[nf-next PATCH] netfilter: nft_exthdr: Search chunks in SCTP packets only

[nf-next PATCH] netfilter: nft_exthdr: Search chunks in SCTP packets only

[Phil Sutter]

Since user space does not generate a payload dependency, plain sctp
chunk matches cause searching in non-SCTP packets, too. Avoid this
potential mis-interpretation of packet data by checking pkt->tprot.

Fixes: 133dc203d77df ("netfilter: nft_exthdr: Support SCTP chunks")
Signed-off-by: Phil Sutter <phil@nwl.cc>
---
 net/netfilter/nft_exthdr.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/net/netfilter/nft_exthdr.c b/net/netfilter/nft_exthdr.c
index 7f705b5c09de8..1093bb83f8aeb 100644
--- a/net/netfilter/nft_exthdr.c
+++ b/net/netfilter/nft_exthdr.c
@@ -312,6 +312,9 @@ static void nft_exthdr_sctp_eval(const struct nft_expr *expr,
 	const struct sctp_chunkhdr *sch;
 	struct sctp_chunkhdr _sch;
 
+	if (!pkt->tprot_set || pkt->tprot != IPPROTO_SCTP)
+		goto err;
+
 	do {
 		sch = skb_header_pointer(pkt->skb, offset, sizeof(_sch), &_sch);
 		if (!sch || !sch->length)
@@ -334,7 +337,7 @@ static void nft_exthdr_sctp_eval(const struct nft_expr *expr,
 		}
 		offset += SCTP_PAD4(ntohs(sch->length));
 	} while (offset < pkt->skb->len);
-
+err:
 	if (priv->flags & NFT_EXTHDR_F_PRESENT)
 		nft_reg_store8(dest, false);
 	else
-- 
2.31.1

Re: [nf-next PATCH] netfilter: nft_exthdr: Search chunks in SCTP packets only

[Pablo Neira Ayuso]

Hi Phil,

On Thu, Jun 10, 2021 at 04:23:16PM +0200, Phil Sutter wrote:
> Since user space does not generate a payload dependency, plain sctp
> chunk matches cause searching in non-SCTP packets, too. Avoid this
> potential mis-interpretation of packet data by checking pkt->tprot.
> 
> Fixes: 133dc203d77df ("netfilter: nft_exthdr: Support SCTP chunks")
> Signed-off-by: Phil Sutter <phil@nwl.cc>
> ---
>  net/netfilter/nft_exthdr.c | 5 ++++-
>  1 file changed, 4 insertions(+), 1 deletion(-)
> 
> diff --git a/net/netfilter/nft_exthdr.c b/net/netfilter/nft_exthdr.c
> index 7f705b5c09de8..1093bb83f8aeb 100644
> --- a/net/netfilter/nft_exthdr.c
> +++ b/net/netfilter/nft_exthdr.c
> @@ -312,6 +312,9 @@ static void nft_exthdr_sctp_eval(const struct nft_expr *expr,
>  	const struct sctp_chunkhdr *sch;
>  	struct sctp_chunkhdr _sch;
>  
> +	if (!pkt->tprot_set || pkt->tprot != IPPROTO_SCTP)
> +		goto err;

nft_set_pktinfo_unspec() already initializes pkt->tprot to zero.

I think it's safe to simplify this to:

	if (pkt->tprot != IPPROTO_SCTP)

Re: [nf-next PATCH] netfilter: nft_exthdr: Search chunks in SCTP packets only

[Phil Sutter]

