On Thu, May 09, 2024 at 08:19:40PM -0600, Jens Axboe wrote:
> On 5/9/24 7:14 PM, Ming Lei wrote:
> > Support to inject result for NOP so that we can inject failure from
> > userspace. It is very helpful for covering failure handling code in
> > io_uring core change.
> >
> > With nop flags, it could be possible to add more test feature for NOP in
> > future, but the NOP behavior of direct completion has to be kept.
> >
> > Cleared NOP SQE is required, look both liburing and Rust io-uring crate
> > clears SQE, and it shouldn't be one big deal for raw, especially it is
> > just NOP.
>
> I think this implementation looks fine, but probably would be best if
> you first had a prep patch that adds nop_flags to io_uring_sqe, and
> checks it in io_nop_prep() and fails if it's non-zero. Then we can mark
> that for stable, rather than need to do the whole thing.
>
> Then patch 2 adds the actual meat of this patch, and now adds the proper
> check in io_nop_prep() for whether any unknown flags are set.
>
> > diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
> > index 922f29b07ccc..5db3a209b302 100644
> > --- a/include/uapi/linux/io_uring.h
> > +++ b/include/uapi/linux/io_uring.h
> > @@ -48,7 +48,10 @@ struct io_uring_sqe {
> > __u32 optname;
> > };
> > };
> > - __u32 len; /* buffer size or number of iovecs */
> > + union {
> > + __u32 len; /* buffer size or number of iovecs */
> > + __s32 result; /* for NOP to inject result only */
> > + };
> > union {
>
> And I'd drop that, just use 'len' throughout.
>
> Rest of the patch looks fine as-is.
OK, will do both two in V2.
Thanks,
Ming
