Re: increase the number of routing tables

[Stephen Hemminger <shemminger@vyatta.com>]

On Mon, 30 Jan 2012 19:57:20 +0100 (CET)
Jan Engelhardt <jengelh@medozas.de> wrote:

> On Monday 2012-01-30 17:14, Stephen Hemminger wrote:
> 
> >> >> > There are 2147483647.
> >> >> 
> >> >> Any reason why it's not an unsigned 32bit int? (surely there is a corner 
> >> >> case where this is useful...)
> >> >
> >> >The 8 bit value is enshrined in the API for 'struct rtmsg' and therefore
> >> >increasing it would break existing applications. 
> >> 
> >> Actually, what Ed (and me too) was wondering about was:
> >> 
> >> why does `ip route show table $[0x80000000]` not print an empty table, 
> >> i.e. where is it that some code uses int/s32 during parsing of 
> >> the argument and/or the RTA_TABLE attribute?
> >
> >There are lots of places internally in ip utilities that use u32
> >for route table. But the route input/output message format is still limited
> >to 8 bits.
> 
> But then, adding a rule to table 258 would make it show up in table 2 as 
> well, which is not the case.

Ok, drilling deeper, overlooked the extended table support.
Tables >= 256 are handled by setting rtm_table to RT_TABLE_UNSPEC and
sending a RTA_TABLE attribute.
Therefore it has worked for quite a while.

But, I also see code in other utilities (like net-snmp and quagga)
which don't know about how to handle tables other than main. But
that is a different issue.