Re: 15.1 diskless hangs

> On 22 May 2026, at 15:38, Rick Macklem <[email protected]> wrote:
> 
> On Thu, May 21, 2026 at 10:24 PM Daniel Braniss <[email protected] <mailto:[email protected]>> wrote:
>> 
>> 
>> 
>>> On 21 May 2026, at 22:02, Bjoern A. Zeeb <[email protected]> wrote:
>>> 
>>> On Thu, 21 May 2026, Daniel Braniss wrote:
>>> 
>>>> have several bhives and my workstation, running diskless, the server is also freebsd,
>>>> and very often, they hang on nfs, (I assume trying to access the root).
>>>> having them boot locally, and everything else, ie. /usr/local, home directory are nfs mounted without issues.
>>>> 
>>>> i’ll try and do a tcpdump but in the meantime any isights are welcome.
>>> 
>>> When do they hang?  During boot? Or during operation?
>> 
>> during normal operation.
>>> 
>>> If they hang on an interactive command try ^T and see.
>> 
>> since root is nfs mounted, and it’s hung, nothing but power cycling works.
>>> 
>>> If tcpdump doesn't help much I'd start turning off checksum offloads and the like
>>> along the path (unclear if your host is your nfs server or not) to see if that helps.
>> 
>> some more info:
>> the server is running 14.3 (i don’t think this is relevant)
>> the server us also providing /usr/local
>> if the host is running with a local root, all is fine.
>> virtual hosts, i.e bhive, also hang under similar configuration.
>> 
>> so now i'm running a tcpdump on the server, and will probably have more info.
> That should give you more information. A few things to note:
> - 15.1 has the capability of using NFSv4 for root, but that requires some
>   careful configuration.
> - Assuming your root is NFSv3 (you'll see that in the packet capture
>  in wireshark), make sure you are not running nfsuserd(8) or gssd(8).
only NFS3 
>  Both of these will try and look up user/group names in the passwd/group
>  file and this can hang the system when they do the upcall from within
>  the NFS client (which would have to access the files).
> 
> rick

so it hanged,
im attaching the relevant tcpdump,
what is weird, and I must have forgotten much, 
nrnd is the server, chamsa is the client.
it seems nrnd is requesting arp too often …
is the TTL on the arp reseted when a packet is received, or
it just times out?
here is the abridged version:

> 
>> 
>> thanks,
>>        danny
>> 
>>> 
>>> /bz
>>> 
>>> --
>>> Bjoern A. Zeeb                                                     r15:7