Re: [PATCH net 2/2] selftests: add openvswitch selftest suite

From: Ilya Maximets
Date: Thu Oct 20 2022 - 12:33:35 EST


On 10/20/22 17:32, Aaron Conole wrote:
> Hi Ilya,
>
> Ilya Maximets <i.maximets@xxxxxxx> writes:
>
>> On 10/19/22 20:30, Aaron Conole wrote:
>>> Previous commit resolves a WARN splat that can be difficult to reproduce,
>>> but with the ovs-dpctl.py utility, it can be trivial. Introduce a test
>>> case which creates a DP, and then downgrades the feature set. This will
>>> include a utility 'ovs-dpctl.py' that can be extended to do additional
>>> work.
>>>
>>> Signed-off-by: Aaron Conole <aconole@xxxxxxxxxx>
>>> Signed-off-by: Kevin Sprague <ksprague0711@xxxxxxxxx>
>>> ---
>>> MAINTAINERS | 1 +
>>> tools/testing/selftests/Makefile | 1 +
>>> .../selftests/net/openvswitch/Makefile | 13 +
>>> .../selftests/net/openvswitch/openvswitch.sh | 216 +++++++++
>>> .../selftests/net/openvswitch/ovs-dpctl.py | 411 ++++++++++++++++++
>>> 5 files changed, 642 insertions(+)
>>> create mode 100644 tools/testing/selftests/net/openvswitch/Makefile
>>> create mode 100755 tools/testing/selftests/net/openvswitch/openvswitch.sh
>>> create mode 100644 tools/testing/selftests/net/openvswitch/ovs-dpctl.py
>>>
>>> diff --git a/MAINTAINERS b/MAINTAINERS
>>> index abbe88e1c50b..295a6b0fbe26 100644
>>> --- a/MAINTAINERS
>>> +++ b/MAINTAINERS
>>> @@ -15434,6 +15434,7 @@ S: Maintained
>>> W: http://openvswitch.org
>>> F: include/uapi/linux/openvswitch.h
>>> F: net/openvswitch/
>>> +F: tools/testing/selftests/net/openvswitch/
>>>
>>> OPERATING PERFORMANCE POINTS (OPP)
>>> M: Viresh Kumar <vireshk@xxxxxxxxxx>
>>
>> ...
>>
>>> +exit ${exitcode}
>>> diff --git a/tools/testing/selftests/net/openvswitch/ovs-dpctl.py
>>> b/tools/testing/selftests/net/openvswitch/ovs-dpctl.py
>>> new file mode 100644
>>> index 000000000000..791d76b7adcd
>>> --- /dev/null
>>> +++ b/tools/testing/selftests/net/openvswitch/ovs-dpctl.py
>>> @@ -0,0 +1,411 @@
>>> +#!/usr/bin/env python3
>>> +# SPDX-License-Identifier: GPL-2.0
>>> +
>>> +# Controls the openvswitch module. Part of the kselftest suite, but
>>> +# can be used for some diagnostic purpose as well.
>>> +
>>> +import logging
>>> +import multiprocessing
>>> +import socket
>>> +import struct
>>> +import sys
>>> +
>>> +try:
>>> + from libnl.attr import NLA_NESTED, NLA_STRING, NLA_U32, NLA_UNSPEC
>>> + from libnl.attr import nla_get_string, nla_get_u32
>>> + from libnl.attr import nla_put, nla_put_string, nla_put_u32
>>> + from libnl.attr import nla_policy
>>> +
>>> + from libnl.error import errmsg
>>> +
>>> + from libnl.genl.ctrl import genl_ctrl_resolve
>>> + from libnl.genl.genl import genl_connect, genlmsg_parse, genlmsg_put
>>> +
>>> + from libnl.handlers import nl_cb_alloc, nl_cb_set
>>> + from libnl.handlers import NL_CB_CUSTOM, NL_CB_MSG_IN, NL_CB_VALID
>>> + from libnl.handlers import NL_OK, NL_STOP
>>> +
>>> + from libnl.linux_private.netlink import NLM_F_ACK, NLM_F_DUMP
>>> + from libnl.linux_private.netlink import NLM_F_REQUEST, NLMSG_DONE
>>> +
>>> + from libnl.msg import NL_AUTO_SEQ, nlmsg_alloc, nlmsg_hdr
>>> +
>>> + from libnl.nl import NLMSG_ERROR, nl_recvmsgs_default, nl_send_auto
>>> + from libnl.socket_ import nl_socket_alloc, nl_socket_set_cb
>>> + from libnl.socket_ import nl_socket_get_local_port
>>> +except ModuleNotFoundError:
>>> + print("Need to install the python libnl3 library.")
>>
>>
>> Hey, Aaron and Kevin. Selftests sounds like a very important and
>> long overdue thing to add. Thanks for working on this!
>>
>> I have some worries about the libnl3 library though. It doesn't
>> seem to be maintained well. It it maintained by a single person,
>> it it was at least 3 different single persons over the last 7
>> years via forks. It didn't get any significant development done
>> since 2015 as well and no commits at all for a last 1.5 years.
>> It is not packaged by any major distributions.
>
> :-/ On my fedora:
>
> 11:12:24 aconole@RHTPC1VM0NT ~$ dnf search python3-libnl3
> Last metadata expiration check: 1 day, 0:25:11 ago on Wed 19 Oct 2022 10:47:21 AM EDT.
> ===================== Name Exactly Matched: python3-libnl3 =====================
> python3-libnl3.x86_64 : libnl3 binding for Python 3
>
>
> And I can use it:
>
> 11:18:39 aconole@RHTPC1VM0NT {(6a5c83bdd991...)} ~/git/linux/tools/testing/selftests/net/openvswitch$ sudo python3 ./ovs-dpctl.py show
> foop
> Lookups: Hit: 0 Missed: 0 Lost: 0
> Flows: 0
> Masks: Hit: 0 Total: 0
> Cache: Hit: 0
> Caches:
> Masks-cache: size: 256
> Port 0: foop (internal)
> 11:18:43 aconole@RHTPC1VM0NT {(6a5c83bdd991...)} ~/git/linux/tools/testing/selftests/net/openvswitch$ rpm -qa | grep python3-libnl3
> python3-libnl3-3.5.0-6.fc34.x86_64
> 11:19:01 aconole@RHTPC1VM0NT {(6a5c83bdd991...)} ~/git/linux/tools/testing/selftests/net/openvswitch$
>
> Was there some place you did not find it?

