On Wed, 28 Feb 2024 at 14:16, Bernd Schubert <bernd.schubert@xxxxxxxxxxx> wrote:
> Could you still apply your previous patch? I think that definitely makes
> sense as well.
Half of it is trivial (s/make_bad_inode/fuse_make_bad/).
The other half is probably broken in that form (see 775c5033a0d1
("fuse: fix live lock in fuse_iget()")).
Things get complicated, though, because the root of submounts can have
nodeid != FUSE_ROOT_ID, yet they should have the same rules as the
original root. Needs more thought...
> I think what we also need is notification message from kernel to server
> that it does something wrong - instead of going to kernel logs, such
> messages should go to server side logs.
That's generally not possible. We can return -EIO to the application,
but not the server. In this case I think it's better to just fall
back to old behavior of ignoring the generation.
Thanks,
Miklos
