Re: [PATCH v6 3/3] NFSD: add rpc_status entry in nfsd debug filesystem

[Lorenzo Bianconi <lorenzo@kernel.org>]

[-- Attachment #1: Type: text/plain, Size: 4247 bytes --]

[...]
> > +#ifdef CONFIG_NFSD_V4
> > +			if (rqstp->rq_vers == NFS4_VERSION &&
> > +			    rqstp->rq_proc == NFSPROC4_COMPOUND) {
> > +				/* NFSv4 compund */
> > +				struct nfsd4_compoundargs *args = rqstp->rq_argp;
> > +				int j;
> > +
> > +				opcnt = args->opcnt;
> > +				for (j = 0; j < opcnt; j++) {
> > +					struct nfsd4_op *op = &args->ops[j];
> > +
> > +					rqstp_info.opnum[j] = op->opnum;
> > +				}
> > +			}
> > +#endif /* CONFIG_NFSD_V4 */
> > +
> > +			/*
> > +			 * Acquire rq_status_counter before reporting the rqst
> > +			 * fields to the user.
> > +			 */
> > +			if (smp_load_acquire(&rqstp->rq_status_counter) != status_counter)
> > +				continue;
> > +
> > +			seq_printf(m,
> > +				   "%04u %04ld NFSv%d %s %016lld",
> > +				   be32_to_cpu(rqstp_info.rq_xid),
> 
> It's proper to display XIDs as 8-hexit hexadecimal values, as you
> did before. "0x%08x" is the correct format, as that matches the
> XID display format used by Wireshark and our tracepoints.

ops, I misunderstood your previous comments. I will address them in v7 if there
are no other comments.

Regards,
Lorenzo

> 
> 
> > +				   rqstp_info.rq_flags,
> 
> I didn't mean for you to change the flags format to decimal. I was
> trying to point out that the content of this field will need to be
> displayed symbolically if we care about an easy user experience.
> 
> Let's stick with hex here. A clever user can read the bits directly
> from that. All others should have a tool that parses this field and
> prints the list of bits in it symbolically.
> 
> 
> > +				   rqstp_info.rq_vers,
> > +				   rqstp_info.pc_name,
> > +				   ktime_to_us(rqstp_info.rq_stime));
> > +			seq_printf(m, " %s",
> > +				   __svc_print_addr(&rqstp_info.saddr, buf,
> > +						    sizeof(buf), false));
> > +			seq_printf(m, " %s",
> > +				   __svc_print_addr(&rqstp_info.daddr, buf,
> > +						    sizeof(buf), false));
> > +			if (opcnt) {
> > +				int j;
> > +
> > +				seq_puts(m, " ");
> > +				for (j = 0; j < opcnt; j++)
> > +					seq_printf(m, "%s%s",
> > +						   nfsd4_op_name(rqstp_info.opnum[j]),
> > +						   j == opcnt - 1 ? "" : ":");
> > +			} else {
> > +				seq_puts(m, " -");
> > +			}
> 
> This looks correct to me.
> 
> I'm leaning towards moving this to a netlink API that can be
> extended over time to handle other stats and also act as an NFSD
> control plane, similar to other network subsystems.
> 
> Any comments, complaints or rotten fruit from anyone?
> 
> 
> > +			seq_puts(m, "\n");
> > +		}
> > +	}
> > +
> > +	rcu_read_unlock();
> > +
> > +	return 0;
> > +}
> > +
> > +/**
> > + * nfsd_rpc_status_open - open routine for nfsd_rpc_status handler
> > + * @inode: entry inode pointer.
> > + * @file: entry file pointer.
> > + *
> > + * nfsd_rpc_status_open is the open routine for nfsd_rpc_status procfs handler.
> > + * nfsd_rpc_status dumps pending RPC requests info queued into nfs server.
> > + */
> > +int nfsd_rpc_status_open(struct inode *inode, struct file *file)
> > +{
> > +	int ret = nfsd_stats_open(inode);
> > +
> > +	if (ret)
> > +		return ret;
> > +
> > +	return single_open(file, nfsd_rpc_status_show, inode->i_private);
> > +}
> > diff --git a/include/linux/sunrpc/svc.h b/include/linux/sunrpc/svc.h
> > index 7838b37bcfa8..b49c0470b4fe 100644
> > --- a/include/linux/sunrpc/svc.h
> > +++ b/include/linux/sunrpc/svc.h
> > @@ -251,6 +251,7 @@ struct svc_rqst {
> >  						 * net namespace
> >  						 */
> >  	void **			rq_lease_breaker; /* The v4 client breaking a lease */
> > +	unsigned int		rq_status_counter; /* RPC processing counter */
> >  };
> >  
> >  /* bits for rq_flags */
> > diff --git a/net/sunrpc/svc.c b/net/sunrpc/svc.c
> > index af692bff44ab..83bee19df104 100644
> > --- a/net/sunrpc/svc.c
> > +++ b/net/sunrpc/svc.c
> > @@ -1656,7 +1656,7 @@ const char *svc_proc_name(const struct svc_rqst *rqstp)
> >  		return rqstp->rq_procinfo->pc_name;
> >  	return "unknown";
> >  }
> > -
> > +EXPORT_SYMBOL_GPL(svc_proc_name);
> >  
> >  /**
> >   * svc_encode_result_payload - mark a range of bytes as a result payload
> > -- 
> > 2.41.0
> > 
> 
> -- 
> Chuck Lever

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]