[Samba] Do software raids don't dance Samba?

Mike Quest mike at quest.de
Fri Sep 24 22:38:45 GMT 2004


Hey all,

well, first of all (even if it's quite late as it's in version 3.0.7 already): thanks a lot to all who 
are developing and supporting Samba! I'm new to Linux and I really like the way things
are happening on this OS. But now, I've reached a point where I have to raise the white
flag... 

It seems to me like - at least my - software raid can't dance Samba. I've got a box 
with SUSE 9.1 (Linux datagarden 2.6.4-54.5-default #1 Fri May 7 16:47:49 UTC 2004 
x86_64 x86_64 x86_64 GNU/Linux) and I've dld and compiled Samba version 3.0.7-1.1-SUSE. 
And as you might guess from the subject, I'd like to use it together software raid (levl5)
to store this and that. I'm using the default raid software package that gets shipped 
with 9.1 together with a Highpoint RocketRAID 454 ATA card with 8 disks connected 
to it:

datagarden:/local # cat /proc/mdstat
Personalities : [raid5]
md0 : active raid5 hdi1[7] hdh1[6] hdg1[5] hdf1[4] hde1[3] hdd1[2] hdc1[1] hdb1[0]
      2050671616 blocks level 5, 128k chunk, algorithm 2 [8/8] [UUUUUUUU]

hdi is on the motherboard's build-in controller as I couldn't convice the box that it 
should boot from the motherboard, but that's another story. But I hope, that this 
isn't causing the problem.

As I am a Linux beginner, I can't really tell you what version of raid software the box 
is using. If I run raid0run -V it tells me "mkraid version 0.90.0" and if I run raidstart 
-V it says "raidstart v0.3d compiled for md raidtools-1.00.3". Hmmm....

However, whenever I run Samba together with the raid, the clients (Win XP SVP2) do 
hang after a few minutes. In the middle of a copy process they tell me "Disk is full" 
(believe me, it isn't!). If I check the smbd log, I find the following:

[2004/09/22 18:47:00, 0] smbd/server.c:main(760)
  smbd version 3.0.7-1.1-SUSE started.
  Copyright Andrew Tridgell and the Samba Team 1992-2004
[2004/09/22 19:13:15, 0] smbd/server.c:main(760)
  smbd version 3.0.7-1.1-SUSE started.
  Copyright Andrew Tridgell and the Samba Team 1992-2004
[2004/09/22 19:14:26, 1] smbd/service.c:make_connection_snum(648)
  object1 (192.168.0.21) connect to service share1 initially as user smbguest (uid=1002, gid=100) (pid 4328)
[2004/09/22 19:14:30, 1] smbd/service.c:make_connection_snum(648)
  object1 (192.168.0.21) connect to service share2 initially as user smbguest (uid=1002, gid=100) (pid 4328)
[2004/09/22 19:18:32, 1] smbd/service.c:make_connection_snum(648)
  laterna-magica (192.168.0.22) connect to service share2 initially as user smbguest (uid=1002, gid=100) (pid 4347)
[2004/09/22 19:23:23, 1] smbd/service.c:make_connection_snum(648)
  laterna-magica (192.168.0.22) connect to service share2 initially as user smbguest (uid=1002, gid=100) (pid 4348)
[2004/09/22 19:23:38, 1] smbd/service.c:make_connection_snum(648)
  object1 (192.168.0.21) connect to service share2 initially as user smbguest (uid=1002, gid=100) (pid 4349)
[2004/09/22 19:23:55, 0] smbd/oplock.c:request_oplock_break(1055)
  request_oplock_break: no response received to oplock break request to pid 4347 on port 32769 for dev = 900, inode = 
21003, file_id = 1170
[2004/09/22 19:23:55, 0] smbd/open.c:open_mode_check(734)
  open_mode_check: exlusive oplock left by process 4347 after break ! For file [some file], dev = 900, inode = 21003. 
Deleting it to continue...
[2004/09/22 19:23:55, 0] smbd/open.c:open_mode_check(738)
 
"ps -ef | grep mbd" shows many smbd processes, which I can't kill. Even a "kill -9 [pid]" doesn't 
work. I am not able to shut it down with the usual command. Even "halt -f" doesn't work and I have 
to cut the power. Afterwards the raid goes starts to resnyc for some hours.

I've tested it with a good-ole-stand-alone-harddisk and then Samba works without any problem. 
I'm actually "stress-testing" it with various machines, which are copying files to it and Samba 
is stable as a rock.

Here is my smb.conf (actually with the "good-ole-stand-alone-harddisk", raid dirs are in /local/raid/):

[global]
   workgroup                 = Workgroup
   interfaces                = 127.0.0.1 eth0
   socket options            = TCP_NODELAY
   wins support              = yes
   bind interfaces only      = true
   printer admin             = @ntadmin, root, administrator
   map to guest              = Bad User
   guest account             = smbguest
   security                  = user
   encrypt passwords         = yes
   server string             = Jede Menge Daten
   netbios name              = DATAGARDEN
   add machine script        =
   domain master             = false
   domain logons             = no
   local master              = no
   preferred master          = auto
   ldap suffix               = dc=example,dc=com
   passdb backend            = smbpasswd
   os level                  = 65
   kernel oplocks            = false
   oplocks                   = no
   use sendfile              = no

[share1]
   path = /local/wd/share1/
   valid users = smbguest
   browsable = yes
   guest ok = yes
   printable = no
   available = yes
   read only = no
   writeable = yes
   create mask = 0775

[share2]
   path = /local/wd/share2/
   valid users = smbguest
   browsable = yes
   guest ok = yes
   printable = no
   available = yes
   read only = no
   writeable = yes
   create mask = 0775

I've googled around and I found a some people who seem to have a similar problem,
but no answers...

I would really appreciate if you have any experience / idea why it runs cool with a
stand-alone disk and does hang with a software raid. May be something special
for the raid to config?

Thanks a lot and happy dancing!
Mike
 



More information about the samba mailing list