Additional run-tme check [Run-time kfree check for correct cache]

From: Evgeniy Polyakov <johnpol@2ka.mipt.ru>
To: Eric Dumazet <dada1@cosmosbay.com>
Cc: David Miller <davem@davemloft.net>,
	paul.moore@hp.com, kaber@trash.net, netdev@vger.kernel.org,
	acme@ghostprotocols.net,
	Eric Sesterhenn / Snakebyte <snakebyte@gmx.de>
Subject: Additional run-tme check [Run-time kfree check for correct cache]
Date: Thu, 1 Mar 2007 19:17:44 +0300	[thread overview]
Message-ID: <20070301161744.GB16344@2ka.mipt.ru> (raw)
In-Reply-To: <20070228141652.GA3033@2ka.mipt.ru>

Due to slab poisoning after origianl freeing, skb->users will not allow
subsequent freeing to succeed and fire slab debugging, which ends up
with obscure bugs.

Patch based on Eric Sesterhenn's <snakebyte@gmx.de> one, I changed
check to skb>truesize instead of skb->users.counter, since the latter in
theory can be different than simple integer containing structure.

--- linux/net/core/skbuff.c.orig	2007-02-28 11:34:13.865540564 +0100
+++ linux/net/core/skbuff.c	2007-03-01 13:42:05.633136591 +0100
@@ -407,6 +407,11 @@ void kfree_skb(struct sk_buff *skb)
 {
 	if (unlikely(!skb))
 		return;
+#ifdef CONFIG_DEBUG_SLAB
+	/* check for double kfree_skb() */
+	WARN_ON(unlikely((skb->truesize & 0xFFFFFFFF) == 0x6b6b6b6b));
+#endif
+
 	if (likely(atomic_read(&skb->users) == 1))
 		smp_rmb();
 	else if (likely(!atomic_dec_and_test(&skb->users)))

-- 
	Evgeniy Polyakov
