On Tue, Aug 11, 2015 at 11:47:25PM +0200, Linus Lüssing wrote:
> On Tue, Aug 11, 2015 at 10:51:40PM +0200, Linus Lüssing wrote:
> > On Mon, Aug 10, 2015 at 02:56:12PM -0700, Brenden Blanco wrote:
> > > Doing some code reading with Alexei, we found a suspect commit, which
> > > introduces an skb_get and skb_may_pull of the same skb, which leads to
> > > the BUG
> > > when skb->len == len.
> >
> > Urgh, didn't know that pskb_may_pull() doesn't like an skb with a
> > reference count greater than one... But yes, the BUG() call in
> > skbuff.c:1128 / pskb_expand_head() says that (though in this case
> > the BUG() in skbuff.c call actually seems kinda weird (/"wrong"?), as
> > it isn't shared between different code paths).
>
> The more I think about it, I'm tending to remove the BUG() call in
> pskb_expand_head() as in this case it obviously isn't a bug.
>
> The skb_get() allows a simple and in my opinion easy to read cleanup
> part of skb_trimmed for any caller of ip{v6,}_mc_check_mld(). No need
> to check whether skb == skb_trimmed for a caller for instance,
> simply checking whether skb_trimmed exists is enough.
>
>
> Any objections to remove the "if (skb_shared(skb)) BUG()" part in
> pskb_expand_head()? Or would there be any other undesired side
> effects in utilising skb_get() like that?
That fundamental check was there for 10+ years and cannot be removed.
bridge already did skb_share_check() before reaching this
__ipv6_mc_check_mld() path.
There is no reason to do skb_get() there.
It wasn't there before commit 9afd85c9e4552 which claims to do:
'Some small refactoring was done to enhance readibility',
but doing skb_get()+pskb_may_pull() which is incorrect.
Avoiding unnecessary skb_clone() is a good thing, but it should be
done without messing with skb->users, since this code path
already owns skb.
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
