Subject: Re: userauth_publickey_fromfile() fails occasionally for no reason

Re: userauth_publickey_fromfile() fails occasionally for no reason

From: Simon Josefsson <simon_at_josefsson.org>
Date: Thu, 17 Dec 2009 08:34:03 +0100

Peter Stuge <peter_at_stuge.se> writes:

> Simon Josefsson wrote:
>> > ping
>>
>> I pushed your patch, please test whether what is in git now works
>> for both RSA and DSA.
>
> So I was looking into this. The reason I didn't immediately apply the
> patch to bring back the previous behavior of the libgcrypt code is
> that it makes some assumptions on how r and s are stored internally
> by libgcrypt and that doesn't seem as reliable as can be.

Do we have any reports that the old code didn't work? If we don't, and
assuming the new code was broken, I don't see any reason not to revert
back to the old code meanwhile.

It would be good to improve the code to not make those assumptions, but
I'm not sure what you are referring to here. What I think the code is
assuming is that we know exactly how the DSA S-exp outputs looks like,
which may not be ideal, but S-exp's is a fairly well-known format [1] so
the assumption doesn't seem all that strange to me. But patches
welcome. :-)

/Simon

[1] http://people.csail.mit.edu/rivest/Sexp.txt
_______________________________________________
libssh2-devel http://cool.haxx.se/cgi-bin/mailman/listinfo/libssh2-devel
Received on 2009-12-17