2.4.2 on AMD64 Linux: assumes 4-byte long

2.4.2 on AMD64 Linux: assumes 4-byte long

[Guy Rouillier]

I'm new here.  I hope this is the right place to post this question;
please point me to the correct place if I've gotten it wrong.  I've been
trying to get 2.4.2 working on Mandrake 10.0 for AMD64, using kernel
2.6.3.  I've downloaded this version, along with a corresponding 1.4.0
version of pptp, from the pptp-client site.  I'm getting a "Fatal signal
11", which I've traced to the SHA1_Final function on sha1.c during CHAP
Challenge phase of connection setup.

Looking in sha1.h and sha1.c source code, I see data structures and
functions with longs in them, which sets off alarms.  I think there was
an assumption when this code was written that a long is 4 bytes long. 
On an AMD64 64-bit OS, a long is 8 bytes.  What leads me to think this
is #define in sha1.c:

#define rol(value, bits) (((value) << (bits)) | ((value) >> (32 -
(bits))))

Clearly, on a 64-bit long, this won't produce a correct rotate-left.  
I've googled on sha1.c, and I've found several instances where the code
has switched to using u_int32 instead of long, including one by the same
author (Steve Reid) here:

http://cvsweb.stop1984.com/index.cgi/entropy/src/sha1.c?rev=1.12&content-type=text/x-cvsweb-markup

As a test, I tried simply changing all the longs to ints in sha1.h and
sha1.c.  Unfortunately, I went from getting segfaults to getting kernel
panics.  I'm suspecting this is not a valid thing to do, since the
corresponding files that the kernel was built from still have long
declarations.  I'm guessing these need to match, correct?  Or is pppd
standalone, and can I change sha1.c/h without worrying about the kernel
(which I compiled myself from Mandrake sources)?

I also tried simply changing the definition of rol above to use 64
instead of 32.  Predictably, that didn't work.

I'm not an expert in networking.  I'd appreciate any and all suggestions
anyone can provide to get this working.  Thanks.

-- 
Guy Rouillier

Re: 2.4.2 on AMD64 Linux: assumes 4-byte long - FIXED!

[Guy Rouillier]

Sorry for the top post, it's late and I'm tired and I'm replying to my
own message anyway.  I fixed this problem, and I'm now able to use
pptp/ppp/mppe to connect to a Microsoft server at work.   The fix
requires changes to the sha1 code both in ppp-2.4.2 and in the kernel. 
To whom should I send the fixes?  This is my first kernel fix so I don't
really know how to proceed.   Hmm, I guess half the fixes go to the ppp
maintainers and the other half go to the kernel people.

On Sun, 6 Jun 2004 02:19:36 -0400
Guy Rouillier <guy-rouillier@speakeasy.net> wrote:

> I'm new here.  I hope this is the right place to post this question;
> please point me to the correct place if I've gotten it wrong.  I've
> been trying to get 2.4.2 working on Mandrake 10.0 for AMD64, using
> kernel 2.6.3.  I've downloaded this version, along with a
> corresponding 1.4.0 version of pptp, from the pptp-client site.  I'm
> getting a "Fatal signal 11", which I've traced to the SHA1_Final
> function on sha1.c during CHAP Challenge phase of connection setup.
> 
> Looking in sha1.h and sha1.c source code, I see data structures and
> functions with longs in them, which sets off alarms.  I think there
> was an assumption when this code was written that a long is 4 bytes
> long. On an AMD64 64-bit OS, a long is 8 bytes.  What leads me to
> think this is #define in sha1.c:
> 
> #define rol(value, bits) (((value) << (bits)) | ((value) >> (32 -
> (bits))))
> 
> Clearly, on a 64-bit long, this won't produce a correct rotate-left.  
> I've googled on sha1.c, and I've found several instances where the
> code has switched to using u_int32 instead of long, including one by
> the same author (Steve Reid) here:
> 
> http://cvsweb.stop1984.com/index.cgi/entropy/src/sha1.c?rev=1.12&content-type=text/x-cvsweb-markup
> 
> As a test, I tried simply changing all the longs to ints in sha1.h and
> sha1.c.  Unfortunately, I went from getting segfaults to getting
> kernel panics.  I'm suspecting this is not a valid thing to do, since
> the corresponding files that the kernel was built from still have long
> declarations.  I'm guessing these need to match, correct?  Or is pppd
> standalone, and can I change sha1.c/h without worrying about the
> kernel(which I compiled myself from Mandrake sources)?
> 
> I also tried simply changing the definition of rol above to use 64
> instead of 32.  Predictably, that didn't work.
> 
> I'm not an expert in networking.  I'd appreciate any and all
> suggestions anyone can provide to get this working.  Thanks.
> 
> -- 
> Guy Rouillier
> -
> To unsubscribe from this list: send the line "unsubscribe linux-ppp"
> in the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html


-- 
Guy Rouillier


-- 
Guy Rouillier

Re: 2.4.2 on AMD64 Linux: assumes 4-byte long - FIXED!

