> On Jul 11, 2024, at 12:00 PM, Benjamin Coddington <bcodding@xxxxxxxxxx> wrote:
>
> On 11 Jul 2024, at 11:52, Chuck Lever III wrote:
>
>>> On Jul 11, 2024, at 11:48 AM, Benjamin Coddington <bcodding@xxxxxxxxxx> wrote:
>>>
>>> On 11 Jul 2024, at 11:28, Chuck Lever wrote:
>>>
>>>> On Thu, Jul 11, 2024 at 11:24:01AM -0400, Benjamin Coddington wrote:
>>>>> The GSS routine errors are values, not flags.
>>>>
>>>> My reading of kernel and user space GSS code is that these are
>>>> indeed flags and can be combined. The definitions are found in
>>>> include/linux/sunrpc/gss_err.h:
>>>>
>>>> To wit:
>>>>
>>>> 116 /*
>>>> 117 * Routine errors:
>>>> 118 */
>>>> 119 #define GSS_S_BAD_MECH (((OM_uint32) 1ul) << GSS_C_ROUTINE_ERROR_OFFSET)
>>>> 120 #define GSS_S_BAD_NAME (((OM_uint32) 2ul) << GSS_C_ROUTINE_ERROR_OFFSET)
>>>
>>> I read this as just values shifted left by a constant.
>>>
>>> No where in-kernel are they bitwise combined.
>>
>> The kernel gets GSS status values from user space code too.
>>
>>
>>> I noticed this problem in practice
>>> while reading the tracepoint output from corrupted GSS hash routines.
>>
>> Can you describe the problem?
>
> It was a week ago or so, and I don't have the test setup any longer, but the
> tracepoint would not print the actual error returned, rather the bitwise
> combination of that error.
>
> Look closer at the values - it makes no sense that these are bits, else
> GSS_S_BAD_NAMETYPE is the same as GSS_S_BAD_MECH|GSS_S_BAD_NAME.
Understood. Please add:
Fixes: 0c77668ddb4e ("SUNRPC: Introduce trace points in rpc_auth_gss.ko")
Reviewed-by: Chuck Lever <chuck.lever@xxxxxxxxxx>
--
Chuck Lever
