[Devel] [PATCH vz10 3/5] fixup! vhost/vsock: only refuse connection when guest has never been ready

[Eva Kurchatova]

From: Andrey Drobyshev <andrey.drobyshev at virtuozzo.com>

Commit 4ff28534c799 ("ms/vhost/vsock: Refuse the connection immediately
when guest isn't ready") added a check which immediately returns
EHOSTUNREACH when the guest isn't ready yet.  Namely, we check that guest
hasn't enabled the RX vq yet, i.e. virtio-vsock has never beed enabled.

However, the check also affects the transient state when backend is
temporarily set to NULL during VHOST_VSOCK_SET_RUNNING(0).  Notably,
this is the case with qemu-update operation, during which we perform
VHOST_RESET_OWNER.  In this case sendmsg()/connect() on otherwise healthy
connection gets EHOSTUNREACH.

Gate the fast-fail on a sticky started_once bit set in
vhost_vsock_start() and never cleared.  Once the guest has brought
up virtio-vsock at least once, a NULL backend means a transient stop
window and the packet must be queued for vhost_vsock_start() to drain
on re-attach.

Fixes: 4ff28534c799 ("ms/vhost/vsock: Refuse the connection immediately when guest isn't ready")
https://virtuozzo.atlassian.net/browse/VSTOR-131956
Signed-off-by: Andrey Drobyshev <andrey.drobyshev at virtuozzo.com>
---
 drivers/vhost/vsock.c | 31 ++++++++++++++-----------------
 1 file changed, 14 insertions(+), 17 deletions(-)

diff --git a/drivers/vhost/vsock.c b/drivers/vhost/vsock.c
index d4c3f94308db..f652f47956d7 100644
--- a/drivers/vhost/vsock.c
+++ b/drivers/vhost/vsock.c
@@ -59,6 +59,7 @@ struct vhost_vsock {
 
 	u32 guest_cid;
 	bool seqpacket_allow;
+	bool started_once;	/* latched in vhost_vsock_start(); never cleared */
 	bool cpr_paused;	/* between stop and next start; queues sends */
 };
 
@@ -289,24 +290,17 @@ vhost_transport_send_pkt(struct sk_buff *skb, struct net *net)
 		return -ENODEV;
 	}
 
-	/* Fast-fail if the guest hasn't enabled the RX vq yet. Queuing the packet
-	 * and making the caller wait is pointless: even if the guest manages to init
-	 * within the timeout, it'll immediately reply with RST, because there's no
-	 * listener on the port yet.
-	 *
-	 * vhost_vq_get_backend() without vq->mutex is acceptable here: locking
-	 * the mutex would be too expensive in this hot path, and we already have
-	 * all the outcomes covered: if the backend becomes NULL right after the check,
-	 * vhost_transport_do_send_pkt() will check it under the mutex anyway.
+	/*
+	 * Fast-fail only when the guest has never enabled virtio-vsock.
+	 * Once it has, a NULL backend means a transient SET_RUNNING(0)
+	 * window (e.g. VHOST_RESET_OWNER); the packet must be
+	 * queued for vhost_vsock_start() to drain on re-attach.
 	 */
-	/* cpr_paused: queue across CPR; else NULL backend means not ready. */
-	if (unlikely(!data_race(vhost_vq_get_backend(&vsock->vqs[VSOCK_VQ_RX])))) {
-		smp_rmb();	/* pairs with smp_wmb() in start/drop_backends */
-		if (!READ_ONCE(vsock->cpr_paused)) {
-			rcu_read_unlock();
-			kfree_skb(skb);
-			return -EHOSTUNREACH;
-		}
+	if (unlikely(!READ_ONCE(vsock->started_once)) &&
+	    !data_race(vhost_vq_get_backend(&vsock->vqs[VSOCK_VQ_RX]))) {
+		rcu_read_unlock();
+		kfree_skb(skb);
+		return -EHOSTUNREACH;
 	}
 
 	if (virtio_vsock_skb_reply(skb))
@@ -637,6 +631,9 @@ static int vhost_vsock_start(struct vhost_vsock *vsock)
 	 */
 	vhost_vq_work_queue(&vsock->vqs[VSOCK_VQ_RX], &vsock->send_pkt_work);
 
+	/* See vhost_transport_send_pkt(); never cleared. */
+	WRITE_ONCE(vsock->started_once, true);
+
 	mutex_unlock(&vsock->dev.mutex);
 	return 0;
 
-- 
2.54.0