Re: [TIP] BUG kmalloc-4096: Poison overwritten (ath5k_rx_skb_alloc)

[Bob Copeland <me@bobcopeland.com>]

On Wed, Feb 25, 2009 at 02:01:39PM +0000, Sitsofe Wheeler wrote:
> NAS is the name of a nearby access point. This is really tough to
> reproduce but if there's anything (e.g. ftrace) I can always turn on
> that will help you track this thing down let me know. Unfortunately the
> machine is an EeePC 900 so any logging would have to be cyclic and held
> in RAM...

Hmm, I suppose there could be some error paths under memory pressure that
aren't quite right.  Here's one, but I don't think it can cause any problems,
at least without error spew.

Looking over the code, perhaps there's a race with ath5k_rx_start,
namely sc->rxlink is changed out from under the rxbuflock.  Still can't
immediately see a use-after-free there.

============

ath5k: don't overwrite bf->skbaddr unless pcu mapping fails.

diff --git a/drivers/net/wireless/ath5k/base.c b/drivers/net/wireless/ath5k/base.c
index 1d77ee9..6d91335 100644
--- a/drivers/net/wireless/ath5k/base.c
+++ b/drivers/net/wireless/ath5k/base.c
@@ -1140,12 +1140,14 @@ ath5k_rxbuf_setup(struct ath5k_softc *sc, struct ath5k_buf *bf)
 	struct ath5k_hw *ah = sc->ah;
 	struct sk_buff *skb = bf->skb;
 	struct ath5k_desc *ds;
+	dma_addr_t dma_addr;
 
 	if (!skb) {
-		skb = ath5k_rx_skb_alloc(sc, &bf->skbaddr);
+		skb = ath5k_rx_skb_alloc(sc, &dma_addr);
 		if (!skb)
 			return -ENOMEM;
 		bf->skb = skb;
+		bf->skbaddr = dma_addr;
 	}
 
 	/*

-- 
Bob Copeland %% www.bobcopeland.com