On Thu, Jul 3, 2025 at 4:01 AM Olivier Certner <[email protected]> wrote:
>
> > It is not the right usage of the MNT_ flags. They are for feature
> > controlling, not for the feature reporting.
>
> That is not the case today (MNT_LOCAL, MNT_QUOTA, MNT_ROOTFS, MNT_USER, 
> etc.). I also would like this to become true at some point.
>
> > VOP_PATHCONF() is exactly for reporting some mnt features.
>
> No, it is to report features that depend on a path.  It can be (ab)used to 
> report mount features (calling it on any file inside the mount), but that 
> obfuscates the fact that an information is per-mount, whereas 
> statfs()/statvfs() are exactly designed for per-mount information.
Note that, for a NFSv4 client mount, support for hidden/system could
change somewhere in
the mounted server tree, since for NFSv4 the tree can consist of multiple file
systems of different types. For example:
If an ext2 fs was mounted under a ufs file system, the hidden/system support
would go away when the mount point is crossed into the ext2 subtree.

So, I think pathconf is the best way to go (and avoids using up flag bits).

rick

>
> > Also, this namespace should not be used frivolously, we already had to
> > extend flags to 64bit, and again we are not too far from exhausting it.
>
> I don't disagree.  That said, at some point, we'll have to dodge the bullet, 
> including separating control flags from report ones in the MNT_* namespace 
> (and provide compat' functionality), and probably extending control flags 
> with a second 64-bit field.
>
> I suspected from the start that that was the main reason behind you 
> recommending VOP_PATHCONF().  My point here is that it is a slightly 
> non-optimal design choice (as explained just above) that will stay and was 
> made out of a practical issue that anyway will have to be solved at some 
> point, and I find this a bit unfortunate.  Solving the issue now is certainly 
> much more work, and in any case I didn't mean to object to not doing it now.  
> But if someone(tm) wants to do it now, that would be great (perhaps me, but 
> not sure yet).
>
> Regards.
>
> --
> Olivier Certner

Reply via email to