On Tue, 2008-08-26 at 08:39 +1000, Andrew Bartlett wrote:
On Mon, 2008-08-25 at 20:04 +0200, Hans de Goede wrote:
> Hi All,
>
> The story begins here:
>
https://bugzilla.redhat.com/show_bug.cgi?id=446860
>
> The problem is, or I believe it to be, that when an application uses
> libgnutls-openssl (meant for easy use of porting openssl programs to gnutls)
> for example because of the openssl license issues, and then a library uses the
> real openssl (for example glibc through nss_ldap) then in the nss_ldap example,
> the layer dlopen's nss_ldap starts using the openssl symbols from gnutls
> instead of those from openssl, but they are not ABI compatible -> boom.
>
> Luckily the list of libgnutls-openssl users seems small:
>
> [hans@localhost src]$ repoquery -q --whatrequires
> 'libgnutls-openssl.so.26()(64bit)'
> mcabber-0:0.9.7-1.fc10.x86_64
> gnutls-0:2.4.1-1.fc10.x86_64
> gnutls-devel-0:2.4.1-1.fc10.x86_64
> zoneminder-0:1.23.3-1.fc10.x86_64
> gkrellm-0:2.3.1-4.fc10.x86_64
>
> So only 3 programs are affected, given that the same may happen when any used
> library uses the real openssl and the application or any other library uses
> gnutls-openssl, I would like to suggest the removal of libgnutls-openssl from
> Fedora, as long as we have this openssl libraries mess (which we unfortunately
> do) we should make sure that the various ssl libraries do not have symbol
> clashes, as changes are that through a mix of libraries an application may be
> using 2 (or even 3) different ssl libs.
>
> So whats your 2 cents on this?
How does the NSS (Mozilla SSL) OpenSSL wrapper avoid this problem?
Completely uninformed guess: by using macros or other techniques to add
a prefix to the symbols that end up in the binary?
Nils
--
Nils Philippsen "Those who would give up Essential Liberty to purchase
Red Hat a little Temporary Safety, deserve neither Liberty
nils(a)redhat.com nor Safety." -- Benjamin Franklin, 1759
PGP fingerprint: C4A8 9474 5C4C ADE3 2B8F 656D 47D8 9B65 6951 3011