Hi Samuel,
> I don't really see these macros being widely used?
> Better implement something that is a largely-known interface,
> for instance a common BSD interface.
Agreed on dropping SIOCGIFSTATS as the target shape. Two BSD
candidates fit pfinet's per-iface model and either has live
userland on the consumer side:
(a) getifaddrs(3) growing a populated ifa_data field of type
`struct if_data` (sys/net/if.h on the BSDs; fields
ifi_ibytes / ifi_obytes / ifi_ipackets / ifi_opackets /
ifi_ierrors / ifi_oerrors / ifi_iqdrops, plus a handful
more). This is the modern BSD path; ifconfig on the BSDs,
`netstat -i`, and the BSD-flavoured rust / go netlink
stand-ins all read from it.
(b) sysctl(NET_RT_IFLIST2) populating `struct if_msghdr2` with
the same if_data block. Older, heavier on the pfinet side
because it adds a sysctl surface that does not exist today.
Provisionally I would go (a) since it requires the smaller new
surface (one populated field on existing API rather than a new
sysctl tree). I have some open questions before drafting:
- Does Hurd's glibc getifaddrs currently leave ifa_data NULL
on pfinet-backed interfaces, or is there already a partial
surface I have not noticed? If NULL today, the patch is
"pfinet emits if_data via a new MIG routine + glibc-side
getifaddrs back-end reads it". If partially populated, the
patch is "extend the existing emission with counter fields".
- Either (a) or (b), do you have a preference on the MIG-side
shape, an array-return routine vs a per-iface routine that
callers iterate?
Happy to test once a draft lands. I will not block on writing
the patch myself in the short term but will pivot the original
ask off SIOCGIFSTATS in the bug record either way.
Thanks,
Borja Tarraso
[email protected]
El dom, 31 may 2026 a las 12:28, Samuel Thibault
(<[email protected]>) escribió:
>
> Hello,
>
> Borja Tarraso, le sam. 30 mai 2026 22:53:54 +0300, a ecrit:
> > On Debian GNU/Hurd 0.9 (pfinet as shipped in the hurd source package,
> > hurd 1:0.9.git20230520-7), there is no RPC, no ioctl, and no procfs
> > file that returns per-interface byte / packet counters.
>
> Contribution welcome!
>
> > 1. /usr/include/x86_64-gnu/bits/ioctls.h defines SIOC* constants
> > No SIOCGIFSTATS, no SIOCGIFMIB, no SIOCGIFDATA, no
> > SIOCGIFCOUNTERS macro anywhere under /usr/include.
>
> I don't really see these macros being widely used?
> Better implement something that is a largely-known interface, for
> instance a common BSD interface.
>
> (the Linux way, /proc, won't be helpful to us since that only defers the
> question to the pfinet interface question anyway)
>
> Samuel
