get-linux-source: always do git fetch

[theihor]

Recently proof-of-work proxy was rolled out for git.kernel.org, which prevents wget of the bpf-next snapshots.

This lead to failures of get-linux-source action:

• https://github.com/kernel-patches/vmtest/actions/runs/14313967784/job/40115985369
• https://github.com/libbpf/libbpf/actions/runs/14316467048/job/40123652831

Simplify the download script to always run git init + git fetch sequence.

[1] https://social.kernel.org/notice/AsgziNL6zgmdbta3lY

[theihor]

Tesed on kernel-patches/vmtest https://github.com/kernel-patches/vmtest/actions/runs/14322033825

[shunghsiyu]

@theihor the root cause here actually seems to be the good manner of setting the user agent to BPFCIBot/1.0 (bpf@vger.kernel.org). If that was dropped wget actually downloads the file just fine.

I'm asking in the announcement whether there's a better approach.

[shunghsiyu]

The suggestion is to use BPFCI/1.0 (bpf@vger.kernel.org) as the user agent instead. Testing that out locally it seems to work just fine.

[theihor]

@shunghsiyu Hi, thank you for the tip.
In practice downloading a snapshot vs fetching doesn't make much of a difference in terms of download time. So switching back to snapshots only makes sense if kernel.org collects statistics on external hits. Do you know if it does?

[anakryiko]

@theihor Konstantin (kernel.org admin at that time) did have a preference for snapshot downloading, it's probably causing much less load on their servers. I forwarded you some emails we exchanged a while ago for some context. So yes, snapshots are preferred.

[theihor]

@theihor Konstantin (kernel.org admin at that time) did have a preference for snapshot downloading, it's probably causing much less load on their servers. I forwarded you some emails we exchanged a while ago for some context. So yes, snapshots are preferred.

Thanks, I wasn't aware. I find it strange to trigger proof-of-work challenge on "bot" string in the user-agent though.