Upstreaming to net-next

Upstreaming to net-next

[Mat Martineau]

[-- Attachment #1: Type: text/plain, Size: 2460 bytes --]


As Matthieu noted in today's meeting, we need to figure out what goes 
upstream to netdev this week:

On Wed, 5 Jan 2022, Matthieu Baerts wrote:

> Patches to send to netdev:
>    - net:
>        - probably best to send:
>            - mptcp: fix opt size when sending DSS + MP_FAIL
>            - mptcp: fix a DSS option writing error → but we might need
> to change
>        - But if they are accepted after the last PR from Net maintainer
> to Linus would delay stuff.

I noticed that Jakub sent a "Networking for 5.16-final" pull request to 
Linus today, so the 5.16 ship has sailed.

>        - So if we don't want to rush to look at the tests, we can also
> send to net-next for the future -net (to be check which tree is merged
> in which one first?

The next pull requests from Jakub to Linus will be from net-next, so 
that's the fastest route to Linus' tree for fixes that will be added to 
the stable tree.


>
>    - net-next:
>        - patches from Paolo could go there, fixing issues, see GH
> issues/250:
>            - *@Paolo*: OK for you?
>
>        - + checksum fixes probably
>
>        - best to skip MP_FAIL series for now as we don't have a stable
> selftest yet
>
>        - and no need to send Mat's series as it is linked to new
> features in the Netlink PM part, not ready yet

Here are the commits that look ready (commit ids from 
export/20220105T054758):

Paolo:

9f6720406819 mptcp: keep snd_una updated for fallback socket
95bc19b21e00 mptcp: implement fastclose xmit path
37b1cee6f04f mptcp: full disconnect implementation
916957cdccd6 mptcp: cleanup accept and poll
6358b829913e mptcp: implement support for user-space disconnect
adba2e970980 selftests: mptcp: add disconnect tests
afef858f5fa4 mptcp: fix per socket endpoint accounting
9caba598fe7a mptcp: clean-up MPJ option writing
f584047ecf9a mptcp: keep track of local endpoint still available for each msk
fc51a4526843 mptcp: do not block subflows creation on errors
893d4cb7ccb9 selftests: mptcp: add tests for subflow creation failure
128e78db18bf mptcp: cleanup MPJ subflow list handling
e470b365da5f mptcp: avoid atomic bit manipulation when possible

3cdd8e644985 selftests: mptcp: more stable join tests-cases

Geliang:

e493c8de5c58 mptcp: change the parameter of __mptcp_make_csum
b79d45e020fd mptcp: reuse __mptcp_make_csum in validate_data_csum


Sound good? I may have to send in two batches so that patch count is not 
too large.

--
Mat Martineau
Intel

Re: Upstreaming to net-next

[Geliang Tang]

On Wed, Jan 05, 2022 at 05:54:12PM -0800, Mat Martineau wrote:
> 
> As Matthieu noted in today's meeting, we need to figure out what goes
> upstream to netdev this week:
> 
> On Wed, 5 Jan 2022, Matthieu Baerts wrote:
> 
> > Patches to send to netdev:
> >    - net:
> >        - probably best to send:
> >            - mptcp: fix opt size when sending DSS + MP_FAIL
> >            - mptcp: fix a DSS option writing error → but we might need
> > to change
> >        - But if they are accepted after the last PR from Net maintainer
> > to Linus would delay stuff.
> 
> I noticed that Jakub sent a "Networking for 5.16-final" pull request to
> Linus today, so the 5.16 ship has sailed.
> 
> >        - So if we don't want to rush to look at the tests, we can also
> > send to net-next for the future -net (to be check which tree is merged
> > in which one first?
> 
> The next pull requests from Jakub to Linus will be from net-next, so that's
> the fastest route to Linus' tree for fixes that will be added to the stable
> tree.
> 
> 
> > 
> >    - net-next:
> >        - patches from Paolo could go there, fixing issues, see GH
> > issues/250:
> >            - *@Paolo*: OK for you?
> > 
> >        - + checksum fixes probably
> > 
> >        - best to skip MP_FAIL series for now as we don't have a stable
> > selftest yet
> > 
> >        - and no need to send Mat's series as it is linked to new
> > features in the Netlink PM part, not ready yet
> 
> Here are the commits that look ready (commit ids from
> export/20220105T054758):
> 
> Paolo:
> 
> 9f6720406819 mptcp: keep snd_una updated for fallback socket
> 95bc19b21e00 mptcp: implement fastclose xmit path

Hi Mat,

There's a squash-to patch for this commit in the ML, ([mptcp-next,
v8,1/8] Squash to "mptcp: implement fastclose xmit path"). It works
on my test and looks good to me.

