[PATCH net v4] netpoll: Fix deadlock in memory allocation under spinlock

[Breno Leitao <leitao@debian.org>]

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(&skb_pool->lock)     <- deadlock attempt

This bug was exposed by commit 248f6571fd4c51 ("netpoll: Optimize skb
refilling on critical path") which removed refill_skbs() from the
critical path (where nested printk was being deferred), letting nested
printk being called from inside refill_skbs()

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

Another possible solution to fix this problem is protecting the
refill_skbs() from nested printks, basically calling
printk_deferred_{enter,exit}() in refill_skbs(), then, any nested
pr_warn() would be deferred.

I prefer this approach, given I _think_ it might be a good idea to move
the alloc_skb() from GFP_ATOMIC to GFP_KERNEL in the future, so, having
the alloc_skb() outside of the lock will be necessary step.

There is a possible TOCTOU issue when checking for the pool length, and
queueing the new allocated skb, but, this is not an issue, given that
an extra SKB in the pool is harmless and it will be eventually used.

Signed-off-by: Breno Leitao <leitao@debian.org>
Fixes: 248f6571fd4c51 ("netpoll: Optimize skb refilling on critical path")
Reviewed-by: Simon Horman <horms@kernel.org>
---
Changes in v4:
- Check for the pool length using READ_ONCE() instead of the proper lock
  (Jakub)
- Remove the additional TOUTOC logic, given a possible extra SKB to the
  pool is harmless. (Jakub)
- Link to v3: https://lore.kernel.org/r/20251014-fix_netpoll_aa-v3-1-bff72762294e@debian.org

Changes in v3:
- Removed the "return" before the exit labels. (Simon)
- Link to v2: https://lore.kernel.org/r/20251014-fix_netpoll_aa-v2-1-dafa6a378649@debian.org

Changes in v2:
- Added a return after the successful path (Rik van Riel)
- Changed the Fixes tag to point to the commit that exposed the problem.
- Link to v1: https://lore.kernel.org/r/20251013-fix_netpoll_aa-v1-1-94a1091f92f0@debian.org
---
 net/core/netpoll.c | 7 ++-----
 1 file changed, 2 insertions(+), 5 deletions(-)

diff --git a/net/core/netpoll.c b/net/core/netpoll.c
index 60a05d3b7c249..c85f740065fc6 100644
--- a/net/core/netpoll.c
+++ b/net/core/netpoll.c
@@ -228,19 +228,16 @@ static void refill_skbs(struct netpoll *np)
 {
 	struct sk_buff_head *skb_pool;
 	struct sk_buff *skb;
-	unsigned long flags;
 
 	skb_pool = &np->skb_pool;
 
-	spin_lock_irqsave(&skb_pool->lock, flags);
-	while (skb_pool->qlen < MAX_SKBS) {
+	while (READ_ONCE(skb_pool->qlen) < MAX_SKBS) {
 		skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
 		if (!skb)
 			break;
 
-		__skb_queue_tail(skb_pool, skb);
+		skb_queue_tail(skb_pool, skb);
 	}
-	spin_unlock_irqrestore(&skb_pool->lock, flags);
 }
 
 static void zap_completion_queue(void)

---
base-commit: 51e5ad549c43b557c7da1e4d1a1dcf061b4a5f6c
change-id: 20251013-fix_netpoll_aa-c991ac5f2138

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