Re: [PATCH] net: xfrm: fix return value check in ipcomp_compress

[Florian Westphal]

Florian Westphal <fw@xxxxxxxxx> wrote:
> Ma Ke <make_ruc2021@xxxxxxx> wrote:
> > In ipcomp_compress, to avoid an unexpected result returned by
> > pskb_trim, we should check the return value of pskb_trim().
> > 
> > Signed-off-by: Ma Ke <make_ruc2021@xxxxxxx>
> > ---
> >  net/xfrm/xfrm_ipcomp.c | 4 +++-
> >  1 file changed, 3 insertions(+), 1 deletion(-)
> > 
> > diff --git a/net/xfrm/xfrm_ipcomp.c b/net/xfrm/xfrm_ipcomp.c
> > index 9c0fa0e1786a..5f2e6edadf48 100644
> > --- a/net/xfrm/xfrm_ipcomp.c
> > +++ b/net/xfrm/xfrm_ipcomp.c
> > @@ -144,7 +144,9 @@ static int ipcomp_compress(struct xfrm_state *x, struct sk_buff *skb)
> >  	memcpy(start + sizeof(struct ip_comp_hdr), scratch, dlen);
> >  	local_bh_enable();
> >  
> > -	pskb_trim(skb, dlen + sizeof(struct ip_comp_hdr));
> > +	err = pskb_trim(skb, dlen + sizeof(struct ip_comp_hdr));
> > +	if (unlikely(err))
> > +		goto out;
> 
> This can't be right, this now calls local_bh_enable() twice.

Furthermore, looking at this:

1. skb went through skb_linearize_cow() before, so no paged data anymore

2. Right before there is a check to bail in case compression inflated
packet size.

IOW, this pskb_trim cannot fail, it boils down to __skb_trim().