[RFC net-next 1/1] : sctp: denoise deprecation log on SCTP_EVENTS

[RFC net-next 1/1] : sctp: denoise deprecation log on SCTP_EVENTS

[Jamal Hadi Salim]

[-- Attachment #1: Type: text/plain, Size: 642 bytes --]



In the newer kernels this message is extremely noisy. After a quick
discussion with Daniel it seems to me it will be very hard to get
existing apps that nobody is going to update to continue to work
(i.e no forward compat). And newer apps that desire to play in both
older kernels and new kernels will have to play some tricks to work
(i.e weak backward compatibility).  These are good reasons
to totally get rid of this message. At minimal to neutre it.
The attached change tries to do that. However, if you had multiple
apps, you will only get warning for the first one.

Will send proper patch when theres some consensus.

cheers,
jamal

[-- Attachment #2: sctp-denoise --]
[-- Type: text/plain, Size: 753 bytes --]

diff --git a/net/sctp/socket.c b/net/sctp/socket.c
index 5f6c4e6..0e8c937 100644
--- a/net/sctp/socket.c
+++ b/net/sctp/socket.c
@@ -2207,9 +2207,10 @@ static int sctp_setsockopt_events(struct sock *sk, char __user *optval,
 		return -EFAULT;
 
 	if (sctp_sk(sk)->subscribe.sctp_data_io_event)
-		pr_warn_ratelimited(DEPRECATED "%s (pid %d) "
-				    "Requested SCTP_SNDRCVINFO event.\n"
-				    "Use SCTP_RCVINFO through SCTP_RECVRCVINFO option instead.\n",
+		pr_warn_once(DEPRECATED "%s (pid %d) "
+				    "SCTP_EVENTS Option - DEPRECATED\n"
+				    "Use SCTP_EVENT Option instead\n"
+				    "Refer to RFC6458 section 6.2\n",
 				    current->comm, task_pid_nr(current));
 
 	/* At the time when a user app subscribes to SCTP_SENDER_DRY_EVENT,

Re: [RFC net-next 1/1] : sctp: denoise deprecation log on SCTP_EVENTS

[Daniel Borkmann]

On 07/09/2015 12:38 PM, Jamal Hadi Salim wrote:
...
> In the newer kernels this message is extremely noisy. After a quick
> discussion with Daniel it seems to me it will be very hard to get
> existing apps that nobody is going to update to continue to work
> (i.e no forward compat). And newer apps that desire to play in both
> older kernels and new kernels will have to play some tricks to work
> (i.e weak backward compatibility).  These are good reasons
> to totally get rid of this message. At minimal to neutre it.

To put some more context into this, here was the past discussion
on this topic:

   http://www.spinics.net/lists/linux-sctp/msg03596.html

This begs the question, which we need to clarify:

   Can we *ever* get rid of the old UAPI that the RFC proposed
   and later on deprecated?

In the thread, there was a mentioning of 5-6 years schedule, but
I honestly doubt that this would be enough, i.e. for people only
upgrading their kernels, but not being able to touch their legacy
SCTP apps (maybe due to proprietary appliance crap, etc).

A pr_warn_once() would only warn for the first app triggering this
(and not for follow-ups), it would surely take incentives away to
update their code since less annoying, but if the uapi in fact
could never be removed, then why bothering spamming the klog in
the first place? ;) I can understand the deprecation dilemma, but
personally, I would have no problem if we just want to remove the
pr_warn_ratelimited() altogether ...

> The attached change tries to do that. However, if you had multiple
> apps, you will only get warning for the first one.
>
> Will send proper patch when theres some consensus.
>
> cheers,
> jamal

Re: [RFC net-next 1/1] : sctp: denoise deprecation log on SCTP_EVENTS

[Marcelo Ricardo Leitner]

On Thu, Jul 09, 2015 at 01:54:37PM +0200, Daniel Borkmann wrote:
> On 07/09/2015 12:38 PM, Jamal Hadi Salim wrote:
> ...
> >In the newer kernels this message is extremely noisy. After a quick
> >discussion with Daniel it seems to me it will be very hard to get
> >existing apps that nobody is going to update to continue to work
> >(i.e no forward compat). And newer apps that desire to play in both
> >older kernels and new kernels will have to play some tricks to work
> >(i.e weak backward compatibility).  These are good reasons
> >to totally get rid of this message. At minimal to neutre it.
> 
> To put some more context into this, here was the past discussion
> on this topic:
> 
>   http://www.spinics.net/lists/linux-sctp/msg03596.html
> 
> This begs the question, which we need to clarify:
> 
>   Can we *ever* get rid of the old UAPI that the RFC proposed
>   and later on deprecated?
> 
> In the thread, there was a mentioning of 5-6 years schedule, but
> I honestly doubt that this would be enough, i.e. for people only
> upgrading their kernels, but not being able to touch their legacy
> SCTP apps (maybe due to proprietary appliance crap, etc).
> 
> A pr_warn_once() would only warn for the first app triggering this
> (and not for follow-ups), it would surely take incentives away to
> update their code since less annoying, but if the uapi in fact
> could never be removed, then why bothering spamming the klog in
> the first place? ;) I can understand the deprecation dilemma, but
> personally, I would have no problem if we just want to remove the
> pr_warn_ratelimited() altogether ...
> 
> >The attached change tries to do that. However, if you had multiple
> >apps, you will only get warning for the first one.
> >
> >Will send proper patch when theres some consensus.

