1 |
Alex Unleashed wrote: |
2 |
> On 3/19/07, *Nathan Smith* <ndansmith@×××××.com |
3 |
> <mailto:ndansmith@×××××.com>> wrote: |
4 |
> |
5 |
> |
6 |
> So anyhow, I am making due, but it is a frustrating problem to say the |
7 |
> least. Any suggestions are greatly appreciated. :-) |
8 |
> |
9 |
> |
10 |
> I would start capturing a reasonable amount of network traffic showing |
11 |
> normal and error conditions, if at all possible on both ends to confirm |
12 |
> there's nothing wrong with some machine within the path. FWIW, I recall |
13 |
> hearing people having problems with FreeBSD's SACK in the past, though |
14 |
> that should be fixed by now. Anyway, you have good hints on how to |
15 |
> approach a closer analysis in this mail thread: |
16 |
> http://lists.freebsd.org/pipermail/freebsd-net/2006-December/012830.html |
17 |
> |
18 |
> Alex |
19 |
|
20 |
I've attached the highlights of what I've been able to tell with tcpdump |
21 |
as of yet. These packets were captured by a Gentoo Linux machine |
22 |
running apache (all of the requests are HTTP). The transmitting |
23 |
computer in each case is the same machine (IBM Thinkpad Z60t - Pentium |
24 |
M), but either the OS, or some feature of Gentoo/FreeBSD is altered for |
25 |
each sample. These two machines were at the same location with no |
26 |
firewall and few routers separating them. |
27 |
|
28 |
There is a little annotation in the file, but in essence the Linux box |
29 |
replied to packets which were sent from non-Gentoo/FreeBSD systems |
30 |
(Linux and FreeSBIE), and did not reply to the Gentoo/FreeBSD system. I |
31 |
varied the tcp window size, and I also tried turning rfc1323 on and off. |
32 |
I've also altered the mtu and max-mss with no success. The common |
33 |
denominator in the packets which failed to get a response is that they |
34 |
have bad checksums. If you compare the first packet listed (G/FBSD) and |
35 |
the packet from the FreeSBIE liveCD, you will see that they are |
36 |
identical in every pertinent way except that the packets coming from the |
37 |
G/FBSD machine were mangled. |
38 |
|
39 |
Apparently these packets are objectionable to some routers and not |
40 |
others. Depending on where I am, I can connect to some sites and not |
41 |
others (and never gentoo-sunrise.org :-( ). I set up a packet capture |
42 |
on my home machine and send HTTP requests from the G/FBSD box at work, |
43 |
and the packets never arrived. |
44 |
|
45 |
Now it is just a matter of finding out _why_ they packets are getting |
46 |
messed up. It is apparently not a TCP window scaling problem (toggling |
47 |
that has no effect), and it is not a FreeBSD proble, (FreeSBIE on the |
48 |
same hardware works great), it is apparently a breakage in Gentoo/FBSD. |
49 |
|
50 |
-- |
51 |
Nathan Smith |
52 |
ndansmith@×××××.com |