xfrm: Use nested-BH locking for nat_keepalive_sk_ipv[46]
authorSebastian Andrzej Siewior <bigeasy@linutronix.de>
Mon, 12 May 2025 09:27:27 +0000 (11:27 +0200)
committerPaolo Abeni <pabeni@redhat.com>
Thu, 15 May 2025 13:23:31 +0000 (15:23 +0200)
nat_keepalive_sk_ipv[46] is a per-CPU variable and relies on disabled BH
for its locking. Without per-CPU locking in local_bh_disable() on
PREEMPT_RT this data structure requires explicit locking.

Use sock_bh_locked which has a sock pointer and a local_lock_t. Use
local_lock_nested_bh() for locking. This change adds only lockdep
coverage and does not alter the functional behaviour for !PREEMPT_RT.

Cc: Steffen Klassert <steffen.klassert@secunet.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Link: https://patch.msgid.link/20250512092736.229935-7-bigeasy@linutronix.de
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
net/xfrm/xfrm_nat_keepalive.c

index 82f0a301683f0fcf11ea97a6be2a5bd6c7bd012d..ebf95d48e86c149ceabd6e059d85d39660df5b8c 100644 (file)
@@ -9,9 +9,13 @@
 #include <net/ip6_checksum.h>
 #include <net/xfrm.h>
 
-static DEFINE_PER_CPU(struct sock *, nat_keepalive_sk_ipv4);
+static DEFINE_PER_CPU(struct sock_bh_locked, nat_keepalive_sk_ipv4) = {
+       .bh_lock = INIT_LOCAL_LOCK(bh_lock),
+};
 #if IS_ENABLED(CONFIG_IPV6)
-static DEFINE_PER_CPU(struct sock *, nat_keepalive_sk_ipv6);
+static DEFINE_PER_CPU(struct sock_bh_locked, nat_keepalive_sk_ipv6) = {
+       .bh_lock = INIT_LOCAL_LOCK(bh_lock),
+};
 #endif
 
 struct nat_keepalive {
@@ -56,10 +60,12 @@ static int nat_keepalive_send_ipv4(struct sk_buff *skb,
 
        skb_dst_set(skb, &rt->dst);
 
-       sk = *this_cpu_ptr(&nat_keepalive_sk_ipv4);
+       local_lock_nested_bh(&nat_keepalive_sk_ipv4.bh_lock);
+       sk = this_cpu_read(nat_keepalive_sk_ipv4.sock);
        sock_net_set(sk, net);
        err = ip_build_and_send_pkt(skb, sk, fl4.saddr, fl4.daddr, NULL, tos);
        sock_net_set(sk, &init_net);
+       local_unlock_nested_bh(&nat_keepalive_sk_ipv4.bh_lock);
        return err;
 }
 
@@ -89,15 +95,19 @@ static int nat_keepalive_send_ipv6(struct sk_buff *skb,
        fl6.fl6_sport = ka->encap_sport;
        fl6.fl6_dport = ka->encap_dport;
 
-       sk = *this_cpu_ptr(&nat_keepalive_sk_ipv6);
+       local_lock_nested_bh(&nat_keepalive_sk_ipv6.bh_lock);
+       sk = this_cpu_read(nat_keepalive_sk_ipv6.sock);
        sock_net_set(sk, net);
        dst = ipv6_stub->ipv6_dst_lookup_flow(net, sk, &fl6, NULL);
-       if (IS_ERR(dst))
+       if (IS_ERR(dst)) {
+               local_unlock_nested_bh(&nat_keepalive_sk_ipv6.bh_lock);
                return PTR_ERR(dst);
+       }
 
        skb_dst_set(skb, dst);
        err = ipv6_stub->ip6_xmit(sk, skb, &fl6, skb->mark, NULL, 0, 0);
        sock_net_set(sk, &init_net);
+       local_unlock_nested_bh(&nat_keepalive_sk_ipv6.bh_lock);
        return err;
 }
 #endif
@@ -202,7 +212,7 @@ static void nat_keepalive_work(struct work_struct *work)
                                      (ctx.next_run - ctx.now) * HZ);
 }
 
-static int nat_keepalive_sk_init(struct sock * __percpu *socks,
+static int nat_keepalive_sk_init(struct sock_bh_locked __percpu *socks,
                                 unsigned short family)
 {
        struct sock *sk;
@@ -214,22 +224,22 @@ static int nat_keepalive_sk_init(struct sock * __percpu *socks,
                if (err < 0)
                        goto err;
 
-               *per_cpu_ptr(socks, i) = sk;
+               per_cpu_ptr(socks, i)->sock = sk;
        }
 
        return 0;
 err:
        for_each_possible_cpu(i)
-               inet_ctl_sock_destroy(*per_cpu_ptr(socks, i));
+               inet_ctl_sock_destroy(per_cpu_ptr(socks, i)->sock);
        return err;
 }
 
-static void nat_keepalive_sk_fini(struct sock * __percpu *socks)
+static void nat_keepalive_sk_fini(struct sock_bh_locked __percpu *socks)
 {
        int i;
 
        for_each_possible_cpu(i)
-               inet_ctl_sock_destroy(*per_cpu_ptr(socks, i));
+               inet_ctl_sock_destroy(per_cpu_ptr(socks, i)->sock);
 }
 
 void xfrm_nat_keepalive_state_updated(struct xfrm_state *x)