Re: [PATCH] nfs: introduce mount option '-olocal_lock' to make locks local

[Suresh Jayaraman <sjayaraman@suse.de>]

On 09/18/2010 03:45 AM, Trond Myklebust wrote:
> On Thu, 2010-09-16 at 11:44 +0530, Suresh Jayaraman wrote:
>> Hi Trond,
>>
>> I'm resending this patch as a separate email as I have not heard from you.
>>
>> Changes since last post:
>>   - remove unneeded NFS_MOUNT_NONLM flag checks from do_getlk()/do_setlk()/do_unlck
>>   - update comments to include the new mount option
>>
> 
> Hi Suresh,
> 
> Just a couple of comments:
> 
>       * Firstly the legacy binary mount interface uses
>         NFS_MOUNT_NONLM=0x200, so we cannot change that value. What I
>         was suggesting was rather that we just parse it to mean
>         NFS_MOUNT_LOCAL_FLOCK | NFS_MOUNT_LOCAL_FCNTL.

Ah, ok. Will fix it in the next spin.

>       * Have you tested this with NFSv4? I think I asked you this
>         previously, but I haven't seen an answer. I ask because -onolock
>         used to cause NFSv4 reboot recovery to Oops since the struct
>         file_lock didn't have any associated nfsv4 lock state. I just
>         want to make sure we're not re-enabling that bug.
> 

I had earlier tested the patch with NFSv4 but I didn't test the reboot
recovery. And, you guessed it right - with "-olocal_lock=all", the patch
causes an oops in _nfs4_do_setlk() when the server recovers after a
reboot. Clearing the new flags in nfs4_validate_mount_flags() fixes the
problem. Also, I have updated the patch to take into account the nfsroot
case.

I also noticed Chuck's recent patchset will move nfsroot option
handling. I'll rebase by patch against nfs-2.6.git and resend it.

Thanks,

-- 
Suresh Jayaraman