[PATCH] rtnetlink: add needed_{head,tail}room attributes

[PATCH] rtnetlink: add needed_{head,tail}room attributes

[Alasdair McWilliam]

Various network interface types make use of needed_{head,tail}room values
to efficiently reserve buffer space for additional encapsulation headers,
such as VXLAN, Geneve, IPSec, etc. However, it is not currently possible
to query these values in a generic way.

Introduce ability to query the needed_{head,tail}room values of a network
device via rtnetlink, such that applications that may wish to use these
values can do so.

Signed-off-by: Alasdair McWilliam <alasdair@mcwilliam.dev>
---
 include/uapi/linux/if_link.h |  2 ++
 net/core/rtnetlink.c         | 10 +++++++++-
 2 files changed, 11 insertions(+), 1 deletion(-)

diff --git a/include/uapi/linux/if_link.h b/include/uapi/linux/if_link.h
index 45f56c9f95d9..3b491d96e52e 100644
--- a/include/uapi/linux/if_link.h
+++ b/include/uapi/linux/if_link.h
@@ -379,6 +379,8 @@ enum {
 	IFLA_DPLL_PIN,
 	IFLA_MAX_PACING_OFFLOAD_HORIZON,
 	IFLA_NETNS_IMMUTABLE,
+	IFLA_HEADROOM,
+	IFLA_TAILROOM,
 	__IFLA_MAX
 };
 
diff --git a/net/core/rtnetlink.c b/net/core/rtnetlink.c
index 094b085cff20..c68e20a36daa 100644
--- a/net/core/rtnetlink.c
+++ b/net/core/rtnetlink.c
@@ -1326,6 +1326,8 @@ static noinline size_t if_nlmsg_size(const struct net_device *dev,
 	       + rtnl_devlink_port_size(dev)
 	       + rtnl_dpll_pin_size(dev)
 	       + nla_total_size(8)  /* IFLA_MAX_PACING_OFFLOAD_HORIZON */
+	       + nla_total_size(2)  /* IFLA_HEADROOM */
+	       + nla_total_size(2)  /* IFLA_TAILROOM */
 	       + 0;
 }
 
@@ -2091,7 +2093,11 @@ static int rtnl_fill_ifinfo(struct sk_buff *skb,
 	    nla_put_u32(skb, IFLA_CARRIER_UP_COUNT,
 			atomic_read(&dev->carrier_up_count)) ||
 	    nla_put_u32(skb, IFLA_CARRIER_DOWN_COUNT,
-			atomic_read(&dev->carrier_down_count)))
+			atomic_read(&dev->carrier_down_count)) ||
+	    nla_put_u16(skb, IFLA_HEADROOM,
+			READ_ONCE(dev->needed_headroom)) ||
+	    nla_put_u16(skb, IFLA_TAILROOM,
+			READ_ONCE(dev->needed_tailroom)))
 		goto nla_put_failure;
 
 	if (rtnl_fill_proto_down(skb, dev))
@@ -2243,6 +2249,8 @@ static const struct nla_policy ifla_policy[IFLA_MAX+1] = {
 	[IFLA_GSO_IPV4_MAX_SIZE]	= NLA_POLICY_MIN(NLA_U32, MAX_TCP_HEADER + 1),
 	[IFLA_GRO_IPV4_MAX_SIZE]	= { .type = NLA_U32 },
 	[IFLA_NETNS_IMMUTABLE]	= { .type = NLA_REJECT },
+	[IFLA_HEADROOM]		= { .type = NLA_U16 },
+	[IFLA_TAILROOM]		= { .type = NLA_U16 },
 };
 
 static const struct nla_policy ifla_info_policy[IFLA_INFO_MAX+1] = {
-- 
2.47.3

Re: [PATCH] rtnetlink: add needed_{head,tail}room attributes

[Daniel Borkmann]

