On Mon, 23 Oct 2023 at 17:40, Dave Chinner <david@xxxxxxxxxxxxx> wrote:
> >
> > Maybe we don't even need a mode, and could just decide that atime
> > updates aren't i_version updates at all?
>
> We do that already - in memory atime updates don't bump i_version at
> all. The issue is the rare persistent atime update requests that
> still happen - they are the ones that trigger an i_version bump on
> XFS, and one of the relatime heuristics tickle this specific issue.
Yes, yes, but that's what I kind of allude to - maybe people still
want the on-disk atime updates, but do they actually want the
i_version updates just because they want more aggressive atime
updates?
> > Or maybe i_version can update, but callers of getattr() could have two
> > bits for that STATX_CHANGE_COOKIE, one for "I care about atime" and
> > one for others, and we'd pass that down to inode_query_version, and
> > we'd have a I_VERSION_QUERIED and a I_VERSION_QUERIED_STRICT, and the
> > "I care about atime" case ould set the strict one.
>
> This makes correct behaviour reliant on the applicaiton using the
> query mechanism correctly. I have my doubts that userspace
> developers will be able to understand the subtle difference between
> the two options and always choose correctly....
I don't think we _have_ a user space interface.
At least the STATX_CHANGE_COOKIE bit isn't exposed to user space at
all. Not in the uapi headers, but not even in xstat():
/* STATX_CHANGE_COOKIE is kernel-only for now */
tmp.stx_mask = stat->result_mask & ~STATX_CHANGE_COOKIE;
So the *only* users of STATX_CHANGE_COOKIE seem to be entirely
in-kernel, unless there is something I'm missing where somebody uses
i_version through some other interface (random ioctl?).
End result: splitting STATX_CHANGE_COOKIE into a "I don't care about
atime" and a "give me all change events" shouldn't possibly break
anything that I can see.
The only other uses of inode_query_iversion() seem to be the explicit
directory optimizations (ie the "I'm caching the offset and don't want
to re-check that it's valid unless required, so give me the inode
version for the directory as a way to decide if I need to re-check"
thing).
And those *definitely* don't want i_version updates on any atime updates.
There might be some other use of inode_query_iversion() that I missed,
of course. But from a quick look, it really looks to me like we can
relax our i_version updates.
Linus
