Re: [PATCH] NFSD: Guard admin state-revocation walks with NFSD_NET_UP

On Mon, 2026-06-22 at 09:14 -0400, Chuck Lever wrote:
> On Mon, 2026-06-22 at 07:57 -0400, Jeff Layton wrote:
> > Can nn->nfsd_serv be non-NULL while the NFSv4 state tables are still
> > NULL?
> > ...
> > Would checking nn->nfsd_net_up instead of nn->nfsd_serv be more
> > accurate here, since nfsd_net_up is set at the tail of
> > nfsd_startup_net() exactly when the state tables become valid?
> 
> Yep, that's exactly the gap. The hunk you've quoted is 1/3 of the
> earlier "post-shutdown use-after-free" series, which only guarded the
> shutdown side of the window. nn->nfsd_serv is also set at service
> creation, before nfsd_startup_net() allocates conf_id_hashtbl, so the
> startup side was still exposed.
> 
> The thread you're replying to is the follow-on that closes it: it
> switches all three unlock paths (write_unlock_fs,
> nfsd_nl_unlock_filesystem_doit, nfsd_nl_unlock_export_doit) from
> nn->nfsd_serv to test_bit(NFSD_NET_UP, &nn->flags) and rewrites the
> walkers' Context: notes to match. NFSD_NET_UP is set at the tail of
> nfsd_startup_net() and cleared in nfsd_shutdown_net() after the tables
> are freed, so it brackets precisely the window you describe.
> 
> I assume your comment was meant for 1/3 of that series rather than
> this thread... but we've reached the same conclusion, so no action
> needed beyond this patch, correct?


Ahh right. I missed that that had been fixed in the other series. This
is fine then.

Reviewed-by: Jeff Layton <[email protected]>