You're right, I missed that somehow. But this is not an
https://github.com/coolshou/libnl3 project. :)
These are python bindings for the C libnl library:

$ dnf info python3-libnl3
Available Packages
Name : python3-libnl3
Version : 3.7.0
Release : 1.fc36
Architecture : x86_64
Size : 153 k
Source : libnl3-3.7.0-1.fc36.src.rpm
Repository : updates
Summary : libnl3 binding for Python 3
URL : http://www.infradead.org/~tgr/libnl/
License : LGPLv2
Description : Python 3 bindings for libnl3

>
>> I'm talking about https://github.com/coolshou/libnl3 . Please,
>> correct me if that is not the right one. There are too many
>> libraies with the name libnl out there... That is also not a great
>> sign.
>
> Yes, this is the project.

Doensn't look like it...

> We did look at some of the ones you
> mentioned, but didn't find much.
>
> It is a sparse landscape of projects that provide netlink support in
> python.
>
>> The C library libnl (https://github.com/thom311/libnl) seems to
>> be well maintained in general. It has experimental python
>> bindings which are not really supported much. Python bindings
>> received only 2 actual code-changing commits in the last 7 years.
>> Both of them are just python 2/3 compatibility changes.
>> Maybe that is not that big of a deal since it's not really a
>> real python library, but a wrapper on top of a main C library.
>> However, I didn't find these python bindings to be packaged in
>> distributions. And they seem to be not available in pip as well.
>> So, building them is kind of a pain.
>
> Well, the python libnl3 should be installable via pip3. Ex:
>
> 11:27:15 aconole@RHTPC1VM0NT ~$ pip3 install libnl3
> Defaulting to user installation because normal site-packages is not writeable
> Collecting libnl3
> Using cached libnl3-0.3.0-py3-none-any.whl (89 kB)
> Installing collected packages: libnl3
> Successfully installed libnl3-0.3.0

And this is https://pypi.org/project/libnl3/, which is the
https://github.com/coolshou/libnl3 project. So, by installing
libnl3 via pip and installing python3-libnl3 from the fedora
you're getting two completely different libraries.

