Re: [PATCH 1/2] nfsd/sunrpc: add svc_rqst->rq_private pointer and remove rq_lease_breaker

[Chuck Lever]

On Tue, Jan 13, 2026, at 2:31 PM, Jeff Layton wrote:
> On Tue, 2026-01-13 at 14:21 -0500, Chuck Lever wrote:
>> 
>> On Tue, Jan 13, 2026, at 1:37 PM, Jeff Layton wrote:
>> > diff --git a/fs/nfsd/nfssvc.c b/fs/nfsd/nfssvc.c
>> > index eee8c3f4a251a3fae6e41679de0ec34c76caf198..8ce366c9e49220e8baf475c2e5f3424fedc1cec1 100644
>> > --- a/fs/nfsd/nfssvc.c
>> > +++ b/fs/nfsd/nfssvc.c
>> > @@ -900,6 +900,7 @@ nfsd(void *vrqstp)
>> > struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
>> > struct net *net = perm_sock->xpt_net;
>> > struct nfsd_net *nn = net_generic(net, nfsd_net_id);
>> > + struct nfsd_thread_local_info ntli = { };
>> > bool have_mutex = false;
>> >  
>> > /* At this point, the thread shares current->fs
>> > @@ -914,6 +915,8 @@ nfsd(void *vrqstp)
>> >  
>> > set_freezable();
>> >  
>> > + rqstp->rq_private = &ntli;
>> > +
>> > /*
>> > * The main request loop
>> > */
>> 
>> Thanks for tackling this one. Nits below...
>> 
>> This assumes sizeof(structure nfsd_thread_local_info) will always
>> be small enough that it is reasonable to keep on the stack. I
>> can't say that would be a good bet in the long run.
>> 
>> And we don't need the perfect reliability of not doing a dynamic
>> allocation here. If kmalloc(sizeof(struct nfsd_thread_local_info))
>> fails, the thread exits immediately, no harm.
>> 
>
> Not sure how much space Ben will need (if any).
>
> We certainly could have nfsd allocate this separately. I didn't see the
> point for something that is only a few bytes though.

If we are designing for today, another approach would be to set up
a BUILD_WARN_ON or other type of static build failure if this
structure grows larger than, say 256 bytes -- then add dynamic
allocation at that point.


>> Also, scatter-gather lists could not be stored directly in this
>> object because it is on the stack. If Ben wanted to stick a
>> 32-byte buffer in struct nfsd_thread_local_info to be used with
>> the crypto API, it would have to be a pointer to it, not the
>> buffer itself.
>> 
>
> *nod*
> -- 
> Jeff Layton <jlayton@xxxxxxxxxx>

-- 
Chuck Lever