Re: [PATCH] ltp/fsx.c: Ignore unsupported keep_size flag even if it's specified by --replay-ops

[Eryu Guan <eguan@linux.alibaba.com>]

On Fri, Jan 03, 2020 at 03:54:51PM +0800, Xiao Yang wrote:
> Current fsx cannot drop keep_size flag if it is specified by --replay-ops but not
> supported by fallocate().  For example, running generic/469 got the following
> error on NFSv4.2:
> ----------------------------------------------------
> main: filesystem does not support fallocate mode FALLOC_FL_KEEP_SIZE, disabling!
> ...
> fallocate: 0x0 to 0x1000
> do_preallocate: fallocate: Operation not supported
> LOG DUMP (1 total operations):
> 1(  1 mod 256): FALLOC   0x0 thru 0x1000        (0x1000 bytes) PAST_EOF
> ...
> ----------------------------------------------------
> 
> Avoid calling fallocate(FALLOC_FL_KEEP_SIZE) in this case by ignoring keep_size flag.
> 
> BTW: NFSv4.2 doesn't support single keep_size flag，as below:
> ----------------------------------------------------
> if ((mode != 0) && (mode != (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE)))
>         return -EOPNOTSUPP;
> ----------------------------------------------------
> 
> Signed-off-by: Xiao Yang <yangx.jy@cn.fujitsu.com>
> ---
>  ltp/fsx.c | 11 +++++++++++
>  1 file changed, 11 insertions(+)
> 
> diff --git a/ltp/fsx.c b/ltp/fsx.c
> index 06d08e4e..4da9d7d8 100644
> --- a/ltp/fsx.c
> +++ b/ltp/fsx.c
> @@ -1948,7 +1948,18 @@ test(void)
>  			size = log_entry.args[1];
>  			offset2 = log_entry.args[2];
>  			closeopen = !!(log_entry.flags & FL_CLOSE_OPEN);
> +
> +			/*
> +			 * Ignore unsupported keep_size flag even if it's specified
> +			 * by --replay-ops
> +			 */
>  			keep_size = !!(log_entry.flags & FL_KEEP_SIZE);
> +			if (!keep_size_calls && keep_size) {
> +				if (!quiet)
> +					prt("Ignoring unsupported keep_size flag\n");
> +				keep_size = 0;
> +			}
> +

I proposed a similar fix back in 2017[1], but Amir didn't like it
because my patch (and your patch) ignored keep_size silently (newly
generated ops list was different than the one fed to --replay-ops).

I proposed a new version in last Oct[2], which will record such ops as
skipped in the generated ops list, and Brian suggested to add a switch
to control the behavior if we want to fail fsx on unsupported ops. But I
didn't have time to work on it since then..

Would you like to help implement the switch Brian suggested based on my
second patch? Then I can apply my v2 patch & your switch patch.

Thanks!

Eryu

[1] https://spinics.net/lists/fstests/msg07953.html
[2] https://lore.kernel.org/fstests/20191023130934.GD59518@bfoster/
>  			goto have_op;
>  		}
>  		return 0;
> -- 
> 2.21.0
> 
>