Requested information attached. It was collected with "apport-cli -f --save" (for linux and zfsutils-linux) on a clean Ubuntu 26.04 install that reproduces the hang: linux-image-7.0.0-34-generic 7.0.0-34.34, zfsutils-linux 2.4.1-1ubuntu5.1. The production host where it first happened can't upload directly. The zfsutils-linux report and the reproducer follow in the next two comments.
zfs-livelock-reproducer.tar.gz holds the reproducer and results behind the Test Plan above. On the stock module, igrab() fails 1.5M+ times per 10 s and the arc_prune thread blocks in zfs_znode_hold_enter <- zfs_zinactive <- ... <- prune_icache_sb. With the two upstream commits that drops to about 100 per 10 s. This isn't hardware- or driver-specific. It's a logic bug in module/os/linux/zfs (zfs_prune() / zfs_zget()), fixed upstream by e95c55285dc8 and 98527e086f0d, which cherry-pick cleanly onto 2.4.1. Setting both tasks back to Confirmed. ** Attachment added: "apport-cli -f -p linux --save (clean 26.04 VM, 7.0.0-34-generic)" https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2170426/+attachment/6014073/+files/linux.apport -- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/2170426 Title: Livelock between arc_prune and zfs_zget() hangs ZFS on large pools (fixed upstream, openzfs#19266) To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2170426/+subscriptions -- ubuntu-bugs mailing list [email protected] https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