Could we squash it to "mptcp: implement fastclose xmit path", then
send them together to net-next if this squash-to patch looks good
to you too? Or could we send this commit to net-next in the next
turn next week?

WDYT?

Thanks,
-Geliang

> 37b1cee6f04f mptcp: full disconnect implementation
> 916957cdccd6 mptcp: cleanup accept and poll
> 6358b829913e mptcp: implement support for user-space disconnect
> adba2e970980 selftests: mptcp: add disconnect tests
> afef858f5fa4 mptcp: fix per socket endpoint accounting
> 9caba598fe7a mptcp: clean-up MPJ option writing
> f584047ecf9a mptcp: keep track of local endpoint still available for each msk
> fc51a4526843 mptcp: do not block subflows creation on errors
> 893d4cb7ccb9 selftests: mptcp: add tests for subflow creation failure
> 128e78db18bf mptcp: cleanup MPJ subflow list handling
> e470b365da5f mptcp: avoid atomic bit manipulation when possible
> 
> 3cdd8e644985 selftests: mptcp: more stable join tests-cases
> 
> Geliang:
> 
> e493c8de5c58 mptcp: change the parameter of __mptcp_make_csum
> b79d45e020fd mptcp: reuse __mptcp_make_csum in validate_data_csum
> 
> 
> Sound good? I may have to send in two batches so that patch count is not too
> large.
> 
> --
> Mat Martineau
> Intel

Re: Upstreaming to net-next

[Paolo Abeni]

On Wed, 2022-01-05 at 17:54 -0800, Mat Martineau wrote:
> As Matthieu noted in today's meeting, we need to figure out what goes 
> upstream to netdev this week:
> 
> On Wed, 5 Jan 2022, Matthieu Baerts wrote:
> 
> > Patches to send to netdev:
> >    - net:
> >        - probably best to send:
> >            - mptcp: fix opt size when sending DSS + MP_FAIL
> >            - mptcp: fix a DSS option writing error → but we might need
> > to change
> >        - But if they are accepted after the last PR from Net maintainer
> > to Linus would delay stuff.
> 
> I noticed that Jakub sent a "Networking for 5.16-final" pull request to 
> Linus today, so the 5.16 ship has sailed.
> 
> >        - So if we don't want to rush to look at the tests, we can also
> > send to net-next for the future -net (to be check which tree is merged
> > in which one first?
> 
> The next pull requests from Jakub to Linus will be from net-next, so 
> that's the fastest route to Linus' tree for fixes that will be added to 
> the stable tree.

I'm a bit lost about all the above. Not sure if due to lack of mtg
background, lack of coffee, or lack of time to digest the above ;)


> 
> > 
> >    - net-next:
> >        - patches from Paolo could go there, fixing issues, see GH
> > issues/250:
> >            - *@Paolo*: OK for you?
> > 
> >        - + checksum fixes probably
> > 
> >        - best to skip MP_FAIL series for now as we don't have a stable
> > selftest yet
> > 
> >        - and no need to send Mat's series as it is linked to new
> > features in the Netlink PM part, not ready yet
> 
> Here are the commits that look ready (commit ids from 
> export/20220105T054758):
> 
> Paolo:
> 
> 9f6720406819 mptcp: keep snd_una updated for fallback socket
> 95bc19b21e00 mptcp: implement fastclose xmit path
> 37b1cee6f04f mptcp: full disconnect implementation
> 916957cdccd6 mptcp: cleanup accept and poll
> 6358b829913e mptcp: implement support for user-space disconnect
> adba2e970980 selftests: mptcp: add disconnect tests
> afef858f5fa4 mptcp: fix per socket endpoint accounting
> 9caba598fe7a mptcp: clean-up MPJ option writing
> f584047ecf9a mptcp: keep track of local endpoint still available for each msk
> fc51a4526843 mptcp: do not block subflows creation on errors
> 893d4cb7ccb9 selftests: mptcp: add tests for subflow creation failure
> 128e78db18bf mptcp: cleanup MPJ subflow list handling
> e470b365da5f mptcp: avoid atomic bit manipulation when possible
> 
> 3cdd8e644985 selftests: mptcp: more stable join tests-cases

... anyhow this LGTM for net-next.

