Text of people.kernel.org › read as Styx crawled it on Nov 19, 2026. The page may have changed since. Only its words are kept, without pictures or layout, and only the first part of a long page.

Open the live page

people.kernel.org Reader

Reader Read the latest posts from people.kernel.org. Gmail's non-compliant DKIM verifier November 4, 2026 from Konstantin Ryabitsev The short version Gmail fails to verify valid DKIM signatures when both of the following conditions are true: the DKIM-Signature header is folded right after b= , so the signature value starts on the next line, and the value is followed by a ; , i.e. the header ends with ...==; . This perfectly valid DKIM signature fails at Gmail: DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=example.org; s=sel; h=From:To:Subject:Date:Message-ID; bh=...=; b= QQH0CQIUShsLqiZt5ogRFg... ...BkOrlFyCm90Kxrk3tQcHPA==; These pass, even though they differ only in where the line break happens or if the last ; is omitted: s=sel; h=From:To:Subject:Date:Message-ID; bh=...=; b=yWQqOWPUIZHdO ...Xt0M+hbzzbWUm1F/SOr9ztNinrNRaOtKvPvS6jqvwxBv1OLBsEnFceGvZ5hg==; s=sel; h=From:To:Subject:Date:Message-ID; bh=...=; b= QQH0CQIUShsLqiZt5ogRFg... ...BkOrlFyCm90Kxrk3tQcHPA== RFC 6376 allows folding whitespace around the = of a tag ( section 3.2 ) and allows a ; after the last tag.

When a verifier computes the header hash, it must remove the b= value “including all surrounding whitespace” ( section 3.7 ). Compliant verifiers accept all three. Gmail doesn't. How this came up Héctor at Igalia contacted us a few weeks ago when he saw dkim=fail in DMARC reports from Google. It was for a subset of mail that reached Gmail through kernel.org list relays, but the failures were sporadic and without an easily discernible pattern.

On our side, we use a set of distributed relays to deliver list messages, and the problem appeared to manifest itself at each one of them, seemingly at random. After a lot of back-and-forth, we eventually ruled out our relays and narrowed things down to how Igalia's MTA DKIM-signs their messages — Exim folds the header at 78 columns, ending the signature with ; . Under certain conditions this triggers the bug. Whether b= lands at the end of a line depends on the length of everything before it — mostly the h= enumeration of headers that are being signed.

It just so happens that the set of headers produced by git-send-email hits the exact length to cause the b= header wrapping, which is why the problem manifests itself when Igalia's developers send patches to the kernel.org mailing lists. What Gmail probably does It looks like Gmail removes the b= value that follows the = , but keeps the folding whitespace in front of it. After relaxed canonicalization, that becomes: ...; b= ; while what the signer hashed was: ...; b=; As a result, the signature doesn't match.

That trailing ; in the original DKIM header is important, because without it, the leftover space ends up at the end of the header and gets trimmed, so the bug doesn't get triggered. Corner case known since 2008 In his research, Héctor came across a similar bug filed with Exim — the same b= ; vs. b=; already caused problems as far back as 2018 ( Exim bug 2278 ), and when I was running my own tests with dkimpy, I came across an interesting comment in its source code, committed in April 2012: apparently, some large receivers mishandle a folded b= value.

Following the breadcrumbs of that commit, we find a mention of the RFC 4871 erratum 1596 , describing this exact situation as far back as 2008. The updated DKIM spec, RFC 6376 , includes specific wording to clarify and address it. I'm not sure if Gmail's DKIM implementation follows the pre-2011 spec or if this is merely a regression, but as things stand right now, Gmail can incorrectly reject or quarantine perfectly valid forwarded mail.

What you can do If you work at Google, maybe poke someone on the Gmail team and show them this post? If you're running Exim, you may want to consider doing what the Igalia folks did: they started adding a t= timestamp tag to their DKIM signatures, which moves b= away from the end of the line. It's not a guaranteed solution, but it appears to work in their case. Acknowledgements Many thanks to Héctor, who did the hard work of finding the trigger and reproducing it.

Read more... Using shallow-git tarballs for CI September 23, 2026 from Konstantin Ryabitsev If your CI system starts every job with git clone --depth=1 from git.kernel.org, then you're probably going to get banned. Sorry, not sorry. Every shallow clone makes git build a fresh pack, compress it on the fly, and then keep it in RAM until you're done downloading — and it has to do this again for every job in your swarm.

I've previously recommended that you run your own mirror instead, but we now also offer an alternative: every night we publish shallow, single-branch tarballs of mainline and the stable branches here: https://www.kernel.org/pub/linux/kernel/shallow-tar/ These are tarballs of the shallow clone of the git repo, precisely for CI use. They are also fronted by a CDN, so they should be cached at your local Fastly PoP. Important safety warning: using a .git directory that somebody else made bypasses a lot of git's defences, so we recommend a few extra safety steps (they are listed in the .git/shallow-tar.readme inside each tarball): rm -rf .git/hooks .git/config .git/objects/info/alternates git init git remote add -t [branch] --no-tags origin [the URL you trust] git fsck To make this easier, we have a wrapper script that does all of it for you.

Grab it from here: https://git.kernel.org/pub/scm/linux/kernel/git/mricon/korg-helpers.git/plain/get-verified-shallow-tar It should work “out of the box”: ./get-verified-shallow-tar torvalds master This checks the signed checksums we publish, downloads the newest tarball for that branch, runs the safety cleanups, and then fetches and checks out the latest tip. You end up with ./linux , ready to use for building. For stable, you need to specify the exact branch that you want: ./get-verified-shallow-tar stable linux-6.12.y <commit> You should always pass the commit SHA.

The tarballs are signed only by the kernel.org autosigner, which just attests that these are the bytes kernel.org published. Kernel.org has been breached before, and you should assume it can happen again, so don't blindly trust content you get from us. The commit ID is what really tells you that you got what you expected. The CI system usually knows the exact commit it wants anyway. For CI, you probably don't want to look up the signing key on every run.

Export it once: gpg --export autosigner@kernel.org > keyring.gpg and set USEKEYRING at the top of the script to its full path. To update the tree at any point, use this command: git fetch --depth=1 origin <branchname> git checkout FETCH_HEAD Do not use git remote update inside a shallow checkout, as it will proceed to download the whole history, which for mainline is about 2.6 GB and potentially even larger for stable.

It'll take ages and probably get you banned, because this is very heavy server-side. Read more... Creepy crawlies August 29, 2026 from Konstantin Ryabitsev You've probably heard me complain about the “AI crawlers” before, but now I actually have some hard numbers I can put up to show their impact. In a few words, it's bad enough to create a constant “background radiation” of system load, permanently tying up a chunk of capacity spent on producing output that is only useful for a single purpose — feeding a learning model.

TL;DR: we spend more CPU cycles rendering commits for scrapers than we spend on all other kinds of legitimate access, including git clones. At any one time, across 5 geo-distributed nodes, there are 14 CPU cores doing nothing but rendering git commits as html. Why is git.kernel.org “interesting” to crawlers Linux development happens in the open — from git repositories you can clone, to discussion archives you can follow in real time.

To a large language model, this is a goldmine of learning data, because all of this is not only immediately available, but is easy to filter in order to guarantee pure unadulterated pre-AI

The rest of this page is on the live site.