tcp: fix another uninit-value (sk_rx_queue_mapping)

KMSAN is still not happy [1].

I missed that passive connections do not inherit their
sk_rx_queue_mapping values from the request socket,
but instead tcp_child_process() is calling
sk_mark_napi_id(child, skb)

We have many sk_mark_napi_id() callers, so I am providing
a new helper, forcing the setting sk_rx_queue_mapping
and sk_napi_id.

Note that we had no KMSAN report for sk_napi_id because
passive connections got a copy of this field from the listener.
sk_rx_queue_mapping in the other hand is inside the
sk_dontcopy_begin/sk_dontcopy_end so sk_clone_lock()
leaves this field uninitialized.

We might remove dead code populating req->sk_rx_queue_mapping
in the future.

[1]

BUG: KMSAN: uninit-value in __sk_rx_queue_set include/net/sock.h:1924 [inline]
BUG: KMSAN: uninit-value in sk_rx_queue_update include/net/sock.h:1938 [inline]
BUG: KMSAN: uninit-value in sk_mark_napi_id include/net/busy_poll.h:136 [inline]
BUG: KMSAN: uninit-value in tcp_child_process+0xb42/0x1050 net/ipv4/tcp_minisocks.c:833
 __sk_rx_queue_set include/net/sock.h:1924 [inline]
 sk_rx_queue_update include/net/sock.h:1938 [inline]
 sk_mark_napi_id include/net/busy_poll.h:136 [inline]
 tcp_child_process+0xb42/0x1050 net/ipv4/tcp_minisocks.c:833
 tcp_v4_rcv+0x3d83/0x4ed0 net/ipv4/tcp_ipv4.c:2066
 ip_protocol_deliver_rcu+0x760/0x10b0 net/ipv4/ip_input.c:204
 ip_local_deliver_finish net/ipv4/ip_input.c:231 [inline]
 NF_HOOK include/linux/netfilter.h:307 [inline]
 ip_local_deliver+0x584/0x8c0 net/ipv4/ip_input.c:252
 dst_input include/net/dst.h:460 [inline]
 ip_sublist_rcv_finish net/ipv4/ip_input.c:551 [inline]
 ip_list_rcv_finish net/ipv4/ip_input.c:601 [inline]
 ip_sublist_rcv+0x11fd/0x1520 net/ipv4/ip_input.c:609
 ip_list_rcv+0x95f/0x9a0 net/ipv4/ip_input.c:644
 __netif_receive_skb_list_ptype net/core/dev.c:5505 [inline]
 __netif_receive_skb_list_core+0xe34/0x1240 net/core/dev.c:5553
 __netif_receive_skb_list+0x7fc/0x960 net/core/dev.c:5605
 netif_receive_skb_list_internal+0x868/0xde0 net/core/dev.c:5696
 gro_normal_list net/core/dev.c:5850 [inline]
 napi_complete_done+0x579/0xdd0 net/core/dev.c:6587
 virtqueue_napi_complete drivers/net/virtio_net.c:339 [inline]
 virtnet_poll+0x17b6/0x2350 drivers/net/virtio_net.c:1557
 __napi_poll+0x14e/0xbc0 net/core/dev.c:7020
 napi_poll net/core/dev.c:7087 [inline]
 net_rx_action+0x824/0x1880 net/core/dev.c:7174
 __do_softirq+0x1fe/0x7eb kernel/softirq.c:558
 run_ksoftirqd+0x33/0x50 kernel/softirq.c:920
 smpboot_thread_fn+0x616/0xbf0 kernel/smpboot.c:164
 kthread+0x721/0x850 kernel/kthread.c:327
 ret_from_fork+0x1f/0x30

