On Thu, Apr 21, 2022 at 9:05 PM Christian Brauner <brauner@xxxxxxxxxx> wrote: > > On Thu, Apr 21, 2022 at 05:37:17PM +0200, Christoph Hellwig wrote: > > On Thu, Apr 21, 2022 at 11:35:13PM +0800, Zorro Lang wrote: > > > Sure, I won't do that wilfully, just try to ask how we can improve this > > > huge and 'keep growing' idmapped-mounts.c, not tend to remove the whole > > > idmapped-mount testing coverage :) > > > > It might just be time to split that file up into a few ones if there > > is a sensible split. I'll let Christian think about that, though. > > Yep, I agree. I think we need to at least rename it to reflect is vfs > generic nature and then split it into separate test binaries. > I'll think about a good approach. The majority of test cases still do require_fs_allow_idmap and from those who don't, most of them are variants for test cases that run with and without idmapped mounts and possibly also in_userns. And this new test case is no exception - there is still idmapping involved, just not idmapped-mounts. However you decide to break it up and/or rename the test binary (I am not sure you must split the binary - only the source files), I think we need to be more consistent about the groups that the tests that run this binary are in. 'perms' group is adequate, but adding to the 'idmapped' group and maybe also to a new 'userns' group would be useful. BTW, the tests that use src/nsexec should also belong to the userns group as does overlay/020, the only test that uses the 'unshare' tool. Thanks, Amir.
