On Sunday, 7 December 2025 12:34:24 CET Warner Losh wrote:
> On Sat, Dec 6, 2025, 10:12 AM Andrey Erokhin <language.lawyer@gmail.com>
> wrote:
> > On 03/12/2025 15:33, Christian Schoenebeck wrote:
> > > On Monday, 1 December 2025 19:00:53 CET Andrey Erokhin wrote:
[...]
> > > But for passthrough it is not of any use, is it?
> >
> > Prolly none, just a side effect of how it's implemented.
> > Can either make it an error when used with passthrough, or ignore them
> > (use default -1 value) when copying options to 9p fs context (with or
> > without a warning)
> >
> > > Also while it is very handy to have a short option name like "uid" and
> >
> > "gid", for the sake of long term progression and clarity an option name
> > like "default-uid" would be more appropriate.
> >
> > Or rather default_uid, to match other options style? But uid/gid also
> > kinda match fmode/dmode :\

Right, that would render it strange having default_uid/default_gid vs. fmod/
gmode when all of them actually mean default values.

OK, as fmode/dmode are already there, then let's stick to your initial
suggestion of just using uid/gid.

But similar to fmode/dmode it should be made clear on documentation level that
uid/gid are only useful for mapped security models.

> FreeBSD has a mode where you can build the image where the files in the
> filesystem are owned by the user with random permission bits, but the
> actual owners / modes are in an mtree formatted file. The nopriv imagers
> combine the two when making images. It would be nice to have p9 do a
> simular mapping for the guest so I can boot test these images more directly
> w/o the copyout to the "bootable image". The set the uid feature would
> help, true, but leaves me wanting more.

And a host level (not yet existing) tool like qemu-9p-chown, qemu-9p-chmod
would be less appropriate for your use case?