Gentoo Websites Logo
Go to: Gentoo Home Documentation Forums Lists Bugs Planet Store Wiki Get Gentoo!
Bug 66656 - Clients lock when accessing Samba-shared files from a sparc-based server
Summary: Clients lock when accessing Samba-shared files from a sparc-based server
Status: RESOLVED WONTFIX
Alias: None
Product: Gentoo Linux
Classification: Unclassified
Component: [OLD] Server (show other bugs)
Hardware: Sparc Linux
: High major (vote)
Assignee: Gentoo's SAMBA Team
URL: http://www.devrandom.org/~jason/samba...
Whiteboard:
Keywords:
Depends on:
Blocks:
 
Reported: 2004-10-07 08:23 UTC by Jason McCormick
Modified: 2004-10-08 06:47 UTC (History)
1 user (show)

See Also:
Package list:
Runtime testing required: ---


Attachments

Note You need to log in before you can comment on or make changes to this bug.
Description Jason McCormick 2004-10-07 08:23:24 UTC
(Note: This bug report is taken from http://forums.gentoo.org/viewtopic.php?p=1622152#1622152)

I'm running into an issue with Samba on Linux/Sparc. I have the latest samba package emerged onto my sparc. Samba is configured for a PDC role with all appropriate machine and user accounds. A Win2K client can authenticate perfectly against the box, get credentials, login to the workstation, etc. A user can browse through shares and folders just fine. However whenever the Win2K client attempts to open a file, the entire workstation locks up, requiring a kill of explorer.exe. Standard logging seems to reveal no obvious problem. Talking with some people on the IRC channel last night reveals that others are seeing similiar problems.

I enabled logging at level 4 and through all the spew it appears that whenever the Win2K box tries to access the file, Samba performs a series of file operations, including attempts at oplocks and then freezes and finally tries to send back an message of "NT_STATUS_OBJECT_PATH_NOT_FOUND" however this does not seem to unlock the client. For reference Samba appears to "see" the file exists (and it does exist as -rw-rw-rw). For reference, this same file structure is exported via NFS as well and file access works as expected there. I tried disabling NFS to rule out any lock problems but that didn't seem to help.

  I've looked at the traffic flow, the TCP steam seems to be working just fine in both directions between the server and the workstation. I've dumped a complete "transaction" of the Win2k box trying to open a sample file. To setup the scenario a little better, the file location /data/usr is shared as \\cirdan\usr. Within that is /data/usr/jason/learnvi.tar.gz. The user logged in to the Win2K box authenticates to a UNIX user that has read permisions to that file. I've posted a log of the attempt to open the file over Samba at http://www.devrandom.org/~jason/samba.log.

My USE flags are only "-X -java" and there's nothing beyond the default settings in make.conf for CCARGS.

The kernel is 2.6.7 and samba is 3.0.7-r1.  My smb.conf can be viewed at the forum posting above (didn't want to clutter up bugzilla too much).

Another user on the forums has reported the same problem and chatting with people in #gentoo-sparc seems to indicate that others are seeing "problems" with Samba that haven't been fully identified.

Reproducible: Always
Steps to Reproduce:
Comment 1 Jason McCormick 2004-10-07 21:52:14 UTC
I've tracked this down, from a helpful suggestion on the Samba mailing list, to enabling the option 'use sendfile = no'.  It appears that this allows Samba to use a more efficient system call for reading the files from the local OS to serve to clients.  This is yes by default.  Setting this to no makes my system behave normally.

So is this a bug in Samba or the underlying system?  I'm not enough of a C programmer to determine how to find out.
Comment 2 Christian Andreetta (RETIRED) gentoo-dev 2004-10-08 06:44:53 UTC
Seems like a O.S. implementation bug. I'm writing a note abouth this in the latest ebuild. thanks :-)
Comment 3 Michael Glauche (RETIRED) gentoo-dev 2004-10-08 06:47:55 UTC
It also seems that 3.0.7 has some bugs, see http://forums.gentoo.org/viewtopic.php?t=228147

A couple of users did report that they got timeouts in 3.0.7 (which were not exisiting in 3.0.5 and 3.0.6), but which are fixed by using "use sendfile = no"