[Gluster-devel] NFS problem (write error: Operation not permitted)

[gpvcs@tiscali.fr]

Hi everybody,

I tried the tla patch 98 and found it brokened for me (fcntl
freeze).
The nfs still doesnt work (see previous mail
http://lists.gnu.org/archive/html/gluster-devel/2007-04/msg00051.html):

context:
-----------

mount:
glusterfs:31850 on /gfs/fs1 on /export/fs1 type none (rw,bind)
/gfs/fs1 type fuse (rw,allow_other,default_permissions)
d610:/export/fs1 on /gfs/nfs1 type nfs
(rw,nfsvers=3,addr=192.80.221.32)

	(-- pts/27) ls > /gfs/nfs1/zz
	ls: write error: Operation not permitted
(-- pts/27) ls -lrt /gfs/nfs1/zz
-rw-r--r--  1 fturi ft 0 Apr 10 16:28 /gfs/nfs1/zz

I've been trying to debug this NFS bug for the last few days
 and all I found is:
nfsd is requesting attribut (fuse_getattr)
nfsd open the file (fuse_open) successfully
nfsd refuse to write to the file
nfsd relase the file (fuse_release)


According to man open:
       On NFS file systems with UID mapping enabled, open
may return a file descrip-
       tor but e.g. read(2) requests are denied with EACCES.
 This  is  because  the client  performs  open  by  checking
the permissions,but UID mapping is per-
       formed by the server upon read and write requests.

I did research on fuse project on found:
	http://sourceforge.net/mailarchive/message.php?msg_id=E1ELKsj-00066E-00%40dorka.pomaz.szeredi.hu
where miklos states:
	There's a difference between exporting and "exporting
	 properly", meaning that you provide proper
	 implementations for export operations that would let
	 you create an inode based on the inode number offered
	 by the client and connect it back into the dentry tree
	 by walking up.
-- and later:
	I believe I have tracked down the problem that I'm
	 getting. fuse_lookup calls d_find_alias for a
	 directory to check if an alias is being created. 
	 There is a check:
 	 if (alias && !(alias->d_flags & DCACHE_DISCONNECTED))
Another mail
	http://sourceforge.net/mailarchive/message.php?msg_id=1109580348.19900.1123.camel%40ipcus
explain the problematic of persistent inode for NFS

I understand that I-node are computed by glusterfs wich
might be a problem.


I dont really know whats happening on my box.
Does anybody have the same problem with exporting a
filesystem using NFS ?
Is there any success story around it ?

I will really appreciate some help on this issue as I can
only use NFS to access my glusterfs from AIX.
My need is to be able to NFS export glusterfs from both
member of the cluster.
Mount them on two separate AIX partitions, and use fcntl to
synchronize writting to the glusterfs cluster.
[AIXpart1] --> [mountNFS1] --> [export fs1 ] -->
[client1.vol] --> [AFR server1.vol server2.vol]
                                                           
                 /\
                                                           
             glusterfs
                                                           
                 \/
[AIXpart2] --> [mountNFS2] --> [export fs2] -->
[client2.vol] --> [AFR server1.vol server2.vol]
This way both AIX partitions will share a high avaibility
parallel file system.

Regards Francois

---------- Initial Header -----------

From      : "gpvcs at tiscali.fr" <gpvcs at tiscali.fr>
Subject : Re: [Gluster-devel] Is lock (fcntl) supported by
multilple concurentclient ?




------------------------ ALICE C'EST ENCORE MIEUX AVEC CANAL+ LE BOUQUET ! ---------------
Découvrez vite l'offre exclusive ALICEBOX et CANAL+ LE BOUQUET, en cliquant ici http://alicebox.fr
Soumis à conditions.