[Guy Rouillier]

On Wed, 9 Jun 2004 16:24:08 +1000
James Cameron <james.cameron@hp.com> wrote:

> You're right.
> 
> The patches to the PPP maintainers, chuck 'em at the mailing list, and
> ask for a response.  If that fails, and it has to do with MPPE, send
> it to Frank Cusack <fcusack@fcusack.com>.
> 
> The kernel, if it has to do with the PPP modules, again, chuck 'em at
> the linux-ppp mailing list.  Paul Mackerras is ultimately responsible
> to promote them to the kernel team.  Or you could explain it very
> carefully and try the LKML (Linux Kernel Mailing List).

James, thank you for the detailed reply.  I mentioned this is my first
fix to the kernel, right?  So I won't be to embarrassed to admit that
I'm a little confused about the kernel part.  The fix I made definitely
worked.   But I took the time to download the stock kernel from
kernel.org, and none of the files I changed are in there!  I looked at
both the 2.6.3 and 2.6.6 versions.  The version I'm working with is
Mandrake 2.6.3-9.  The files I changed are drivers/net/sha1.c and
/drivers/net/sha1.h.  In the Mandrake kernel source, /drivers/net also
has some files related to mppe: ppp_mppe_compress.c and ppp_mppe.mod.c. 
The stock kernels from kernel.org don't have any of these four files!

Does that mean these files were added by Mandrake, and I should file a
bug report with them instead of ppp-linux?  I'm surprised to learn that
mppe is not supported by the stock kernel.  Appreciate your guiding a
newbie through this process.

> 
> http://www.tldp.org/HOWTO/Software-Release-Practice-HOWTO/patching.html
> has some good hints to help your patch be accepted.
> 
> I'm just an interested bystander.  ;-)
> 
> -- 
> James Cameron                         http://quozl.netrek.org/
> HP Open Source, Volunteer             http://opensource.hp.com/
> PPTP Client Project, Release Engineer
> http://pptpclient.sourceforge.net/


-- 
Guy Rouillier

Re: 2.4.2 on AMD64 Linux: assumes 4-byte long - FIXED!

[James Cameron]

Guy,

Yes, Mandrake's kernel source includes MPPE support, and the kernel.org
sources do not.  Other distributions also add MPPE support in this way;
Debian and SUSE included.  Red Hat do not.

I know Debian get their kernel-patch-mppe package from PPP CVS.  I
suspect Mandrake and SUSE did as well.  You should check that the
sources match what is in PPP CVS.  To get PPP CVS onto your system,

	cd /usr/src  # (or some other reasonable directory)
	cvs -d :pserver:cvs@pserver.samba.org:/cvsroot co ppp/linux/mppe

If you've made changes to drivers/net/sha1.{c,h} and Mandrake got these
from Frank Cusack's work that is in PPP CVS (directory linux/mppe), then
you could post them here or to Frank.  

Matt Domsch, Jan Dubiec, and some others have also been working on the
MPPE patch; they have posted to the PPTP mailing lists.

If you logged a bug report with Mandrake, one would hope that they would
elevate this report back to Frank.  ;-)  There is a risk that they will
adjust their patch instead.

-- 
James Cameron                         http://quozl.netrek.org/
HP Open Source, Volunteer             http://opensource.hp.com/
PPTP Client Project, Release Engineer http://pptpclient.sourceforge.net/

Re: 2.4.2 on AMD64 Linux: assumes 4-byte long - FIXED!

[Guy Rouillier]

On Thu, 10 Jun 2004 14:43:46 +1000
James Cameron <james.cameron@hp.com> wrote:

> Guy,
> 
> Yes, Mandrake's kernel source includes MPPE support, and the
> kernel.org sources do not.  Other distributions also add MPPE support
> in this way; Debian and SUSE included.  Red Hat do not.
> 
> I know Debian get their kernel-patch-mppe package from PPP CVS.  I
> suspect Mandrake and SUSE did as well.  You should check that the
> sources match what is in PPP CVS.  To get PPP CVS onto your system,
> 
> 	cd /usr/src  # (or some other reasonable directory)
> 	cvs -d :pserver:cvs@pserver.samba.org:/cvsroot co ppp/linux/mppe
> 
> If you've made changes to drivers/net/sha1.{c,h} and Mandrake got
> these from Frank Cusack's work that is in PPP CVS (directory
> linux/mppe), then you could post them here or to Frank.  