So, which one users should use?

I can't find python bindings for the C libnl (which is the
python3-libnl3 package) in pypi, so it can't be installed
with pip.


>
> So I guess that is worth something.
>
> At least on Fedora it is installable from distribution as well.
>
>> There is another option which is pyroute2. It positions itself
>> primarily as a netlink library and it does include an
>> pyroute2.netlink module indeed:
>> https://github.com/svinota/pyroute2/tree/master/pyroute2/netlink
>> See the __init__.py for usage examples.
>>
>> This one looks to me like the most trustworthy. It is actively
>> used by everyone in the python networking world, e.g. by OpenStack.
>> And it is actively developed and maintained unlike other
>> netlink-related python projects. It is also packaged in most of the
>> main distributions, so it's easy to install and use. Many people
>> already have it installed for other purposes.
>>
>> TBH, I didn't compare the functionality, but I'd expect that most
>> of the things we need are implemented.
>>
>> What do you think?
>
> We can certainly look at switching, but having a quick glance, it seems
> pyroute2 expects to provide the genl commands as well, so they would
> want us to create an ovs module in pyroute2 that includes all of the ovs
> family support. Of course, we can always do this just in our module,
> but I think it isn't the way pyroute2 project wants to be structured.
> More like a library that provides all the command functionality.

What I was thinking is to import pyroute2.netlink and the
pyroute2.netlink.generic and go from there. But I didn't
look too deep on how to actually implement the functionality.

The python bindings for the C libnl (python3-libnl3) sounds
like a fine option since they are actually packaged in
distributions (missed that in my initial reply). However,
the fact that you can not install them via pip and actually
you will install something but completely different is kind
of weird. This has to be at least better documented, so
users will know what to install and they will not try to use
pip for that.

>
>> On the other note, I'm not a real python developer, but the code
>> looks more like a C than a python even for me. Firstly, I'd say
>> that it would be great to maintain some coding style, e.g. by
>> checking with flake8 and/or black. See some issues/suggestions
>> provided by these tools below.
>
> Agreed. BTW, on the rhel8 system I developed on:
>
> [root@wsfd-netdev60 openvswitch]# flake8 ./ovs-dpctl.py
> [root@wsfd-netdev60 openvswitch]#
>
> So, I guess it is probably that I should have used a different system to
> do the flake8 checks.

Maybe the python version is different... I was running on f36
with python 3.10. Also, the list of defaults might be different.
flake8 doesn't use default ignore list if one is explicitly provided.

>
>> Secondly, we shouldd at least use argparse for argument parsing.
>> It's part of the standard library since python 3.2, so doens't
>> require any special dependencies to be installed.
>
> Okay - I can switch to argparse. TBH, I haven't kept up with python
> standard library for some time.

Well, 3.2 was released 11 years ago. :)

>
>> Some parts of the code can probably be re-written to be more
>> "pythonic" as well, but I won't dive into that for now. I didn't
>> review the code deep enough for that.
>
> I have difficulty sometimes understanding what it means to be "Real
> Python (tm)" - I don't plan to change things too much. I can certainly
> switch to using argparse, but unless you give something you want to
> change, I would not change anything.

I breifly looked through code and though I don't fully
understand what this piece supposed to do:

+ segment = hdrval.find(":")
+ if segment == -1:
+ segment = len(hdrval)
+ hdrver = int(hdrval[:segment], 0)
+ if len(hdrval[:segment]):
+ userfeatures = int(hdrval[:segment], 0)

but I have a strong feeling that this part can benefit
from use of hdrval.split(':').

I won't insist on that too much. :)

Best regards, Ilya Maximets.