Re: [PATCH] staging: lustre: Fix unneeded byte-ordering cast

[Justin Skists]

On 20 March 2018 at 01:06, NeilBrown <neilb@xxxxxxxx> wrote:
> On Sat, Mar 17 2018, Justin Skists wrote:
>
>> Fix sparse warning:
>>
>>   CHECK   drivers/staging//lustre/lnet/lnet/acceptor.c
>> drivers/staging//lustre/lnet/lnet/acceptor.c:243:30: warning: cast to
>> restricted __le32
>>
>> LNET_PROTO_TCP_MAGIC, as a define, is already CPU byte-ordered when
>> compared to 'magic', so no need for a cast.
>>
>> Signed-off-by: Justin Skists <j.skists@xxxxxxxxx>
>> ---
>>  drivers/staging/lustre/lnet/lnet/acceptor.c | 2 +-
>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/drivers/staging/lustre/lnet/lnet/acceptor.c b/drivers/staging/lustre/lnet/lnet/acceptor.c
>> index fb478e20e204..13e981781b9a 100644
>> --- a/drivers/staging/lustre/lnet/lnet/acceptor.c
>> +++ b/drivers/staging/lustre/lnet/lnet/acceptor.c
>> @@ -240,7 +240,7 @@ lnet_accept(struct socket *sock, __u32 magic)
>>                       return -EPROTO;
>>               }
>>
>> -             if (magic == le32_to_cpu(LNET_PROTO_TCP_MAGIC))
>> +             if (magic == LNET_PROTO_TCP_MAGIC)
>>                       str = "'old' socknal/tcpnal";
>>               else
>>                       str = "unrecognised";
>
> This code is almost completely irrelevant (it just choose which error
> message to use when failing), but we may as well get it right and I
> cannot see why your change is a fix.

I admit that the change is trivial, and, in hindsight, the word fix is
a little "strong".
The rationale was that the if-statement, as it was, probably wouldn't work as
intented on big-endian systems.

I chose this sparse warning to test the waters, as it was an isolated change,
before I thought about proposing a bigger change: There are quite a few
sparse warning in regards to struct lnet_hdr with regards to __u32 vs. __le32
(etc.) restricted castings.

> I suspect a more correct fix would be to use
>   lnet_accept_magic(magic, LNET_PROTO_TCP_MAGIC)
> as the condition of the if().  This is consistent with other code that
> tests magic, and it is consistent with the general understanding that
> "magic" should be in host-byte-order for the peer which sent the
> message.
>
> Could you resubmit with that change?

I agree that your suggestion is a much better fix.

As Greg has already accepted the patch in question into staging-next,
would the correct course of action be for me to submit a new patch with a
"fixes" tag based on staging-next? Or would Greg prefer to drop the
previous one for a fresh v2?

Regards,
Justin.