On Thu, Aug 29, 2024 at 5:05 PM Miklos Szeredi <miklos@xxxxxxxxxx> wrote:
>
> On Thu, 29 Aug 2024 at 16:39, Aleksandr Mikhalitsyn
> <aleksandr.mikhalitsyn@xxxxxxxxxxxxx> wrote:
>
> > So, your point is to set uid/gid to -1 for FUSE_{READ,WRITE,LOOKUP,RELEASE,...}?
>
> Yes. Not sure what will happen with those servers that check
> permissions based on these values, but my guess is it's still better
> than sending the unmapped value.
That's an interesting and a bit unexpected proposal. I have not
considered that option before,
because it feels like a bit of a radical change in the fuse protocol semantic,
while I was trying to be extremely careful and not make a revolution
in stable fuse protocol and code ;-)
We even have old commit:
c9582eb0ff7 ("fuse: Fail all requests with invalid uids or gids")
https://github.com/torvalds/linux/commit/c9582eb0ff7d2b560be60eafab29183882cdc82b
At the same time, if *you* propose that, I can't find a single word
against that. :)
You idea also solves problem with ->rename operation:
https://lore.kernel.org/all/20240815092429.103356-10-aleksandr.mikhalitsyn@xxxxxxxxxxxxx/
because if we just reuse existing fuse header fields and remap values
in them then we don't need
any extension for FUSE_RENAME2.
Let's think about it a bit more and if you confirm that we want to go
this way, then I'll rework my patches.
Kind regards,
Alex
>
> Thanks,
> Miklos
