On Sat, Nov 23, 2024 at 01:06:14PM +0100, Christian Brauner wrote: > On Fri, Nov 22, 2024 at 09:21:58PM -0800, Linus Torvalds wrote: > > On Fri, 22 Nov 2024 at 01:57, Amir Goldstein <amir73il@xxxxxxxxx> wrote: > > > > > > - Introduction and use of revert/override_creds_light() helpers, that were > > > suggested by Christian as a mitigation to cache line bouncing and false > > > sharing of fields in overlayfs creator_cred long lived struct cred copy. > > > > So I don't actively hate this, but I do wonder if this shouldn't have > > been done differently. > > > > In particular, I suspect *most* users of override_creds() actually > > wants this "light" version, because they all already hold a ref to the > > cred that they want to use as the override. > > > > We did it that safe way with the extra refcount not because most > > people would need it, but it was expected to not be a big deal. > > > > Now you found that it *is* a big deal, and instead of just fixing the > > old interface, you create a whole new interface and the mental burden > > of having to know the difference between the two. > > > So may I ask that you look at perhaps just converting the (not very > > many) users of the non-light cred override to the "light" version? > > I think that could be a good idea in general. > > But I have to say I'm feeling a bit defensive after having read your > message even though I usually try not to. :) It was just pointed out to me that this was written like I'm not reading you messages - which is obviously not the case. What I means it that I usually try to not be defensive when valid criticism is brought up. :)
