Fix a AA deadlock in refill_skbs() where memory allocation while holding
skb_pool->lock can trigger a recursive lock acquisition attempt.

The deadlock scenario occurs when the system is under severe memory
pressure:

1. refill_skbs() acquires skb_pool->lock (spinlock)
2. alloc_skb() is called while holding the lock
3. Memory allocator fails and calls slab_out_of_memory()
4. This triggers printk() for the OOM warning
5. The console output path calls netpoll_send_udp()
6. netpoll_send_udp() attempts to acquire the same skb_pool->lock
7. Deadlock: the lock is already held by the same CPU

Call stack:
  refill_skbs()
    spin_lock_irqsave(&skb_pool->lock)    <- lock acquired
    __alloc_skb()
      kmem_cache_alloc_node_noprof()
        slab_out_of_memory()
          printk()
            console_flush_all()
              netpoll_send_udp()
                skb_dequeue()
                  spin_lock_irqsave()     <- deadlock attempt

Refactor refill_skbs() to never allocate memory while holding
the spinlock.

Signed-off-by: Breno Leitao <leitao@debian.org>
Fixes: 1da177e4c3f41 ("Linux-2.6.12-rc2")
---
 net/core/netpoll.c | 18 +++++++++++++++---
 1 file changed, 15 insertions(+), 3 deletions(-)

diff --git a/net/core/netpoll.c b/net/core/netpoll.c
index 60a05d3b7c249..788cec4d527f8 100644
--- a/net/core/netpoll.c
+++ b/net/core/netpoll.c
@@ -232,14 +232,26 @@ static void refill_skbs(struct netpoll *np)
 
 	skb_pool = &np->skb_pool;
 
-	spin_lock_irqsave(&skb_pool->lock, flags);
-	while (skb_pool->qlen < MAX_SKBS) {
+	while (1) {
+		spin_lock_irqsave(&skb_pool->lock, flags);
+		if (skb_pool->qlen >= MAX_SKBS)
+			goto unlock;
+		spin_unlock_irqrestore(&skb_pool->lock, flags);
+
 		skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
 		if (!skb)
-			break;
+			return;
 
+		spin_lock_irqsave(&skb_pool->lock, flags);
+		if (skb_pool->qlen >= MAX_SKBS)
+			/* Discard if len got increased (TOCTOU) */
+			goto discard;
 		__skb_queue_tail(skb_pool, skb);
+		spin_unlock_irqrestore(&skb_pool->lock, flags);
 	}
+discard:
+	dev_kfree_skb_any(skb);
+unlock:
 	spin_unlock_irqrestore(&skb_pool->lock, flags);
 }
 

---
base-commit: 0b4b77eff5f8cd9be062783a1c1e198d46d0a753
change-id: 20251013-fix_netpoll_aa-c991ac5f2138

Best regards,
--  
Breno Leitao <leitao@debian.org>

On Mon, Oct 13, 2025 at 02:42:29AM -0700, Breno Leitao wrote:
> Fix a AA deadlock in refill_skbs() where memory allocation while holding
> skb_pool->lock can trigger a recursive lock acquisition attempt.
> 
> The deadlock scenario occurs when the system is under severe memory
> pressure:
> 
> 1. refill_skbs() acquires skb_pool->lock (spinlock)
> 2. alloc_skb() is called while holding the lock
> 3. Memory allocator fails and calls slab_out_of_memory()
> 4. This triggers printk() for the OOM warning
> 5. The console output path calls netpoll_send_udp()
> 6. netpoll_send_udp() attempts to acquire the same skb_pool->lock
> 7. Deadlock: the lock is already held by the same CPU
> 
> Call stack:
>   refill_skbs()
>     spin_lock_irqsave(&skb_pool->lock)    <- lock acquired
>     __alloc_skb()
>       kmem_cache_alloc_node_noprof()
>         slab_out_of_memory()
>           printk()
>             console_flush_all()
>               netpoll_send_udp()
>                 skb_dequeue()
>                   spin_lock_irqsave()     <- deadlock attempt
> 
> Refactor refill_skbs() to never allocate memory while holding
> the spinlock.
> 
> Signed-off-by: Breno Leitao <leitao@debian.org>
> Fixes: 1da177e4c3f41 ("Linux-2.6.12-rc2")
> ---
>  net/core/netpoll.c | 18 +++++++++++++++---
>  1 file changed, 15 insertions(+), 3 deletions(-)
> 
> diff --git a/net/core/netpoll.c b/net/core/netpoll.c
> index 60a05d3b7c249..788cec4d527f8 100644
> --- a/net/core/netpoll.c
> +++ b/net/core/netpoll.c
> @@ -232,14 +232,26 @@ static void refill_skbs(struct netpoll *np)
>  
>  	skb_pool = &np->skb_pool;
>  
> -	spin_lock_irqsave(&skb_pool->lock, flags);
> -	while (skb_pool->qlen < MAX_SKBS) {
> +	while (1) {
> +		spin_lock_irqsave(&skb_pool->lock, flags);
> +		if (skb_pool->qlen >= MAX_SKBS)
> +			goto unlock;
> +		spin_unlock_irqrestore(&skb_pool->lock, flags);
> +
>  		skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
>  		if (!skb)
> -			break;
> +			return;
>  
> +		spin_lock_irqsave(&skb_pool->lock, flags);
> +		if (skb_pool->qlen >= MAX_SKBS)
> +			/* Discard if len got increased (TOCTOU) */
> +			goto discard;
>  		__skb_queue_tail(skb_pool, skb);
> +		spin_unlock_irqrestore(&skb_pool->lock, flags);
>  	}

We probably want to return in here, as pointed out by Rik van Riel
offline.

If there are no more concerns, I will wait the 24-hours period and send
a v2.