Re: [PATCH] Fix bug nfslock request sending fail will be process as blocked

[Mi Jinlong <mijinlong@cn.fujitsu.com>]

Trond Myklebust =E5=86=99=E9=81=93:
> On Tue, 2010-03-16 at 18:20 +0800, Mi Jinlong wrote:=20
>> Trond Myklebust =E5=86=99=E9=81=93:
>>> On Fri, 2010-03-12 at 18:17 +0800, Mi Jinlong wrote:=20
>>>> If local reason cause nfslock request send fail(means status < 0,
>>>> resp->status not be reset), the request will be process as blocked=
=20
>>>> at first now.
>>>>
>>>> This patch initialize resp->status to nlm_lck_denied_nolocks, it=20
>>>> can make the following process correctly.
>>>>
>>>> Signed-off-by: Mi Jinlong <mijinlong@cn.fujitsu.com>
>>>> ---
>>>>  fs/lockd/clntproc.c |    2 +-
>>>>  1 files changed, 1 insertions(+), 1 deletions(-)
>>>>
>>>> diff --git a/fs/lockd/clntproc.c b/fs/lockd/clntproc.c
>>>> index c81249f..a631582 100644
>>>> --- a/fs/lockd/clntproc.c
>>>> +++ b/fs/lockd/clntproc.c
>>>> @@ -535,7 +535,7 @@ again:
>>>>  	 * Initialise resp->status to a valid non-zero value,
>>>>  	 * since 0 =3D=3D nlm_lck_granted
>>>>  	 */
>>>> -	resp->status =3D nlm_lck_blocked;
>>>> +	resp->status =3D nlm_lck_denied_nolocks;
>>>>  	for(;;) {
>>>>  		/* Reboot protection */
>>>>  		fl->fl_u.nfs_fl.state =3D host->h_state;
>>> We _want_ to process it as being blocked if the RPC call was
>>> interrupted. The above patch will cause the client to just abandon =
the
>>> interrupted lock request without sending the CANCEL request...
>>   No, the above patch don't just abandon the interrupted lock reques=
t,
>>   when the RPC call was interrupted, client will send an UNLOCK requ=
est for status < 0.
>>    ...
>>    583         if (status < 0)
>>    584                 goto out_unlock;
>>    ...
>>
>>   And, I think an UNLOCK request is more advisable than a CANCEL req=
uest.
>>   If a LOCK request was succeed when CANCEL request coming, it's use=
less;
>>   the lock should be unlocked.
>=20
> An unlock does not guarantee that the server will cancel the blocked
> lock request.

  When server receive an UNLOCK request, it will try to cancel the lock=
 at first,
  codes as follows:
    ...
    567 nlmsvc_unlock(struct nlm_file *file, struct nlm_lock *lock)
    568 {
    ....
    578         /* First, cancel any lock that might be there */
    579         nlmsvc_cancel_blocked(file, lock);
    ...

  Server don't process the nlmsvc_cancel_blocked's return value, is it =
the problem=20
  that you said above ? or some other reason?

  Server cannot cancel the blocked lock request, it's acceptable someti=
mes, IMO,
  but client should work strongly itself.

thanks,
Mi Jinlong