Uninit was created at:
 __alloc_pages+0xbc7/0x10a0 mm/page_alloc.c:5409
 alloc_pages+0x8a5/0xb80
 alloc_slab_page mm/slub.c:1810 [inline]
 allocate_slab+0x287/0x1c20 mm/slub.c:1947
 new_slab mm/slub.c:2010 [inline]
 ___slab_alloc+0xbdf/0x1e90 mm/slub.c:3039
 __slab_alloc mm/slub.c:3126 [inline]
 slab_alloc_node mm/slub.c:3217 [inline]
 slab_alloc mm/slub.c:3259 [inline]
 kmem_cache_alloc+0xbb3/0x11c0 mm/slub.c:3264
 sk_prot_alloc+0xeb/0x570 net/core/sock.c:1914
 sk_clone_lock+0xd6/0x1940 net/core/sock.c:2118
 inet_csk_clone_lock+0x8d/0x6a0 net/ipv4/inet_connection_sock.c:956
 tcp_create_openreq_child+0xb1/0x1ef0 net/ipv4/tcp_minisocks.c:453
 tcp_v4_syn_recv_sock+0x268/0x2710 net/ipv4/tcp_ipv4.c:1563
 tcp_check_req+0x207c/0x2a30 net/ipv4/tcp_minisocks.c:765
 tcp_v4_rcv+0x36f5/0x4ed0 net/ipv4/tcp_ipv4.c:2047
 ip_protocol_deliver_rcu+0x760/0x10b0 net/ipv4/ip_input.c:204
 ip_local_deliver_finish net/ipv4/ip_input.c:231 [inline]
 NF_HOOK include/linux/netfilter.h:307 [inline]
 ip_local_deliver+0x584/0x8c0 net/ipv4/ip_input.c:252
 dst_input include/net/dst.h:460 [inline]
 ip_sublist_rcv_finish net/ipv4/ip_input.c:551 [inline]
 ip_list_rcv_finish net/ipv4/ip_input.c:601 [inline]
 ip_sublist_rcv+0x11fd/0x1520 net/ipv4/ip_input.c:609
 ip_list_rcv+0x95f/0x9a0 net/ipv4/ip_input.c:644
 __netif_receive_skb_list_ptype net/core/dev.c:5505 [inline]
 __netif_receive_skb_list_core+0xe34/0x1240 net/core/dev.c:5553
 __netif_receive_skb_list+0x7fc/0x960 net/core/dev.c:5605
 netif_receive_skb_list_internal+0x868/0xde0 net/core/dev.c:5696
 gro_normal_list net/core/dev.c:5850 [inline]
 napi_complete_done+0x579/0xdd0 net/core/dev.c:6587
 virtqueue_napi_complete drivers/net/virtio_net.c:339 [inline]
 virtnet_poll+0x17b6/0x2350 drivers/net/virtio_net.c:1557
 __napi_poll+0x14e/0xbc0 net/core/dev.c:7020
 napi_poll net/core/dev.c:7087 [inline]
 net_rx_action+0x824/0x1880 net/core/dev.c:7174
 __do_softirq+0x1fe/0x7eb kernel/softirq.c:558

Fixes: 342159ee394d ("net: avoid dirtying sk->sk_rx_queue_mapping")
Fixes: a37a0ee4d25c ("net: avoid uninit-value from tcp_conn_request")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Tested-by: Alexander Potapenko <glider@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

2 files changed, 15 insertions, 2 deletions

diff --git a/include/net/busy_poll.h b/include/net/busy_poll.h
index 7994455ec714..c4898fcbf923 100644
--- a/include/net/busy_poll.h
+++ b/include/net/busy_poll.h
@@ -136,6 +136,19 @@ static inline void sk_mark_napi_id(struct sock *sk, const struct sk_buff *skb)
 	sk_rx_queue_update(sk, skb);
 }
 
+/* Variant of sk_mark_napi_id() for passive flow setup,
+ * as sk->sk_napi_id and sk->sk_rx_queue_mapping content
+ * needs to be set.
+ */
+static inline void sk_mark_napi_id_set(struct sock *sk,
+				       const struct sk_buff *skb)
+{
+#ifdef CONFIG_NET_RX_BUSY_POLL
+	WRITE_ONCE(sk->sk_napi_id, skb->napi_id);
+#endif
+	sk_rx_queue_set(sk, skb);
+}
+
 static inline void __sk_mark_napi_id_once(struct sock *sk, unsigned int napi_id)
 {
 #ifdef CONFIG_NET_RX_BUSY_POLL
diff --git a/net/ipv4/tcp_minisocks.c b/net/ipv4/tcp_minisocks.c
index cf913a66df17..7c2d3ac2363a 100644
--- a/net/ipv4/tcp_minisocks.c
+++ b/net/ipv4/tcp_minisocks.c
@@ -829,8 +829,8 @@ int tcp_child_process(struct sock *parent, struct sock *child,
 	int ret = 0;
 	int state = child->sk_state;
 
-	/* record NAPI ID of child */
-	sk_mark_napi_id(child, skb);
+	/* record sk_napi_id and sk_rx_queue_mapping of child. */
+	sk_mark_napi_id_set(child, skb);
 
 	tcp_segs_in(tcp_sk(child), skb);
 	if (!sock_owned_by_user(child)) {