Thanks.  I checked out ppp, and the Mandrake 2.6.3-9 kernel did indeed
get sha1.c/h from linux/mppe.  I will post two new messages here, one
for the fix to ppp/pppd and the other for the kernel fix in linux/mppe. 
 I appreciate your help.

> 
> Matt Domsch, Jan Dubiec, and some others have also been working on the
> MPPE patch; they have posted to the PPTP mailing lists.
> 
> If you logged a bug report with Mandrake, one would hope that they
> would elevate this report back to Frank.  ;-)  There is a risk that
> they will adjust their patch instead.
> 
> -- 
> James Cameron                         http://quozl.netrek.org/
> HP Open Source, Volunteer             http://opensource.hp.com/
> PPTP Client Project, Release Engineer
> http://pptpclient.sourceforge.net/
> -
> To unsubscribe from this list: send the line "unsubscribe linux-ppp"
> in the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html


-- 
Guy Rouillier

Re: 2.4.2 on AMD64 Linux: assumes 4-byte long - FIXED!

[mole]

Guy Rouillier wrote:

>Sorry for the top post, it's late and I'm tired and I'm replying to my
>own message anyway.  I fixed this problem, and I'm now able to use
>pptp/ppp/mppe to connect to a Microsoft server at work.   
>

me too :)

>The fix
>requires changes to the sha1 code both in ppp-2.4.2 and in the kernel. 
>  
>

Yes, and I have provided the patch to the list for both changes.

Have you tried my patch (as you wrote were going to do) and found it 
doesn't work?

>To whom should I send the fixes?  
>

Please CC it to me. I would like to know what else but two-liner you 
think is needed.

>This is my first kernel fix so I don't
>really know how to proceed.   Hmm, I guess half the fixes go to the ppp
>maintainers and the other half go to the kernel people.
>
>On Sun, 6 Jun 2004 02:19:36 -0400
>Guy Rouillier <guy-rouillier@speakeasy.net> wrote:
>
>  
>
>>I'm new here.  I hope this is the right place to post this question;
>>please point me to the correct place if I've gotten it wrong.  I've
>>been trying to get 2.4.2 working on Mandrake 10.0 for AMD64, using
>>kernel 2.6.3.  I've downloaded this version, along with a
>>corresponding 1.4.0 version of pptp, from the pptp-client site.  I'm
>>getting a "Fatal signal 11", which I've traced to the SHA1_Final
>>function on sha1.c during CHAP Challenge phase of connection setup.
>>
>>Looking in sha1.h and sha1.c source code, I see data structures and
>>functions with longs in them, which sets off alarms.  I think there
>>was an assumption when this code was written that a long is 4 bytes
>>long. On an AMD64 64-bit OS, a long is 8 bytes.  What leads me to
>>think this is #define in sha1.c:
>>
>>#define rol(value, bits) (((value) << (bits)) | ((value) >> (32 -
>>(bits))))
>>
>>Clearly, on a 64-bit long, this won't produce a correct rotate-left.  
>>I've googled on sha1.c, and I've found several instances where the
>>code has switched to using u_int32 instead of long, including one by
>>the same author (Steve Reid) here:
>>
>>http://cvsweb.stop1984.com/index.cgi/entropy/src/sha1.c?rev=1.12&content-type=text/x-cvsweb-markup
>>
>>As a test, I tried simply changing all the longs to ints in sha1.h and
>>sha1.c.  Unfortunately, I went from getting segfaults to getting
>>kernel panics.  I'm suspecting this is not a valid thing to do, since
>>the corresponding files that the kernel was built from still have long
>>declarations.  I'm guessing these need to match, correct?  Or is pppd
>>standalone, and can I change sha1.c/h without worrying about the
>>kernel(which I compiled myself from Mandrake sources)?
>>
>>I also tried simply changing the definition of rol above to use 64
>>instead of 32.  Predictably, that didn't work.
>>
>>I'm not an expert in networking.  I'd appreciate any and all
>>suggestions anyone can provide to get this working.  Thanks.
>>
>>-- 
>>Guy Rouillier
>>-
>>To unsubscribe from this list: send the line "unsubscribe linux-ppp"
>>in the body of a message to majordomo@vger.kernel.org
>>More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>    
>>
>
>
>  
>