[Linux-v4.4-LTS] ppp: Backport of rtnetlink device handling

[Linux-v4.4-LTS] ppp: Backport of rtnetlink device handling

[Sedat Dilek]

Hi Guillaume,

which patches do I need to backport "ppp: rtnetlink device handling"
to Linux v4.4 which will be a LongTerm-Supported (LTS) Linux-kernel
[0]?

I tried [1] and [2] on top of recent net-next Git tree which will be
in Linux v4.5.
Currently, your patches are not included in net-next.git#master.

In the thread "[net-next] ppp: rtnetlink device handling" [4] you
explained the benefits and use-case etc.

Checking with git-log shows me these commits...

$ git log --oneline --no-merges v4.4-rc8.. drivers/net/ppp net/l2tp
19e8c5713e78 l2tp: rely on ppp layer for skb scrubbing
645eee4eba45 ppp: implement rtnetlink device handling
3a9bce0ae138 ppp: define reusable device creation functions
69d9728d00c7 ppp: declare ppp devices as enumerated interfaces
94dbffe16eb1 ppp: define "ppp" device type
681b4d88ad8e pppox: use standard module auto-loading feature
a8acce6aa584 ppp: remove PPPOX_ZOMBIE socket state
8734e485fed5 ppp: don't set sk_state to PPPOX_ZOMBIE in pppoe_disc_rcv()

...is that sufficient for a backport?

Where there changes to net(-nex)t infrastructure which I need as well?

Thanks in advance.

Kind regards,
- Sedat -

[0] https://www.kernel.org/category/releases.html
[1] https://patchwork.ozlabs.org/patch/560702/
[2] https://patchwork.ozlabs.org/patch/560703/
[3] https://patchwork.ozlabs.org/patch/561540/
[4] http://marc.info/?t=145154824300001&r=1&w=2

Re: [Linux-v4.4-LTS] ppp: Backport of rtnetlink device handling

[Guillaume Nault]

On Mon, Jan 04, 2016 at 08:47:30AM +0100, Sedat Dilek wrote:
> Hi Guillaume,
> 
> which patches do I need to backport "ppp: rtnetlink device handling"
> to Linux v4.4 which will be a LongTerm-Supported (LTS) Linux-kernel
> [0]?
> 
Quite frankly, backporting this series doesn't look like a good idea.
It only provides a new ABI for creating ppp devices and your control
plane most likely hasn't been updated to use it. So it won't bring any
benefit.

> I tried [1] and [2] on top of recent net-next Git tree which will be
> in Linux v4.5.
> Currently, your patches are not included in net-next.git#master.
> 
Indeed, and that's why no control plane should rely on them (yet).

> In the thread "[net-next] ppp: rtnetlink device handling" [4] you
> explained the benefits and use-case etc.
> 
> Checking with git-log shows me these commits...
> 
> $ git log --oneline --no-merges v4.4-rc8.. drivers/net/ppp net/l2tp
> 19e8c5713e78 l2tp: rely on ppp layer for skb scrubbing
> 645eee4eba45 ppp: implement rtnetlink device handling
> 3a9bce0ae138 ppp: define reusable device creation functions
> 69d9728d00c7 ppp: declare ppp devices as enumerated interfaces
> 94dbffe16eb1 ppp: define "ppp" device type
> 681b4d88ad8e pppox: use standard module auto-loading feature
> a8acce6aa584 ppp: remove PPPOX_ZOMBIE socket state
> 8734e485fed5 ppp: don't set sk_state to PPPOX_ZOMBIE in pppoe_disc_rcv()
> 
> ...is that sufficient for a backport?
> 
Applying the series directly on v4.4-rc8 should work (with a few
conflicts, but other patches are unrelated). But still, you probably
don't want to maintain backports unless strictly required. BTW, this
one has no chance to hit any -stable tree anyway.

Re: [Linux-v4.4-LTS] ppp: Backport of rtnetlink device handling

[Sedat Dilek]

On Mon, Jan 4, 2016 at 1:55 PM, Guillaume Nault <g.nault@alphalink.fr> wrote:
> On Mon, Jan 04, 2016 at 08:47:30AM +0100, Sedat Dilek wrote:
>> Hi Guillaume,
>>
>> which patches do I need to backport "ppp: rtnetlink device handling"
>> to Linux v4.4 which will be a LongTerm-Supported (LTS) Linux-kernel
>> [0]?
>>
> Quite frankly, backporting this series doesn't look like a good idea.
> It only provides a new ABI for creating ppp devices and your control
> plane most likely hasn't been updated to use it. So it won't bring any
> benefit.
>

What do you mean by "control plane"?

