3 Bugs to Report: OpenSSH V1.2pre13

[Robert Hardy]

Three possibly related bugs to report. N.B.  The test machines in question
are in peak form (with the exception of different kernel versions) and were
working 100% under the old ssh 1.2.x.  The two clients we tested from are
machines running 2.2.13 & 2.2.14preX Linux kernels. The server where the
problems appeared is running 2.2.12.

1. sshd dies periodically. The crash occurred just after a connect
   immediately followed by a hang-up. It is unclear if that is relevant.  It
   could be coincidental. I have only examined the logs for this one failure
   (there have been 2-3 others but we just restarted the daemon). This
   failure was seen after BB (Big Brother) has been probing ssh for several
   days.  BB probes sshd to see if it responds and when it does it promptly
   hangs up without negotiating a connection. In response to this rude
   hang-up sshd usually logs a warning and goes back to waiting... For some
   reason every couple of days it decides to die. I grabbed the log except
   below at the last crash. /dev/urandom is in use by other things on the
   system without difficulties (to my knowledge anyways...).

Nov 21 20:59:20 aserver sshd[4059]: Connection from x.x.x.170 port 2222
Nov 21 20:59:20 aserver sshd[4059]: fatal: Bad protocol version identification: quit
Nov 21 20:59:56 aserver sshd[4047]: Closing connection to x.x.x.18
Nov 21 21:04:28 aserver sshd[4092]: Connection from x.x.x.170 port 2258
Nov 21 21:04:28 aserver sshd[4092]: fatal: Bad protocol version identification: quit
Nov 21 21:04:28 aserver sshd[24736]: fatal: Couldn't read from random pool "/dev/urandom":Interrupted system call

^^ After this we get a page from BB indicating ssh has given up the ghost...

2. sshd will sometimes hang when disconnecting from a server. 
   -ssh host
   -we do some work
   -we hit CTRL-D to disconnect
   -we logout on remote system
   -ssh does not disconnect from remote system and will stay hung indefinitely
    (an ps -axuww shows an sshd process still running on the pty.)

3. For no rhyme or reason, we occasionally get an warning message just
   before we get a shell prompt when connecting to some of our servers
   through openssh. All our test servers are running the same software build
   (distribution) and the same version of openssh yet only some of them
   occasionally see the problem.  This is the message we get:
   chan_shutdown_read failed for #0/fd4: Transport endpoint is not connected

   It is not clear what relation the warning message may have to the other 2
   bugs.  The warning message does not seem to indicate that shell will
   either hang or kill the parent sshd.

I am willing to test various things to try and help isolate the problem(s). 
I'm open to suggestions...

Regards,
Rob

--
----------------"Linux the choice of a GNU Generation!"-----------------
Robert Hardy                                          C.E.O. Webcon Inc.
rhardy at webcon.net      PGP Key available by finger        (613) 276-6206