1 |
2007/4/20, Nathan Smith <ndansmith@×××××.com>: |
2 |
> |
3 |
> The symptoms (broken TCP traffic to some sites) |
4 |
> depend entirely upon the sites you are connecting to and how your |
5 |
> connections are routed. |
6 |
|
7 |
|
8 |
We saw that affected users are sending TCP data with bad checksum |
9 |
(this could apply to udp and/or icmp too? anyone to check?). |
10 |
|
11 |
Then, some router on the Internet can decide to drop such bad packet. I |
12 |
noticed Linux doesn't drop but Checkpoint does. |
13 |
|
14 |
Another point, note that the bad checksum is always good_checksum+0100 |
15 |
""" |
16 |
[bad tcp cksum 3de5 (->3ce5)!] |
17 |
[bad tcp cksum 322d (->312d)!] |
18 |
[bad tcp cksum 25ad (->24ad)!] |
19 |
""" |
20 |
|
21 |
I guess this is the checksum algorithm that is incompatible with -O2. Even |
22 |
if I won't check the source code :p |
23 |
|
24 |
However, using tcpdump I can tell that some machines apparently do not |
25 |
> experience this problem, so it may be processor-specific (which makes |
26 |
> sense given that it is an optimization problem). I use a Pentium M. |
27 |
> What are other people using that have been affected by the problem? |
28 |
|
29 |
|
30 |
Mine is pentium-4 |
31 |
|
32 |
-- |
33 |
Camille Huot |