On Thu, Jun 10, 2021 at 07:43:34PM +0200, Pablo Neira Ayuso wrote:
> On Thu, Jun 10, 2021 at 04:23:16PM +0200, Phil Sutter wrote:
> > Since user space does not generate a payload dependency, plain sctp
> > chunk matches cause searching in non-SCTP packets, too. Avoid this
> > potential mis-interpretation of packet data by checking pkt->tprot.
> > 
> > Fixes: 133dc203d77df ("netfilter: nft_exthdr: Support SCTP chunks")
> > Signed-off-by: Phil Sutter <phil@nwl.cc>
> > ---
> >  net/netfilter/nft_exthdr.c | 5 ++++-
> >  1 file changed, 4 insertions(+), 1 deletion(-)
> > 
> > diff --git a/net/netfilter/nft_exthdr.c b/net/netfilter/nft_exthdr.c
> > index 7f705b5c09de8..1093bb83f8aeb 100644
> > --- a/net/netfilter/nft_exthdr.c
> > +++ b/net/netfilter/nft_exthdr.c
> > @@ -312,6 +312,9 @@ static void nft_exthdr_sctp_eval(const struct nft_expr *expr,
> >  	const struct sctp_chunkhdr *sch;
> >  	struct sctp_chunkhdr _sch;
> >  
> > +	if (!pkt->tprot_set || pkt->tprot != IPPROTO_SCTP)
> > +		goto err;
> 
> nft_set_pktinfo_unspec() already initializes pkt->tprot to zero.
> 
> I think it's safe to simplify this to:
> 
> 	if (pkt->tprot != IPPROTO_SCTP)

Are you sure? Checking the spots that (should) initialize
tprot/tprot_set, in nft_do_chain_inet() it seems that if state->pf is
neither NFPROTO_IPV4 nor NFPROTO_IPV6, nft_do_chain() is called without
prior init. Maybe default case should call nft_set_pktinfo_unspec()?

BTW: The final return call in nft_do_chain_inet_ingress() is dead code,
right?

Thanks, Phil

Re: [nf-next PATCH] netfilter: nft_exthdr: Search chunks in SCTP packets only

[Pablo Neira Ayuso]

On Fri, Jun 11, 2021 at 12:26:24PM +0200, Phil Sutter wrote:
> On Thu, Jun 10, 2021 at 07:43:34PM +0200, Pablo Neira Ayuso wrote:
> > On Thu, Jun 10, 2021 at 04:23:16PM +0200, Phil Sutter wrote:
> > > Since user space does not generate a payload dependency, plain sctp
> > > chunk matches cause searching in non-SCTP packets, too. Avoid this
> > > potential mis-interpretation of packet data by checking pkt->tprot.
> > > 
> > > Fixes: 133dc203d77df ("netfilter: nft_exthdr: Support SCTP chunks")
> > > Signed-off-by: Phil Sutter <phil@nwl.cc>
> > > ---
> > >  net/netfilter/nft_exthdr.c | 5 ++++-
> > >  1 file changed, 4 insertions(+), 1 deletion(-)
> > > 
> > > diff --git a/net/netfilter/nft_exthdr.c b/net/netfilter/nft_exthdr.c
> > > index 7f705b5c09de8..1093bb83f8aeb 100644
> > > --- a/net/netfilter/nft_exthdr.c
> > > +++ b/net/netfilter/nft_exthdr.c
> > > @@ -312,6 +312,9 @@ static void nft_exthdr_sctp_eval(const struct nft_expr *expr,
> > >  	const struct sctp_chunkhdr *sch;
> > >  	struct sctp_chunkhdr _sch;
> > >  
> > > +	if (!pkt->tprot_set || pkt->tprot != IPPROTO_SCTP)
> > > +		goto err;
> > 
> > nft_set_pktinfo_unspec() already initializes pkt->tprot to zero.
> > 
> > I think it's safe to simplify this to:
> > 
> > 	if (pkt->tprot != IPPROTO_SCTP)
> 
> Are you sure? Checking the spots that (should) initialize
> tprot/tprot_set, in nft_do_chain_inet() it seems that if state->pf is
> neither NFPROTO_IPV4 nor NFPROTO_IPV6, nft_do_chain() is called without
> prior init. Maybe default case should call nft_set_pktinfo_unspec()?

state->pf in nft_do_chain_inet() can only be either NFPROTO_IPV4 or
NFPROTO_IPV6.

pkt->tprot_set is there to deal with a corner case: IPPROTO_IP (0).
If pkt->tprot_set == true and pkt->tprot == 0, it means: "match on
IPPROTO_IP". For other IPPROTO_*, checking pkt->tprot looks safe to me.

