[PATCH 1/1] bonding: crypto offload enabled, non-offload slave failover, rekey failed

[PATCH 1/1] bonding: crypto offload enabled, non-offload slave failover, rekey failed

[David Dai]

Create a bonding device (i.e. bond0) in active-backup mode, 2 slaves.
Active slave: offload capable interface (i.e. eth1), primary interface.
Backup slave: non-offload capable interface(i.e. eth2).
Configure strongswan service swantl.conf child SA "hw_offload = crypto"
Start strongswan service
IPSec Crytpo Offload is enabled on top of bond0. i.e. 
ip xfrm state |grep offload
        crypto offload parameters: dev bond0 dir out mode crypto
	crypto offload parameters: dev bond0 dir in mode crypto

Active slave eth1 takes adavantage of IPSec Crypto Offload capability. 

If active slave eth1 is down for any reason (i.e. eth1 link down):
ip link set down dev eth1
non-offload capable interface eth2 failover to becomes active slave. 
The existing SAs can continue use software IPsec after failover.
Traffic still keeps going properly.

However if eth1 link had not recovered yet, strongswan service does
new child SA rekey, or uses swanctl command to do new child SA rekey, 
it will fail because active slave eth2 doesn't support crypto offload.
In bond_ipsec_add_sa routine, it returns -EINVAL now, which is 
treated as fatal error by xfrm_dev_state_add routine in kernel xfrm.

To make the non-offload active slave survive the child SA rekey, need 
to make bond_ipsec_add_sa routine returns -EOPNOTSUPP instead when 
active slave doesn't support IPsec Crypto offload, the xfrm will 
gracefully fallback to create new SA using Software IPsec. 
Network traffic can keep going.

After offload capable interface eth1 link is up, becomes active slave,
next time strongswan child SA rekey will create a new SA which enables
crypto offload again.

This way, network traffic is never interrupted, always keeps going.

Signed-off-by: David Dai <zdai@linux.ibm.com>
Tested-by: David Dai <zdai@linux.ibm.com>
---
 drivers/net/bonding/bond_main.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/bond_main.c
index ef9eb0c53c66..79fc892dab07 100644
--- a/drivers/net/bonding/bond_main.c
+++ b/drivers/net/bonding/bond_main.c
@@ -490,7 +490,7 @@ static int bond_ipsec_add_sa(struct net_device *bond_dev,
 	    !real_dev->xfrmdev_ops->xdo_dev_state_add ||
 	    netif_is_bond_master(real_dev)) {
 		NL_SET_ERR_MSG_MOD(extack, "Slave does not support ipsec offload");
-		err = -EINVAL;
+		err = -EOPNOTSUPP;
 		goto out;
 	}
 
-- 
2.55.0

Re: [PATCH 1/1] bonding: crypto offload enabled, non-offload slave failover, rekey failed

[Hangbin Liu]

On Thu, Sep 03, 2026 at 05:01:23PM -0500, David Dai wrote:
> Create a bonding device (i.e. bond0) in active-backup mode, 2 slaves.
> Active slave: offload capable interface (i.e. eth1), primary interface.
> Backup slave: non-offload capable interface(i.e. eth2).
> Configure strongswan service swantl.conf child SA "hw_offload = crypto"
> Start strongswan service
> IPSec Crytpo Offload is enabled on top of bond0. i.e. 
> ip xfrm state |grep offload
>         crypto offload parameters: dev bond0 dir out mode crypto
> 	crypto offload parameters: dev bond0 dir in mode crypto
> 
> Active slave eth1 takes adavantage of IPSec Crypto Offload capability. 
> 
> If active slave eth1 is down for any reason (i.e. eth1 link down):
> ip link set down dev eth1
> non-offload capable interface eth2 failover to becomes active slave. 
> The existing SAs can continue use software IPsec after failover.
> Traffic still keeps going properly.
> 
> However if eth1 link had not recovered yet, strongswan service does
> new child SA rekey, or uses swanctl command to do new child SA rekey, 
> it will fail because active slave eth2 doesn't support crypto offload.
> In bond_ipsec_add_sa routine, it returns -EINVAL now, which is 
> treated as fatal error by xfrm_dev_state_add routine in kernel xfrm.
> 
> To make the non-offload active slave survive the child SA rekey, need 
> to make bond_ipsec_add_sa routine returns -EOPNOTSUPP instead when 
> active slave doesn't support IPsec Crypto offload, the xfrm will 
> gracefully fallback to create new SA using Software IPsec. 
> Network traffic can keep going.
> 
> After offload capable interface eth1 link is up, becomes active slave,
> next time strongswan child SA rekey will create a new SA which enables
> crypto offload again.
> 
> This way, network traffic is never interrupted, always keeps going.
> 
> Signed-off-by: David Dai <zdai@linux.ibm.com>
> Tested-by: David Dai <zdai@linux.ibm.com>
> ---
>  drivers/net/bonding/bond_main.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/bond_main.c
> index ef9eb0c53c66..79fc892dab07 100644
> --- a/drivers/net/bonding/bond_main.c
> +++ b/drivers/net/bonding/bond_main.c
> @@ -490,7 +490,7 @@ static int bond_ipsec_add_sa(struct net_device *bond_dev,
>  	    !real_dev->xfrmdev_ops->xdo_dev_state_add ||
>  	    netif_is_bond_master(real_dev)) {
>  		NL_SET_ERR_MSG_MOD(extack, "Slave does not support ipsec offload");
> -		err = -EINVAL;
> +		err = -EOPNOTSUPP;
>  		goto out;
>  	}
>  
> -- 
> 2.55.0
> 

The patch looks good to me. I am just not sure whether we should shorten
the description. Let's wait for others' opinions.

Thanks
Hangbin