From: David Miller Date: Fri, 28 Sep 2007 21:20:15 +0000 Subject: Re: PATCH: tcp rfc 2385 security/bugfix for sparc64 Message-Id: <20070928.142015.82358847.davem@davemloft.net> List-Id: References: <23966.84.62.25.8.1191012145.squirrel@cs1.dlh.net> In-Reply-To: <23966.84.62.25.8.1191012145.squirrel@cs1.dlh.net> MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: pl@dlh.net Cc: sparclinux@vger.kernel.org, alan@lxorguk.ukuu.org.uk, torvalds@linux-foundation.org, linux-kernel@vger.kernel.org From: "Peter Lieven" Date: Fri, 28 Sep 2007 22:42:25 +0200 (CEST) > TCP MD5 signatures on sparc64 (big-endian) completely fail on current > kernel releases in interoperability with Cisco/Foundry or other > little-endian linux systems. > > The root cause is a cast in the return statement of tcp_v4_md5_do_lookup, > where a tcp4_md5sig_key is casted onto tcp_md5sig_key without proper > conversion. On little-endian systems the upper 8 bits are cut of which > yields the expected behaviour. However, on big-endian systems (like > sparc64) only the most significant 8 bits are preserved. Since > TCP_MD5SIG_MAXKEYLEN is 80, this always yields 0. > > In the calculation of the md5 signature afterwards the key is therefore > not appended to the tcp segment which could result in a security problem > since only the presence of a md5 signature is checked, and the key itself > doesn't matter. Thanks for finding this bug. > --- linux.old/include/net/tcp.h 2007-09-28 21:43:26.000000000 +0200 +++ > linux/include/net/tcp.h 2007-09-28 21:45:35.000000000 +0200 @@ -1055,6 > +1055,7 @@ static inline void clear_all_retrans_hin I'll have to apply this patch by hand because your email client completely corrupted the patch. > Signed-off-by: Peter Lieven > Signed-off-by: Matthias M. Dellweg <2500@gmx.de> Thanks again.