[conntrack-tools PATCH] conntrack: Fix CIDR to mask conversion on Big Endian

[conntrack-tools PATCH] conntrack: Fix CIDR to mask conversion on Big Endian

[Phil Sutter]

Code assumed host architecture to be Little Endian. Instead produce a
proper mask by pushing the set bits into most significant position and
apply htonl() on the result.

Fixes: 3f6a2e90936bb ("conntrack: add support for CIDR notation")
Signed-off-by: Phil Sutter <phil@nwl.cc>
---
 src/conntrack.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/conntrack.c b/src/conntrack.c
index c980a13f33d2c..baafcbd869c12 100644
--- a/src/conntrack.c
+++ b/src/conntrack.c
@@ -2210,7 +2210,7 @@ nfct_build_netmask(uint32_t *dst, int b, int n)
 			dst[i] = 0xffffffff;
 			b -= 32;
 		} else if (b > 0) {
-			dst[i] = (1 << b) - 1;
+			dst[i] = htonl(((1 << b) - 1) << (32 - b));
 			b = 0;
 		} else {
 			dst[i] = 0;
-- 
2.22.0

Re: [conntrack-tools PATCH] conntrack: Fix CIDR to mask conversion on Big Endian

[Pablo Neira Ayuso]

On Mon, Sep 02, 2019 at 06:44:31PM +0200, Phil Sutter wrote:
> Code assumed host architecture to be Little Endian. Instead produce a
> proper mask by pushing the set bits into most significant position and
> apply htonl() on the result.
> 
> Fixes: 3f6a2e90936bb ("conntrack: add support for CIDR notation")
> Signed-off-by: Phil Sutter <phil@nwl.cc>
> ---
>  src/conntrack.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/src/conntrack.c b/src/conntrack.c
> index c980a13f33d2c..baafcbd869c12 100644
> --- a/src/conntrack.c
> +++ b/src/conntrack.c
> @@ -2210,7 +2210,7 @@ nfct_build_netmask(uint32_t *dst, int b, int n)
>  			dst[i] = 0xffffffff;
>  			b -= 32;
>  		} else if (b > 0) {
> -			dst[i] = (1 << b) - 1;
> +			dst[i] = htonl(((1 << b) - 1) << (32 - b));

Simply this instead?

                        dst[i] = htonl(((1 << b) - 1);

>  			b = 0;
>  		} else {
>  			dst[i] = 0;
> -- 
> 2.22.0
> 

Re: [conntrack-tools PATCH] conntrack: Fix CIDR to mask conversion on Big Endian

[Phil Sutter]

On Tue, Sep 03, 2019 at 10:34:47PM +0200, Pablo Neira Ayuso wrote:
> On Mon, Sep 02, 2019 at 06:44:31PM +0200, Phil Sutter wrote:
> > Code assumed host architecture to be Little Endian. Instead produce a
> > proper mask by pushing the set bits into most significant position and
> > apply htonl() on the result.
> > 
> > Fixes: 3f6a2e90936bb ("conntrack: add support for CIDR notation")
> > Signed-off-by: Phil Sutter <phil@nwl.cc>
> > ---
> >  src/conntrack.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/src/conntrack.c b/src/conntrack.c
> > index c980a13f33d2c..baafcbd869c12 100644
> > --- a/src/conntrack.c
> > +++ b/src/conntrack.c
> > @@ -2210,7 +2210,7 @@ nfct_build_netmask(uint32_t *dst, int b, int n)
> >  			dst[i] = 0xffffffff;
> >  			b -= 32;
> >  		} else if (b > 0) {
> > -			dst[i] = (1 << b) - 1;
> > +			dst[i] = htonl(((1 << b) - 1) << (32 - b));
> 
> Simply this instead?
> 
>                         dst[i] = htonl(((1 << b) - 1);

You got me confused, so I played with different options. To see the
results, I used:

| union {
|         uint32_t i;
|         char b[4];
| } u;

What we need in b is 'ff ff ff 00' for a prefix length of 24. Your
suggested alternative does not compile, so I tried both options for the
closing brace:

| htonl((1 << 24) - 1)

This turns into '00 ff ff ff' for both LE and BE, the opposite of what
we need.

| htonl((1 << 24)) - 1

This turns into '00 00 00 00' on LE and '00 ff ff ff' on BE.

My code leads to correct result on either architecture and I don't see a
simpler way of doing it.

Cheers, Phil

Re: [conntrack-tools PATCH] conntrack: Fix CIDR to mask conversion on Big Endian

[Florian Westphal]

Phil Sutter <phil@nwl.cc> wrote:
> What we need in b is 'ff ff ff 00' for a prefix length of 24. Your
> suggested alternative does not compile, so I tried both options for the
> closing brace:
> 
> | htonl((1 << 24) - 1)
> 
> This turns into '00 ff ff ff' for both LE and BE, the opposite of what
> we need.
> 
> | htonl((1 << 24)) - 1
> 
> This turns into '00 00 00 00' on LE and '00 ff ff ff' on BE.
> 
> My code leads to correct result on either architecture and I don't see a
> simpler way of doing it.

htonl(~0u << (32 - i)) would work, assuming i > 0 and <= 32.

Re: [conntrack-tools PATCH] conntrack: Fix CIDR to mask conversion on Big Endian

[Phil Sutter]

Hi,

On Wed, Sep 04, 2019 at 10:21:28AM +0200, Florian Westphal wrote:
> Phil Sutter <phil@nwl.cc> wrote:
> > What we need in b is 'ff ff ff 00' for a prefix length of 24. Your
> > suggested alternative does not compile, so I tried both options for the
> > closing brace:
> > 
> > | htonl((1 << 24) - 1)
> > 
> > This turns into '00 ff ff ff' for both LE and BE, the opposite of what
> > we need.
> > 
> > | htonl((1 << 24)) - 1
> > 
> > This turns into '00 00 00 00' on LE and '00 ff ff ff' on BE.
> > 
> > My code leads to correct result on either architecture and I don't see a
> > simpler way of doing it.
> 
> htonl(~0u << (32 - i)) would work, assuming i > 0 and <= 32.

Ah, indeed! Left-shifting all-ones didn't come to mind. I'll send a v2.

Thanks, Phil