Re: [RFC] metadata updates vs. fetches (was Re: [PATCH v4] fs: Fix data race in inode_set_ctime_to_ts)

From: Linus Torvalds
Date: Sun Nov 24 2024 - 20:26:53 EST


On Sun, 24 Nov 2024 at 17:15, Matthew Wilcox <willy@xxxxxxxxxxxxx> wrote:
>
> I literally said that.

Bah, I misunderstood you.

Then yes, if all the writers are always in order, and you don't
actually care about exact time matching but only ordering, I guess it
might work.

But since you need all the same barriers that you would need for just
doing it right with a seqcount, it's not much of an advantage, and it
doesn't give you consistency for any other kind of action.

Linus