Re: BUG in 2.6.29 final: broken network connection

Re: BUG in 2.6.29 final: broken network connection

[Niel Lambrechts]

On 03/24/2009 07:50 PM, Eric Dumazet wrote:
> Alan Cox a écrit :
>> Wouldn't it have been simpler to wait when you found a problem and read
>> Ingo's email on the subject from a few hours ago ?
> 
> This guy had his life depending on linux-2.6.29 apparently, and failed
> to read netdev or lkml. Go figure...
> 
> It sounds like a troll, dont you think ?

Sounds like Uwe Bugla to me.

Niel

Re: BUG in 2.6.29 final: broken network connection

[markus reichelt]

[-- Attachment #1: Type: text/plain, Size: 229 bytes --]

* Niel Lambrechts <niel.lambrechts@gmail.com> wrote:

> > It sounds like a troll, dont you think ?
> 
> Sounds like Uwe Bugla to me.

It's him for sure, his gmx authentication gave him away.

-- 
left blank, right bald

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

BUG in 2.6.29 final: broken network connection

[sinter]

Hi everybody,

ftp, ping, IP forwarding, IP masquerading etc., in short: everything having to 
do with networking is broken using kernel 2.6.29.

My thousand thanks for that again go to Mr. David S. Miller, kernel 
contributor of the netdev section, who again and again is adding his SOB under 
untested crap code that itself breaks networking hardware by reaching the 
kernel mainline via git patches.

In the past exactly that happened again and again, and it is particularly Mr. 
David S. Miller himself who, proven by facts, is not willing to learn to stay 
away from sending in malicious code via the git path.
I am really wondering why Linus Torvalds is still trusting him, as it is a 
fact that not everybody can send in code just like that, i. e. without 
restrictions.

It cost me almost 1 complete day (and again I do state that this is not the 
first time that some unqualified hacker operating with the responsibility of Mr. 
David S. Miller deeply sucks and steals my nerves and wastes my time) to find 
out that the following patch needs to be reverted to make networking services 
in 2.6.29 final available again for everybody.

--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -2588,9 +2588,9 @@ static int process_backlog(struct napi_struct *napi, int 
quota)
                local_irq_disable();
                skb = __skb_dequeue(&queue->input_pkt_queue);
                if (!skb) {
-                       __napi_complete(napi);
                        local_irq_enable();
-                       break;
+                       napi_complete(napi);
+                       goto out;
                }
                local_irq_enable();
 
@@ -2599,6 +2599,7 @@ static int process_backlog(struct napi_struct *napi, int 
quota)
 
        napi_gro_flush(napi);
 
+out:
        return work;
 }
 
@@ -2671,7 +2672,7 @@ void netif_napi_del(struct napi_struct *napi)
        struct sk_buff *skb, *next;
 
        list_del_init(&napi->dev_list);
-       kfree(napi->skb);
+       kfree_skb(napi->skb);
 
        for (skb = napi->gro_list; skb; skb = next) {
                next = skb->next;

Best wishes

Uwe

Re: BUG in 2.6.29 final: broken network connection

[Alan Cox]

> It cost me almost 1 complete day 

Wouldn't it have been simpler to wait when you found a problem and read
Ingo's email on the subject from a few hours ago ?

Re: BUG in 2.6.29 final: broken network connection

[Eric Dumazet]

Alan Cox a écrit :
>> It cost me almost 1 complete day 
> 
> Wouldn't it have been simpler to wait when you found a problem and read
> Ingo's email on the subject from a few hours ago ?

This guy had his life depending on linux-2.6.29 apparently, and failed
to read netdev or lkml. Go figure...

It sounds like a troll, dont you think ?

Re: BUG in 2.6.29 final: broken network connection

[sinter]

Am Dienstag 24 März 2009 18:25:03 schrieben Sie:
> > It cost me almost 1 complete day
>
> Wouldn't it have been simpler to wait when you found a problem and read
> Ingo's email on the subject from a few hours ago ?

Wouldn't it be simpler to restrictively avoid crap code in final kernel 
versions? And if that does not work by appeal shouldn't the netdev maintainer 
simply be substituted?

Who needs a maintainer adding his SOB with closed eyes under untested crap 
code? Who needs someone like that for kernel development?

Re: BUG in 2.6.29 final: broken network connection

[David Miller]

From: sinter <sinter.salt@gmx.de>
Date: Tue, 24 Mar 2009 19:18:29 +0100

> Who needs a maintainer adding his SOB with closed eyes under untested crap 
> code? Who needs someone like that for kernel development?

Hey Uwe, nothing interesting going on in your life so you have to make
other people miserable?

This patch was tested, and it made a bug reporter's kernel crash go
away.

We'll fix the problems we're seeing now, just like we always do.

Re: BUG in 2.6.29 final: broken network connection

[Ingo Molnar]

* sinter <sinter.salt@gmx.de> wrote:

> Am Dienstag 24 März 2009 18:25:03 schrieben Sie:
> > > It cost me almost 1 complete day
> >
> > Wouldn't it have been simpler to wait when you found a problem and read
> > Ingo's email on the subject from a few hours ago ?
> 
> Wouldn't it be simpler to restrictively avoid crap code in final 
> kernel versions? And if that does not work by appeal shouldn't the 
> netdev maintainer simply be substituted?
> 
> Who needs a maintainer adding his SOB with closed eyes under 
> untested crap code? Who needs someone like that for kernel 
> development?

Nonsense. 303c6a0 "gro: Fix legacy path napi_complete crash" was an 
obvious looking bug fix for a real crash that was reported, against 
a commit that went upstream in 2.6.29-rc1.

It was a fix for a serious regression and i'd have applied it too, 
and would have pushed it to Linus.

Furthermore, even on affected systems it took certain configs and 
certain conditions for the bug to trigger under high load. So there 
was no real good way to find this bug quickly.

Such kind of bugs can happen anytime, to any maintainer. Unless we 
100% freeze the kernel for a full month before release, this risk 
simply cannot be avoided. It is far more problematic if maintainers 
dont push known fixes or ignore fixes. The networking tree is 
exemplary in picking up fixes addressing bug reports quickly. A 
proposed fix was posted 33 minutes after i sent my bugreport.

Things breaking is simply the nature of software changes - get used 
to it. When new software with 1 million lines of changes over the 
previous version comes out, dont jump on it immediately, if your 
peace of mind depends on a 100% working system. Only try it if you 
want to help out developing it.

And there's an easy solution for you in any case: you can wait for 
.29.1 which i'm sure will be out quickly.

Thanks,

	Ingo

Re: BUG in 2.6.29 final: broken network connection

[Ingo Molnar]

* sinter <sinter.salt@gmx.de> wrote:

> > And there's an easy solution for you in any case: you can wait 
> > for .29.1 which i'm sure will be out quickly.
> 
> And what top-class advice would you give me if I would have found 
> that serious bug in .29.8 or in .29.9? Due to your answer every 
> shit is possible and nothing is impossible, or did I get something 
> wrong?

You got it exactly right.

This world is ruled by probabilities, and generally everything is 
possible - even a bug slipping into a .0 release that you could 
trigger personally, even though -rc1 through -rc8 worked just fine 
for you.

Such things happened before, and they can happen again. Yes, it can 
hit you - and it did hit me too and i'm not complaining.

We still try our best to reduce the probabilities. We cannot, 
however, unfortunately, bring them down to zero. The fix looked 
reasonable and the justification of it fixing a live crash looked 
reasonable too.

What we can do, and what we did do, was to provide people (you 
included) with a fix within 12 hours after the code having been 
released.

That's the deal really.

	Ingo