On Mon, May 22, 2023 at 11:27:12PM +0200, Anthony Iliopoulos wrote:
> Commit a27e6e6f4c18 introduced xfs health checking on no-repair fuzz,
> which in turn requires scrub to be run before that. The health checks
> are done only if scrub returns with an error (which is expected as an
> indication that fuzzed metadata errors were picked up), but the code
> does not discern between xfs_scrub returning an error because of
> uncorrected metadata vs failing because the kernel does not support
> scrub at all.
>
> This causes all tests that do fuzzing with no-repair strategy to fail on
> kernels compiled without online scrub support (CONFIG_XFS_ONLINE_SCRUB).
>
> Skip scrub and health checks altogether, if the kernel does not support
> it, since the tests are still valuable.
>
> Fixes: a27e6e6f4c18 ("common: check xfs health after doing an online scrub")
> Signed-off-by: Anthony Iliopoulos <ailiop@xxxxxxxx>
At first I thought "Doesn't the _require_scrub in (say) xfs/466 prevent
this test from running at all?" But then I remembered that helper only
checks for the existence of the xfs_scrub program; it doesn't actually
verify that the kernel works.
I probably need to do more work making sure that the
online/offline/bothrepair tests actually go away if the kernel support
isn't there. In the meantime, I think this is correct so
Reviewed-by: Darrick J. Wong <djwong@xxxxxxxxxx>
--D
> ---
> common/fuzzy | 6 ++++++
> 1 file changed, 6 insertions(+)
>
> diff --git a/common/fuzzy b/common/fuzzy
> index 4365e7343e92..a78a354142df 100644
> --- a/common/fuzzy
> +++ b/common/fuzzy
> @@ -422,6 +422,12 @@ __scratch_xfs_fuzz_field_norepair() {
> return 1
> fi
>
> + # Skip scrub and health check if scrub is not supported
> + if ! _supports_xfs_scrub $SCRATCH_MNT $SCRATCH_DEV; then
> + __scratch_xfs_fuzz_unmount
> + return 0
> + fi
> +
> # Make sure online scrub will catch whatever we fuzzed
> __fuzz_notify "++ Detect fuzzed field (online)"
> _scratch_scrub -n -a 1 -e continue 2>&1
> --
> 2.40.0
>
