Re: [PATCH 3/3] NFSD: Add trace point for SCSI fencing operation.

[Dai Ngo <dai.ngo@oracle.com>]

On 11/2/25 7:40 AM, Chuck Lever wrote:
> On 11/1/25 2:51 PM, Dai Ngo wrote:
>> +	TP_STRUCT__entry(
>> +		__string(dev, dev)
>> +		__array(unsigned char, addr, sizeof(struct sockaddr_in6))
>> +		__field(u32, error)
>> +	),
>> +	TP_fast_assign(
>> +		memcpy(__entry->addr, &clp->cl_addr,
>> +			sizeof(struct sockaddr_in6));
>> +		__assign_str(dev);
>> +		__entry->error = error;
>> +	),
>> +	TP_printk("client=%pISpc dev=%s error=%d",
>> +		__entry->addr,
>> +		__get_str(dev),
>> +		__entry->error
>> +	)
> Have a look at the nfsd_cs_slot_class event class (fs/nfsd/trace.h) to
> see how to use the trace subsystem's native sockaddr handling.

The native sockaddr handling used by nfsd_cs_slot_class seems
to be broken:

[opc]# trace-cmd record -e nfsd_slot_seqid_conf -e nfsd_slot_seqid_unconf
[opc]# trace-cmd report
CPU 0 is empty
...
CPU 15 is empty
cpus=16
             nfsd-2784  [011]  1205.355839: nfsd_slot_seqid_unconf: [FAILED TO PARSE] seqid=1 slot_seqid=0 cl_boot=1762214337 cl_id=3172331794 addr=
[opc]#

After changing 'nfsd_pnfs_class' to use native sockaddr handling:

[opc]# trace-cmd record -e nfsd_pnfs_fence
kworker/u65:0-7458  [001] 933.253811: nfsd_pnfs_fence: [FAILED TO PARSE] addr=ARRAY[02, 00, 02, f9, 64, 64, fa, 2b, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00] netns_ino=4026531833 dev=sdb error=24
[opc]#

-Dai

>
> Do you want to record anything else here? The cl_boot/cl_id, perhaps? I
> guess there's no XID available... but is there a relevant state ID?
>