Thanks!

Paolo

Re: Upstreaming to net-next

[Matthieu Baerts]

Hi Geliang,

On 06/01/2022 03:17, Geliang Tang wrote:
> On Wed, Jan 05, 2022 at 05:54:12PM -0800, Mat Martineau wrote:
>>
>> As Matthieu noted in today's meeting, we need to figure out what goes
>> upstream to netdev this week:
>>
>> On Wed, 5 Jan 2022, Matthieu Baerts wrote:
>>
>>> Patches to send to netdev:
>>>    - net:
>>>        - probably best to send:
>>>            - mptcp: fix opt size when sending DSS + MP_FAIL
>>>            - mptcp: fix a DSS option writing error → but we might need
>>> to change
>>>        - But if they are accepted after the last PR from Net maintainer
>>> to Linus would delay stuff.
>>
>> I noticed that Jakub sent a "Networking for 5.16-final" pull request to
>> Linus today, so the 5.16 ship has sailed.
>>
>>>        - So if we don't want to rush to look at the tests, we can also
>>> send to net-next for the future -net (to be check which tree is merged
>>> in which one first?
>>
>> The next pull requests from Jakub to Linus will be from net-next, so that's
>> the fastest route to Linus' tree for fixes that will be added to the stable
>> tree.
>>
>>
>>>
>>>    - net-next:
>>>        - patches from Paolo could go there, fixing issues, see GH
>>> issues/250:
>>>            - *@Paolo*: OK for you?
>>>
>>>        - + checksum fixes probably
>>>
>>>        - best to skip MP_FAIL series for now as we don't have a stable
>>> selftest yet
>>>
>>>        - and no need to send Mat's series as it is linked to new
>>> features in the Netlink PM part, not ready yet
>>
>> Here are the commits that look ready (commit ids from
>> export/20220105T054758):
>>
>> Paolo:
>>
>> 9f6720406819 mptcp: keep snd_una updated for fallback socket
>> 95bc19b21e00 mptcp: implement fastclose xmit path
> 
> Hi Mat,
> 
> There's a squash-to patch for this commit in the ML, ([mptcp-next,
> v8,1/8] Squash to "mptcp: implement fastclose xmit path"). It works
> on my test and looks good to me.
> 
> Could we squash it to "mptcp: implement fastclose xmit path", then
> send them together to net-next if this squash-to patch looks good
> to you too? Or could we send this commit to net-next in the next
> turn next week?

Yes sorry, I forgot to write that in my notes but the idea is to first
apply the Squash-to patch once it has been reviewed by Mat and then send
Paolo's series.

But to have more validation done, I think it would be good if today I
apply the 3 first patches of the series I sent yesterday: Squash-to
patch + 2 fixes. These 3 patches would go in our next batch of patches
to netdev. We can still fix them with squash-to patch later.

Cheers,
Matt
-- 
Tessares | Belgium | Hybrid Access Solutions
www.tessares.net

Re: Upstreaming to net-next

[Mat Martineau]

[-- Attachment #1: Type: text/plain, Size: 2995 bytes --]

On Thu, 6 Jan 2022, Paolo Abeni wrote:

> On Wed, 2022-01-05 at 17:54 -0800, Mat Martineau wrote:
>> As Matthieu noted in today's meeting, we need to figure out what goes
>> upstream to netdev this week:
>>
>> On Wed, 5 Jan 2022, Matthieu Baerts wrote:
>>
>>> Patches to send to netdev:
>>>    - net:
>>>        - probably best to send:
>>>            - mptcp: fix opt size when sending DSS + MP_FAIL
>>>            - mptcp: fix a DSS option writing error → but we might need
>>> to change
>>>        - But if they are accepted after the last PR from Net maintainer
>>> to Linus would delay stuff.
>>
>> I noticed that Jakub sent a "Networking for 5.16-final" pull request to
>> Linus today, so the 5.16 ship has sailed.
>>
>>>        - So if we don't want to rush to look at the tests, we can also
>>> send to net-next for the future -net (to be check which tree is merged
>>> in which one first?
>>
>> The next pull requests from Jakub to Linus will be from net-next, so
>> that's the fastest route to Linus' tree for fixes that will be added to
>> the stable tree.
>
> I'm a bit lost about all the above. Not sure if due to lack of mtg
> background, lack of coffee, or lack of time to digest the above ;)
>

That's ok - I'm not sure I explained it clearly in the meeting either. 
Just trying to get fixes into Linus' tree ASAP, without getting stuck in 
the net branch until after 5.17-rc1. This happened around a recent 
release, maybe 5.14-ish.

