Squid and the Year 2000

From: Julian Anderson <Julian.Anderson@dont-contact.us>
Date: Sat, 28 Mar 1998 19:26:09 +1200 (NZST)

Our IT group has recently been catching up on it's buzzword quota
and attempting to determine which mission-critical network infra-
structure elements are Year 2000 compliant, and for those that aren't,
who to point the figure at. Hard questions have been asked about
Squid, mainly because there is no-one to point the finger at if it
does break, so said finger has been pointed at me.

Has anyone taken a good, serious look at Squid for Y2K compliance? My
first thought is that Squid is probably pretty safe, as most (all?) of
it's internal dates are handled as "time_t"s or "struct tm"s, so the
main place it could get into strife is with dates handed to it by
browsers and servers. Unfortunately, our Y2K auditors frown at
phrases like "probably pretty safe".

Now of course we all know how closely browser and server implementors
follow the specs, right? So I took a look at the relevant RFCs. RFC
1945 (HTTP/1.0) and RFC 2068 (HTTP/1.1) both allow the same three
formats:
       Sun, 06 Nov 1994 08:49:37 GMT ; RFC 822, updated by RFC 1123
       Sunday, 06-Nov-94 08:49:37 GMT ; RFC 850, obsoleted by RFC 1036
       Sun Nov 6 08:49:37 1994 ; ANSI C's asctime() format

A quick perusal of lib/rfc1123.c (from the 1.2beta11 source I happened
to have lying around) suggests that squid probably screw up if handed
a bogus RFC 850 date such as
        Saturday, 01-Jan-100 08:49:37 GMT
because it depends on positional referencing into the array, and it
assumes that years are two digits. It will cope with
        Saturday, 01-Jan-00 08:49:37 GMT
but will then start making some incorrect assumptions about expiry
time, etc. This won't make squid fail, just be perhaps a little less
efficient as it may muck up IMS computations and pseudo-expiry-time
generation.

rfc1123.c does the Right Thing in generating Common Log Format entries
and other date headers (it uses strict RFC1123 4-digit date format for
that), assuming your OS strftime behaves correctly.

Of course, if your OS is broken all bets are off. In that case,
however, there often is someone you can point the proverbial finger
at.

Does anyone else have any suggestions about other places to look for
potential problems? I haven't examined the cachemgr, dnsserver,
client, unlinkd or ftpget source, nor have I checked if the
"acl time" day-abbrevs code correctly handles Y2K leap-years.
However, I don't use "acl time", so I don't care :-).

Does anyone *know* of any other problems, or have had to make the case
for Squid Y2K compliance to their bosses and would like to share their
report?

I don't think it's worth investigating 1.1 or 1.0 for compliance; by
the time problems would have struck you should have upgraded to
something newer.

Comments, anyone?

Jules Anderson
Received on Fri Mar 27 1998 - 23:32:38 MST

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:39:28 MST