Re: [PATCH net] udp: fix segmentation crash for GRO packet without fraglist

[Willem de Bruijn]

Lena Wang (王娜) wrote:
> On Tue, 2024-04-16 at 19:14 -0400, Willem de Bruijn wrote:
> >  	 
> > External email : Please do not click links or open attachments until
> > you have verified the sender or the content.
> >  > > > > Personally, I think bpf_skb_pull_data() should have
> > automatically
> > > > > > (ie. in kernel code) reduced how much it pulls so that it
> > would pull
> > > > > > headers only,
> > > > >
> > > > > That would be a helper that parses headers to discover header
> > length.
> > > >
> > > > Does it actually need to?  Presumably the bpf pull function could
> > > > notice that it is
> > > > a packet flagged as being of type X (UDP GSO FRAGLIST) and reduce
> > the pull
> > > > accordingly so that it doesn't pull anything from the non-linear
> > > > fraglist portion???
> > > >
> > > > I know only the generic overview of what udp gso is, not any
> > details, so I am
> > > > assuming here that there's some sort of guarantee to how these
> > packets
> > > > are structured...  But I imagine there must be or we wouldn't be
> > hitting these
> > > > issues deeper in the stack?
> > > 
> > > Perhaps for a packet of this type we're already guaranteed the
> > headers
> > > are in the linear portion,
> > > and the pull should simply be ignored?
> > > 
> > > >
> > > > > Parsing is better left to the BPF program.
> > 
> > I do prefer adding sanity checks to the BPF helpers, over having to
> > add then in the net hot path only to protect against dangerous BPF
> > programs.
> > 
> Is it OK to ignore or decrease pull length for udp gro fraglist packet?
> It could save the normal packet and sent to user correctly.
> 
> In common/net/core/filter.c
> static inline int __bpf_try_make_writable(struct sk_buff *skb,
>               unsigned int write_len)
> { 
> +	if (skb_is_gso(skb) && (skb_shinfo(skb)->gso_type &
> +		(SKB_GSO_UDP  |SKB_GSO_UDP_L4)) {

The issue is not with SKB_GSO_UDP_L4, but with SKB_GSO_FRAGLIST.

> +		return 0;

Failing for any pull is a bit excessive. And would kill a sane
workaround of pulling only as many bytes as needed.
 
> +	     or if (write_len > skb_headlen(skb))
> +			write_len = skb_headlen(skb);

Truncating requests would be a surprising change of behavior
for this function.

Failing for a pull > skb_headlen is arguably reasonable, as
the alternative is that we let it go through but have to drop
the now malformed packets on segmentation.


> +	}
> 	return skb_ensure_writable(skb, write_len);
> }
>  
> 
> > In this case, it would be detecting this GSO type and failing the
> > operation if exceeding skb_headlen().
> > > > >
> > > > > > and not packet content.
> > > > > > (This is assuming the rest of the code isn't ready to deal
> > with a longer pull,
> > > > > > which I think is the case atm.  Pulling too much, and then
> > crashing or forcing
> > > > > > the stack to drop packets because of them being malformed
> > seems wrong...)
> > > > > >
> > > > > > In general it would be nice if there was a way to just say
> > pull all headers...
> > > > > > (or possibly all L2/L3/L4 headers)
> > > > > > You in general need to pull stuff *before* you've even looked
> > at the packet,
> > > > > > so that you can look at the packet,
> > > > > > so it's relatively hard/annoying to pull the correct length
> > from bpf
> > > > > > code itself.
> > > > > >
> > > > > > > > > BPF needs to modify a proper length to do pull data.
> > However kernel
> > > > > > > > > should also improve the flow to avoid crash from a bpf
> > function
> > > > > > > > call.
> > > > > > > > > As there is no split flow and app may not decode the
> > merged UDP
> > > > > > > > packet,
> > > > > > > > > we should drop the packet without fraglist in
> > skb_segment_list
> > > > > > > > here.
> > > > > > > > >
> > > > > > > > > Fixes: 3a1296a38d0c ("net: Support GRO/GSO fraglist
> > chaining.")
> > > > > > > > > Signed-off-by: Shiming Cheng <
> > shiming.cheng@xxxxxxxxxxxx>
> > > > > > > > > Signed-off-by: Lena Wang <lena.wang@xxxxxxxxxxxx>
> > > > > > > > > ---
> > > > > > > > >  net/core/skbuff.c | 3 +++
> > > > > > > > >  1 file changed, 3 insertions(+)
> > > > > > > > >
> > > > > > > > > diff --git a/net/core/skbuff.c b/net/core/skbuff.c
> > > > > > > > > index b99127712e67..f68f2679b086 100644
> > > > > > > > > --- a/net/core/skbuff.c
> > > > > > > > > +++ b/net/core/skbuff.c
> > > > > > > > > @@ -4504,6 +4504,9 @@ struct sk_buff
> > *skb_segment_list(struct
> > > > > > > > sk_buff *skb,
> > > > > > > > >  if (err)
> > > > > > > > >  goto err_linearize;
> > > > > > > > >
> > > > > > > > > +if (!list_skb)
> > > > > > > > > +goto err_linearize;
> > > > > > > > > +
> > > > >
> > > > > This would catch the case where the entire data frag_list is
> > > > > linearized, but not a pskb_may_pull that only pulls in part of
> > the
> > > > > list.
> > > > >
> > > > > Even with BPF being privileged, the kernel should not crash if
> > BPF
> > > > > pulls a FRAGLIST GSO skb.
> > > > >
> > > > > But the check needs to be refined a bit. For a UDP GSO packet,
> > I
> > > > > think gso_size is still valid, so if the head_skb length does
> > not
> > > > > match gso_size, it has been messed with and should be dropped.
> > > > >
> Is it OK as below? Is it OK to add log to record the error for easy
> checking issue.
> 
> In net/core/skbuff.c skb_segment_list
> +unsigned int mss = skb_shinfo(head_skb)->gso_size;
> +bool err_len = false;
> 
> +if ( mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb)) {
> +	pr_err("skb is dropped due to messed data. gso size:%d,
> +		hdrlen:%d", mss, skb_headlen(head_skb)

Such logs should always be rate limited. But no need to log cases
where we well understood how we get there.

I would stick with one approach: either in the BPF func or in
segmentation, not both. And then I find BPF preferable, as explained
before.

> +	if (!list_skb)
> +		goto err_linearize;
> +	else
> +		err_len = true;
> +}
> 
> ...
> +if (err_len) {
> +	goto err_linearize;
> +}
> 
> skb_get(skb);
> ...