Hi Alasdair,

On 9/15/25 6:32 PM, Alasdair McWilliam wrote:
> Various network interface types make use of needed_{head,tail}room values
> to efficiently reserve buffer space for additional encapsulation headers,
> such as VXLAN, Geneve, IPSec, etc. However, it is not currently possible
> to query these values in a generic way.
> 
> Introduce ability to query the needed_{head,tail}room values of a network
> device via rtnetlink, such that applications that may wish to use these
> values can do so.
> 
> Signed-off-by: Alasdair McWilliam <alasdair@mcwilliam.dev>
> ---
>   include/uapi/linux/if_link.h |  2 ++
>   net/core/rtnetlink.c         | 10 +++++++++-
>   2 files changed, 11 insertions(+), 1 deletion(-)
> 
> diff --git a/include/uapi/linux/if_link.h b/include/uapi/linux/if_link.h
> index 45f56c9f95d9..3b491d96e52e 100644
> --- a/include/uapi/linux/if_link.h
> +++ b/include/uapi/linux/if_link.h
> @@ -379,6 +379,8 @@ enum {
>   	IFLA_DPLL_PIN,
>   	IFLA_MAX_PACING_OFFLOAD_HORIZON,
>   	IFLA_NETNS_IMMUTABLE,
> +	IFLA_HEADROOM,
> +	IFLA_TAILROOM,
>   	__IFLA_MAX
>   };
>   
> diff --git a/net/core/rtnetlink.c b/net/core/rtnetlink.c
> index 094b085cff20..c68e20a36daa 100644
> --- a/net/core/rtnetlink.c
> +++ b/net/core/rtnetlink.c
> @@ -1326,6 +1326,8 @@ static noinline size_t if_nlmsg_size(const struct net_device *dev,
>   	       + rtnl_devlink_port_size(dev)
>   	       + rtnl_dpll_pin_size(dev)
>   	       + nla_total_size(8)  /* IFLA_MAX_PACING_OFFLOAD_HORIZON */
> +	       + nla_total_size(2)  /* IFLA_HEADROOM */
> +	       + nla_total_size(2)  /* IFLA_TAILROOM */
>   	       + 0;
>   }
>   
> @@ -2091,7 +2093,11 @@ static int rtnl_fill_ifinfo(struct sk_buff *skb,
>   	    nla_put_u32(skb, IFLA_CARRIER_UP_COUNT,
>   			atomic_read(&dev->carrier_up_count)) ||
>   	    nla_put_u32(skb, IFLA_CARRIER_DOWN_COUNT,
> -			atomic_read(&dev->carrier_down_count)))
> +			atomic_read(&dev->carrier_down_count)) ||
> +	    nla_put_u16(skb, IFLA_HEADROOM,
> +			READ_ONCE(dev->needed_headroom)) ||
> +	    nla_put_u16(skb, IFLA_TAILROOM,
> +			READ_ONCE(dev->needed_tailroom)))
>   		goto nla_put_failure;
>   
>   	if (rtnl_fill_proto_down(skb, dev))
> @@ -2243,6 +2249,8 @@ static const struct nla_policy ifla_policy[IFLA_MAX+1] = {
>   	[IFLA_GSO_IPV4_MAX_SIZE]	= NLA_POLICY_MIN(NLA_U32, MAX_TCP_HEADER + 1),
>   	[IFLA_GRO_IPV4_MAX_SIZE]	= { .type = NLA_U32 },
>   	[IFLA_NETNS_IMMUTABLE]	= { .type = NLA_REJECT },
> +	[IFLA_HEADROOM]		= { .type = NLA_U16 },
> +	[IFLA_TAILROOM]		= { .type = NLA_U16 },

Given this is for dumping only, we'd need to replace NLA_U16 above with NLA_REJECT
like in case of IFLA_NETNS_IMMUTABLE.

Also the Documentation/netlink/specs/rt_link.yaml needs an extension, otherwise lgtm.
$subj should have [PATCH net-next] to indicate the target tree.

