Re: [PATCH] ibmvnic: fix OOB array access in ibmvnic_xmit on queue count reduction

[Simon Horman]

On Fri, Mar 20, 2026 at 10:54:39PM -0500, Tyllis Xu wrote:
> When the number of TX queues is reduced (e.g., via ethtool -L), the
> Qdisc layer retains previously enqueued skbs with queue mappings from
> before the reduction. After the reset completes and tx_queues_active is
> set to true, netif_tx_start_all_queues() drains these stale skbs through
> ibmvnic_xmit(). The queue index from skb_get_queue_mapping() may exceed
> the newly allocated array bounds, causing out-of-bounds reads on
> tx_scrq[] and tx_pool[]/tso_pool[], and out-of-bounds writes on
> tx_stats_buffers[] in the function's exit path.
> 
> The existing tx_queues_active guard does not help here: it is set to
> true by __ibmvnic_open() before netif_tx_start_all_queues() restarts
> queue draining, so stale skbs pass the check with an invalid queue index.
> 
> Add a bounds check against num_active_tx_scrqs immediately after the
> tx_queues_active guard. Use a dedicated out_unlock label to skip the
> per-queue stats updates (which also index tx_stats_buffers[queue_num])
> when the queue index is invalid.
> 
> Fixes: 4219196d1f66 ("ibmvnic: fix race between xmit and reset")
> Reported-by: Yuhao Jiang <danisjiang@xxxxxxxxx>
> Cc: stable@xxxxxxxxxxxxxxx
> Signed-off-by: Tyllis Xu <LivelyCarpet87@xxxxxxxxx>
> ---
>  drivers/net/ethernet/ibm/ibmvnic.c | 8 ++++++++
>  1 file changed, 8 insertions(+)
> 
> diff --git a/drivers/net/ethernet/ibm/ibmvnic.c b/drivers/net/ethernet/ibm/ibmvnic.c
> index 5a510eed335e..c939391474cb 100644
> --- a/drivers/net/ethernet/ibm/ibmvnic.c
> +++ b/drivers/net/ethernet/ibm/ibmvnic.c
> @@ -2453,6 +2453,11 @@ static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
>  		goto out;
>  	}
>  
> +	if (unlikely(queue_num >= adapter->num_active_tx_scrqs)) {
> +		dev_kfree_skb_any(skb);
> +		goto out_unlock;
> +	}
> +

This doesn't seem quite right. Shouldn't it be as per other
blocks in this function that drop packets. In which case
it could re-use the existing handling in the conditional immediately above
this hunk.

Also, I don't think unlikely() seems in keeping with the existing
implementation of this function.

I'm suggesting something like (completely untested):

diff --git a/drivers/net/ethernet/ibm/ibmvnic.c b/drivers/net/ethernet/ibm/ibmvnic.c
index 5a510eed335e..67e1e62631e3 100644
--- a/drivers/net/ethernet/ibm/ibmvnic.c
+++ b/drivers/net/ethernet/ibm/ibmvnic.c
@@ -2457,7 +2457,8 @@ static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
 	txq = netdev_get_tx_queue(netdev, queue_num);
 	ind_bufp = &tx_scrq->ind_buf;
 
-	if (ibmvnic_xmit_workarounds(skb, netdev)) {
+	if (ibmvnic_xmit_workarounds(skb, netdev) ||
+	    queue_num >= adapter->num_active_tx_scrqs) {
 		tx_dropped++;
 		tx_send_failed++;
 		ret = NETDEV_TX_OK;

Where the next line is:

		goto out;

...

> @@ -2672,6 +2677,9 @@ static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
>  	adapter->tx_stats_buffers[queue_num].bytes += tx_bytes;
>  	adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped;
>  
> +	return ret;
> +out_unlock:
> +	rcu_read_unlock();
>  	return ret;
>  }

My previous comment not, withstanding:

The RCU read side critical section is already enormous.
So perhaps making it slightly better doesn't make a difference.

If so, can we go for this slightly flow here (completely untested).

diff --git a/drivers/net/ethernet/ibm/ibmvnic.c b/drivers/net/ethernet/ibm/ibmvnic.c
index 5a510eed335e..1e1cd8c11cf9 100644
--- a/drivers/net/ethernet/ibm/ibmvnic.c
+++ b/drivers/net/ethernet/ibm/ibmvnic.c
@@ -2664,14 +2664,14 @@ static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
 		netif_carrier_off(netdev);
 	}
 out:
-	rcu_read_unlock();
 	adapter->tx_send_failed += tx_send_failed;
 	adapter->tx_map_failed += tx_map_failed;
 	adapter->tx_stats_buffers[queue_num].batched_packets += tx_bpackets;
 	adapter->tx_stats_buffers[queue_num].direct_packets += tx_dpackets;
 	adapter->tx_stats_buffers[queue_num].bytes += tx_bytes;
 	adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped;
-
+out_unlock:
+	rcu_read_unlock();
 	return ret;
 }
 

-- 
pw-bot: changes-requested