On Tue, 29 Oct 2024, Jeff Layton wrote: > On Mon, 2024-10-28 at 09:04 +1100, NeilBrown wrote: > > refcount_t, by design, checks for unwanted situations and provides > > warnings. It is rarely useful to have explicit warnings with refcount > > usage. > > > > In this case we have an explicit warning if a refcount_t reaches zero > > when decremented. Simply using refcount_dec() will provide a similar > > warning and also mark the refcount_t as saturated to avoid any possible > > use-after-free. > > > > This patch drops the warning and uses refcount_dec() instead of > > refcount_dec_and_test(). > > > > Signed-off-by: NeilBrown <neilb@xxxxxxx> > > --- > > fs/nfsd/filecache.c | 2 +- > > 1 file changed, 1 insertion(+), 1 deletion(-) > > > > diff --git a/fs/nfsd/filecache.c b/fs/nfsd/filecache.c > > index 1408166222c5..c16671135d17 100644 > > --- a/fs/nfsd/filecache.c > > +++ b/fs/nfsd/filecache.c > > @@ -1050,7 +1050,7 @@ nfsd_file_do_acquire(struct svc_rqst *rqstp, struct net *net, > > * the last one however, since we should hold another. > > */ > > if (nfsd_file_lru_remove(nf)) > > - WARN_ON_ONCE(refcount_dec_and_test(&nf->nf_ref)); > > + refcount_dec(&nf->nf_ref); > > The existing code threw a warning when the counter reached 0. Your > change will make the potential warning fire later, after we try to put > the last reference and the counter goes to -1. That's probably fine as > it should happen later in this function either way. The code in refcount_dec() contains: if (unlikely(old <= 1)) refcount_warn_saturate(r, REFCOUNT_DEC_LEAK); NeilBrown > > > goto wait_for_construction; > > } > > > > > > base-commit: 7fa861d5df402b2327f45e0240c1b842f71fec11 > > Reviewed-by: Jeff Layton <jlayton@xxxxxxxxxx> >
