On Wed, Dec 31, 2003 at 10:08:03AM -0500, John Von Essen wrote: > I can do two rsh's back to back with no problems, its the third (and 4th > and so on) that hang. > > On the FreeBSD side, after the first rsh, netstat shows: > > tcp4 0 0 mx100.851 embryo.bluebell..1021 > TIME_WAIT > tcp4 0 0 mx100.shell embryo.bluebell..1022 > TIME_WAIT > > Those connections stay around for awhile, about 30 seconds. Only when they > disappear does the next rsh work.
OK. Some progress. This rules out problems due to limitations in the number of possible connections you can have open at any one time -- if the limit is just two, then there would be a lot more things complaining than just rcp(1). And you'ld have to try exceedingly hard to get a FBSD system that limited. Hmmm... What flags are you invoking inetd(8) with on the FreeBSD side? Specifically are you using any of these (quoting from the manual page): -c maximum Specify the default maximum number of simultaneous invocations of each service; the default is unlimited. May be overridden on a per-service basis with the "max-child" parameter. -C rate Specify the default maximum number of times a service can be invoked from a single IP address in one minute; the default is unlimited. May be overridden on a per-service basis with the "max-connections-per-ip-per-minute" parameter. -R rate Specify the maximum number of times a service can be invoked in one minute; the default is 256. A rate of 0 allows an unlimited number of invocations. -s maximum Specify the default maximum number of simultaneous invocations of each service from a single IP address; the default is unlimited. May be overridden on a per-service basis with the "max-child-per- ip" parameter. The symptoms you describe could be caused eg. by running with '-s 2' in the inetd flags (you're getting two socket connections per rsh or rcp invocation because a second channel is opened to carry the stderr from the invoked command, but that doesn't count towards inetd's connection limits). The default for all of these is unlimited (ie. inetd_flags="-wW") and there aren't any per-service limits on the rsh (shell) service in the default inetd.conf. I generally use: inetd_flags="-wWl -R 1024 -c 128 -a ${hostname}" in my /etc/rc.conf on internet facing machines where I run inetd(8) -- if this is a purely internal machine (which it certainly should be if you're using rsh(1) on it) then I wouldn't bother with any sort of connection rate-limiting, at least for the time being. Hmmm... As well -- what's the output of: % sysctl net.inet.tcp.delayed_ack You might try setting that to zero to turn off delayed ack. That's where the system will wait for about a minute before sending an ACK in order to try and coalesce it with a data packet. Usually that's a win performance-wise. See tcp(4). There's also the RFC1644 support you might want to try toggling: see ttcp(4) -- I'm unable to find any definitive statement on the net about Solaris support for this, so no idea if it will actually help or not. > As for the rcp, I was missing a trailing slash, apparently rcp -r syntax > between Solaris and FreeBSD is a little different. So the rcp's work, but > that take just as long as the rsh calls. That's one problem down then. Good. > As for name resolution, the Solaris box uses dns, and so does FreeBSD. > Both have some entries in the hosts file. That's good too. Rules out some more areas which could be causing the trouble. Cheers, Matthew -- Dr Matthew J Seaman MA, D.Phil. 26 The Paddocks Savill Way PGP: http://www.infracaninophile.co.uk/pgpkey Marlow Tel: +44 1628 476614 Bucks., SL7 1TH UK
pgp00000.pgp
Description: PGP signature