> > Yes, that was something that crossed my mind too. However,
 > > if we are going to allow some of these settings from ipadm
 > > (e.g., the nofailover flag), then in the interest of being
 > > symmetric, it would be nice to have some basic information
 > > displayed in the corresponding ipadm show-* sub-command.
 > 
 > I'd be tempted to draw the line at the flags, otherwise you're on
 > a slipperly slope whereby future work on IPMP needs to be done
 > on both tools.
 >
 > If this were to be remodeled, would things like the nofailover flag
 > be a part of the generic flags for an interface or would they belong
 > in a sub structure/field that was owned by IPMP?

To be clear, the IFF_NOFAILOVER flag is an address flag, not an interface
flag.  The name is a remnant from the old model that we got stuck with; it
should really be IFF_TESTADDR since it indicates that the address is only
for use in sending and receiving ICMP probes via in.mpathd.  Nonetheless,
the address belongs in ipadm/ifconfig because it is one of the IP
addresses the system will recognize as its own.  However, it needs to be
specially marked (ideally in a manner more obvious than IFF_NOFAILOVER is)
that the IP address is not suitable for general use.  There are no other
IPMP things like the IFF_NOFAILOVER flag; all other IPMP-specific flags
are phyint flags.

More generally, IPMP is a core part of IP and should be managed via ipadm,
in the same manner that link aggregation is a core part of L2 and managed
via dladm.  One of the advantages of the new IPMP model is that each group
has its own IP interface and thus any per-group properties can be
administered as IP properties on that group's IPMP IP interface -- and
indeed, we intentionally avoided creating ipmpadm since the plan was for
ipadm to fill this need.

-- 
meem

Reply via email to