Some odd TCP behavior (timeouts)

[Jens Schweikhardt]

Hello,

I'm having a TCP connectivity issue stumping me and driving me crazy.
The TLDR version: No more printing on my LaserJet (but ping works),
while printing and the web interfacw works fine under Windows.
Other TCP connections apparently flawless.
Wireshark shows FreeBSD performing the three way handshake but
then ignoring some packets sent from the printer (no ACKs)
and retransmitting until timeout hits.

The details with the wireshark logs are on
https://superuser.com/questions/1649435/mysterious-tcp-http-issue-accessing-laserjet-web-interface

I have seen a few commits to CURRENT's tcp and timeout code recently.
I went back to stable/13 as of two days ago, but still no joy.

Has anyone else noticed something going on with TCP?
Especially HTTP/HTTPS, like connections timing out in rare cases?
If you have a LaserJet (preferably 551dn), can you still access
its web interface, or get a reply with

printf "GET / HTTP/1.0\r\n\r\n" | nc LaserJet 80 # or 443.

Regards,

Jens

[Rodney W. Grimes]

Post by Jens Schweikhardt
Hello,
I'm having a TCP connectivity issue stumping me and driving me crazy.
The TLDR version: No more printing on my LaserJet (but ping works),
while printing and the web interfacw works fine under Windows.
Other TCP connections apparently flawless.
Wireshark shows FreeBSD performing the three way handshake but
then ignoring some packets sent from the printer (no ACKs)
and retransmitting until timeout hits.
The details with the wireshark logs are on
https://superuser.com/questions/1649435/mysterious-tcp-http-issue-accessing-laserjet-web-interface
I have seen a few commits to CURRENT's tcp and timeout code recently.
I went back to stable/13 as of two days ago, but still no joy.
Has anyone else noticed something going on with TCP?
Especially HTTP/HTTPS, like connections timing out in rare cases?
If you have a LaserJet (preferably 551dn), can you still access
its web interface, or get a reply with
printf "GET / HTTP/1.0\r\n\r\n" | nc LaserJet 80 # or 443.
Regards,
Jens

One thing that did change that comes to mind is the
fact FreeBSD now drops certain packets if timestamps
got negotiated. Locally I have that turned off as
an unconfirmed suspect of some connection problems:

sysctl net.inet.tcp.tolerate_missing_ts=1

You can try that, if it fixes the problem PLEASE let
us know!

Post by Jens Schweikhardt
Jens Schweikhardt http://www.schweikhardt.net/

[Jens Schweikhardt]

Post by Rodney W. Grimes
One thing that did change that comes to mind is the
fact FreeBSD now drops certain packets if timestamps
got negotiated. Locally I have that turned off as
sysctl net.inet.tcp.tolerate_missing_ts=1
You can try that, if it fixes the problem PLEASE let us know!

YES! I've got my sanity back! Web interface and printing are working again
with this sysctl (I verified it was 0 before).

Thanks a ton, Rodney!

Jens

[Michael Tuexen]

Post by Jens Schweikhardt
Hello,
I'm having a TCP connectivity issue stumping me and driving me crazy.
The TLDR version: No more printing on my LaserJet (but ping works),
while printing and the web interfacw works fine under Windows.
Other TCP connections apparently flawless.
Wireshark shows FreeBSD performing the three way handshake but
then ignoring some packets sent from the printer (no ACKs)
and retransmitting until timeout hits.
The details with the wireshark logs are on
https://superuser.com/questions/1649435/mysterious-tcp-http-issue-accessing-laserjet-web-interface

Hi Jens,

I only see screen shots, but not a link a the actual log files. Could
you share the a .pcap or .pcapng file? You can send them to me via e-mail,
if you prefer.

Best regards
Michael

Post by Jens Schweikhardt
I have seen a few commits to CURRENT's tcp and timeout code recently.
I went back to stable/13 as of two days ago, but still no joy.
Has anyone else noticed something going on with TCP?
Especially HTTP/HTTPS, like connections timing out in rare cases?
If you have a LaserJet (preferably 551dn), can you still access
its web interface, or get a reply with
printf "GET / HTTP/1.0\r\n\r\n" | nc LaserJet 80 # or 443.
Regards,
Jens
--
Jens Schweikhardt http://www.schweikhardt.net/
SIGSIG -- signature too long (core dumped)