static inline int unix_sysctl_register(struct net *net) { return 0; }
 static inline void unix_sysctl_unregister(struct net *net) {}
 #endif
+
+#ifdef CONFIG_BPF_SYSCALL
+extern struct proto unix_proto;
+
+int unix_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore);
+void __init unix_bpf_build_proto(void);
+#else
+static inline void __init unix_bpf_build_proto(void)
+{}
+#endif
 #endif
 
        release_sock(sk);
        saved_close(sk, timeout);
 }
+EXPORT_SYMBOL_GPL(sock_map_close);
 
 static int sock_map_iter_attach_target(struct bpf_prog *prog,
                                       union bpf_iter_link_info *linfo,
 
         */
 }
 
-static struct proto unix_proto = {
+struct proto unix_proto = {
        .name                   = "UNIX",
        .owner                  = THIS_MODULE,
        .obj_size               = sizeof(struct unix_sock),
        .close                  = unix_close,
+#ifdef CONFIG_BPF_SYSCALL
+       .psock_update_sk_prot   = unix_bpf_update_proto,
+#endif
 };
 
 static struct sock *unix_create1(struct net *net, struct socket *sock, int kern)
 
        sock_register(&unix_family_ops);
        register_pernet_subsys(&unix_net_ops);
+       unix_bpf_build_proto();
 out:
        return rc;
 }
 
--- /dev/null
+// SPDX-License-Identifier: GPL-2.0
+/* Copyright (c) 2021 Cong Wang <cong.wang@bytedance.com> */
+
+#include <linux/skmsg.h>
+#include <linux/bpf.h>
+#include <net/sock.h>
+#include <net/af_unix.h>
+
+static struct proto *unix_prot_saved __read_mostly;
+static DEFINE_SPINLOCK(unix_prot_lock);
+static struct proto unix_bpf_prot;
+
+static void unix_bpf_rebuild_protos(struct proto *prot, const struct proto *base)
+{
+       *prot        = *base;
+       prot->close  = sock_map_close;
+}
+
+static void unix_bpf_check_needs_rebuild(struct proto *ops)
+{
+       if (unlikely(ops != smp_load_acquire(&unix_prot_saved))) {
+               spin_lock_bh(&unix_prot_lock);
+               if (likely(ops != unix_prot_saved)) {
+                       unix_bpf_rebuild_protos(&unix_bpf_prot, ops);
+                       smp_store_release(&unix_prot_saved, ops);
+               }
+               spin_unlock_bh(&unix_prot_lock);
+       }
+}
+
+int unix_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore)
+{
+       if (restore) {
+               sk->sk_write_space = psock->saved_write_space;
+               WRITE_ONCE(sk->sk_prot, psock->sk_proto);
+               return 0;
+       }
+
+       unix_bpf_check_needs_rebuild(psock->sk_proto);
+       WRITE_ONCE(sk->sk_prot, &unix_bpf_prot);
+       return 0;
+}
+
+void __init unix_bpf_build_proto(void)
+{
+       unix_bpf_rebuild_protos(&unix_bpf_prot, &unix_proto);
+}