Re: [tor-bugs] #22203 [Core Tor/Tor]: Should a hup reload the geoip files?

["Tor Bug Tracker & Wiki" <blackhole@xxxxxxxxxxxxxx>]

#22203: Should a hup reload the geoip files?
--------------------------+---------------------
 Reporter:  arma          |          Owner:
     Type:  enhancement   |         Status:  new
 Priority:  Medium        |      Milestone:
Component:  Core Tor/Tor  |        Version:
 Severity:  Normal        |     Resolution:
 Keywords:                |  Actual Points:
Parent ID:                |         Points:
 Reviewer:                |        Sponsor:
--------------------------+---------------------
Description changed by arma:

Old description:

> In config.c in config_maybe_load_geoip_files_() we have this line:
> {{{
>   /* XXXX Reload GeoIPFile on SIGHUP. -NM */
> }}}
>
> and sure enough, it looks like there's nothing in main.c's do_hup() that
> would make us reload the geoip files.
>
> It would be relatively easy to do I think:
> * In do_hup(), right around the call to routerlist_reset_warnings(), we
> call something in geoip.c that tells it to no longer consider itself
> initialized. Maybe that call is something like clear_geoip_db().
> * Then in config_maybe_load_geoip_files_(), since geoip_is_loaded()
> returns 0, it loads them.
>
> Things get tricky though: catalyst asked if reloading the geoip files
> messes up the statistics gathering.
>
> If reloading the geoip files *does* mess up statistics gathering, we have
> ourselves a minor bug, since config_maybe_load_geoip_files_() does reload
> them if the GeoIPFile location changes.
>
> But it turns out to be more complicated than that, since
> geoip_load_file() only clears selective things: it clears
> geoip_ipv4_entries and geoip_ipv6_entries, but it leaves geoip_countries
> alone! And I see elsewhere, at the bottom of geoip_note_client_seen(),
> that we are keeping statistics directly in the geoip_countries smartlist:
> {{{
>       geoip_country_t *country = smartlist_get(geoip_countries,
> country_idx);
>       ++country->n_v3_ns_requests;
> }}}
>
> So we would not want to call clear_geoip_db() on hup, or we'd lose some
> stats.
>
> I guess that means if we want to make do_hup reload the geoip stats file,
> the better (non-invasive) plan is to have a boolean
> want_to_reload_geoip{4,6} in geoip.c that we turn on in do_hup, and then
> we check the right boolean in geoip_is_loaded().
>
> I think there's a big argument for growing some good unit tests here,
> around "what happens when we reload this, collect those stats, reload
> that, examine those other stats, etc", so things are either subtly broken
> right now, or awfully fragile.

New description:

 In config.c in config_maybe_load_geoip_files_() we have this line:
 {{{
   /* XXXX Reload GeoIPFile on SIGHUP. -NM */
 }}}

 and sure enough, it looks like there's nothing in main.c's do_hup() that
 would make us reload the geoip files.

 It would be relatively easy to do I think:
 * In do_hup(), right around the call to routerlist_reset_warnings(), we
 call something in geoip.c that tells it to no longer consider itself
 initialized. Maybe that call is something like clear_geoip_db().
 * Then in config_maybe_load_geoip_files_(), since geoip_is_loaded()
 returns 0, it loads them.

 Things get tricky though: catalyst asked if reloading the geoip files
 messes up the statistics gathering.

 If reloading the geoip files *does* mess up statistics gathering, we have
 ourselves a minor bug, since config_maybe_load_geoip_files_() does reload
 them if the GeoIPFile location changes.

 But it turns out to be more complicated than that, since geoip_load_file()
 only clears selective things: it clears geoip_ipv4_entries and
 geoip_ipv6_entries, but it leaves geoip_countries alone! And I see
 elsewhere, at the bottom of geoip_note_client_seen(), that we are keeping
 statistics directly in the geoip_countries smartlist:
 {{{
       geoip_country_t *country = smartlist_get(geoip_countries,
 country_idx);
       ++country->n_v3_ns_requests;
 }}}

 So we would not want to call clear_geoip_db() on hup, or we'd lose some
 stats.

 I guess that means if we want to make do_hup reload the geoip stats file,
 the better (non-invasive) plan is to have a boolean
 want_to_reload_geoip{4,6} in geoip.c that we turn on in do_hup, and then
 we check the right boolean in geoip_is_loaded().

 I think there's a big argument for growing some good unit tests here,
 around "what happens when we reload this, collect those stats, reload
 that, examine those other stats, etc", since things are either subtly
 broken right now, or awfully fragile.

--

--
Ticket URL: <https://trac.torproject.org/projects/tor/ticket/22203#comment:1>
Tor Bug Tracker & Wiki <https://trac.torproject.org/>
The Tor Project: anonymity online
_______________________________________________
tor-bugs mailing list
tor-bugs@xxxxxxxxxxxxxxxxxxxx
https://lists.torproject.org/cgi-bin/mailman/listinfo/tor-bugs