On Mon, 2015-03-30 at 22:01 +0200, Johannes Berg wrote:

On Mon, 2015-03-30 at 09:37 -0700, Joe Perches wrote:

quoted

@@ -89,11 +90,11 @@ struct ieee80211_fragment_entry {
 	unsigned int last_frag;
 	unsigned int extra_len;
 	struct sk_buff_head skb_list;
-	int ccmp; /* Whether fragments were encrypted with CCMP */
-	u8 last_pn[6]; /* PN of the last fragment if CCMP was used */
+	/* for CCMP fragments */
+	bool ccmp;				/* encrypted with CCMP */
+	u8 last_pn[IEEE80211_CCMP_PN_LEN];	/* PN of the last fragment */

I took your patch as an opportunity to check into this, and it turns out
all of this logic is also going to be needed for GCMP. As a result, I'm
not going to take this patch but instead we'll fix it up for GCMP (where
using CCMP_PN_LEN would not be appropriate anyway - perhaps we need a
union or just keep '8' which is the right size for both anyway, or we'll
go to a u64 value or something)

No worries.  Whatever's right.