Re: [PATCH net] net/dccp: fix use-after-free in dccp_invalid_packet

[Eric Dumazet]

On Mon, 2016-11-28 at 12:05 -0300, Arnaldo Carvalho de Melo wrote:
> Em Mon, Nov 28, 2016 at 06:47:14AM -0800, Eric Dumazet escreveu:
> > On Mon, 2016-11-28 at 11:40 -0300, Arnaldo Carvalho de Melo wrote:
> > > Em Mon, Nov 28, 2016 at 06:26:49AM -0800, Eric Dumazet escreveu:
> > > > From: Eric Dumazet <edumazet@xxxxxxxxxx>
> > > > 
> > > > pskb_may_pull() can reallocate skb->head, we need to reload dh pointer
> > > > in dccp_invalid_packet() or risk use after free.
> > > > 
> > > > Bug found by Andrey Konovalov using syzkaller.
> > > > 
> > > > Signed-off-by: Eric Dumazet <edumazet@xxxxxxxxxx>
> > > > Reported-by: Andrey Konovalov <andreyknvl@xxxxxxxxxx>
> > > 
> > > Acked-by: Arnaldo Carvalho de Melo <acme@xxxxxxxxxx>
> > > 
> > > I was about to send exactly this patch, and while looking at it I think
> > > the patch below needs to go in as well, no? To follow the advice of that
> > > Warning line there :-)
> > > 
> > > From: Arnaldo Carvalho de Melo <acme@xxxxxxxxxx>
> > > 
> > > pskb_may_pull() can reallocate skb->head, so we can't access
> > > iph->frag_off or risk use after free, save it to a variable and us that
> > > later.
> > > 
> > > Cc: Andrey Konovalov <andreyknvl@xxxxxxxxxx>
> > > Cc: Eric Dumazet <edumazet@xxxxxxxxxx>
> > > Signed-off-by: Arnaldo Carvalho de Melo <acme@xxxxxxxxxx>
> > > 
> > > diff --git a/net/ipv4/af_inet.c b/net/ipv4/af_inet.c
> > > index 5ddf5cda07f4..9462070561a3 100644
> > > --- a/net/ipv4/af_inet.c
> > > +++ b/net/ipv4/af_inet.c
> > > @@ -1198,6 +1198,7 @@ struct sk_buff *inet_gso_segment(struct sk_buff *skb,
> > >  	struct iphdr *iph;
> > >  	int proto, tot_len;
> > >  	int nhoff;
> > > +	u16 frag_off;
> > >  	int ihl;
> > >  	int id;
> > >  
> > > @@ -1213,6 +1214,7 @@ struct sk_buff *inet_gso_segment(struct sk_buff *skb,
> > >  
> > >  	id = ntohs(iph->id);
> > >  	proto = iph->protocol;
> > > +	frag_off = iph->frag_off;
> > >  
> > >  	/* Warning: after this point, iph might be no longer valid */
> > >  	if (unlikely(!pskb_may_pull(skb, ihl)))
> > > @@ -1233,7 +1235,7 @@ struct sk_buff *inet_gso_segment(struct sk_buff *skb,
> > >  		fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
> > >  
> > >  		/* fixed ID is invalid if DF bit is not set */
> > > -		if (fixedid && !(iph->frag_off & htons(IP_DF)))
> > > +		if (fixedid && !(frag_off & htons(IP_DF)))
> > >  			goto out;
> > >  	}
> > >  
> > 
> > 
> > I do not see why this patch would be needed ?
> 
> Where is iph being reloaded after that pskb_may_pull() and thus at line 1236 we
> could use after free? The warning at line 1217?
> 
> 1209         iph = ip_hdr(skb);
> 1210         ihl = iph->ihl * 4;
> 1211         if (ihl < sizeof(*iph))
> 1212                 goto out;
> 1213 
> 1214         id = ntohs(iph->id);
> 1215         proto = iph->protocol;
> 1216 
> 1217         /* Warning: after this point, iph might be no longer valid */
> 1218         if (unlikely(!pskb_may_pull(skb, ihl)))
> 1219                 goto out;
> 1220         __skb_pull(skb, ihl);
> 1221 
> 1222         encap = SKB_GSO_CB(skb)->encap_level > 0;
> 1223         if (encap)
> 1224                 features &= skb->dev->hw_enc_features;
> 1225         SKB_GSO_CB(skb)->encap_level += ihl;
> 1226 
> 1227         skb_reset_transport_header(skb);
> 1228 
> 1229         segs = ERR_PTR(-EPROTONOSUPPORT);
> 1230 
> 1231         if (!skb->encapsulation || encap) {
> 1232                 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
> 1233                 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
> 1234 
> 1235                 /* fixed ID is invalid if DF bit is not set */
> 1236                 if (fixedid && !(iph->frag_off & htons(IP_DF)))
> 1237                         goto out;
> 1238         }
> 

Arg, I was looking at an old tree.

Please then add

Fixes: cbc53e08a793b ("GSO: Add GSO type for fixed IPv4 ID")

To ease stable backports.

Also, it looks comments are not read, we might kill this one and reload
iph.

( Saving 3 fields is now more expensive than simply reloading iph )

Thanks.