The new message is misleading as currently we don't have SCTP_EVENT 
available to use. So if someone tries to actually follow that lead, it
will lead to a dead and confusing end.

I'm okay with just switching it to pr_warn_once, though. It makes sense
to not spam klog but to have the message there, because even if we keep
it available for backwards compatibility, it might suffer from some
limitations in the future and the usage of the new one is always
recommended.

  Marcelo

RE: [RFC net-next 1/1] : sctp: denoise deprecation log on SCTP_EVENTS

[David Laight]

From: Jamal Hadi Salim
> Sent: 09 July 2015 11:38
> 
> In the newer kernels this message is extremely noisy. After a quick
> discussion with Daniel it seems to me it will be very hard to get
> existing apps that nobody is going to update to continue to work
> (i.e no forward compat). And newer apps that desire to play in both
> older kernels and new kernels will have to play some tricks to work
> (i.e weak backward compatibility).  These are good reasons
> to totally get rid of this message. At minimal to neutre it.
> The attached change tries to do that. However, if you had multiple
> apps, you will only get warning for the first one.

Never mind spanning the kernel log, there will be performance
issues with systems that are sending significant amounts of SCTP data.
I'm not sure you'll get 10000+/sec through the trace system.

I'm going to have to find which version this trace went into and
make sure we tell customers not to use the relevant kernel versions.

	David

Re: [RFC net-next 1/1] : sctp: denoise deprecation log on SCTP_EVENTS

[Daniel Borkmann]

On 07/22/2015 01:48 PM, David Laight wrote:
> From: Jamal Hadi Salim
>> Sent: 09 July 2015 11:38
>>
>> In the newer kernels this message is extremely noisy. After a quick
>> discussion with Daniel it seems to me it will be very hard to get
>> existing apps that nobody is going to update to continue to work
>> (i.e no forward compat). And newer apps that desire to play in both
>> older kernels and new kernels will have to play some tricks to work
>> (i.e weak backward compatibility).  These are good reasons
>> to totally get rid of this message. At minimal to neutre it.
>> The attached change tries to do that. However, if you had multiple
>> apps, you will only get warning for the first one.
>
> Never mind spanning the kernel log, there will be performance
> issues with systems that are sending significant amounts of SCTP data.
> I'm not sure you'll get 10000+/sec through the trace system.

What do you mean by "trace system" (there's no backtrace)?

Anyway, as previously stated, I doubt we'd get rid of the deprecated
SCTP uapi from the RFC any time soon, so, imho, we should just remove
these rate-limited messages altogether to stop spamming the klog ...

> I'm going to have to find which version this trace went into and
> make sure we tell customers not to use the relevant kernel versions.
>
> 	David
>

Re: [RFC net-next 1/1] : sctp: denoise deprecation log on SCTP_EVENTS

[Michael Tuexen]

> On 22 Jul 2015, at 14:04, Daniel Borkmann <daniel@iogearbox.net> wrote:
> 
> On 07/22/2015 01:48 PM, David Laight wrote:
>> From: Jamal Hadi Salim
>>> Sent: 09 July 2015 11:38
>>> 
>>> In the newer kernels this message is extremely noisy. After a quick
>>> discussion with Daniel it seems to me it will be very hard to get
>>> existing apps that nobody is going to update to continue to work
>>> (i.e no forward compat). And newer apps that desire to play in both
>>> older kernels and new kernels will have to play some tricks to work
>>> (i.e weak backward compatibility).  These are good reasons
>>> to totally get rid of this message. At minimal to neutre it.
>>> The attached change tries to do that. However, if you had multiple
>>> apps, you will only get warning for the first one.
>> 
>> Never mind spanning the kernel log, there will be performance
>> issues with systems that are sending significant amounts of SCTP data.
>> I'm not sure you'll get 10000+/sec through the trace system.
> 
> What do you mean by "trace system" (there's no backtrace)?
> 
> Anyway, as previously stated, I doubt we'd get rid of the deprecated
> SCTP uapi from the RFC any time soon, so, imho, we should just remove
> these rate-limited messages altogether to stop spamming the klog ...
... and we will not remove it in FreeBSD either.

Best regards
Michael
> 
>> I'm going to have to find which version this trace went into and
>> make sure we tell customers not to use the relevant kernel versions.
>> 
>> 	David
>> 
> 
> 

Re: [RFC net-next 1/1] : sctp: denoise deprecation log on SCTP_EVENTS

[Daniel Borkmann]

On 07/22/2015 02:30 PM, Michael Tuexen wrote:
>> On 22 Jul 2015, at 14:04, Daniel Borkmann <daniel@iogearbox.net> wrote:
...
>> Anyway, as previously stated, I doubt we'd get rid of the deprecated
>> SCTP uapi from the RFC any time soon, so, imho, we should just remove
>> these rate-limited messages altogether to stop spamming the klog ...
> ... and we will not remove it in FreeBSD either.

Right, I think same here.

> Best regards
> Michael