Re: [nft PATCH] tests/py: Set a fixed timezone in nft-test.py

[Phil Sutter <phil@nwl.cc>]

Hi Pablo,

On Tue, Nov 19, 2019 at 11:33:18PM +0100, Pablo Neira Ayuso wrote:
> On Tue, Nov 19, 2019 at 11:20:23PM +0100, Phil Sutter wrote:
> > Hi,
> > 
> > On Tue, Nov 19, 2019 at 11:12:36PM +0100, Pablo Neira Ayuso wrote:
> > > On Tue, Nov 19, 2019 at 12:06:23PM +0100, Ander Juaristi wrote:
> > > > El 2019-11-18 19:34, Pablo Neira Ayuso escribió:
> > > > > Hi Phil,
> > > > > 
> > > > > On Sat, Nov 16, 2019 at 10:32:18PM +0100, Phil Sutter wrote:
> > > > > > Payload generated for 'meta time' matches depends on host's timezone
> > > > > > and
> > > > > > DST setting. To produce constant output, set a fixed timezone in
> > > > > > nft-test.py. Choose UTC-2 since most payloads are correct then, adjust
> > > > > > the remaining two tests.
> > > > > 
> > > > > This means that the ruleset listing for the user changes when daylight
> > > > > saving occurs, right? Just like it happened to our tests.
> > > > 
> > > > It shouldn't, as the date is converted to a timestamp that doesn't take DST
> > > > into account (using timegm(3), which is Linux-specific).
> > > > 
> > > > The problem is that payloads are hard-coded in the tests.
> > > > 
> > > > Correct me if I'm missing something.
> > > 
> > > I see, so it's just the _snprintf() function in the library. I
> > > remember we found another problem with these on big endian, it would
> > > be probably to move them to libnftables at some point.
> > > 
> > > Acked-by: Pablo Neira Ayuso <pablo@netfilter.org>
> > 
> > Ah, now that I reread your question, I finally got it. And you're right:
> > If DST occurs, time values will change. This is clear from looking at
> > hour_type_print(): Whatever the kernel returned gets cur_tm->tm_gmtoff
> > added to it. Here, this is either 3600 or 7200 depending on whether DST
> > is active or not.
> > 
> > The other alternative would be to make kernel DST-aware, I don't think
> > that's the case.
> 
> Hm, so the ruleset listing changes after DST.

Yes, but only because payload (and hence kernel data) remains the same.

> If the user specifies the time according to the userspace timezone,
> then ruleset listing should not change?
> 
> I'm still trying to understand if this is a real problem.

IMHO, it's a controversy of whether we want to make the time match
respect DST changes or not:

It should, because when I restrict internet access to my kids after 8pm,
I don't want to touch my firewall's ruleset twice a year.

It needs not, because upon the next reboot things are correct again,
anyway.

It should not because otherwise once a year there are two spooky hours in
which things run twice that should never ever run more than once per
night. Apart from that, description of --kerneltz in
iptables-extensions(8) speaks for itself.

So long story short, it is always a problem no matter how it is solved.
:(

Cheers, Phil