Bug in HP-UX 11.31/NFSv4 fcntl(F_SETLKW)?

Hi,

debugging a Samba problem where the database is on NFSv4, I see that
fcntl(F_SETLKW) returns EAGAIN. However, according to the manual, F_SETLKW is
supposed to wait until the lock is available. Despite of that I'm almost sure
that the lock was not busy at that time. Yet another bug in HP-UX NFSv4
(B.11.31.04 (ONC+ 2.3))?

Regards,
Ulrich

Ulrich Windl wrote:
> Hi,
>
> debugging a Samba problem where the database is on NFSv4, I see that
> fcntl(F_SETLKW) returns EAGAIN. However, according to the manual, F_SETLKW is
> supposed to wait until the lock is available. Despite of that I'm almost sure
> that the lock was not busy at that time. Yet another bug in HP-UX NFSv4
> (B.11.31.04 (ONC+ 2.3))?
>
> Regards,
> Ulrich

I know from Solaris that there was such a problem with > 256 locks on a file;
Samba uncovered it.

But here it looks looks more like a bug in NFSv4 on HP-UX.

Also for performance reason - have the Samba DBs on a local disk!

If you must have locks on NFS, mounting it with the "llock" option is
probably a work-around.
"llock" is special on HP-UX: locks are handled locally,
not via the NFS server, so do not work with other NFS clients.

--
echo imhcea\.lophc.tcs.hmo |
sed 's2\(....\)\(.\{5\}\)2\2\122;s1\(.\)\(.\)1\2\11g;1 s;\.;::;2'

Michael Tosch writes:

> Ulrich Windl wrote:
>> Hi,
>>
>> debugging a Samba problem where the database is on NFSv4, I see that
>> fcntl(F_SETLKW) returns EAGAIN. However, according to the manual, F_SETLKW is
>> supposed to wait until the lock is available. Despite of that I'm almost sure
>> that the lock was not busy at that time. Yet another bug in HP-UX NFSv4
>> (B.11.31.04 (ONC+ 2.3))?
>>
>> Regards,
>> Ulrich
>
> I know from Solaris that there was such a problem with > 256 locks on a file;
> Samba uncovered it.
>
> But here it looks looks more like a bug in NFSv4 on HP-UX.
>
> Also for performance reason - have the Samba DBs on a local disk!

Hi!

Honestly I don't know how CIFS handles broken TCP connections regarding locks:
If the locks are lost anyway, your suggestion makes sense. Now the locks are
on a filesystem that is mounted where the IP service address for CIFS Server
is. For some historical reasons that filesystem is mounted as NFS(v4).

>
> If you must have locks on NFS, mounting it with the "llock" option is
> probably a work-around.
> "llock" is special on HP-UX: locks are handled locally,
> not via the NFS server, so do not work with other NFS clients.

That's a no-no here: The filesystem is exported/mounted read-write to several
clients (not the samba directory, of course).

Regards,
Ulrich