Re: [PATCH] Fix up truesize after pskb_expand_head() in wireless stack

[Johannes Berg <johannes-cdvu00un1VgdHxzADdlk8Q@public.gmane.org>]

[-- Attachment #1: Type: text/plain, Size: 1549 bytes --]

On Sun, 2009-01-04 at 17:28 +0100, Andi Kleen wrote:
> > Thanks, but I'll need to look at this in more detail, we need to make
> > sure that we orphan the skb before 
> 
> What do you mean with orphaning the skb? 

Well, touching truesize is absolutely not allowed while the skb is
charged to a socket. This is what causes the truesize warning. The thing
we need to do is figure out is why the skb has a wrong truesize.

> etc. And then, we need to check
> > whether it makes sense to do this in pskb_expand_head().
> 
> Well whatever you do this short term patch is needed, there's no 
> reason to delay it.

Given that we've had this problem for a very long time now I think
there's no reason to rush a workaround now. I realise that we already
have a workaround like this in the tx path which we added because I
thought the tx path was the problem, but it still doesn't make much
sense to work around it at all spots until we know why it is required.

I think this patch similarly just papers over the problem with
pskb_expand_head(). I haven't looked through all the code yet, but if
anything then I think pskb_expand_head() should fix up truesize
afterwards, and we should audit all other callers too. Similar problems
exist in net/core/pktgen.c, drivers/net/wireless/libertas/rx.c,
net/ipv4/netfilter.c and many more, though those seem to not run into
trouble. Only a few users adjust truesize.

Any proper fix should also verify that the skb isn't charged to a socket
while it's being reallocated.

johannes

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 836 bytes --]