Re: [PATCH] generic/478: should wait until dup success or fail

[Eryu Guan <guaneryu@gmail.com>]

On Thu, Apr 11, 2019 at 05:35:21PM +0800, ZhangXiaoxu wrote:
> When test on nfs, maybe failed as below:
>   wait sem0 0: Resource temporarily unavailable
>   set sem0 0: Invalid argument
>   exit rmid: Invalid argument
> 
> It means getlk routine wait sem0 timeout, because the setlk routine
> maybe take too long time on ‘fcntl’ or 'dup'.
> 
> So, let wait on sem0 in getlk routine until the setlk routine success
> or fail.
> 
> Signed-off-by: ZhangXiaoxu <zhangxiaoxu5@huawei.com>
> ---
>  src/t_ofd_locks.c | 4 +---
>  1 file changed, 1 insertion(+), 3 deletions(-)
> 
> diff --git a/src/t_ofd_locks.c b/src/t_ofd_locks.c
> index e3b15dd..32e90c0 100644
> --- a/src/t_ofd_locks.c
> +++ b/src/t_ofd_locks.c
> @@ -414,9 +414,7 @@ int main(int argc, char **argv)
>  		sop.sem_num = 0;
>  		sop.sem_op = 0;
>  		sop.sem_flg = 0;
> -		ts.tv_sec = 5;
> -		ts.tv_nsec = 0;
> -		if (semtimedop(semid, &sop, 1, &ts) == -1)
> +		if (semtimedop(semid, &sop, 1, NULL) == -1)

We set a timeout value on purpose in the first place, to avoid hanging
the whole test if the setlk process went wrong.

And I didn't reproduce it on NFS. Do I need any special NFS
configuration?

Thanks,
Eryu

>  			err_exit("wait sem0 0", errno);
>  
>  		if (fcntl(fd, getlk_macro, &flk) < 0)
> -- 
> 2.7.4
>