lkml.org 
[lkml]   [2024]   [Apr]   [27]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
SubjectRe: [PATCH net 0/4] net: qede: avoid overruling error codes
From
Hi Simon,

Thank you for your review effort.

On 4/27/24 11:48 AM, Simon Horman wrote:
> On Fri, Apr 26, 2024 at 09:12:22AM +0000, Asbjørn Sloth Tønnesen wrote:
>> This series fixes the qede driver, so that
>> qede_parse_flow_attr() and it's subfunctions
>> doesn't get their error codes overruled
>> (ie. turning -EOPNOTSUPP into -EINVAL).
>>
>> ---
>> I have two more patches along the same lines,
>> but they are not yet causing any issues,
>> so I have them destined for net-next.
>> (those are for qede_flow_spec_validate_unused()
>> and qede_flow_parse_ports().)
>>
>> After that I have a series for converting to
>> extack + the final one for validating control
>> flags.
>
> Hi,
>
> I'm fine with these patches so far as the code changes go.
> But it is not clear to me that they are fixing a bug.
>
> If so, I think some explanation should go in the commit messages.
> If not, I think these should be targeted at net-next
> (and not have Fixes tags.

Since I don't have the hardware I didn't try to construct commands, showing
the wrong error code being returned. I could make up some hypothetical commands,
and simulate how they would error. I assumed that the bug, was clear based on
the list of possible return values for each function.

As an example, in qede_parse_flow_attr() it validates dissector->used_keys,
and if an unsupported FLOW_DISSECTOR_KEY_* is set, then ede_parse_flow_attr()
returns -EOPNOTSUPP, which is returned to qede_add_tc_flower_fltr(),
and only check for non-zero, and since -EOPNOTSUPP is non zero,
then it returns -EINVAL. So if you try to match on a vlan tag,
then FLOW_DISSECTOR_KEY_VLAN would be set, and cause a -EOPNOTSUPP
to be returned, which then gets converted into a -EINVAL.

All drivers generally returns -EOPNOTSUPP in their used_keys checks, and
this driver clearly intended to do that as well.

The -EINVAL override was introduced in the same commit as the above check,
so it was broken from the start.

Another example is 319a1d19471e (blamed in 4th patch), Jiri added
a call to flow_action_basic_hw_stats_types_check() across multiple drivers,
and since -EINVAL was returned only a few lines above, then he assumed
that he could just return -EOPNOTSUPP, but that return value gets overruled
into a -EINVAL. It is clear from the commit that Jiri intended to return
-EOPNOTSUPP, but this part of the driver didn't follow the principle of
least astonishment, so that function could only fail with -EINVAL.

I think it's a bug, when another error code is returned than the one that
was clearly intended, but it's properly a low impact one.


> Also, if you do end posting a v2, blamed, is misspelt several
> times in commit messages.

Sorry about that, will fix that if a v2 turns out to be needed.

--
Best regards
Asbjørn Sloth Tønnesen
Network Engineer
Fiberby - AS42541

\
 
 \ /
  Last update: 2024-05-27 18:06    [W:0.063 / U:0.024 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site