MIME-Version: 1.0
From:   Daire Byrne <daire@dneg.com>
Date:   Fri, 17 Feb 2023 11:54:12 +0000
Message-ID: <CAPt2mGNEYUk5u8V4abe=5MM5msZqmvzCVrtCP4Qw1n=gCHCnww@mail.gmail.com>
Subject: /proc/PID/io/read_bytes accounting regression?
To:     linux-nfs <linux-nfs@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

Hi,

Maybe someone here can quickly point me in the right direction for
this oddity that we noticed.

On newer kernels, it looks like the task io accounting is not
incrementing the read_bytes when reading from a NFS mount? This was
definitely working on v5.16 downwards, but has not been working since
v5.18 up to v6.2 (I haven't tested v5.17 yet).

If I read from a local filesystem, then the read_bytes for that PID is
incremented as expected.

If I read over NFS using directIO, then the read_bytes is also
correctly incremented for that PID. It's just when reading normally
without directIO that it is not.

The write_bytes and rchar are also still both correct in all situations.

I have checked the kernel config and I'm fairly sure I have all the
right things enabled:

CONFIG_TASK_XACCT=y
CONFIG_TASK_IO_ACCOUNTING=y
CONFIG_TASKSTATS=y

Unless there was some extra config introduced specific to the nfs
client in later kernels that I missed?

Cheers,

Daire