Query regarding multiple comparisons

When you perform a permutation test on a two-factor within-subject repeated measures ANOVA, how should you relabel the category information?

For example, let's say you're using a 3x2 treatment by condition experiment, with 10 subjects.

Do you relabel completely randomly, or do you relabel within each subject/treatment/condition category?

As an example of what I mean, if you relabel completely randomly, that would mean you can end up with all data from subjects 1:5 being condition 1, and all data from subjects 6:10 as condition 2.

Should you relabel randomly? Should you relabel only within subjects? Should you relabel within condition for treatments, and within treatments for condition (two vectors of category information are needed to perform the ANOVA)? A mix of these?