On Thu, Jan 22, 2026 at 11:33:42AM +0100, Steffen Klassert via Devel wrote:

On Mon, Jan 19, 2026 at 09:54:55AM +0100, Antony Antony wrote:

quoted

+
+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.

yes. I added a lock, x->lock, and within the lock call __xfrm_state_delete()
That seems to work, not encountring a deadlock.

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.