> BTW: The final return call in nft_do_chain_inet_ingress() is dead code,
> right?

You mean the default case of nft_do_chain_inet_ingress()? inet/ingress
is special, it allows you to filter IPv4 and IPv6 traffic only.
Anything else from ingress is accepted (you should filter it via
netdev family).

Re: [nf-next PATCH] netfilter: nft_exthdr: Search chunks in SCTP packets only

[Phil Sutter]

On Fri, Jun 11, 2021 at 01:56:54PM +0200, Pablo Neira Ayuso wrote:
> On Fri, Jun 11, 2021 at 12:26:24PM +0200, Phil Sutter wrote:
> > On Thu, Jun 10, 2021 at 07:43:34PM +0200, Pablo Neira Ayuso wrote:
> > > On Thu, Jun 10, 2021 at 04:23:16PM +0200, Phil Sutter wrote:
> > > > Since user space does not generate a payload dependency, plain sctp
> > > > chunk matches cause searching in non-SCTP packets, too. Avoid this
> > > > potential mis-interpretation of packet data by checking pkt->tprot.
> > > > 
> > > > Fixes: 133dc203d77df ("netfilter: nft_exthdr: Support SCTP chunks")
> > > > Signed-off-by: Phil Sutter <phil@nwl.cc>
> > > > ---
> > > >  net/netfilter/nft_exthdr.c | 5 ++++-
> > > >  1 file changed, 4 insertions(+), 1 deletion(-)
> > > > 
> > > > diff --git a/net/netfilter/nft_exthdr.c b/net/netfilter/nft_exthdr.c
> > > > index 7f705b5c09de8..1093bb83f8aeb 100644
> > > > --- a/net/netfilter/nft_exthdr.c
> > > > +++ b/net/netfilter/nft_exthdr.c
> > > > @@ -312,6 +312,9 @@ static void nft_exthdr_sctp_eval(const struct nft_expr *expr,
> > > >  	const struct sctp_chunkhdr *sch;
> > > >  	struct sctp_chunkhdr _sch;
> > > >  
> > > > +	if (!pkt->tprot_set || pkt->tprot != IPPROTO_SCTP)
> > > > +		goto err;
> > > 
> > > nft_set_pktinfo_unspec() already initializes pkt->tprot to zero.
> > > 
> > > I think it's safe to simplify this to:
> > > 
> > > 	if (pkt->tprot != IPPROTO_SCTP)
> > 
> > Are you sure? Checking the spots that (should) initialize
> > tprot/tprot_set, in nft_do_chain_inet() it seems that if state->pf is
> > neither NFPROTO_IPV4 nor NFPROTO_IPV6, nft_do_chain() is called without
> > prior init. Maybe default case should call nft_set_pktinfo_unspec()?
> 
> state->pf in nft_do_chain_inet() can only be either NFPROTO_IPV4 or
> NFPROTO_IPV6.

Shouldn't there be a WARN_ON_ONCE or something in the default case then?
Looking at nf_hook(), it seems entirely possible to me that state->pf
might be NFPROTO_ARP, for instance. That's probably just me not getting
it, but things we rely upon shouldn't be hidden that well, right?

> pkt->tprot_set is there to deal with a corner case: IPPROTO_IP (0).
> If pkt->tprot_set == true and pkt->tprot == 0, it means: "match on
> IPPROTO_IP". For other IPPROTO_*, checking pkt->tprot looks safe to me.

Ah, thanks for clarifying! So whenever I check a specific value that's
non-zero, tprot_set doesn't matter. Should I send a patch for the same
change in nft_tcp_header_pointer(), too? (That's where I copied the code
from. ;)

> > BTW: The final return call in nft_do_chain_inet_ingress() is dead code,
> > right?
> 
> You mean the default case of nft_do_chain_inet_ingress()? inet/ingress
> is special, it allows you to filter IPv4 and IPv6 traffic only.
> Anything else from ingress is accepted (you should filter it via
> netdev family).

Oh, sorry. Looks like I had tomatoes on the eyes[1]: I missed that the
non-default cases just 'break' and therefore hit the function's last
line.