Thanks,
Daniel

Re: [PATCH] rtnetlink: add needed_{head,tail}room attributes

[Stephen Hemminger]

On Mon, 15 Sep 2025 18:49:55 +0200
Daniel Borkmann <daniel@iogearbox.net> wrote:

> Hi Alasdair,
> 
> On 9/15/25 6:32 PM, Alasdair McWilliam wrote:
> > Various network interface types make use of needed_{head,tail}room values
> > to efficiently reserve buffer space for additional encapsulation headers,
> > such as VXLAN, Geneve, IPSec, etc. However, it is not currently possible
> > to query these values in a generic way.
> > 
> > Introduce ability to query the needed_{head,tail}room values of a network
> > device via rtnetlink, such that applications that may wish to use these
> > values can do so.
> > 
> > Signed-off-by: Alasdair McWilliam <alasdair@mcwilliam.dev>

Why is this info needed in user space?
How would an application use this?
Seems like a hardware specific optimization that should be hidden in driver.

Re: [PATCH] rtnetlink: add needed_{head,tail}room attributes

[Daniel Borkmann]

On 9/15/25 7:20 PM, Stephen Hemminger wrote:
> On Mon, 15 Sep 2025 18:49:55 +0200
> Daniel Borkmann <daniel@iogearbox.net> wrote:
>> On 9/15/25 6:32 PM, Alasdair McWilliam wrote:
>>> Various network interface types make use of needed_{head,tail}room values
>>> to efficiently reserve buffer space for additional encapsulation headers,
>>> such as VXLAN, Geneve, IPSec, etc. However, it is not currently possible
>>> to query these values in a generic way.
>>>
>>> Introduce ability to query the needed_{head,tail}room values of a network
>>> device via rtnetlink, such that applications that may wish to use these
>>> values can do so.
>>>
>>> Signed-off-by: Alasdair McWilliam <alasdair@mcwilliam.dev>
> 
> Why is this info needed in user space?
> How would an application use this?
> Seems like a hardware specific optimization that should be hidden in driver.

We would like to use this in Cilium. Basically our Cilium agent iterates over
the present devices based on the Cilium user config (e.g. direct routing/vxlan/
geneve/wireguard/etc) to then configure netkit in order to expose the
needed_{head,tail}room into K8s Pods, see also b9ed315d3c4c ("netkit: Allow
for configuring needed_{head,tail}room").

Thanks,
Daniel

Re: [PATCH] rtnetlink: add needed_{head,tail}room attributes

[Alasdair McWilliam]

Hi Daniel,

On 15/09/2025 17:49, Daniel Borkmann wrote:
> Hi Alasdair,
> 

>> @@ -2243,6 +2249,8 @@ static const struct nla_policy
>> ifla_policy[IFLA_MAX+1] = {
>>       [IFLA_GSO_IPV4_MAX_SIZE]    = NLA_POLICY_MIN(NLA_U32,
>> MAX_TCP_HEADER + 1),
>>       [IFLA_GRO_IPV4_MAX_SIZE]    = { .type = NLA_U32 },
>>       [IFLA_NETNS_IMMUTABLE]    = { .type = NLA_REJECT },
>> +    [IFLA_HEADROOM]        = { .type = NLA_U16 },
>> +    [IFLA_TAILROOM]        = { .type = NLA_U16 },
> 
> Given this is for dumping only, we'd need to replace NLA_U16 above with
> NLA_REJECT
> like in case of IFLA_NETNS_IMMUTABLE.
> 
> Also the Documentation/netlink/specs/rt_link.yaml needs an extension,
> otherwise lgtm.
> $subj should have [PATCH net-next] to indicate the target tree.
> 

Acknowledged. I will revise and re-submit. I will also include
additional context in the commit log as per your later email in this thread.

> Thanks,
> Daniel

Thanks!
Alasdair