topic Re: NFS server problem with Solaris 9 client readdir in Operating System - OpenVMS

NFS server problem with Solaris 9 client readdir

COLLOT_2 — Mon, 27 Sep 2004 05:48:24 GMT

When using VMS/NFS server of TCPIP 5.3 or 5.4 and mounting a share from a Solaris 9 client, the "ls" command on this share hangs, and the VMS NFS Server process loops. Direct access to a specific file ("cat", "rm", etc) works OK. It looks like that only the "readdir" primitive loops.
No such problem with TCPIP V5.1, or with Solaris 8 and any TCPIP version.

Re: NFS server problem with Solaris 9 client readdir

faris_3 — Mon, 27 Sep 2004 06:22:51 GMT

Hi,

Did you test TCPIP V5.1 Eco 4 and Solaris V9 ?
(the same problem was seen with these 2 versions)

Does a mount using nfs V2 work ?

mount -o vers=2 vmsnode:/filesys /mnt

/HF

Re: NFS server problem with Solaris 9 client readdir

COLLOT_2 — Mon, 27 Sep 2004 06:31:01 GMT

I don't know about TCPIP 5.1 ECO 4.

mount -o vers=2 fixes the problem.

Great !!!

Is there any patch available (either Solaris or TCPIP 5.4) to fix it without the "vers=2" turn-around ?

Re: NFS server problem with Solaris 9 client readdir

faris_3 — Mon, 27 Sep 2004 06:43:45 GMT

Are you using
TCP/IP 5.4 ECO 2 ?

Re: NFS server problem with Solaris 9 client readdir

COLLOT_2 — Mon, 27 Sep 2004 09:13:34 GMT

TCPIP V5.4 ECO 2 does not remove the problem.

Thanks anyway

Re: NFS server problem with Solaris 9 client readdir

faris_3 — Tue, 28 Sep 2004 04:10:29 GMT

Hi,

TCP/IP Engineering is aware of the problem.

A workaround given by them is to use
the directory and name caches :

With ovms_xqp_plus_enabled = 0
# ls - hangs

With ovms_xqp_plus_enabled = 7
# ls - DOES NOT hang - works as expected

To show ovms_xqp_plus_enabled:

$ tcpip sysconfig -q nfs ovms_xqp_plus_enabled
nfs:
ovms_xqp_plus_enabled = 0

And to set ovms_xqp_plus_enabled:

$ tcpip sysconfig -r nfs ovms_xqp_plus_enabled=7
ovms_xqp_plus_enabled: reconfigured
$ tcpip sysconfig -q nfs ovms_xqp_plus_enabled
nfs:
ovms_xqp_plus_enabled = 7

For more information on the File Name Cache
see :
http://h71000.www7.hp.com/doc/732final/6526/6526pro_047.html#troub_nfs_file

excerpt :
In addition, you can also use the NFS sysconfig attribute ovms_xqp_plus_enabled to modify the behavior of the NFS server to take advantage of the directory and name caches. This attribute is specified as a bit mask. The default value is 0, or OFF.

The following list describes the mask values:

1 (open directory on LOOKUP)
When an NFS LOOKUP operation is performed on a directory, the directory is accessed. This allows subsequent operations to use the directory cache. If the name cache is enabled, entries will be posted to it.
2 (open directory on READDIR)
When an NFS READDIR operation is received, the directory is accessed. This allows subsequent operations to use the directory cache.
4 (open file on GETATTR)
When the attributes of a file subject to record format conversion are read and the MODUS_OPERANDI mask 512 is enabled, the file's true size (that is, its converted size) is to be returned. If this option is enabled, then the access to convert the file will be cached for up to the number of seconds specified by the subsystem attribute vnode_age . If the ODS name cache is also enabled, the size and conversion information will be retained in the ODS cache until either the file is deleted or the entry is replaced by another, subject to the LRU behavior.
Obtain a combination of choices by adding the desired values. For example, enter 7 for a combination of the three.

When directory caching is enabled, the system must be configured to be able to handle the increased directory cache requirements. The following SYSGEN parameters may need to be increased, depending on the maximum number of files that the NFS server may access at any given time. This maximum is determined by the FILLM quota of the NFS$SERVER account and the SYSGEN parameter CHANNELCNT .

Hth,
/HF

Re: NFS server problem with Solaris 9 client readdir

COLLOT_2 — Tue, 28 Sep 2004 04:34:15 GMT

I tried the solution, but my "ls" on Solaris 9 is still hanging, with exactly the same behaviour (NFS server looping on the VMS side, and lots of network messages between the two).

May be I did something wrong :
1. Start NFS server
2. Execute "tcpip sysconfig -r nfs ovms_xqp_plus_enabled=7". Got the message "ovms_xqp_plus_enabled: reconfigured"
3. "mount vmsnode:/share /mnt" on Solaris client.
4. "ls /mnt" on Solaris client hangs.

Back to "-o vers=2" option ?

Re: NFS server problem with Solaris 9 client readdir

faris_3 — Tue, 28 Sep 2004 04:54:51 GMT

Hi,

Can you open a call with the French CSC ?

/HF