Re: [PATCH v2 0/2] Fix sockaddr handling in NFSD trace points

[Steven Rostedt <rostedt@goodmis.org>]

On Mon, 10 Jan 2022 10:56:49 -0500
Chuck Lever <chuck.lever@oracle.com> wrote:

> The patches in this series address a simple buffer over-read bug in
> the Linux NFS server.
> 
> However I was thinking it would be nice to have trace helpers to
> deal safely with generic socket addresses. I'd like to be able to
> treat them the same way we currently treat strings. So for example:
> 
> 
> #define field_sockaddr(field, len)  __dynamic_array(u8, field, len)
> #define assign_sockaddr(dest, src, len)  memcpy(__get_dynamic_array(dest), src, len)
> #define __get_sockaddr(field)  ((struct sockaddr *)__get_dynamic_array(field))
> 
> TRACE_EVENT(sockaddr_example,
>         TP_PROTO(
>                 const struct sockaddr *sap,
>                 size_t salen
>         ),  
>         TP_ARGS(sap, salen),
>         TP_STRUCT__entry(
>                 __field_sockaddr(addr, salen)
>         ),  
>         TP_fast_assign(
>                 __assign_sockaddr(addr, sap, salen);
>         ),  
>         TP_printk("addr=%pIS", __get_sockaddr(addr))
> );
> 
> 
> should be able to store any address family in a dynamically-sized
> array field (addr).
> 
> I haven't quite been able to figure out how to handle the 
> TP_printk() part of this equation. `trace-cmd report` displays
> something like "addr=ARG TYPE NOT FIELD BUT 7". 
> 
> Thoughts or advice appreciated.

I'll take a look into it.

Thanks,

-- Steve