On 04/08/2023 05:27, Qu Wenruo wrote: > [...] > My concern is still about the "virtual" fsid part. > > If we go virtual fsid, there can be some unexpected problems. > > E.g. the /sys/fs/btrfs/<uuid>/ entry would be the new virtual one. > > And there may be some other problems like user space UUID detection of > mounted fs, thus I'm not 100% sure if this is a good idea. > > However I don't have any better solution either, so this may be the > least worst solution for now. > > Thanks, > Qu Hi Qu, thanks for your analysis! I think the virtual/spoofed fsid part is not without problems but I consider it to be less prone to unexpected issues than not. It's based on the metadata_uuid code, which is stable and present in btrfs for like 5 years. Also, we don't need to "corner-case" a lot of stuff to use that, which would be needed if we went to the pure dup fsid route. I tried that and it breaks in a lot of places, to which we require a lot of if conditionals (I even discussed that briefly in the last thread with you, about sysfs, remember?). So, despite not perfect, I agree with you that seems to be the least worse solution :) Cheers, Guilherme
