On Monday 01 October 2012 22:53:28 Simon Wunderlich wrote:
It is possible that the mac_len is not properly
exported because of
strange device configuration (this behaviour has been observed when
using batman-adv on top of a vlan interface). Therefore it is needed to
explicitly recompute it at the very beginning of the rx path.
This is done by appending the recompute function to the skb_share_mac()
function, hence the "dirty hack" in the subject. We expect this problem to
be fixed in linux 3.8 and above.
Reported-by: Antonio Quartulli <ordex(a)autistici.org>
Signed-off-by: Simon Wunderlich <siwu(a)hrz.tu-chemnitz.de>
this is a rewrite of Antonios patch
"batman-adv: recompute mac_len at the beginning of the rx path". It is
intended to fix the issue for out-of-kernel releases, without hurting
the in-kernel code too much. Obviously, this patch won't go upstream
into the kernel.
compat.h | 15 +++++++++++++++
1 file changed, 15 insertions(+)
diff --git a/compat.h b/compat.h
index 14969e0..dca9685 100644
@@ -159,4 +159,19 @@ static inline void eth_hw_addr_random(struct net_device
#endif /* < KERNEL_VERSION(3, 5, 0) */
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 8, 0)
+/* hack for not correctly set mac_len. This may happen for some special
+ * configurations like batman-adv on VLANs.
+ * This is pretty dirty, but we only use skb_share_check() in main.c right
+ * before mac_len is checked, and the recomputation shouldn't hurt too
much. + */
+#define skb_share_check(skb, b) \
+ skb_share_check(skb, b); \
+ if (skb) \
+#endif /* < KERNEL_VERSION(3, 8, 0) */
#endif /* _NET_BATMAN_ADV_COMPAT_H_ */
Can we try a more sane solution like
#define skb_share_check(skb, b) \
struct sk_buff *_t_skb; \
_t_skb = skb_share_check(skb, b); \
if (_t_skb) \
Please test whether this thing really works and compiles. I just wrote it
doing something else and never compiled it.