealFatCat]

Probably, there is no issue with tcpi_delivery_rate_app_limited.
I'll update this PR after some tests.

[RealFatCat]

This is probably not the most beautiful solution, but should work on every kernel.

[aboch]

Is it feasible to add a UT for this ?

[RealFatCat]

Is it feasible to add a UT for this ?

I really don't know how to do it properly. It seems like we need to mock kernel answers somehow, to compare them with expected stats data. But how.

[vishvananda]

This looks good? How about a unit test to make sure that the request to the kernel for stats returns data without actually checking the values to see if they are valid? That would at least validate that the call works.

[RealFatCat]

Will do.