Lately I'm seeing bizarre problems with the SAMBA server I'm using in production. For no rhyme or reason the connections get dropped, the same set of users who previously were able to access the shares, now get permission denied. Users (not all but some) are having trouble opening the folders, files and so on. This was not the case few days ago, it started happening lately with amazing inconsistency. Inconsistent in it works some times, it simply doesn't sometimes. I have two samba servers in the setup one on Solaris-10 and the other on Solaris-9. On Solaris 10 I'm using stock SUN Samba packages. It would be real easy, if I could isolate the problem by limiting it to one, but it occurs on both the servers. The samba versions are different on both servers. Before any can suggest, I did shutdown one server and pointed all the users to remaining one. No luck. Shut down the other server, re-pointed the users, no luck. I'm going nuts trying to isolate the problem, if only it wasn't happening with such astonishing inconsistency. Trussing the smbd shows the user access is stuck in fcntl system calls like this, and these users do have all the proper permissions for messing with these files Fcntl(10, F_SETLKW64, 0xFFBFF750) . (sleeping) Fcntl (27,F_GETLKW64,0xFFBFF840)..(sleeping) They never get out of this. Tried the usual options of oplocks = no kernel oplocks = no and even faking oplocks in the smb.conf, I can't get out of this. Anyone can suggest something I can muck with? I know earlier Solaris versions had a kernel bug with fcntl and it was patched. So, what else could be the issue here? Thanks. Ravi K. Channavajhala http://www.dciera.com
On Tue, Aug 11, 2009 at 11:49:09PM +0530, ravi channavajhala wrote:> Lately I'm seeing bizarre problems with the SAMBA server I'm using in > production. For no rhyme or reason the connections get dropped, the same > set of users who previously were able to access the shares, now get > permission denied. Users (not all but some) are having trouble opening the > folders, files and so on. This was not the case few days ago, it started > happening lately with amazing inconsistency. Inconsistent in it works some > times, it simply doesn't sometimes. > > > > I have two samba servers in the setup one on Solaris-10 and the other on > Solaris-9. On Solaris 10 I'm using stock SUN Samba packages. It would be > real easy, if I could isolate the problem by limiting it to one, but it > occurs on both the servers. The samba versions are different on both > servers. Before any can suggest, I did shutdown one server and pointed all > the users to remaining one. No luck. Shut down the other server, > re-pointed the users, no luck. I'm going nuts trying to isolate the > problem, if only it wasn't happening with such astonishing inconsistency. > > > > Trussing the smbd shows the user access is stuck in fcntl system calls like > this, and these users do have all the proper permissions for messing with > these files > > > > Fcntl(10, F_SETLKW64, 0xFFBFF750) . (sleeping) > > Fcntl (27,F_GETLKW64,0xFFBFF840)..(sleeping) > > > > They never get out of this. Tried the usual options of oplocks = no kernel > oplocks = no and even faking oplocks in the smb.conf, I can't get out of > this. Anyone can suggest something I can muck with? I know earlier Solaris > versions had a kernel bug with fcntl and it was patched. So, what else > could be the issue here? Thanks.What filename is the smbd stuck on the lock in ? Jeremy.
On Wed, Aug 12, 2009 at 08:51:51AM +0530, ravi channavajhala wrote:> Hi Jeremy, > > Why would the file name be an issue here? It is not just one file, several > files are getting affected. I followed your earlier discussion on the issue > way back in 2002, can you suggest something to try. Really, I will take > whatever I can get.Firstly, let's keep the list CC:ed so we keep everyone up to date. I'm asking what file name the fcntl is blocked in, as I want to know if this is a Samba tdb, or a file the server is trying to access. The Samba server should never make a blocking fcntl lock call on a user data file, but will make such calls on tdb files. So please let me know what file the fcntl syscall is blocked in. Thanks, Jeremy.