>> I tried [1] and [2] on top of recent net-next Git tree which will be
>> in Linux v4.5.
>> Currently, your patches are not included in net-next.git#master.
>>
> Indeed, and that's why no control plane should rely on them (yet).
>

Again, what do you mean by "control plane"?

Does anything speak against to have these patches for upcoming Linux v4.5?
AFAICS, the merge-window will open next week.

>> In the thread "[net-next] ppp: rtnetlink device handling" [4] you
>> explained the benefits and use-case etc.
>>
>> Checking with git-log shows me these commits...
>>
>> $ git log --oneline --no-merges v4.4-rc8.. drivers/net/ppp net/l2tp
>> 19e8c5713e78 l2tp: rely on ppp layer for skb scrubbing
>> 645eee4eba45 ppp: implement rtnetlink device handling
>> 3a9bce0ae138 ppp: define reusable device creation functions
>> 69d9728d00c7 ppp: declare ppp devices as enumerated interfaces
>> 94dbffe16eb1 ppp: define "ppp" device type
>> 681b4d88ad8e pppox: use standard module auto-loading feature
>> a8acce6aa584 ppp: remove PPPOX_ZOMBIE socket state
>> 8734e485fed5 ppp: don't set sk_state to PPPOX_ZOMBIE in pppoe_disc_rcv()
>>
>> ...is that sufficient for a backport?
>>
> Applying the series directly on v4.4-rc8 should work (with a few
> conflicts, but other patches are unrelated). But still, you probably
> don't want to maintain backports unless strictly required. BTW, this
> one has no chance to hit any -stable tree anyway.

I know about the "rules" for -stable.

As far as I understood you, no pre-conditional changes to the
net-infrastructure are required.
I haven't tested to apply above series.

Thanks for your explanations.

- Sedat -

Re: [Linux-v4.4-LTS] ppp: Backport of rtnetlink device handling

[Guillaume Nault]

On Fri, Jan 08, 2016 at 10:28:39AM +0100, Sedat Dilek wrote:
> On Mon, Jan 4, 2016 at 1:55 PM, Guillaume Nault <g.nault@alphalink.fr> wrote:
> > On Mon, Jan 04, 2016 at 08:47:30AM +0100, Sedat Dilek wrote:
> >> Hi Guillaume,
> >>
> >> which patches do I need to backport "ppp: rtnetlink device handling"
> >> to Linux v4.4 which will be a LongTerm-Supported (LTS) Linux-kernel
> >> [0]?
> >>
> > Quite frankly, backporting this series doesn't look like a good idea.
> > It only provides a new ABI for creating ppp devices and your control
> > plane most likely hasn't been updated to use it. So it won't bring any
> > benefit.
> >
> 
> What do you mean by "control plane"?
> 
For the purpose of this discussion, I mean the PPP daemon running in
user space (programs like pppd or accel-ppp).

They implement control protocols like LCP, authentication and various
NCPs, used to negociate PPP encapsulation parameters. They also have to
report on these parameters to the kernel, so that it can properly
handle data exchanged over the established PPP connections. This
generally includes creating a ppp device.

My patch set allows user space to create ppp devices using rtnetlink,
which is more flexible than the current PPP specific ioctl(). But the
PPP daemon needs to be updated or it won't know about the new rtnetlink
and will continue to use the ioctl (which is fine as long as the added
features brought by rtnetlink aren't necessary). That's why I wonder
why do you want to backport this series. I brings no value if user
space isn't able to take advantage of PPP's rtnetlink API.

FYI, the plan is to use the rtnl interface to handle things like
network isolation (with network namespaces) and deterministic ppp
device names in accel-ppp.

> Does anything speak against to have these patches for upcoming Linux v4.5?
> AFAICS, the merge-window will open next week.
> 
The patch set has been deferred due to lack of external reviews. I'll
repost after the merge window, so that interested people will have
enough time to comment on it. Therefore it won't be in v4.5. If
everything goes fine with the new submission, it could go into v4.6.