Re: [devel-ipsec] Re: [PATCH ipsec-next v4 5/5] xfrm: add XFRM_MSG_MIGRATE_STATE for single SA migration

[Steffen Klassert <steffen.klassert@secunet.com>]

On Fri, Jan 23, 2026 at 06:58:22AM +0100, Antony Antony wrote:
> On Thu, Jan 22, 2026 at 11:50:16AM +0100, Steffen Klassert via Devel wrote:
> > On Thu, Jan 22, 2026 at 11:33:42AM +0100, Steffen Klassert wrote:
> > > On Mon, Jan 19, 2026 at 09:54:55AM +0100, Antony Antony wrote:
> > > > +
> > > > +static int xfrm_do_migrate_state(struct sk_buff *skb, struct nlmsghdr *nlh,
> > > > +				 struct nlattr **attrs, struct netlink_ext_ack *extack)
> > > > +{
> > > > +	int err = -ESRCH;
> > > > +	struct xfrm_state *x;
> > > > +	struct net *net = sock_net(skb->sk);
> > > > +	struct xfrm_encap_tmpl *encap = NULL;
> > > > +	struct xfrm_user_offload *xuo = NULL;
> > > > +	struct xfrm_migrate m = { .old_saddr.a4 = 0,};
> > > > +	struct xfrm_user_migrate_state *um = nlmsg_data(nlh);
> > > > +
> > > > +	if (!um->id.spi) {
> > > > +		NL_SET_ERR_MSG(extack, "Invalid SPI 0x0");
> > > > +		return -EINVAL;
> > > > +	}
> > > > +
> > > > +	err = copy_from_user_migrate_state(&m, um);
> > > > +	if (err)
> > > > +		return err;
> > > > +
> > > > +	x = xfrm_user_state_lookup(net, &um->id, attrs, &err);
> > > > +
> > > > +	if (x) {
> > > > +		struct xfrm_state *xc;
> > > > +
> > > > +		if (!x->dir) {
> > > > +			NL_SET_ERR_MSG(extack, "State direction is invalid");
> > > > +			err = -EINVAL;
> > > > +			goto error;
> > > > +		}
> > > > +
> > > > +		if (attrs[XFRMA_ENCAP]) {
> > > > +			encap = kmemdup(nla_data(attrs[XFRMA_ENCAP]),
> > > > +					sizeof(*encap), GFP_KERNEL);
> > > > +			if (!encap) {
> > > > +				err = -ENOMEM;
> > > > +				goto error;
> > > > +			}
> > > > +		}
> > > > +		if (attrs[XFRMA_OFFLOAD_DEV]) {
> > > > +			xuo = kmemdup(nla_data(attrs[XFRMA_OFFLOAD_DEV]),
> > > > +				      sizeof(*xuo), GFP_KERNEL);
> > > > +			if (!xuo) {
> > > > +				err = -ENOMEM;
> > > > +				goto error;
> > > > +			}
> > > > +		}
> > > > +		xc = xfrm_state_migrate(x, &m, encap, net, xuo, extack);
> > > > +		if (xc) {
> > > > +			xfrm_state_delete(x);
> > > 
> > > This xfrm_state_delete() comes too late. You need to synchronize this
> > > with xfrm_state_clone_and_setup() by holding the state lock.
> > > Otherwise we might hand out the same sequence number etc. more than
> > > one time. Unfortunately xfrm_migrate() has the same problem, the
> > > error recovery mechanism in xfrm_migrate() looks broken.
> > 
> > The xfrm_migrate() problem is nasty, looks like we can't recover
> > if the state migration fails. We probably have to delete the
> > states and wait for the userspace to insert new ones.
> 
> userspace should notice the inconsistany. Most of them are ENOMEM or EMSGSIZE
> 
> > 
> > Another thing, xfrm_user_state_lookup() returns a reference on x,
> > you should drop it somewhere.
> 
> There is a call a few lines bellow, the label was "error:"  I renamed it to  
> "out:" for clariy.
> 
> +out:
> +       xfrm_state_put(x);

Yes, this makes it more clear that we hit this even without error.

Maybe you should do a 'if (!X)' check on the retuen value of
xfrm_user_state_lookup(). This woulds make the structure of
the function even clearer.