[1/2] rtl8xxxu: Fix memory leak in handling rxdesc16 packets

Commit Message

Jes Sorensen Sept. 29, 2016, 7:40 p.m. UTC

From: Jes Sorensen <[e-mail]>

A device running without RX package aggregation could return more data
in the USB packet than the actual network packet. In this case the
could would clone the skb but then determine that that there was no
packet to handle and exit without freeing the cloned skb first.

This has so far only been observed with 8188eu devices, but could
affect others.

Signed-off-by: Jes Sorensen <[e-mail]>
---
 drivers/net/wireless/realtek/rtl8xxxu/rtl8xxxu_core.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

Comments

Kalle Valo Oct. 3, 2016, 7:24 a.m. UTC | #1

[e-mail] writes:

> From: Jes Sorensen <[e-mail]>
>
> A device running without RX package aggregation could return more data
> in the USB packet than the actual network packet. In this case the
> could would clone the skb but then determine that that there was no
> packet to handle and exit without freeing the cloned skb first.
>
> This has so far only been observed with 8188eu devices, but could
> affect others.
>
> Signed-off-by: Jes Sorensen <[e-mail]>

Before I commit I'll add:

Cc: [e-mail] # 4.8+

Kalle Valo Oct. 3, 2016, 7:25 a.m. UTC | #2

[e-mail] writes:

> From: Jes Sorensen <[e-mail]>
>
> A device running without RX package aggregation could return more data
> in the USB packet than the actual network packet. In this case the
> could would clone the skb but then determine that that there was no
> packet to handle and exit without freeing the cloned skb first.

s/case the/case we/? I can edit that before applying the patch.

Jes Sorensen Oct. 3, 2016, 12:34 p.m. UTC | #3

Kalle Valo <[e-mail]> writes:
> [e-mail] writes:
>
>> From: Jes Sorensen <[e-mail]>
>>
>> A device running without RX package aggregation could return more data
>> in the USB packet than the actual network packet. In this case the
>> could would clone the skb but then determine that that there was no
>> packet to handle and exit without freeing the cloned skb first.
>
> s/case the/case we/? I can edit that before applying the patch.

Sounds good - thanks!

Jes

Kalle Valo Oct. 7, 2016, 11:22 a.m. UTC | #4

Jes Sorensen <[e-mail]> wrote:
> From: Jes Sorensen <[e-mail]>
> 
> A device running without RX package aggregation could return more data
> in the USB packet than the actual network packet. In this case we
> could would clone the skb but then determine that that there was no
> packet to handle and exit without freeing the cloned skb first.
> 
> This has so far only been observed with 8188eu devices, but could
> affect others.
> 
> Signed-off-by: Jes Sorensen <[e-mail]>
> Cc: [e-mail] # 4.8+

2 patches applied to wireless-drivers.git, thanks.

1e54134ccad0 rtl8xxxu: Fix memory leak in handling rxdesc16 packets
8a55698f2f29 rtl8xxxu: Fix big-endian problem reporting mactime

Patch

diff --git a/drivers/net/wireless/realtek/rtl8xxxu/rtl8xxxu_core.c b/drivers/net/wireless/realtek/rtl8xxxu/rtl8xxxu_core.c
index b2d7f6e..a96ff17 100644
--- a/drivers/net/wireless/realtek/rtl8xxxu/rtl8xxxu_core.c
+++ b/drivers/net/wireless/realtek/rtl8xxxu/rtl8xxxu_core.c
@@ -5197,7 +5197,12 @@  int rtl8xxxu_parse_rxdesc16(struct rtl8xxxu_priv *priv, struct sk_buff *skb)
 		pkt_offset = roundup(pkt_len + drvinfo_sz + desc_shift +
 				     sizeof(struct rtl8xxxu_rxdesc16), 128);
 
-		if (pkt_cnt > 1)
+		/*
+		 * Only clone the skb if there's enough data at the end to
+		 * at least cover the rx descriptor
+		 */
+		if (pkt_cnt > 1 &&
+		    urb_len > (pkt_offset + sizeof(struct rtl8xxxu_rxdesc16)))
 			next_skb = skb_clone(skb, GFP_ATOMIC);
 
 		rx_status = IEEE80211_SKB_RXCB(skb);