i/o stats per inode in /proc/net/tcp

[Robert Hogan <robert@roberthogan.net>]

This may be a more appropriate question for lkml, but:

Why does /proc/net/tcp not give i/o stats per inode? Briefly looking at 
tcp_ipv4.c and co., it seems none of the sock structures associated with a 
tcp sock contain i/o counters so it may be reasons of non-triviality. That 
said, it does seem as if tcp_input.c and tcp_output.c could implement 
counting if there was a counter available in tcp_sock or the like.

Is there another /proc file that can be used in conjuction with /proc/net/tcp 
to gather per inode i/o stats? 

It just seems like such a natural thing to offer - as a user I'm always 
wondering where that little trickle of i/o on my bw monitor is coming from.

Can someone enlighten me?


Thanks,
Robert