[Dnsmasq-discuss] Reg: Info related to leases file

[Simon Kelley]

On 25/09/13 22:48, Vladislav Grishenko wrote:
>> From: Simon Kelley [mailto:simon at thekelleys.org.uk]
>>> I understand that leasefile format is preserved due legacy reasons to
>>> allow dnsmasq precompiled updates.
>>> What if just to add LEASEFILE_LEGACY_FORMAT define to control the subj
>>> format?
>>> Ppl who able to build it and to control lease format dependent apps
>>> would be able to enable/disable it and use on his own, while the rests
>>> can continue to use legacy defaults with no harm.
>>> No code redundancy would be brought this way, new/old lease-format
>>> code would be just not compiled in, anyway don't treat this opinion as
>>> a kind of pushing, just consider it's probably good to have it in
> upstream.
>>
>> The problem is that you're asking for two incompatible things. You want
> the
>> leasefile format to be a defined interface, which implies, amongst other
>> things, that it's stable, and you want to change the leasefile format.
>> BTW, as far as I'm concerned, the leasefile format has never been a
> defined
>> interface, if people are looking in there directly, that's their risk.
>
> Actually not exactly, I see possible leasfile code simplification, and in
> the side effect it could be possible to have file format as defined
> interface, nothing else.
> Ok, if lease format hasn't never been a defined interface, nothing should
> prevent from format change at any time, right? So, why not then.
>
>>>
>>> Thanks for the remind, both cases require either HAVE_SCRIPT or
>> HAVE_DBUS to
>>> be enabled, openwrt folks use scripts from the very start.
>>> So, dnsmasq code, additional not-dnsmasq file sizes&   required mem
>>> increases, complexity grows, possible fault places number too, with the
>> same
>>> desired result as if leasefile contains new lease info.
>>> In my humble opinion, it seems a little bit overkill for embedded world.
>>> As an example - a number of recent mails of Tomato firmware folks here
> in
>>> the list when they decided to update dnsmasq and have accidentally
>> dropped
>>> homegrown leases-save code, that extends the format of the original one.
>>> They have the same reasons - keep dnsmasq&   outter environment as
>> small and
>>> functional as possible.
>>
>> HAVE_SCRIPT isn't very big and this tiny shell script makes a file of
>> (IP-address, MAC-address) pairs. It's trivial to alter it to split IPv4
>> and IPv6 into different files or include any information from the fields
>> exposed by the script interface. The script interface is defined and
>> I'll keep it compatible going forward.
>
> Ok, giving up. Script-based has even-driven pros while adding/deleting
> lease. My next concern is about lease init stage, when dnsmasq needs to be
> restarted due reconfig. Does it need to be implemented, or leasefile in
> whatever format could coexist with script?


The leasefile can co-exist with the script, the script doesn't need to 
do anything for dnsmasq to keep working.

There's one change which needs to be made to the script. When dnsmasq is 
restarted, it won't know the MAC addresses for DHCPv6 (because they're 
not in the leasefile). So at start-up it  will execute "old" script runs 
on each lease without the DNSMASQ_MAC environment variable set. The 
script needs to detect when called with "old" and no DNSMASQ_MAC and not 
delete the relevant line from its file.

Cheers,

Simon.


>
> Best Regards, Vladislav Grishenko
>
>
>