[This is the E-mail I am referring to in the message that now precedes this, since it bounced the first time round, since I included a load of attachments with the original]
I read on the mailing list that AOL are turning on XFF on their proxies. I have been poking around a bit to find out how the AOL proxy information squares up with reports of AOL vandalism.
== The boring details ==
First, I did a DNS reverse lookup on the entire AOL Proxy IP range, as defined in http://webmaster.info.aol.com/proxyinfo.html using dig in batch mode. Out of the 56542 addresses in those ranges, only 6740 give a valid reverse lookup.
I then went through the list of 116 pages flagged with the {{AOL}} template, and extracted their IP addresses, and compared them to the list of reverse lookups generated above.
Out of 116 pages with addresses flagged with {{AOL}}: * 76 were found in the list of reverse lookups generated above, and every one of them had an address of the form *.proxy.aol.com * 40 were not
I then ran the 40 remaining addresses through dig -x.
Of these, 32 had valid reverse lookups, most of which were of the form [8 hex digits, starting with AC].ipt.aol.com (for example, ACA40DC5.ipt.aol.com.) Every single one of these was in the address range 172.128.0.0 - 172.216.255.255, assigned to AOL clients according to the AOL proxy info page.
Five of the addresses flagged with {{AOL}} did not belong to AOL addresses at all, namely:
195.9.72.12.in-addr.arpa. 172768 IN PTR 195.los-angeles-19-20rs.ca.dial-access.att.net. 22.2.25.138.in-addr.arpa. 86369 IN PTR www2.itd.uts.edu.au. 176.72.250.134.in-addr.arpa. 86370 IN PTR elc214-176.lab.suu.edu. 10.219.196.205.in-addr.arpa. 272 IN PTR franc.dreamhost.com. 5.21.196.69.in-addr.arpa. 1773 IN PTR CPE00609425bbe3-CM00080d7f2c84.cpe.net.cable.rogers.com.
The remaining five {{AOL}}-flagged pages, which appear to have no reverse lookup at all, are (with whois lookups):
;163.130.157.152.in-addr.arpa. -> 152.157.130.163 -> Washington School Information Processing Cooperative ;106.209.188.205.in-addr.arpa. -> 205.188.209.106 -> AOL ;76.164.174.149.in-addr.arpa. -> 149.174.164.76 -> Compuserve (ie AOL) ;234.96.12.64.in-addr.arpa. -> 64.12.96.234 -> AOL ;135.209.188.205.in-addr.arpa. -> 205.188.209.135 -> AOL
and of the last four given as AOL addresses, all of them were in the AOL server IP address ranges given by AOL.
== Summary ==
Out of all of the 116 pages flagged as {{AOL}}: * six are bogus non-AOL addresses, and are probably attempts by vandals to confuse anti-vandalism efforts * 32 are from the AOL client range, with *.ipt.aol.com reverse lookups * 76 are from the official AOL proxy range, with *.proxy.aol.com reverse lookups; they all appear to be either of the form cache-XXX-XXXX.proxy.aol.com [74 of them], or spider-XXX-XXXXX.proxy.aol.com. [2 of them] * four have no reverse lookup, but are in the official AOL proxy range
Out of the 458 cache-*.proxy.aol.com servers, only 74 are flagged with {{AOL}} Out of the 2537 spider-*.proxy.aol.com servers, only 2 are flagged with {{AOL}}
== Conclusions ==
* It seems safe to assume that *.proxy.aol.com servers are valid AOL proxies; these account for about two-thirds of all {{AOL}} warnings * It _might well_ be safe to assume that other servers from the AOL server range without reverse lookups are also AOL proxies, but I'm not sure that this is necessarily so; these account for < 5% of the valid {{AOL}} warnings. * But about a third of {{AOL}} warnings are about IPs with reverse lookups of the form *.ipt.aol.com in the AOL client IP address range: are these AOL proxies or not? They might, for example, be dynamically assigned client addresses. If so, we should _definitely not_ be trusting any XFF headers from these.
[On review: not all the figures sum to 100% so I may have dropped a couple in my counting, but I think the overall conclusions still hold up]
-- Neil
Neil Harris wrote:
First, I did a DNS reverse lookup on the entire AOL Proxy IP range, as defined in http://webmaster.info.aol.com/proxyinfo.html using dig in batch mode. Out of the 56542 addresses in those ranges, only 6740 give a valid reverse lookup.
I then went through the list of 116 pages flagged with the {{AOL}} template, and extracted their IP addresses, and compared them to the list of reverse lookups generated above.
[snip]
Out of all of the 116 pages flagged as {{AOL}}:
- six are bogus non-AOL addresses, and are probably attempts by vandals
to confuse anti-vandalism efforts
- 32 are from the AOL client range, with *.ipt.aol.com reverse lookups
- 76 are from the official AOL proxy range, with *.proxy.aol.com reverse
lookups; they all appear to be either of the form cache-XXX-XXXX.proxy.aol.com [74 of them], or spider-XXX-XXXXX.proxy.aol.com. [2 of them]
- four have no reverse lookup, but are in the official AOL proxy range
I think the info at http://webmaster.info.aol.com/proxyinfo.html can be trusted, certainly more than whatever templates random Wikipedians may have slapped onto user talk pages. So no, the 172.* ranges should not be treated as proxies, and the {{AOL}} templates on those pages should probably be removed or replaced with a more accurate template.
[On review: not all the figures sum to 100% so I may have dropped a couple in my counting, but I think the overall conclusions still hold up]
That's probably a _good_ sign. Now who was it who said that any set of figures that add up to exactly 100% is almost certainly doctored?...
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
I don't have the expertise to comment on the results, but I think that Brion fixed XFF headers during the MARMOT case.
Edward Z. Yang wrote:
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
I don't have the expertise to comment on the results, but I think that Brion fixed XFF headers during the MARMOT case.
If XFF is now operational on all AOL proxies, and WP is now configured to trust the XFF headers, we should no longer see any edits as coming from those proxies, since all edits should now be listed under the originating IP address.
Can anyone confirm whether this is the case?
-- Neil
wikitech-l@lists.wikimedia.org