xsk: Fix race in SKB mode transmit with shared cq

Fix a race when multiple sockets are simultaneously calling sendto()
when the completion ring is shared in the SKB case. This is the case
when you share the same netdev and queue id through the
XDP_SHARED_UMEM bind flag. The problem is that multiple processes can
be in xsk_generic_xmit() and call the backpressure mechanism in
xskq_prod_reserve(xs->pool->cq). As this is a shared resource in this
specific scenario, a race might occur since the rings are
single-producer single-consumer.

Fix this by moving the tx_completion_lock from the socket to the pool
as the pool is shared between the sockets that share the completion
ring. (The pool is not shared when this is not the case.) And then
protect the accesses to xskq_prod_reserve() with this lock. The
tx_completion_lock is renamed cq_lock to better reflect that it
protects accesses to the potentially shared completion ring.

Fixes: 35fcde7 ("xsk: support for Tx")
Reported-by: Xuan Zhuo <[email protected]>
Signed-off-by: Magnus Karlsson <[email protected]>
Signed-off-by: Daniel Borkmann <[email protected]>
Acked-by: Björn Töpel <[email protected]>
Link: https://lore.kernel.org/bpf/[email protected]

include/net/xdp_sock.h

@@ -58,10 +58,6 @@ struct xdp_sock {
 
 	struct xsk_queue *tx ____cacheline_aligned_in_smp;
 	struct list_head tx_list;
-	/* Mutual exclusion of NAPI TX thread and sendmsg error paths
-	 * in the SKB destructor callback.
-	 */
-	spinlock_t tx_completion_lock;
 	/* Protects generic receive. */
 	spinlock_t rx_lock;
 

include/net/xsk_buff_pool.h

@@ -73,6 +73,11 @@ struct xsk_buff_pool {
 	bool dma_need_sync;
 	bool unaligned;
 	void *addrs;
+	/* Mutual exclusion of the completion ring in the SKB mode. Two cases to protect:
+	 * NAPI TX thread and sendmsg error paths in the SKB destructor callback and when
+	 * sockets share a single cq when the same netdev and queue id is shared.
+	 */
+	spinlock_t cq_lock;
 	struct xdp_buff_xsk *free_heads[];
 };
 

net/xdp/xsk.c

@@ -423,9 +423,9 @@ static void xsk_destruct_skb(struct sk_buff *skb)
 	struct xdp_sock *xs = xdp_sk(skb->sk);
 	unsigned long flags;
 
-	spin_lock_irqsave(&xs->tx_completion_lock, flags);
+	spin_lock_irqsave(&xs->pool->cq_lock, flags);
 	xskq_prod_submit_addr(xs->pool->cq, addr);
-	spin_unlock_irqrestore(&xs->tx_completion_lock, flags);
+	spin_unlock_irqrestore(&xs->pool->cq_lock, flags);
 
 	sock_wfree(skb);
 }
@@ -437,6 +437,7 @@ static int xsk_generic_xmit(struct sock *sk)
 	bool sent_frame = false;
 	struct xdp_desc desc;
 	struct sk_buff *skb;
+	unsigned long flags;
 	int err = 0;
 
 	mutex_lock(&xs->mutex);
@@ -468,10 +469,13 @@ static int xsk_generic_xmit(struct sock *sk)
 		 * if there is space in it. This avoids having to implement
 		 * any buffering in the Tx path.
 		 */
+		spin_lock_irqsave(&xs->pool->cq_lock, flags);
 		if (unlikely(err) || xskq_prod_reserve(xs->pool->cq)) {
+			spin_unlock_irqrestore(&xs->pool->cq_lock, flags);
 			kfree_skb(skb);
 			goto out;
 		}
+		spin_unlock_irqrestore(&xs->pool->cq_lock, flags);
 
 		skb->dev = xs->dev;
 		skb->priority = sk->sk_priority;
@@ -1303,7 +1307,6 @@ static int xsk_create(struct net *net, struct socket *sock, int protocol,
 	xs->state = XSK_READY;
 	mutex_init(&xs->mutex);
 	spin_lock_init(&xs->rx_lock);
-	spin_lock_init(&xs->tx_completion_lock);
 
 	INIT_LIST_HEAD(&xs->map_list);
 	spin_lock_init(&xs->map_list_lock);

net/xdp/xsk_buff_pool.c

@@ -71,6 +71,7 @@ struct xsk_buff_pool *xp_create_and_assign_umem(struct xdp_sock *xs,
 	INIT_LIST_HEAD(&pool->free_list);
 	INIT_LIST_HEAD(&pool->xsk_tx_list);
 	spin_lock_init(&pool->xsk_tx_list_lock);
+	spin_lock_init(&pool->cq_lock);
 	refcount_set(&pool->users, 1);
 
 	pool->fq = xs->fq_tmp;