On 12/4/19 11:26 AM, Christoph Hellwig wrote: > On Wed, Dec 04, 2019 at 10:24:32AM -0600, Eric Sandeen wrote: >> It'd be great to fix this universally in the kernel but it seems like >> that patch is in discussion for now, and TBH I don't see any real >> drawbacks to looping in mkfs - it would also solve the problem on any >> old kernel w/o the block layer change. > > The problem is that we throw out efficiency for no good reason. The reason, as I stated earlier, is that up to this day, no kernel properly handles this, and people are hitting this problem today. And nobody has shown a significant efficiency loss. (To be fair, nobody has shown that there isn't a loss either, but in my admittedly small test set I didn't see any meaningful overhead from the loop.) >> I'd propose that we go ahead w/ the mkfs change, and if/when the kernel >> handles this better, and it's reasonable to expect that we're running >> on a kernel where it can be interrupted, we could remove the mkfs loop >> at a later date if we wanted to. > > I'd rather not touch mkfs if a trivial kernel patch handles the issue. I don't think a trivial kernel patch to handle the issue even exists yet... -Eric
