Query routes via routesocket

[hurricanehrndz]

On Unix systems, this module currently dumps and caches the complete routing table. Afterwards, when a user submits a query to a specific destination it walks the table looking for the most applicable route found in the cache.

If you compare this functionality to route get it is good, but not exact. Would it not be better to do almost exactly what route get does?

This patch is an attempt at just that. If uses the AF_ROUTE socket to formalate route get messages and parses the output. Until recently this would not have been possible due to bugs in the x/net/route package. I have fixed said bugs, so this approach is very feasible.

closes #56

[willscott]

This does look generally nicer / less invasive than the current strategy, thanks for the PR.
I'll try to leave detailed comments tomorrow

[hurricanehrndz]

@willscott any chance you can re-review

[willscott]

Thanks for the comments, @djdv !

[hurricanehrndz]

Thank you for the review. I will review the comments and update accordingly.

…

On Wed, Apr 2, 2025 at 5:06 PM Will ***@***.***> wrote: Thanks for the comments, @djdv <https://github.com/djdv> ! — Reply to this email directly, view it on GitHub <#57 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABMJBTO3FBQ5S56XLEHKSKD2XRUPNAVCNFSM6AAAAABZQGL5VGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDONZTHEZDEMBXGQ> . You are receiving this because you authored the thread.Message ID: ***@***.***> [image: willscott]*willscott* left a comment (libp2p/go-netroute#57) <#57 (comment)> Thanks for the comments, @djdv <https://github.com/djdv> ! — Reply to this email directly, view it on GitHub <#57 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABMJBTO3FBQ5S56XLEHKSKD2XRUPNAVCNFSM6AAAAABZQGL5VGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDONZTHEZDEMBXGQ> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[djdv]

Thanks

You're welcome. :^]
And good luck with it.

I was planning to look at this for real either Friday or Saturday (if I have free time).
But skimmed it early and didn't want to forget those comments.

I'm assuming some of this code is based around my original PR I'm tagged in, but just to be clear, I wanted to say explicitly that I have no official involvement with this project. So take my review remarks lightly.

[hurricanehrndz]

Thanks

You're welcome. :^] And good luck with it.

I was planning to look at this for real either Friday or Saturday (if I have free time). But skimmed it early and didn't want to forget those comments.

I'm assuming some of this code is based around my original PR I'm tagged in, but just to be clear, I wanted to say explicitly that I have no official involvement with this project. So take my review remarks lightly.

Actually the code came about because this library was used in netbird, I only noticed your PR after I had submitted mine. Upon which point I did take some inspiration for dealing with the possibility of indefinite read loops, so thank you for that.

Prior to that I was developing similar applications to troubleshoot and fix some bugs in the x/net package.

[djdv]

This is for a different OS than I use, so I could not test executing it, but I did read netroute_bsd.go and it uses the same routing protocol.
The code is very nice and makes sense to me. 👍
Only some minor/negligible remarks.

[djdv]

Actually the code came about because this library was used in netbird, I only noticed your PR after I had submitted mine. Upon which point I did take some inspiration for dealing with the possibility of indefinite read loops, so thank you for that.

Prior to that I was developing similar applications to troubleshoot and fix some bugs in the x/net package.

Ahh, makes sense. The code looked familiar while I was reading it, and we're both targeting similar APIs after all.

Related to x/net, the golang.org/x/net/route dependency, does not seem to build on illumos.
Meaning I can't execute the library or test code, but did read it.

$ uname -irsv
SunOS 5.11 illumos-3caf57556b i86pc
$ go test -v .
# github.com/libp2p/go-netroute
package github.com/libp2p/go-netroute
        imports golang.org/x/net/route: build constraints exclude all Go files in /export/home/dd/go/pkg/mod/golang.org/x/[email protected]/route
FAIL    github.com/libp2p/go-netroute [setup failed]
FAIL

[hurricanehrndz]

Actually the code came about because this library was used in netbird, I only noticed your PR after I had submitted mine. Upon which point I did take some inspiration for dealing with the possibility of indefinite read loops, so thank you for that.
Prior to that I was developing similar applications to troubleshoot and fix some bugs in the x/net package.

Ahh, makes sense. The code looked familiar while I was reading it, and we're both targeting similar APIs after all.

Related to x/net, the golang.org/x/net/route dependency, does not seem to build on illumos. Meaning I can't execute the library or test code, but did read it.

$ uname -irsv
SunOS 5.11 illumos-3caf57556b i86pc
$ go test -v .
# github.com/libp2p/go-netroute
package github.com/libp2p/go-netroute
        imports golang.org/x/net/route: build constraints exclude all Go files in /export/home/dd/go/pkg/mod/golang.org/x/[email protected]/route
FAIL    github.com/libp2p/go-netroute [setup failed]
FAIL

I wonder if you vendor and enable the build for illumos if it would work, the route socket doesn't change much from one unix os to another:
https://github.com/hurricanehrndz/golang-net/blob/master/route/address.go#L5

[hurricanehrndz]

@willscott any further thought to this PR?

[djdv]

I wonder if you vendor and enable the build for illumos if it would work, the route socket doesn't change much from one unix os to another:

I'm sure something can be done to make it work on Solaris/Illumos, however-
I'm reluctant to admit this so bluntly, but I'll soon be completely out of money.
Implying the number of users is going to drop from 1 to 0. So it's not going to matter enough for anyone to put time into it.
(；´∀｀)
Honesty aside, that concern can be addressed in a follow up PR anyhow. This one still seems like an improvement for the other systems.

[willscott]

Sorry for my slow response, and sorry to hear that @djdv !

I do wonder if the route socket method here also is worth using in the linux case - it seems like it should work as well there as in BSD-derived cases?

I think i'm okay with merging this and doing the linux part in a followup, but will ideally wait on a next release until both are in the same place. I'm optimistic that the linux one can directly use this implementation even.

[hurricanehrndz]

Let's do the linux in a follow up, it will be some work because Linux has moved the routing IPC to netlink socket, but it should be doable:
https://man7.org/linux/man-pages/man7/rtnetlink.7.html