Re: [PATCH] strparser: remove any offset before parsing messages

[Dominique Martinet <asmadeus@codewreck.org>]

Dave Watson wrote on Wed, Aug 22, 2018:
> > I've tried measuring that overhead as well by writing a more complex bpf
> > program that would fetch the offset in the skb but for some reason I'm
> > reading a 0 offset when it's not zero... well, not like there's much
> > choice for this at this point anyway; I don't think we'll do this
> > without pull, I'll put that on background.
> 
> For what it is worth we checked the offset in bpf, something
> along the lines of

Oh, thanks!

> 	  > struct kcm_rx_msg {   int full_len;  int offset;};
> static inline struct kcm_rx_msg *kcm_rx_msg(struct __sk_buff *skb)
>       { return (struct kcm_rx_msg *)skb->cb;}
> 
> int decode_framing(struct __sk_buff *skb)
> { return load_word(skb, kcm_rx_msg(skb)->offset);}

So you're taking directly the address at skb->cb but the linux code has
this function:
static inline struct strp_msg *strp_msg(struct sk_buff *skb)
{
        return (struct strp_msg *)((void *)skb->cb +
                offsetof(struct qdisc_skb_cb, data));
}
and qdisc_skb_cb.data is another 8 bytes in, that would explain I had
different results (and now I'm trying your snippet it does work), but
I'll have to admit I fail to understand this....

Ok, so 'cb' in __sk_buff is 48 bytes in but 'cb' in sk_buff is 40 bytes
in -- I might just start getting annoyed over this, is there a reason
for the different offset?!


> Although it did puzzle me for a while figuring that out when I ran in
> to it.

Well, at least it means some people were aware of the problem and worked
around it in their own way -- what do you think of pulling instead?
I mean, we could just document that "really well" and provide the
get-offset function in some header that would be made include-able from
bpf.. But right now this isn't really the case.


FWIW now I have this version I also don't notice any performance change
with the pull on my example, it actually looks like the bpf load_word is
slightly slower than pull to access data that is not in the head, but
the noise level is pretty bad.


Thanks,
-- 
Dominique