Re: [PATCH] bpf: return EOPNOTSUPP when JIT is needed and not possible

[Ido Schimmel]

On Thu, Dec 09, 2021 at 11:05:18AM -0800, John Fastabend wrote:
> Thadeu Lima de Souza Cascardo wrote:
> > When a CBPF program is JITed and CONFIG_BPF_JIT_ALWAYS_ON is enabled, and
> > the JIT fails, it would return ENOTSUPP, which is not a valid userspace
> > error code.  Instead, EOPNOTSUPP should be returned.
> > 
> > Fixes: 290af86629b2 ("bpf: introduce BPF_JIT_ALWAYS_ON config")
> > Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@xxxxxxxxxxxxx>
> > ---
> >  kernel/bpf/core.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/kernel/bpf/core.c b/kernel/bpf/core.c
> > index de3e5bc6781f..5c89bae0d6f9 100644
> > --- a/kernel/bpf/core.c
> > +++ b/kernel/bpf/core.c
> > @@ -1931,7 +1931,7 @@ struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err)
> >  		fp = bpf_int_jit_compile(fp);
> >  		bpf_prog_jit_attempt_done(fp);
> >  		if (!fp->jited && jit_needed) {
> > -			*err = -ENOTSUPP;
> > +			*err = -EOPNOTSUPP;
> >  			return fp;
> >  		}
> >  	} else {
> > -- 
> > 2.32.0
> > 
> 
> It seems BPF subsys returns ENOTSUPP in multiple places. This fixes one
> paticular case and is user facing. Not sure we want to one-off fix them
> here creating user facing changes over multiple kernel versions. On the
> fence with this one curious to see what others think. Haven't apps
> already adapted to the current convention or they don't care?

Similar issue was discussed in the past. See:
https://lore.kernel.org/netdev/20191204.125135.750458923752225025.davem@xxxxxxxxxxxxx/