1. 37

  2. 11

    TL;DR: A perfectly fine ext4 commit exposed an issue elsewhere in the kernel. This resulted in problematic system behaviour and had to be reverted.

    The original commit avoids large numbers of small I/O requests. Less I/O resulted in less entropy being available for the getrandom() syscall. At least for some Arch Linux system, systemd was using getrandom() during boot, which waited for entropy, but there was no other I/O to fill the entropy pool again.

    The proper fix (for getrandom()) needs more discussion, so the ext4 commit needs to be delayed until then.

    edit: related: https://lobste.rs/s/okrclo/linux_getrandom_boot_hangs

    1. 6

      To be precise, systemd isn’t the issue. Systemd has a target which purposefully waits for the crng to be initialized, and that doesn’t block the boot; it just looks suspicious in logs, which is why it got blamed. The real issue was that gnome-session blocked on getrandom.

      See https://lore.kernel.org/linux-ext4/20190915065142.GA29681@gardel-login/ where Pottering explains why Linus was incorrect in blaming systemd-random-seed, and https://lore.kernel.org/linux-ext4/CAHk-=wjPDR6_crhmvaoXDo8q6Joz5rD02bZpd2x9rr-LazPxRA@mail.gmail.com/ where Linus acknowledges that systemd is blameless and that the issue is that something is blocking to generate a cookie for X.

      EDIT: Also, fwiw, at https://lore.kernel.org/linux-ext4/20190915084802.GB29771@gardel-login/, Lennart describes why systemd doesn’t default to telling the kernel to account for the randomness from systemd-random-seed, which is what your linked article is complaining about. I don’t know it you agree with his reasons, but it’s good to realize it’s not just a bug or something they didn’t know they should’ve done.

        1. 4

          That’s an impressive URL slug.

      1. 7

        Titling this “Linux 5.3” is not inaccurate, but it fails to point out that there’s a really interesting story about a reverted commit and the reasoning for the revert.

        Thanks for sharing! But please change the title so that people don’t just scroll past what they believe to be a list of bugfixes :)

        1. 3

          Any editorializing of the title would be quickly reverted.

          1. 3

            Because it’s a link directly to the release which has that title? That seems a bit silly but I guess

            Do not editorialize story titles, but when the original story’s title has no context or is unclear, please change it.

            is pretty clear.

            1. 2

              Anyone can suggest a new title. I don’t understand why there needs to be a passive-aggressive call for anonymous / mod action.

              If you want to make the change “stick”, suggest a wording in a comment that people can use.

              My previous wording was unnecessarily harsh, I have struck them out above.

              That said, I don’t think that there has to be an addition to the title - the comments and the supplementary info on precisely what change was added and reverted is interesting enough.

              1. 1

                You misunderstand. I was asking the person posting, and I used the “suggest” functionality as well. I also wanted to gather opinions in discussion of the suggestion, so I made a comment.

                passive-aggressive call

                It was a direct call to action along with my opinion which I believe justifies the action. This is the opposite of passive-aggressive. It was not an anonymous request, nor one to moderators, per the context “Thanks for sharing” which directly addresses the person who posted.

                1. 2

                  Re-reading my comment I realize I misread you and over-reacted. I apologize and will edit my comment to reflect this.

                  1. 3

                    No worries, thank you for being willing to clarify and discuss!

          2. 3

            Thanks for pointing that out, I would have skipped the whole post otherwise.

            (In general I avoid reading the LKM; I think it leads to the ops version of hypochondria)

          3. 4

            Anyone remember that post Linus made about getting help for his anger issues and focusing hard on being compassionate towards other maintainers? Looks like he’s really taken that to heart. Seems like his old tone is basically gone here - very refreshing.

            1. 4

              TBH, Linus’s release emails always had this sort of tone even before he vowed to get help on his anger issues.

            2. 4

              I’d be interested in more info about the reverted commit and the problem exposed.