[v2,net-next] VXLAN: Precompute vin for VXLAN header.

Compute VXLAN vin at time of device create so that there is no need
to translate it on packet send and receive.
This patch do not change userspace interface.

CC: Stephen Hemminger <stephen@networkplumber.org>
CC: David Stevens <dlstevens@us.ibm.com>
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
---
v1-v2:
  - Fixed fdb vni.
---
 drivers/net/vxlan.c |   54 +++++++++++++++++++++++++++++---------------------
 1 files changed, 31 insertions(+), 23 deletions(-)

netdev-owner@vger.kernel.org wrote on 03/26/2013 05:33:14 PM:

> 
> Compute VXLAN vin at time of device create so that there is no need
> to translate it on packet send and receive.
> This patch do not change userspace interface.

        The name of the field is "VNI" (Virtual Network Identifier), not 
"vin")

> -static inline struct hlist_head *vni_head(struct net *net, u32 id)
> +static u32 vni_to_vid(__be32 vni)
> +{
> +   return ntohl(vni) >> 8;
> +}
> +
> +static __be32 vid_to_vni(u32 id)
> +{
> +   return htonl(id << 8);
> +}

        What's a "vid"? I know what you're trying to do here, but like 
"vin",
it is not proper VXLAN terminology. I don't think these need to be
functions at all, but rather simply in-line code. I think renaming
this as a "vid" just obscures what it is.
        Changing it to a bitfield of the right size and place might
be more proper, but probably not worth the complexity.
        But if you want them to be functions (or, I'd prefer macros),
I'd suggest something like hton_vni() and ntoh_vni() -- transforming
between the host and packet (network) formats -- and adding "inline".

> +
> +static inline struct hlist_head *vni_head(struct net *net, __be32 id)
>  {
>     struct vxlan_net *vn = net_generic(net, vxlan_net_id);
> 
> -   return &vn->vni_list[hash_32(id, VNI_HASH_BITS)];
> +   return &vn->vni_list[hash_32((__force u32)id, VNI_HASH_BITS)];
>  }
> 
>  /* Look up VNI in a per net namespace table */
> -static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id)
> +static struct vxlan_dev *vxlan_find_vni(struct net *net, __be32 id)
>  {
>     struct vxlan_dev *vxlan;
> 
> @@ -195,7 +205,7 @@ static int vxlan_fdb_info(struct sk_buff *skb, 
> struct vxlan_dev *vxlan,
>         nla_put_be16(skb, NDA_PORT, rdst->remote_port))
>        goto nla_put_failure;
>     if (rdst->remote_vni != vxlan->vni &&
> -       nla_put_be32(skb, NDA_VNI, rdst->remote_vni))
> +       nla_put_be32(skb, NDA_VNI, vni_to_vid(rdst->remote_vni)))
>        goto nla_put_failure;
>     if (rdst->remote_ifindex &&
>         nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
> @@ -261,7 +271,7 @@ static void vxlan_ip_miss(struct net_device 
> *dev, __be32 ipa)
>     memset(&f, 0, sizeof f);
>     f.state = NUD_STALE;
>     f.remote.remote_ip = ipa; /* goes to NDA_DST */
> -   f.remote.remote_vni = VXLAN_N_VID;
> +   f.remote.remote_vni = vid_to_vni(VXLAN_N_VID);

        This is not correct -- VXLAN_N_VID is an invalid VNI with
value 1<<24. Applying vid_to_vni() makes it 1<<32 (==0 since it's 32
bits wide). To do the same thing correctly here, you'd either need to
make it 64 bits, or now set a bit in the low-order byte and then have
that transform to 1<<24 in host representation.

So, perhaps [again, ought to have a name changes], so:

static u32 ntoh_vni(__be32 vni)
{
        if (vni & 0xff)
                return VXLAN_N_VID;
        return ntohl(vni) >> 8;
}

static __be32 hton_vni(u32 id)
{
        if (vni >= VXLAN_N_VID)
                return 1;
        return htonl(id << 8);
}

        Or, instead of "1", you could use a define of VXLAN_N_VID_NET or
some such.
        Of course, that'll work internally only as long as the reserved
bits in the packet are 0; if they're used for something else, you need
to mask them before calling this.

        All in all, I'm not sure it's worth the trouble for a few 
byte-swaps.

