Feature #4805: Add X509::Name#hash_old for 0.9.X compat

• Status: Closed
• Priority: Normal
• Assignee: nahi (Hiroshi Nakamura)
• Target version: 1.9.3

[nahi (Hiroshi Nakamura)]

X509::Name#hash with OpenSSL 1.0.0 returns different value than with OpenSSL 0.9.X.
Attached patch adds X509::Name#hash_old when you need MD5 based same X509_NAME_hash value as OpenSSL 0.9.X.

Martin, how do you think about adding it?

[MartinBosslet (Martin Bosslet)]

• Assignee changed from MartinBosslet (Martin Bosslet) to nahi (Hiroshi Nakamura)

Hi Hiroshi,

sorry for taking some time to answer, I was on vacation last week...
I think your patch is good! But there is one thing I don't like
about OpenSSL itself here - why do they hardcode the digest algorithm
in the first place?

There are situations where neither MD5 nor SHA-1 fits, OCSP requests
are a good example: The requested CertID is defined as

CertID ::= SEQUENCE {
hashAlgorithm AlgorithmIdentifier,
issuerNameHash OCTET STRING, -- Hash of Issuer's DN
issuerKeyHash OCTET STRING, -- Hash of Issuers public key
serialNumber CertificateSerialNumber }

This implies trouble for any SHA-2 family "hashAlgorithm".

In addition to applying your patch I'd favor a Name#hash implementation
that takes an optional OpenSSL::Digest that specifies the hash algorithm
to be used.
This would of course mean that we would have to implement the functionality
of X509_name_hash on our own. What do you think - would the benefit of a
cleaner solution outweigh the (partial) code duplication?

Regards,
Martin

[nahi (Hiroshi Nakamura)]

• Due date deleted (05/31/2011)

[nahi (Hiroshi Nakamura)]

Hi,

On Thu, Jun 9, 2011 at 06:14, Martin Bosslet
[e-mail] wrote:

I think your patch is good! But there is one thing I don't like
about OpenSSL itself here - why do they hardcode the digest algorithm
in the first place?

They're using the hash of name for c_rehash. You see files something
like hex encoded in certs dir of OpenSSL;

d2adc77d.0@
d537fba6.0@
d78a75c7.0@
d8274e24.0@
ddc328ff.0@

(e.g. /etc/ssl/certs/ in Ubuntu)

For that purpose, algorithm should be fixed so they don't get Digester
as a parameter for X509_NAME_hash and X509_NAME_hash_old I guess.

Besides this, I don't know the reason why they change base digester
from MD5 to SHA1 at the version bump from 0.9.8 to 1.0.0.

There are situations where neither MD5 nor SHA-1 fits, OCSP requests
are a good example: The requested CertID is defined as

CertID ::= SEQUENCE {
hashAlgorithm AlgorithmIdentifier,
issuerNameHash OCTET STRING, -- Hash of Issuer's DN
issuerKeyHash OCTET STRING, -- Hash of Issuers public key
serialNumber CertificateSerialNumber }

This implies trouble for any SHA-2 family "hashAlgorithm".

I can understand it but it should be different problem for them I
think. issuerNameHash has variable length, not fixed to 32bit integer.

In addition to applying your patch I'd favor a Name#hash implementation
that takes an optional OpenSSL::Digest that specifies the hash algorithm
to be used.
This would of course mean that we would have to implement the functionality
of X509_name_hash on our own. What do you think - would the benefit of a
cleaner solution outweigh the (partial) code duplication?

• Hide quoted text -
  I like 'X509::Name#hash' to be a wrapper of 'X509_NAME_hash' and
  'X509::Name#hash_old' is for 'X509::Name::hash_old'. I prefer to have
  another name for hashing X509::Name if it's needed.

Regards,
// NaHi

[nahi (Hiroshi Nakamura)]

• Assignee changed from nahi (Hiroshi Nakamura) to MartinBosslet (Martin Bosslet)

[MartinBosslet (Martin Bosslet)]

• Assignee changed from MartinBosslet (Martin Bosslet) to nahi (Hiroshi Nakamura)

Hiroshi NAKAMURA wrote:

They're using the hash of name for c_rehash. You see files something
like hex encoded in certs dir of OpenSSL;

d2adc77d.0@
d537fba6.0@
d78a75c7.0@
d8274e24.0@
ddc328ff.0@

(e.g. /etc/ssl/certs/ in Ubuntu)

For that purpose, algorithm should be fixed so they don't get Digester
as a parameter for X509_NAME_hash and X509_NAME_hash_old I guess.

OK, I see, thanks for the info! X509_NAME_hash(_old) has a special meaning
internally. I checked, there also exists X509_NAME_digest for the general
purpose usage I was thinking of. It's also what they use internally to
create the CertIDs in their OCSP implementation.

Besides this, I don't know the reason why they change base digester
from MD5 to SHA1 at the version bump from 0.9.8 to 1.0.0.

I could imagine that they changed for some sort of security reasons - a lot
of official recommendations/guidelines advise to refrain from using MD5, so
it could be a political reason...

I like 'X509::Name#hash' to be a wrapper of 'X509_NAME_hash' and
'X509::Name#hash_old' is for 'X509::Name::hash_old'. I prefer to have
another name for hashing X509::Name if it's needed.

Now that I have a better understanding of the context I completely agree.
If we feel the need for a general-purpose method, we could probably also use
X509::Name#digest in analogy to OpenSSL.

If you'd like me to apply the patch, please feel free to reassign to me!

Regards,
Martin

[nahi (Hiroshi Nakamura)]

• Status changed from Open to Closed

On Thu, Jun 23, 2011 at 20:54, Martin Bosslet [e-mail] wrote:

 For that purpose, algorithm should be fixed so they don't get Digester
 as a parameter for X509_NAME_hash and X509_NAME_hash_old I guess.

OK, I see, thanks for the info! X509_NAME_hash(_old) has a special meaning
internally. I checked, there also exists X509_NAME_digest for the general
purpose usage I was thinking of. It's also what they use internally to
create the CertIDs in their OCSP implementation.

Ah, X509::Name#digest sounds good. I didn't know OCSP impl already used. Since ext/openssl is a wrapper, I don't like to break class hierarchy, name, etc. of OpenSSL. Original API should be defined under Security::Crypto or something and it should be OpenSSL free... someday.

I like 'X509::Name#hash' to be a wrapper of 'X509_NAME_hash' and
'X509::Name#hash_old' is for 'X509::Name::hash_old'. I prefer to have
another name for hashing X509::Name if it's needed.

Now that I have a better understanding of the context I completely agree.
If we feel the need for a general-purpose method, we could probably also use
X509::Name#digest in analogy to OpenSSL.

Thanks. Applied it at r32213.