[SOLVED] samba nmbd does not start at boot

[pernel]

New Mageia 5 64 bit on a Dell precision T5500 with two processors, in total 8 cores.
The host name is not visible in network after boot.
Found out that if samba is restarted with webmin, the hostname becomes visible and samba sharing works.

After boot:

Code: Select all

 
# ps -A | grep mbd
 3155 ?        00:00:00 smbd
 3174 ?        00:00:00 smbd


shows that nmbd is not running after boot.

Restarting:

Code: Select all

]
# /etc/init.d/smb restart
Restarting smb (via systemctl):                                 [  OK  ]
[root@lindell ~]# ps -A | grep mbd
16156 ?        00:00:00 smbd
16166 ?        00:00:00 nmbd
16171 ?        00:00:00 smbd


After this the host name appears visible in the network.

I tried uninstall and re-install samba but the result is the same.
I have another Mageia 5 64 bit in virtualbox with samba working normally but that is an upgrade from Mageia 4.

I would appreciate help to pinpoint and fix the problem.
Of course a possible workaround would be to use a script to restart samba some period after the machine has booted but that's not so elegant.

[doktor5000]

Can you please show the output as root of

Code: Select all

systemctl status smb.service -a -n50

edit: Hmmm, will probably not help that much as samba has not been converted yet to systemd unit files.

But the output of

Code: Select all

journalctl -ab | grep -iE "smbd|nmbd|samba"

might be useful.

[pernel]

Code: Select all

[root@lindell ~]# journalctl -ab | grep -iE "smbd|nmbd|samba"
[root@lindell ~]#


Obviously gives no output when the machine was booted recently.

Code: Select all

[root@lindell ~]# ps -A | grep mbd                                               
 3085 ?        00:00:00 smbd
 3103 ?        00:00:00 smbd
[root@lindell ~]# systemctl status smb.service -a -n50                          ● smb.service - LSB: Samba servers (nmbd and smbd)
   Loaded: loaded (/etc/rc.d/init.d/smb)
   Active: active (running) since fre 2015-07-31 19:37:15 CEST; 2min 49s ago
  Process: 3076 ExecStart=/etc/rc.d/init.d/smb start (code=exited, status=0/SUCCESS)
   CGroup: /system.slice/smb.service
           ├─3085 smbd -D
           └─3103 smbd -D

jul 31 19:37:15 lindell smb[3076]: Startar SMB-tjänster: [  OK  ]
jul 31 19:37:15 lindell smb[3076]: Startar NMB-tjänster: [  OK  ]
[root@lindell ~]# ps -A | grep mbd
 3085 ?        00:00:00 smbd
 3103 ?        00:00:00 smbd



This behaviour is not completely consistent. Sometimes also nmbd is started at boot.

[doktor5000]

Obviously gives no output when the machine has booted recently.

Well, I'd say it should at least log when it starts, no?

Apart from that, as the other ouput doesn't provide much insight as it's still a legacy sysV service, can't help that much.
Either it's some kind of race condition and it starts too early, before network stack is already available.
You could try to rule that out by disabling it for system start, and start it manually later on - if it starts OK,
then you might need to enable the network-auth service so that the system waits for the network to be up during boot.

Otherwise you probably need to enable debug logging to see why nmbd does not start normally, only on restart.

[pernel]

I edited last post while you were commenting it. Maybe some useful information now?

[pernel]

Output from this command attached

[root@lindell ~]# journalctl -ab | grep -iE "smb|nmb|eth0|em1" > /home/per/journal.log

[doktor5000]

I edited last post while you were commenting it. Maybe some useful information now?

None that I can see. I understood why the previous journalctl query gave no output, as it does not write the log as smbd but only as smb.
Nonetheless, only the two lines of it starting.

You can try to increase the loglevel of samba log itself, but I'm not sure that will help with your issue.
With default settings you would need to at least uncomment "log level" setting and increase it slightly.

Code: Select all

┌─[doktor5000@Mageia5]─[19:40:15]─[~]
└──╼ grep "log [file,level]" /etc/samba/smb.conf                                                                                                                                                                   
# this tells Samba to use a separate log file for each machine
   log file = /var/log/samba/%m.log
# Put a capping on the size of the log files (in Kb).
# Set the log (verbosity) level (0 <= log level <= 10)
# log level = 3

[pernel]

Have not yet found any solution, only a "dirty" workaround.
I'm more or less sure this is a race condition. Posts in other forums indicate that possibility.
Anyhow I made a script run by cron each minut that checks if nmbd is up, and if not, restarts samba. Not very elegant, but at least it gives a stable network connection.
Would prefer to run this script only once but at a point after startup when the network is guaranteed to be up vith a consistent interface name.
Even better would be to find a way to wait with the first attempt to start samba until network is up and the interface name is stable.
This mix between system V and systemd makes me very confused. I dont understand how to modify start scripts to ensure samba is initiated very late in the startup sequence.
In post 4 you suggest:

if it starts OK,
then you might need to enable the network-auth service so that the system waits for the network to be up during boot.

.
How do I enable that?

[doktor5000]

As root do

Code: Select all

systemctl enable network-auth

If you use networkmanager it would be

Code: Select all

systemctl enable NetworkManager-wait-online.service

[pernel]

Thanks a lot!

systemctl enable network-auth

fixed the problem. Samba (including nmbd) now consistently starts each time. It's obvious that there was a race condition.
In which configuration file is this parameter stored or where is the startup code changed by this command? I ask because I wonder if future updates or upgrades will break this fix.
Is this considered to be a bug that should be reported or is my case too rare?

[doktor5000]

It is too rare, I'd say, never heard before about this issue with nmbd.

As network-auth is one of the legacy sysV services which was not yet converted to a systemd unit,
it's only stored as empty files to signal start/stop in the respective runlevels:

Code: Select all

/etc/rc.d/rc0.d/K50network-auth
/etc/rc.d/rc1.d/K50network-auth
/etc/rc.d/rc2.d/S50network-auth
/etc/rc.d/rc3.d/S50network-auth
/etc/rc.d/rc4.d/S50network-auth
/etc/rc.d/rc5.d/S50network-auth
/etc/rc.d/rc6.d/K50network-auth
/etc/rc.d/rc7.d/S50network-auth

The only thing this does is that the system will wait during boot until the network interface is fully initialised.
Normally this is done in the background and in a parallel way, as it's not that critical during boot for majority of users,
and leads to the graphical desktop that can start much earlier.

The other thing you could do, is to mark any network filesystem you might have in /etc/fstab with the _netdev option,
that might have a similar affected if you mount any network filesystem during boot, but might not be sufficient for your nmbd issue.

Anyways, with mga6 we will finally switch to samba 4

Please mark the thread accordingly by editing the topic of the first post and prefix it by [SOLVED], thanks