How many keys are used for 3DES in PGP?
Tom St Denis
"Mxsmanic" <mxsm...@hotmail.com> wrote in message
news:vipstv0luv1tvcp9i...@4ax.com...
> Does PGP do 3DES with two keys or three keys?
>
> And what should one normally assume in discussions of 3DES if the number
> of keys is not given? What's the most common implementation?
Checking GPG it has both 2key and 3key routines. However, following
"des_get_info" to do_tripledes_setkey it uses the 3key variant.
As an aside... let's do something in /gnupg-1.2.3/cipher/
$ grep "burn_stack (int bytes)" *.c
blowfish.c:burn_stack (int bytes)
cast5.c:burn_stack (int bytes)
des.c:burn_stack (int bytes)
md5.c:burn_stack (int bytes)
random.c:burn_stack (int bytes)
rijndael.c:burn_stack (int bytes)
rmd160.c:burn_stack (int bytes)
sha1.c:burn_stack (int bytes)
sha256.c:burn_stack (int bytes)
sha512.c:burn_stack (int bytes)
tiger.c:burn_stack (int bytes)
twofish.c:burn_stack (int bytes)
That's what I call efficient coding!!! they rewrote burn_stack not once,
not twice, but 12! [count em!] 12 times!!!
That's just F'ing stupid.
> And in the case of implementations using only two keys, why don't they
> use three keys instead?
The lib provides both but the default crypto descriptor [see
/gnupg-1.2.3/cipher/cipher.c] uses 3key.
Tom
Mxsmanic
> That's what I call efficient coding!!! they rewrote burn_stack not once,
> not twice, but 12! [count em!] 12 times!!!
>
> That's just F'ing stupid.
Although it does prevent a compromise in one routine from affecting the
others (at least potentially).
--
Transpose hotmail and mxsmanic in my e-mail address to reach me directly.
Tom St Denis
> Tom St Denis writes:
>
> > That's what I call efficient coding!!! they rewrote burn_stack not
once,
> > not twice, but 12! [count em!] 12 times!!!
> >
> > That's just F'ing stupid.
>
> Although it does prevent a compromise in one routine from affecting the
> others (at least potentially)
Not really. In fact it does quite the opposite. By raising the # of loc
you raise the probability of an error.
I don't know why they did that but I'd think a single burn_stack function
would be ideal. At least they didn't pull an OpenSSL and write a
cbc/cfb/ofb/... for each cipher... ;-)
Tom