2021-02-07 17:30:31

by Rue Mohr

[permalink] [raw]
Subject: nfs kernel server bug


I'm having an issue where the kernel server is dishing out all zeros for file content being read over NFS. The file byte
count is correct.
Writing files is fine.
With the same client, a different server works just fine.

I have been trying to pin this down, so far all I know is that the packets from the server actually contain zeros for
the file contents.

The troubled server kernel is 5.4.1

Has anyone encountered this or is it just me?



2021-02-08 12:05:32

by Benjamin Coddington

[permalink] [raw]
Subject: Re: nfs kernel server bug

On 7 Feb 2021, at 12:18, Rue Mohr wrote:

> I'm having an issue where the kernel server is dishing out all zeros
> for file content being read over NFS. The file byte count is correct.
> Writing files is fine.
> With the same client, a different server works just fine.
>
> I have been trying to pin this down, so far all I know is that the
> packets from the server actually contain zeros for the file contents.
>
> The troubled server kernel is 5.4.1
>
> Has anyone encountered this or is it just me?

Haven't heard of this before, but sounds like a fascinating problem.

What version of nfs, and what security? You're seeing READ operations
returning
with rsize-ed blocks of all-zero data in a network capture?

Ben

2021-02-24 00:53:51

by J. Bruce Fields

[permalink] [raw]
Subject: Re: nfs kernel server bug

On Mon, Feb 08, 2021 at 06:58:38AM -0500, Benjamin Coddington wrote:
> On 7 Feb 2021, at 12:18, Rue Mohr wrote:
>
> >I'm having an issue where the kernel server is dishing out all
> >zeros for file content being read over NFS. The file byte count is
> >correct.
> >Writing files is fine.
> >With the same client, a different server works just fine.
> >
> >I have been trying to pin this down, so far all I know is that the
> >packets from the server actually contain zeros for the file
> >contents.
> >
> >The troubled server kernel is 5.4.1
> >
> >Has anyone encountered this or is it just me?
>
> Haven't heard of this before, but sounds like a fascinating problem.
>
> What version of nfs, and what security? You're seeing READ
> operations returning
> with rsize-ed blocks of all-zero data in a network capture?

It might also be useful to know what filesystem you're exporting, with
what mount and export options.

--b.