Re: [PATCH v5 40/40] 9p: Use netfslib read/write_iter

From: Emanuele Rocca
Date: Thu May 30 2024 - 15:17:13 EST


Hi David,

On 2024-05-23 08:44, David Howells wrote:
> In https://bugs.launchpad.net/ubuntu/+source/autopkgtest/+bug/2056461 you say:
>
> | It seems that kernel 6.8 introduced a regression in the 9pfs related to
> | caching and netfslib, that can cause some user-space apps to read content
> | from files that is not up-to-date (when they are used in a producer/consumer
> | fashion).
>
> Can you clarify how these files are being used?

I don't know the details of the 9pfs operations involved, but still I
wanted to mention that to reliably reproduce the issue on a Debian
system one can run:

autopkgtest-build-qemu unstable /tmp/sid.img
autopkgtest -ddd -B dpdk -- autopkgtest-virt-qemu --debug --show-boot /tmp/sid.img

If the kernel installed in the guest VM is affected by this problem,
after a while the test hangs with something like:

autopkgtest-virt-qemu: DBG: executing copydown /tmp/alog/tests-tree/ /tmp/autopkgtest.uG6tsJ/build.6QA/src/
[...]
autopkgtest-virt-qemu: DBG: +>?

Full logs at https://people.debian.org/~ema/1072004-6.10-rc1.log

Part of the code mounting the 9pfs in case it helps is at:
https://salsa.debian.org/ci-team/autopkgtest/-/blob/master/virt/autopkgtest-virt-qemu#L290

I could reproduce the issue with both 6.9.2 and 6.10-rc1.