A couple of days ago, I was trying to kill Samba processes on an Irix box, but was unable to do so. Any smbd process (but not nmbd) would go into an uninterruptible sleep, waiting for some sort of event which it never received. 'ps' reported that all smbd processes were waiting for the same event number. The file systems on the box seemed fine; this didn't seem to be the symptom of an unresponsive disk. There didn't seem to be any NIC problems; other traffic wasn't affected. What else would smbd get hung up like that on? Why just smbd, not nmbd? Details: Irix 6.2; Samba 2.0.7. Nothing interesting in system log; didn't think to look at Samba log; has never happened before (that I know of). CC me, if you would... Andrew Klaassen
On Tue, Oct 09, 2001 at 09:58:09AM -0700, Herb Lewis wrote:> Were you resharing NFS mounted directories?No. (I can't be 100% sure, but I'm 99.97% sure.) One other odd thing: The init.d startup script ("./samba start" on Irix) would hang, simply adding one more unkillable smbd processes waiting for the same event as all the others. Ditto with a straight "smbd -D" command; it would never return console control, as if it was being hung up before it had a chance to daemonize itself. Andrew Klaassen> Andrew Klaassen wrote: > > > > A couple of days ago, I was trying to kill Samba processes on an > > Irix box, but was unable to do so. Any smbd process (but not > > nmbd) would go into an uninterruptible sleep, waiting for some > > sort of event which it never received. 'ps' reported that all > > smbd processes were waiting for the same event number. > > > > The file systems on the box seemed fine; this didn't seem to be > > the symptom of an unresponsive disk. > > > > There didn't seem to be any NIC problems; other traffic wasn't > > affected. > > > > What else would smbd get hung up like that on? Why just smbd, > > not nmbd? > > > > Details: Irix 6.2; Samba 2.0.7. Nothing interesting in system > > log; didn't think to look at Samba log; has never happened > > before (that I know of). > > > > CC me, if you would... > > > > Andrew Klaassen > > > > -- > =====================================================================> Herb Lewis Silicon Graphics > Networking Engineer 1600 Amphitheatre Pkwy MS-510 > Strategic Software Organization Mountain View, CA 94043-1351 > herb@sgi.com Tel: 650-933-2177 > http://www.sgi.com Fax: 650-932-2177 > =====================================================================> >
On Tue, Oct 09, 2001 at 01:34:41PM -0700, Herb Lewis wrote:> Unfortunately we need a little more info. Is this reproducible?I've only seen it once, ever.> When it happens, does "smbstatus -p" show all the pids that > show up in "ps -ef | grep smbd" (ps should show 1 additional > smbd whose parent is pid 1) If ps shows many more than > smbstatus you may be seeing a problem with pids waiting to > die. Check the log files and see if you are seeing crash > messages for these smbd processes. I would suggest upgrading > to at least 2.2.1a as there were some bugs fixed along these > lines if this is what you are seeing.Unfortunately, the logs seem to be on an extremely short rotation; when it occured to me today to look at Friday's smbd logs, the information was long gone. I didn't know to try "smbstatus -p" on Friday, either, unfortunately. What might cause pids to be so long "waiting to die"? Why would that prevent new smbd processes from being able to daemonize themselves? Andrew Klaassen
On Tue, Oct 09, 2001 at 01:34:41PM -0700, Herb Lewis wrote:> Unfortunately we need a little more info. Is this reproducible? When > it happens, does "smbstatus -p" show all the pids that show up in > "ps -ef | grep smbd" (ps should show 1 additional smbd whose parent > is pid 1) If ps shows many more than smbstatus you may be seeing a > problem with pids waiting to die. Check the log files and see if you > are seeing crash messages for these smbd processes. I would suggest > upgrading to at least 2.2.1a as there were some bugs fixed along > these lines if this is what you are seeing.Well, it looks like it is reproducible - it's happening again. smbstatus reports anywhere from 0 to 3 more processes than ps. I don't see any crash reports in the smbd logs. This time, one additional thing is happening that wasn't happening before - X freezes partway through login on the machine's console. Also - I can telnet to it as root, but it hangs on login if I try to telnet in as a regular user. The NFS exports of the same drive that Samba is exporting seem to be fine. Andrew Klaassen