Re: [PATCH v2 1/1] nf_conntrack_ftp: Added nfct_seqadj_ext_add() for ftp's conntrack.

[Florian Westphal]

Andrii Melnychenko <a.melnychenko@xxxxxxx> wrote:
> There was an issue with NAT'ed ftp and replaced messages
> for PASV/EPSV mode. "New" IP in the message may have a
> different length that would require sequence adjustment.
> 
> Signed-off-by: Andrii Melnychenko <a.melnychenko@xxxxxxx>
> ---
>  net/netfilter/nf_conntrack_ftp.c | 3 +++
>  1 file changed, 3 insertions(+)
> 
> diff --git a/net/netfilter/nf_conntrack_ftp.c b/net/netfilter/nf_conntrack_ftp.c
> index 617f744a2..0216bc099 100644
> --- a/net/netfilter/nf_conntrack_ftp.c
> +++ b/net/netfilter/nf_conntrack_ftp.c
> @@ -25,6 +25,7 @@
>  #include <net/netfilter/nf_conntrack_ecache.h>
>  #include <net/netfilter/nf_conntrack_helper.h>
>  #include <linux/netfilter/nf_conntrack_ftp.h>
> +#include <net/netfilter/nf_conntrack_seqadj.h>
>  
>  #define HELPER_NAME "ftp"
>  
> @@ -390,6 +391,8 @@ static int help(struct sk_buff *skb,
>  	/* Until there's been traffic both ways, don't look in packets. */
>  	if (ctinfo != IP_CT_ESTABLISHED &&
>  	    ctinfo != IP_CT_ESTABLISHED_REPLY) {
> +		if (!nf_ct_is_confirmed(ct))
> +			nfct_seqadj_ext_add(ct);

Still not convinced this is the correct place.

In nf_nat_setup_info we have:

                if (nfct_help(ct) && !nfct_seqadj(ct))
                        if (!nfct_seqadj_ext_add(ct))
                                return NF_DROP;

Looking at your cover letter (some of that info should be in
patch changelog, it provides essential context), I would say
that dnat rule is evaluated before the helper gets attached.

If so, the bug is in nft_ct_helper_obj_eval, and the Fixes
tag should be set to

Fixes: 1a64edf54f55 ("netfilter: nft_ct: add helper set support")

Probably something like:

if ((ct->status & IPS_NAT_MASK) && !!nfct_seqadj(ct))
	if (!nfct_seqadj_ext_add(ct)) ...

Could you please investigate a bit futher? To me it looks
like all other helpers have the same issue, a generic fix would
be better.