Thanks, Phil

[1] Famous German idiom.

Re: [nf-next PATCH] netfilter: nft_exthdr: Search chunks in SCTP packets only

[Pablo Neira Ayuso]

On Fri, Jun 11, 2021 at 02:52:48PM +0200, Phil Sutter wrote:
> On Fri, Jun 11, 2021 at 01:56:54PM +0200, Pablo Neira Ayuso wrote:
> > On Fri, Jun 11, 2021 at 12:26:24PM +0200, Phil Sutter wrote:
> > > On Thu, Jun 10, 2021 at 07:43:34PM +0200, Pablo Neira Ayuso wrote:
> > > > On Thu, Jun 10, 2021 at 04:23:16PM +0200, Phil Sutter wrote:
> > > > > Since user space does not generate a payload dependency, plain sctp
> > > > > chunk matches cause searching in non-SCTP packets, too. Avoid this
> > > > > potential mis-interpretation of packet data by checking pkt->tprot.
> > > > > 
> > > > > Fixes: 133dc203d77df ("netfilter: nft_exthdr: Support SCTP chunks")
> > > > > Signed-off-by: Phil Sutter <phil@nwl.cc>
> > > > > ---
> > > > >  net/netfilter/nft_exthdr.c | 5 ++++-
> > > > >  1 file changed, 4 insertions(+), 1 deletion(-)
> > > > > 
> > > > > diff --git a/net/netfilter/nft_exthdr.c b/net/netfilter/nft_exthdr.c
> > > > > index 7f705b5c09de8..1093bb83f8aeb 100644
> > > > > --- a/net/netfilter/nft_exthdr.c
> > > > > +++ b/net/netfilter/nft_exthdr.c
> > > > > @@ -312,6 +312,9 @@ static void nft_exthdr_sctp_eval(const struct nft_expr *expr,
> > > > >  	const struct sctp_chunkhdr *sch;
> > > > >  	struct sctp_chunkhdr _sch;
> > > > >  
> > > > > +	if (!pkt->tprot_set || pkt->tprot != IPPROTO_SCTP)
> > > > > +		goto err;
> > > > 
> > > > nft_set_pktinfo_unspec() already initializes pkt->tprot to zero.
> > > > 
> > > > I think it's safe to simplify this to:
> > > > 
> > > > 	if (pkt->tprot != IPPROTO_SCTP)
> > > 
> > > Are you sure? Checking the spots that (should) initialize
> > > tprot/tprot_set, in nft_do_chain_inet() it seems that if state->pf is
> > > neither NFPROTO_IPV4 nor NFPROTO_IPV6, nft_do_chain() is called without
> > > prior init. Maybe default case should call nft_set_pktinfo_unspec()?
> > 
> > state->pf in nft_do_chain_inet() can only be either NFPROTO_IPV4 or
> > NFPROTO_IPV6.
> 
> Shouldn't there be a WARN_ON_ONCE or something in the default case then?
> Looking at nf_hook(), it seems entirely possible to me that state->pf
> might be NFPROTO_ARP, for instance. That's probably just me not getting
> it, but things we rely upon shouldn't be hidden that well, right?

nft_do_chain_inet() is called from the NFPROTO_INET hook, which
results in either NFPROTO_IPV4 or NFPROTO_IPV6.

This is hot path, I would not add more code there. The default case is
just there to avoid a warning from gcc.

Probably a comment like /* Should not ever happen */ for the default
case in nft_do_chain_inet() is fine with you? :)

> > pkt->tprot_set is there to deal with a corner case: IPPROTO_IP (0).
> > If pkt->tprot_set == true and pkt->tprot == 0, it means: "match on
> > IPPROTO_IP". For other IPPROTO_*, checking pkt->tprot looks safe to me.
> 
> Ah, thanks for clarifying! So whenever I check a specific value that's
> non-zero, tprot_set doesn't matter. Should I send a patch for the same
> change in nft_tcp_header_pointer(), too? (That's where I copied the code
> from. ;)

I think so, that's fine indeed.