Error_Packet_Parsing with Netscape Comment in Certificate

[nikomi]

Hi,

while testing 1.3.11 we found the following issue when connecting to a server providing a certificate containing a Netscape Comment extension:

TlsExceptionHostPort (HandshakeFailed (Error_Packet_Parsing
  Failed reading: error certificate parsing 0:signed object error:
    "fromASN1: X509.ExtensionRaw: OID=[2,16,840,1,113730,1,13]: cannot decode data: ParsingPartial"
    From: handshake[HandshakeType_Certificate]

See http://www.alvestrand.no/objectid/2.16.840.1.113730.1.13.html

This is waht OpenSSL says the certificate contains:

        X509v3 extensions:
            Netscape Comment: 
                Puppet Ruby/OpenSSL Internal Certificate

Do you think the problem could be in TLS or rather in the ASN.1 libs?

[vincenthz]

it's definitely a low level problem either in x509 or in asn.1 (not tls fault I would assume). Any chance this certificate is public (or worse case can be send to me semi-privately) ?

[nikomi]

Sorry, it's a company cert :-(

[vincenthz]

x509 certs don't contain private information, there's no non-public cryptographic material (AFAIK). In any case, if I can't see the cert or something exhibiting the exact same problem, I won't be able to help/fix the issue. As a tip, you can use the x509-util haskell package to drive your debugging

[nikomi]

Our admins created a certificate showing the same problem which I am allowed to give away.

How could I get this to you?

[vincenthz]

send it to my email (listed on github)

[nikomi]

You have mail :-)

Thanx for looking into this!

[ocheron]

What is the status on this issue?

I couldn't reproduce with a certificate I generated.
But I see that this kind of failure was already reported in haskell-tls/hs-certificate#66:
there we have 23 ParsingPartial exceptions for extension "Subject Alternative Name".

Looking at those 23 certificates, the OctetString containing the ASN.1 of the extension value is padded with an extra 0x00 byte. openssl x509 accepts the certificates but not gnutls-certtool -i.
So maybe it's the same for the Netscape Comment here.

[vincenthz]

@nikomi could you re-try using x509-1.7.1 ? It should be fixed now I think, the netscape comment extension violate all the other extension format, which is a bytestring that contains ASN1, but in the case of netscape comment it just contains a ascii string.

[vdukhovni]

I also have run into an example certificate that exhibits the problem. The Netscape comment is encoded directly as the octet string of the extension, from openssl asn1parse I get:

  688:d=5  hl=2 l=   9 prim: OBJECT            :Netscape Comment
  699:d=5  hl=2 l=  40 prim: OCTET STRING      :Puppet Ruby/OpenSSL Internal Certificate

while when I ask OpenSSL to generate a "Netscape Comment" extension with payload "This is a comment", I get:

  509:d=5  hl=2 l=   9 prim: OBJECT            :Netscape Comment
  520:d=5  hl=2 l=  19 prim: OCTET STRING      [HEX DUMP]:161154686973206973206120636F6D6D656E74

Which has tag(22) and length (17). So it seems that the standard encoding of Netscape Comments is as an IA5String, but openssl also accepts a variant form with no tag/length:

Attaching variant.pem (observed on live SMTP server) and normal.pem generated via OpenSSL conf file with nsComment = "This is a comment"

normal.pem.txt
variant.pem.txt

The parser reports the same value when asked for "text":

$ openssl x509 -text -in /tmp/normal.pem.txt | grep -A1 "Netscape Comment"
            Netscape Comment: 
                This is a comment
$ openssl x509 -text -in /tmp/variant.pem.txt | grep -A1 "Netscape Comment"
            Netscape Comment: 
                Puppet Ruby/OpenSSL Internal Certificate

This variant of an extension we can safely ignore should not I think prevent processing of the certificate as a whole.

[nikomi]

Sorry, I've been on holiday... I'll try to test the new lib as soon as I get hold of a colleague I need to do this.

thanx for your support!

[vdukhovni]

I can confirm that the site with the variant Netscape comment in its certificate works with x509-1.7.1. Will let you know if anything else breaks...

[vincenthz]

let's close for now, and if this is not sorted you're both welcome to create a new issue.

[nikomi]

Sorry, I have not managed to grab the colleague yet... holiday season takes it's toll...

I whole-heartedly agree to close this for now, and thanks again for your support!