Re: [RFC] NFSD: add rpc_status entry in nfsd debug filesystem

[Jeff Layton <jlayton@kernel.org>]

On Mon, 2023-07-10 at 15:55 +0000, Chuck Lever III wrote:
> 
> > On Jul 10, 2023, at 11:33 AM, Lorenzo Bianconi <lorenzo@kernel.org> wrote:
> > 
> > > > + for (i = 0; i < serv->sv_nrpools; i++) {
> > > > + struct svc_rqst *rqstp;
> > > > +
> > > > + seq_puts(m, "XID        | FLAGS      | PROG       |");
> > > > + seq_puts(m, " VERS       | PROC\t|");
> > > > + seq_puts(m, " REMOTE - LOCAL IP ADDR");
> > > > + seq_puts(m, "\t\t\t\t\t\t\t\t   | NFS4 COMPOUND OPS\n");
> > > > + list_for_each_entry_rcu(rqstp,
> > > > + &serv->sv_pools[i].sp_all_threads,
> > > > + rq_all) {
> > > > + if (!test_bit(RQ_BUSY, &rqstp->rq_flags))
> > > > + continue;
> > > > +
> > > > + seq_printf(m,
> > > > +    "0x%08x | 0x%08lx | 0x%08x | NFSv%d      | %s\t|",
> > > > +    be32_to_cpu(rqstp->rq_xid), rqstp->rq_flags,
> > > > +    rqstp->rq_prog, rqstp->rq_vers,
> > > > +    svc_proc_name(rqstp));
> > > > +
> > > > + if (rqstp->rq_addr.ss_family == AF_INET) {
> > > > + seq_printf(m, " %pI4 - %pI4\t\t\t\t\t\t\t   |",
> > > > +    &((struct sockaddr_in *)&rqstp->rq_addr)->sin_addr,
> > > > +    &((struct sockaddr_in *)&rqstp->rq_daddr)->sin_addr);
> > > > + } else if (rqstp->rq_addr.ss_family == AF_INET6) {
> > > > + seq_printf(m, " %pI6 - %pI6 |",
> > > > +    &((struct sockaddr_in6 *)&rqstp->rq_addr)->sin6_addr,
> > > > +    &((struct sockaddr_in6 *)&rqstp->rq_daddr)->sin6_addr);
> > > > + } else {
> > > > + seq_printf(m, " Unknown address family: %hu\n",
> > > > +    rqstp->rq_addr.ss_family);
> > > > + continue;
> > > > + }
> > > > +#ifdef CONFIG_NFSD_V4
> > > > + if (rqstp->rq_vers == NFS4_VERSION &&
> > > > +     rqstp->rq_proc == NFSPROC4_COMPOUND) {
> > > > + /* NFSv4 compund */
> > > > + struct nfsd4_compoundargs *args = rqstp->rq_argp;
> > > > + struct nfsd4_compoundres *resp = rqstp->rq_resp;
> > > > +
> > > > + while (resp->opcnt < args->opcnt) {
> > > > + struct nfsd4_op *op = &args->ops[resp->opcnt++];
> > > > +
> > > > + seq_printf(m, " %s", nfsd4_op_name(op->opnum));
> > > > + }
> > > > + }
> > > > +#endif /* CONFIG_NFSD_V4 */
> > > > + seq_puts(m, "\n");
> > > 
> > > My only quibble here is that the file format needs to be parsable
> > > by scripts as well as readable by humans. I'm not sure I have a
> > > specific comment, but it's something that needs some attention and
> > > verification (with, say, a sample user space tool, hint hint).
> > 
> > maybe we can add a csv hanlder, what do you think? not sure.
> 
> I suggested JSON to Jeff as another option, but I don't think we want
> to swing that far in the other direction.
> 
> There are plenty of examples of /sys files that are both parsable and
> human-friendly. I'll leave it to you to find one or two formats that
> seem capable of the task at hand, and let's pick from one of those.
> 

Are there already kernel libraries for this, or any examples of kernel
interfaces that emit JSON? Most of the kernel interfaces I have
experience with just use well-known fields delimited by spaces.

-- 
Jeff Layton <jlayton@kernel.org>