Re: Allow matching whole DN from a client certificate

From: Andrew Dunstan <andrew(at)dunslane(dot)net>
To: Jacob Champion <pchampion(at)vmware(dot)com>, "daniel(at)yesql(dot)se" <daniel(at)yesql(dot)se>
Cc: "pgsql-hackers(at)postgresql(dot)org" <pgsql-hackers(at)postgresql(dot)org>
Subject: Re: Allow matching whole DN from a client certificate
Date: 2021-01-27 20:42:10
Views: Raw Message | Whole Thread | Download mbox | Resend email
Lists: pgsql-hackers

On 1/27/21 3:14 PM, Jacob Champion wrote:
> On Tue, 2021-01-26 at 18:43 +0000, Jacob Champion wrote:
>> On Tue, 2021-01-26 at 13:49 +0100, Daniel Gustafsson wrote:
>>> The OpenSSL X509_NAME_cmp function use RFC 5280 section 7.1 and RFC 4517
>>> section 4.2.15 (which in turn reference RFC 4514 for the DN string format).
>>> libnss has CERT_AsciiToName which is referencing RFCs 1485, 1779 and 2253 in
>>> lib/certdb/alg1485.c. Comparing the two would be interesting.
>> Yeah. I'll poke around a bit.
> Here's some output from a test program I threw together, which parses
> identical DER using OpenSSL and NSS and writes the corresponding string
> representation.
>> input/basic.conf:
>> ssl: CN=pchampion,OU=VMware
>> nss: CN=pchampion,OU=VMware
>> input/escape.conf:
>> ssl: CN=\,\+\\\;\"\<\>
>> nss: CN=",+\\;\"<>"
>> input/multivalue.conf:
>> ssl: CN=pchampion+SN=Champion+GN=Jacob,OU=VMware
>> nss: givenName=Jacob+SN=Champion+CN=pchampion,OU=VMware
>> input/unicode.conf:
>> ssl: CN=οδυσσέας,OU=VMware
>> nss: CN=οδυσσέας,OU=VMware
>> input/unprintable.conf:
>> ssl: CN=\01\,\02\,\03,OU=\01\02\03
>> nss: CN="\01,\02,\03",OU=\01\02\03
> basic.conf is exactly what it looks like: CN=pchampion,OU=VMware. Both
> implementations agree.
> escape.conf contains a CN with the literal value
> ,+\;"<>
> and you can see that NSS doesn't follow RFC 4514 here; it uses the
> older double-quoting form of escaping. There's a 15-year-old bug on
> this in NSS [1].
> multivalue.conf contains a multivalued AVA with commonName "pchampion",
> givenName "Jacob", and surname "Champion". They aren't sorted in the
> same order, and the implementations even disagree on how to represent
> the givenName attribute. (I'm not convinced that either choice is RFC-
> 4514-compliant; it doesn't look like GN is registered with IANA as a
> short name for givenName.)
> unicode.conf contains a commonName of "οδυσσέας". Both implementations
> agree, but the only way I was able to get OpenSSL to produce this
> (rather than a string of escaped hex) was by using the flags
> in the call to X509_NAME_print_ex(). This should work fine for a
> database encoding of UTF-8, but would we need to convert for other
> encodings? Also, I'm not sure how this would handle certificates that
> aren't UTF-8 encoded. It looks like some UCS variants are legal?
> unprintable.conf contains the bytes 0x01, 0x02, and 0x03 in the
> commonName and organizationalUnit. They're backslash-escaped by both
> implementations, but if you add any other printable escaped characters
> (such as comma, in the CN example here) NSS will still double-quote the
> whole thing.
> --Jacob
> [1]

OK, that bug is a bit ugly.

I'm not sure where we want to go with the present patch. Do we want to
go with what we have and document the limitations more, or try to do
something more elaborate? If so, exactly what?



Andrew Dunstan

In response to


Browse pgsql-hackers by date

  From Date Subject
Next Message Alexey Kondratov 2021-01-27 21:19:06 Re: Allow CLUSTER, VACUUM FULL and REINDEX to change tablespace on the fly
Previous Message Joel Jacobson 2021-01-27 20:37:51 Re: [HACKERS] GSoC 2017: Foreign Key Arrays