net: skbuff: Skip early return in skb_unref when debugging

This patch modifies the skb_unref function to skip the early return
optimization when CONFIG_DEBUG_NET is enabled. The change ensures that
the reference count decrement always occurs in debug builds, allowing
for more thorough checking of SKB reference counting.

Previously, when the SKB's reference count was 1 and CONFIG_DEBUG_NET
was not set, the function would return early after a memory barrier
(smp_rmb()) without decrementing the reference count. This optimization
assumes it's safe to proceed with freeing the SKB without the overhead
of an atomic decrement from 1 to 0.

With this change:
- In non-debug builds (CONFIG_DEBUG_NET not set), behavior remains
  unchanged, preserving the performance optimization.
- In debug builds (CONFIG_DEBUG_NET set), the reference count is always
  decremented, even when it's 1, allowing for consistent behavior and
  potentially catching subtle SKB management bugs.

This modification enhances debugging capabilities for networking code
without impacting performance in production kernels. It helps kernel
developers identify and diagnose issues related to SKB management and
reference counting in the network stack.

Cc: Chris Mason <clm@fb.com>
Suggested-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Breno Leitao <leitao@debian.org>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20240729104741.370327-1-leitao@debian.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>

include/linux/skbuff.h

@@ -1225,7 +1225,7 @@ static inline bool skb_unref(struct sk_buff *skb)
 {
 	if (unlikely(!skb))
 		return false;
-	if (likely(refcount_read(&skb->users) == 1))
+	if (!IS_ENABLED(CONFIG_DEBUG_NET) && likely(refcount_read(&skb->users) == 1))
 		smp_rmb();
 	else if (likely(!refcount_dec_and_test(&skb->users)))
 		return false;