On 8/5/19 12:56 PM, Holger Hoffstätte wrote: > On 8/5/19 12:31 PM, Hans van Kranenburg wrote: >> On 8/5/19 12:20 PM, Holger Hoffstätte wrote: >>> On 8/2/19 6:10 PM, Josef Bacik wrote: >>>> In testing block group removal it's sometimes handy to be able to >>>> create >>>> block groups on demand. Add an ioctl to allow us to force allocation >>>> from userspace. >>> >>> Gave this a try and it works as advertised, though I noticed that the >>> redundancy level is ignored, e.g. adding a single metadata chunk will >>> add a new "single" chunk even when the metadata level is dup. >>> Doing a balance -mconvert dup,soft fixes that right up, but it's IMHO >>> unexpected. Can you put a cherry on top and create the new chunk >>> according >>> to its dup level? >> >> Looking at the code, you should get -EINVAL when you specify anything >> else than single? (because of the != comparisons). >> >> If this gets in, and is available without doing some special debug style >> kernel build, then it will (tm) be (ab)used by users in the future for >> things we didn't imagine today. So, in that case, it makes sense to be >> able to specify any valid combination of flags (type+profile), like >> indeed METADATA|DUP or whatever. > > ..and that's precisely why it (IMHO) should simply do The Right Thing > by default instead of giving people more footguns with unexpected > behaviour. Just observe the existing profile in the kernel ioctl and > act accordingly, no need for more user-supplied flags. Just don't.
You're right. That's better than my suggestion. +1 Hans
