Re: [PATCH 1/1] [RFC] sunrpc: inject process namespace into machinename field

Thanks, Jeff. Yes, Chuck has sent me the link too. Indeed, we have
noticed that while tracing some applications with perf.

Chuck as well suggested using the COMPOUND tag, which sounds more
reasonable and will work with RPCSEC_GSS. I will post it after testing.

Best regards,
   Tigran.

----- Original Message -----
> From: "Jeff Layton" <[email protected]>
> To: "Tigran Mkrtchyan" <[email protected]>, "trondmy" <[email protected]>, "chuck lever"
> <[email protected]>, "anna" <[email protected]>
> Cc: "linux-nfs" <[email protected]>
> Sent: Friday, 26 June, 2026 16:04:36
> Subject: Re: [PATCH 1/1] [RFC] sunrpc: inject process namespace into machinename field

> On Fri, 2026-06-26 at 14:52 +0200, Tigran Mkrtchyan wrote:
>> On large shared machines often multiple jobs of a same user run in
>> parallel. For debugging, it's usually impossible to identify requests
>> coming from different processes.
>> 
>> The batch systems like HTCondor or SLURM start every job in it's own
>> namespace, thus passing namespace info to the server will help by
>> debugging.
>> 
>> Signed-off-by: Tigran Mkrtchyan <[email protected]>
>> ---
>>  net/sunrpc/auth_unix.c | 27 +++++++++++++++++++++++++--
>>  1 file changed, 25 insertions(+), 2 deletions(-)
>> 
>> diff --git a/net/sunrpc/auth_unix.c b/net/sunrpc/auth_unix.c
>> index 6c742a3400c4..b218cfa9871a 100644
>> --- a/net/sunrpc/auth_unix.c
>> +++ b/net/sunrpc/auth_unix.c
>> @@ -15,6 +15,7 @@
>>  #include <linux/sunrpc/clnt.h>
>>  #include <linux/sunrpc/auth.h>
>>  #include <linux/user_namespace.h>
>> +#include <linux/pid_namespace.h>
>>  
>>  
>>  #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
>> @@ -117,6 +118,28 @@ unx_marshal(struct rpc_task *task, struct xdr_stream *xdr)
>>  	struct group_info *gi = cred->cr_cred->group_info;
>>  	struct user_namespace *userns = clnt->cl_cred ?
>>  		clnt->cl_cred->user_ns : &init_user_ns;
>> +	char ns_aware_nodename[UNX_MAXNODENAME + 1];
>> +	int ns_aware_nodename_len;
>> +
>> +	struct pid_namespace *pid_ns = task_active_pid_ns(current);
>> +
>> +	/* the process runs in a dedicated namespace */
>> +	if (pid_ns != &init_pid_ns) {
> 
> FYI, Sashiko has some comments on this that seem valid:
> 
> https://sashiko.dev/#/patchset/20260626125216.1467845-2-tigran.mkrtchyan-T5F83Mi6MZE@public.gmane.org?part=1
> 
> Fetching the user namespace ID might not be correct in some async ops.
> 
> 
>> +		/* Format as: <pid_ns_inum>@<current-hostname> */
>> +		int prefix_len = snprintf(ns_aware_nodename, sizeof(ns_aware_nodename),
>> +			"%u@", pid_ns->ns.inum);
>> +
>> +		if (prefix_len < sizeof(ns_aware_nodename))
>> +			strscpy(ns_aware_nodename + prefix_len, clnt->cl_nodename,
>> +				sizeof(ns_aware_nodename) - prefix_len);
>> +		else
>> +			strscpy(ns_aware_nodename, clnt->cl_nodename, sizeof(ns_aware_nodename));
>> +
>> +		ns_aware_nodename_len = strlen(ns_aware_nodename);
>> +	} else {
>> +		ns_aware_nodename_len = clnt->cl_nodelen;
>> +		strscpy(ns_aware_nodename, clnt->cl_nodename, sizeof(ns_aware_nodename));
>> +	}
>>  
>>  	/* Credential */
>>  
>> @@ -126,8 +149,8 @@ unx_marshal(struct rpc_task *task, struct xdr_stream *xdr)
>>  	*p++ = rpc_auth_unix;
>>  	cred_len = p++;
>>  	*p++ = xdr_zero;	/* stamp */
>> -	if (xdr_stream_encode_opaque(xdr, clnt->cl_nodename,
>> -				     clnt->cl_nodelen) < 0)
>> +	if (xdr_stream_encode_opaque(xdr, ns_aware_nodename,
>> +				     ns_aware_nodename_len) < 0)
>>  		goto marshal_failed;
>>  	p = xdr_reserve_space(xdr, 3 * sizeof(*p));
>>  	if (!p)
> 
> --
> Jeff Layton <[email protected]>