the fast IP checksum update in ip_decrease_ttl appears
to be broken (at least on big endian machines) since 2.2.18.
Even on little endian machines IMO the overflow is incorrect
in two cases:
0xfeff goes to 0x0000 instead of 0xffff
0xffff goes to 0x0000 instead of 0x0100
On big endian machines, the overflow from the high byte
is never carried over correctly:
0xfeff goes to 0x0000 instead of 0xffff
0xff00 goes to 0x0000 instead of 0x0001
0xff01 goes to 0x0001 instead of 0x0002
...
0xffff goes to 0x00ff instead of 0x0100
The following patch reverts the ip_decrease_ttl routine
to the pre-2.2.18 level, which might be less efficient,
but should at least be correct ...
diff -urN linux-2.2.19/include/net/ip.h linux-2.2.19-s390/include/net/ip.h
--- linux-2.2.19/include/net/ip.h Sun Mar 25 18:37:40 2001
+++ linux-2.2.19-s390/include/net/ip.h Wed May 16 14:51:03 2001
@@ -171,8 +171,10 @@
int ip_decrease_ttl(struct iphdr *iph)
{
u16 check = iph->check;
- check += __constant_htons(0x0100);
- iph->check = check + ((check>=0xFFFF) ? 1 : 0);
+ check = ntohs(check) + 0x0100;
+ if ((check & 0xFF00) == 0)
+ check++; /* carry overflow */
+ iph->check = htons(check);
return --iph->ttl;
}
Mit freundlichen Gruessen / Best Regards
Ulrich Weigand
-- Dr. Ulrich Weigand Linux for S/390 Design & Development IBM Deutschland Entwicklung GmbH, Schoenaicher Str. 220, 71032 Boeblingen Phone: +49-7031/16-3727 --- Email: Ulrich.Weigand@de.ibm.com
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/