openssh bugs - Jun 2004 - [Bug 884] DSA keys (id_dsa.pub) with 8192 bits or more aren't correctly recognized

http://bugzilla.mindrot.org/show_bug.cgi?id=884

dmr at gmx.it changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|DSA keys (id_dsa.pub) with  |DSA keys (id_dsa.pub) with
                   |8192 bytes or more aren't   |8192 bits or more
aren't
                   |correctly recognized        |correctly recognized





------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

http://bugzilla.mindrot.org/show_bug.cgi?id=884





------- Additional Comments From dmr at gmx.it  2004-06-24 10:01 -------
Oops, I've written: 'this is what I get using ssh with a 8192 bytes DSA
key...'
but I wanted to say 'this is what I get using ssh with a 8192 bits DSA
key...'.
Sorry.

Unfortunately it seems that the key size has been hard-wired in some other
place, because every attempt to raise the keysize over 8192 bits limit won't
work at all.




------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

http://bugzilla.mindrot.org/show_bug.cgi?id=884

dtucker at zip.com.au changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED



------- Additional Comments From dtucker at zip.com.au  2004-06-24 15:27
-------
> every attempt to raise the keysize over 8192 bits limit won't work
Maybe a limit in OpenSSL's DSA functions?

Your read_whole_line() function seems a lot more complicated than it needs to
be.  xrealloc will leave the file content intact, so instead of ftell/fseek/goto
you can just initialise everthing to zero/NULL then just repeatedly read chunks
until you get a newline, eg,

while (1) {
        buf = xrealloc(buf, size + CHUNKSZ);
        size += CHUNKSZ;
        nextpart = buf + len;
        if (fgets(nextpart, CHUNKSZ, f) == NULL)
                break;
        len += strlen(nextpart);
        if (buf[len - 1] == '\n')
                break;
}

(For bonus points, add some better handling of fgets() == NULL.  Or use
buffer_append instead of xrealloc.)

Anyway, an 8kbit key is just over 4KBytes encoded and I don't think it's
worth
the extra complexity of making it dynamic just to save a couple of KB of stack
space.

Either way key_try_load_public shouldn't pass incomplete lines to key_read,
though.



------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

http://bugzilla.mindrot.org/show_bug.cgi?id=884





------- Additional Comments From dtucker at zip.com.au  2004-06-24 15:30 -------
Created an attachment (id=659)
 --> (http://bugzilla.mindrot.org/attachment.cgi?id=659&action=view)
discard excessively long lines in key_try_load_public




------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

http://bugzilla.mindrot.org/show_bug.cgi?id=884

dtucker at zip.com.au changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
 Attachment #659 is|0                           |1
           obsolete|                            |



------- Additional Comments From dtucker at zip.com.au  2004-06-24 17:16 -------
Created an attachment (id=660)
 --> (http://bugzilla.mindrot.org/attachment.cgi?id=660&action=view)
Allow large keys to work.

Try this patch, tested OK with up to 16 kbit keys.



------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

http://bugzilla.mindrot.org/show_bug.cgi?id=884





------- Additional Comments From dmr at gmx.it  2004-06-24 23:36
-------
> > every attempt to raise the keysize over 8192 bits limit won't work
> 
> Maybe a limit in OpenSSL's DSA functions?
> 
> Your read_whole_line() function seems a lot more complicated than it needs
to
> be.  xrealloc will leave the file content intact, so instead of
ftell/fseek/goto
> you can just initialise everthing to zero/NULL then just repeatedly read
chunks
> until you get a newline, eg,
> 
> while (1) {
>         buf = xrealloc(buf, size + CHUNKSZ);
>         size += CHUNKSZ;
>         nextpart = buf + len;
>         if (fgets(nextpart, CHUNKSZ, f) == NULL)
>                 break;
>         len += strlen(nextpart);
>         if (buf[len - 1] == '\n')
>                 break;
> }
> 
> (For bonus points, add some better handling of fgets() == NULL.  Or use
> buffer_append instead of xrealloc.)
You are right... but that patch was only a quick and dirty trick... that
function has been stripped down from another place only for testing purposes.
However now it shoud be ok.
 
> Anyway, an 8kbit key is just over 4KBytes encoded and I don't think
it's
> worth the extra complexity of making it dynamic just to save a couple of KB
> of stack space.
IMHO the problem here is that the maximum key size is hard coded in openssh,
while it should be _at most_ hard coded in (e.g.) openssl...
 
> Either way key_try_load_public shouldn't pass incomplete lines to
key_read,
  though.

I agree... However, the following patch should fix - it worked pretty fine for
me - any key-size troubles: tested ok with 32K keys, actually in production on
my cluster and several other systems.

Thanks a lot for your support.

Regards,
Alessandro



------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

http://bugzilla.mindrot.org/show_bug.cgi?id=884

dmr at gmx.it changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
 Attachment #658 is|0                           |1
           obsolete|                            |



------- Additional Comments From dmr at gmx.it  2004-06-24 23:40 -------
Created an attachment (id=661)
 --> (http://bugzilla.mindrot.org/attachment.cgi?id=661&action=view)
lets openssh to accept and use very large keys (>8Kbits)




------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

http://bugzilla.mindrot.org/show_bug.cgi?id=884

dtucker at zip.com.au changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
 Attachment #660 is|0                           |1
           obsolete|                            |



------- Additional Comments From dtucker at zip.com.au  2004-06-25 17:40 -------
Created an attachment (id=663)
 --> (http://bugzilla.mindrot.org/attachment.cgi?id=663&action=view)
Allow arbitary sized rsa1, rsa & dsa keys

Current WIP patch for dynamically allocating memory for keys.
Tests OK for 4kbit -> 16kbit keys (I have a regression test for this which
I'll
attach separately).



------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

http://bugzilla.mindrot.org/show_bug.cgi?id=884

dtucker at zip.com.au changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
OtherBugsDependingO|                            |822
              nThis|                            |
           Keywords|                            |openbsd, patch





------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.