Anyway, I will just ask Jakub about how to best handle it.


>
>>
>>>
>>>    - net-next:
>>>        - patches from Paolo could go there, fixing issues, see GH
>>> issues/250:
>>>            - *@Paolo*: OK for you?
>>>
>>>        - + checksum fixes probably
>>>
>>>        - best to skip MP_FAIL series for now as we don't have a stable
>>> selftest yet
>>>
>>>        - and no need to send Mat's series as it is linked to new
>>> features in the Netlink PM part, not ready yet
>>
>> Here are the commits that look ready (commit ids from
>> export/20220105T054758):
>>
>> Paolo:
>>
>> 9f6720406819 mptcp: keep snd_una updated for fallback socket
>> 95bc19b21e00 mptcp: implement fastclose xmit path
>> 37b1cee6f04f mptcp: full disconnect implementation
>> 916957cdccd6 mptcp: cleanup accept and poll
>> 6358b829913e mptcp: implement support for user-space disconnect
>> adba2e970980 selftests: mptcp: add disconnect tests
>> afef858f5fa4 mptcp: fix per socket endpoint accounting
>> 9caba598fe7a mptcp: clean-up MPJ option writing
>> f584047ecf9a mptcp: keep track of local endpoint still available for each msk
>> fc51a4526843 mptcp: do not block subflows creation on errors
>> 893d4cb7ccb9 selftests: mptcp: add tests for subflow creation failure
>> 128e78db18bf mptcp: cleanup MPJ subflow list handling
>> e470b365da5f mptcp: avoid atomic bit manipulation when possible
>>
>> 3cdd8e644985 selftests: mptcp: more stable join tests-cases
>
> ... anyhow this LGTM for net-next.
>

Ok, good! Thank you for checking in :)

--
Mat Martineau
Intel

Re: Upstreaming to net-next

[Mat Martineau]

[-- Attachment #1: Type: text/plain, Size: 1863 bytes --]

On Thu, 6 Jan 2022, Mat Martineau wrote:

> On Thu, 6 Jan 2022, Paolo Abeni wrote:
>
>> On Wed, 2022-01-05 at 17:54 -0800, Mat Martineau wrote:
>>> As Matthieu noted in today's meeting, we need to figure out what goes
>>> upstream to netdev this week:
>>> 
>>> On Wed, 5 Jan 2022, Matthieu Baerts wrote:
>>> 
>>>> Patches to send to netdev:
>>>>    - net:
>>>>        - probably best to send:
>>>>            - mptcp: fix opt size when sending DSS + MP_FAIL
>>>>            - mptcp: fix a DSS option writing error → but we might need
>>>> to change
>>>>        - But if they are accepted after the last PR from Net maintainer
>>>> to Linus would delay stuff.
>>> 
>>> I noticed that Jakub sent a "Networking for 5.16-final" pull request to
>>> Linus today, so the 5.16 ship has sailed.
>>>
>>>>        - So if we don't want to rush to look at the tests, we can also
>>>> send to net-next for the future -net (to be check which tree is merged
>>>> in which one first?
>>> 
>>> The next pull requests from Jakub to Linus will be from net-next, so
>>> that's the fastest route to Linus' tree for fixes that will be added to
>>> the stable tree.
>> 
>> I'm a bit lost about all the above. Not sure if due to lack of mtg
>> background, lack of coffee, or lack of time to digest the above ;)
>> 
>
> That's ok - I'm not sure I explained it clearly in the meeting either. Just 
> trying to get fixes into Linus' tree ASAP, without getting stuck in the net 
> branch until after 5.17-rc1. This happened around a recent release, maybe 
> 5.14-ish.
>
> Anyway, I will just ask Jakub about how to best handle it.
>

I think I was being overly paranoid about this - while it looks like the 
net->net-next merge was skipped for 5.13, the netdev maintainers have been 
consistent about it since then. So I won't complicate things by asking 
about it.


--
Mat Martineau
Intel