openssh bugs - Jan 2007 - [Bug 1263] connection sharing often freezes

http://bugzilla.mindrot.org/show_bug.cgi?id=1263





------- Comment #11 from djm at mindrot.org  2007-01-22 12:48 -------
When do the connections freeze? Is it at random, after a period of
inactivity, during large transfers, etc.?




------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

http://bugzilla.mindrot.org/show_bug.cgi?id=1263





------- Comment #12 from vincent at vinc17.org  2007-01-22 13:19 -------
(In reply to comment #11)
> When do the connections freeze? Is it at random, after a period of
> inactivity, during large transfers, etc.?
At random. The only parameter that seems to have an effect on the
freezes is the debug level of the master: with -vv, it is difficult to
reproduce a freeze, and with -vvv, I couldn't reproduce a single
freeze.

Note: I won't be able to do other tests before Friday.




------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

http://bugzilla.mindrot.org/show_bug.cgi?id=1263





------- Comment #13 from dtucker at zip.com.au  2007-01-22 14:16 -------
Created an attachment (id=1230)
 --> (http://bugzilla.mindrot.org/attachment.cgi?id=1230&action=view)
Revert TCP_NODELAY patch

Since you mentioned that it's only started happening recently, this
patch is worth a try: it reverts the following change:

- dtucker at cvs.openbsd.org 2006/03/13 08:33:00
[packet.c]
Set TCP_NODELAY for all connections not just "interactive" ones. 
Fixes
poor performance and protocol stalls under some network conditions
(mindrot bugs #556 and #981). Patch originally from markus@, ok djm@

Also, you said that the trouble started when you upgraded OpenSSH:
which version did you upgrade from, and was that on the client, server
or both?




------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

http://bugzilla.mindrot.org/show_bug.cgi?id=1263





------- Comment #14 from djm at mindrot.org  2007-01-22 16:52 -------
Created an attachment (id=1231)
 --> (http://bugzilla.mindrot.org/attachment.cgi?id=1231&action=view)
fd passing test program

I *think* (i.e. I'm not sure) that I have seen this on OS X too, when
opening subsequent connections on an existing control master. It seems
to be getting stuck while passing file descriptors from the slave to
the master (which suggests a kernel bug of some sort), however my
attempt to make a test case (attached) never fails.

Perhaps there is some subtlety around non-blocking file descriptors or
running into ulimits?




------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

http://bugzilla.mindrot.org/show_bug.cgi?id=1263





------- Comment #15 from vincent at vinc17.org  2007-02-16 03:15 -------
(In reply to comment #13)
> Also, you said that the trouble started when you upgraded OpenSSH:
> which version did you upgrade from, and was that on the client, server
> or both?
If fact, I had tried to revert to an older version, and the freeze was
still occurring. So, it could be due to another upgrade. I've done
other tests without the patch to see if it still occurs, and I could
reproduce it only once (the first time), whereas it was occurring much
more often before.

I don't know if this is related, but I also had freeze problems with
the man command (the freeze occurred in close(), called from pclose()),
and it seems to occur much less often.

I wonder if the 2007-01-25 Mac OS X security update change anything
concerning the freeze.




------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

http://bugzilla.mindrot.org/show_bug.cgi?id=1263





------- Comment #16 from djm at mindrot.org  2007-02-16 16:11 -------
Since I wrote comment #14, I have definitely seen it a number of times
on OSX. I'm pretty sure it is an OSX kernel bug: I have never seen it
on any other platform, and the problem occurs inside a syscall.

OTOH we could handle fd passing errors more gracefully.




------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

http://bugzilla.mindrot.org/show_bug.cgi?id=1263





------- Comment #17 from vincent at vinc17.org  2007-02-16 18:29 -------
I could reproduce the problem several times yesterday, even after
recompiling ssh. Then I applied the patch. I couldn't reproduce the
problem since, but I haven't tested it extensively.




------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.