Hi all, > On 23. 08. 24 23:59, NeilBrown wrote: > > On Fri, 23 Aug 2024, Petr Vorel wrote: > > > We discussed in v1 how to fix tests. Neil suggested to fix the test the way so > > > that it works on all kernels. As I note [1] > > > 1) either we give up on checking the new functionality still works (if we > > > fallback to old behavior) > > I don't understand. What exactly do you mean by "the new > > functionality". > > As I understand it there is no new functionality. All there was was and > > information leak between network namespaces, and we stopped the leak. > > Do you consider that to be new functionality? Thanks Martin for jumping in. I hoped I was clear, but obviously not. Following are the questions for kernel maintainers and developers. I put my opinion, but it's really up to you what you want to have tested. > The new functionality is that the patches add a new file to network > namespaces: /proc/net/rpc/nfs. This file did not exist outside the root > network namespace at least on some of the kernels where we still need to run > this test. So the question is: How aggressively do we want to enforce > backporting of these NFS patches into distros with older kernels? > We have 3 options how to fix the test depending on the answer: > 1) Don't enforce at all. We'll check whether /proc/net/rpc/nfs exists in the > client namespace and read it only if it does. Otherwise we'll fall back on > the global file. 1) is IMHO the worst case because it's not testing patch gets reverted. > 2) Enforce aggressively. We'll hardcode a minimal kernel version into the > test (e.g. v5.4) and if the procfile doesn't exist on any newer kernel, it's > a bug. I would prefer 2), which is the usual LTP approach (do not hide bugs, we even fail on upstream kernel WONTFIX [1], why we should refuse the policy here?). Whichever older LTS upstream kernel gets fixed would drive the line where new functionality is requested (currently v5.14, I suppose at least 5.10 will also be fixed). LTP also has a way to specify enterprise distro kernel version if older enterprise kernel also gets fixed (this should not be needed, but it'd be possible). > 3) Enforce on new kernels only. We'll set a hard requirement for kernel > v6.9+ as in option 2) and check for existence of the procfile on any older > kernels as in option 1). Due way to specify enterprise distro kernel version and upstream kernel testing expecting people update to the latest stable/LTS we should not worry much about people with older kernels. Kind regards, Petr [1] https://github.com/linux-test-project/ltp/blob/master/testcases/kernel/syscalls/ustat/ustat01.c#L48-L49