                                                        +-DLS

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

On Tue, Mar 26, 2013 at 3:44 PM, David Stevens <dlstevens@us.ibm.com> wrote:
> netdev-owner@vger.kernel.org wrote on 03/26/2013 05:33:14 PM:
>
>>
>> Compute VXLAN vin at time of device create so that there is no need
>> to translate it on packet send and receive.
>> This patch do not change userspace interface.
>
>         The name of the field is "VNI" (Virtual Network Identifier), not
> "vin")
>
>> -static inline struct hlist_head *vni_head(struct net *net, u32 id)
>> +static u32 vni_to_vid(__be32 vni)
>> +{
>> +   return ntohl(vni) >> 8;
>> +}
>> +
>> +static __be32 vid_to_vni(u32 id)
>> +{
>> +   return htonl(id << 8);
>> +}
>
>         What's a "vid"? I know what you're trying to do here, but like
> "vin",
> it is not proper VXLAN terminology. I don't think these need to be
> functions at all, but rather simply in-line code. I think renaming
> this as a "vid" just obscures what it is.
>         Changing it to a bitfield of the right size and place might
> be more proper, but probably not worth the complexity.
>         But if you want them to be functions (or, I'd prefer macros),
> I'd suggest something like hton_vni() and ntoh_vni() -- transforming
> between the host and packet (network) formats -- and adding "inline".
>
>> +
>> +static inline struct hlist_head *vni_head(struct net *net, __be32 id)
>>  {
>>     struct vxlan_net *vn = net_generic(net, vxlan_net_id);
>>
>> -   return &vn->vni_list[hash_32(id, VNI_HASH_BITS)];
>> +   return &vn->vni_list[hash_32((__force u32)id, VNI_HASH_BITS)];
>>  }
>>
>>  /* Look up VNI in a per net namespace table */
>> -static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id)
>> +static struct vxlan_dev *vxlan_find_vni(struct net *net, __be32 id)
>>  {
>>     struct vxlan_dev *vxlan;
>>
>> @@ -195,7 +205,7 @@ static int vxlan_fdb_info(struct sk_buff *skb,
>> struct vxlan_dev *vxlan,
>>         nla_put_be16(skb, NDA_PORT, rdst->remote_port))
>>        goto nla_put_failure;
>>     if (rdst->remote_vni != vxlan->vni &&
>> -       nla_put_be32(skb, NDA_VNI, rdst->remote_vni))
>> +       nla_put_be32(skb, NDA_VNI, vni_to_vid(rdst->remote_vni)))
>>        goto nla_put_failure;
>>     if (rdst->remote_ifindex &&
>>         nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
>> @@ -261,7 +271,7 @@ static void vxlan_ip_miss(struct net_device
>> *dev, __be32 ipa)
>>     memset(&f, 0, sizeof f);
>>     f.state = NUD_STALE;
>>     f.remote.remote_ip = ipa; /* goes to NDA_DST */
>> -   f.remote.remote_vni = VXLAN_N_VID;
>> +   f.remote.remote_vni = vid_to_vni(VXLAN_N_VID);
>
>         This is not correct -- VXLAN_N_VID is an invalid VNI with
> value 1<<24. Applying vid_to_vni() makes it 1<<32 (==0 since it's 32
> bits wide). To do the same thing correctly here, you'd either need to
> make it 64 bits, or now set a bit in the low-order byte and then have
> that transform to 1<<24 in host representation.
>
> So, perhaps [again, ought to have a name changes], so:
>
> static u32 ntoh_vni(__be32 vni)
> {
>         if (vni & 0xff)
>                 return VXLAN_N_VID;
>         return ntohl(vni) >> 8;
> }
>
> static __be32 hton_vni(u32 id)
> {
>         if (vni >= VXLAN_N_VID)
>                 return 1;
>         return htonl(id << 8);
> }
>
>         Or, instead of "1", you could use a define of VXLAN_N_VID_NET or
> some such.
>         Of course, that'll work internally only as long as the reserved
> bits in the packet are 0; if they're used for something else, you need
> to mask them before calling this.
>
>         All in all, I'm not sure it's worth the trouble for a few
> byte-swaps.
>

right, there is no simple solution to avoid byte swaps here, I will
drop this patch for now.
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html