[PATCH] net: usb: smsc95xx: Limit packet length to skb->len

[PATCH] net: usb: smsc95xx: Limit packet length to skb->len

[Szymon Heidrich]

Packet length retrieved from skb data may be larger than
the actual socket buffer length (up to 1526 bytes). In such
case the cloned skb passed up the network stack will leak
kernel memory contents.

Fixes: 2f7ca802bdae ("net: Add SMSC LAN9500 USB2.0 10/100 ethernet adapter driver")
Signed-off-by: Szymon Heidrich <szymon.heidrich@gmail.com>
---
 drivers/net/usb/smsc95xx.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/net/usb/smsc95xx.c b/drivers/net/usb/smsc95xx.c
index 32d2c60d3..ba766bdb2 100644
--- a/drivers/net/usb/smsc95xx.c
+++ b/drivers/net/usb/smsc95xx.c
@@ -1851,7 +1851,8 @@ static int smsc95xx_rx_fixup(struct usbnet *dev, struct sk_buff *skb)
 			}
 		} else {
 			/* ETH_FRAME_LEN + 4(CRC) + 2(COE) + 4(Vlan) */
-			if (unlikely(size > (ETH_FRAME_LEN + 12))) {
+			if (unlikely(size > (ETH_FRAME_LEN + 12) ||
+				     size > skb->len)) {
 				netif_dbg(dev, rx_err, dev->net,
 					  "size err header=0x%08x\n", header);
 				return 0;
-- 
2.39.2

Re: [PATCH] net: usb: smsc95xx: Limit packet length to skb->len

[Jakub Kicinski]

On Mon, 13 Mar 2023 23:01:24 +0100 Szymon Heidrich wrote:
> Packet length retrieved from skb data may be larger than

nit: s/skb data/descriptor/ may be better in terms of terminology

> the actual socket buffer length (up to 1526 bytes). In such

nit: the "up to 1526 bytes" is a bit confusing, I'd remove it.

> case the cloned skb passed up the network stack will leak
> kernel memory contents.



> Fixes: 2f7ca802bdae ("net: Add SMSC LAN9500 USB2.0 10/100 ethernet adapter driver")
> Signed-off-by: Szymon Heidrich <szymon.heidrich@gmail.com>
> ---
>  drivers/net/usb/smsc95xx.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/net/usb/smsc95xx.c b/drivers/net/usb/smsc95xx.c
> index 32d2c60d3..ba766bdb2 100644
> --- a/drivers/net/usb/smsc95xx.c
> +++ b/drivers/net/usb/smsc95xx.c
> @@ -1851,7 +1851,8 @@ static int smsc95xx_rx_fixup(struct usbnet *dev, struct sk_buff *skb)
>  			}
>  		} else {
>  			/* ETH_FRAME_LEN + 4(CRC) + 2(COE) + 4(Vlan) */
> -			if (unlikely(size > (ETH_FRAME_LEN + 12))) {
> +			if (unlikely(size > (ETH_FRAME_LEN + 12) ||
> +				     size > skb->len)) {

We need this check on both sides of the big if {} statement.

In case the error bit is set and we drop the packet we still
end up in skb_pull() which if size > skb->len will panic the
kernel.

So let's do this check right after size and align are calculated?
The patch for smsc75xx has sadly already been applied so you'll
need to prepare a fix to the fix :(

>  				netif_dbg(dev, rx_err, dev->net,
>  					  "size err header=0x%08x\n", header);
>  				return 0;

[PATCH v2] net: usb: smsc95xx: Limit packet length to skb->len

[Szymon Heidrich]

Packet length retrieved from descriptor may be larger than
the actual socket buffer length. In such case the cloned
skb passed up the network stack will leak kernel memory contents.

Fixes: 2f7ca802bdae ("net: Add SMSC LAN9500 USB2.0 10/100 ethernet adapter driver")
Signed-off-by: Szymon Heidrich <szymon.heidrich@gmail.com>
---
V1 -> V2: Move packet length check to prevent kernel panic in skb_pull

 drivers/net/usb/smsc95xx.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/drivers/net/usb/smsc95xx.c b/drivers/net/usb/smsc95xx.c
index 32d2c60d3..563ecd27b 100644
--- a/drivers/net/usb/smsc95xx.c
+++ b/drivers/net/usb/smsc95xx.c
@@ -1833,6 +1833,12 @@ static int smsc95xx_rx_fixup(struct usbnet *dev, struct sk_buff *skb)
 		size = (u16)((header & RX_STS_FL_) >> 16);
 		align_count = (4 - ((size + NET_IP_ALIGN) % 4)) % 4;
 
+		if (unlikely(size > skb->len)) {
+			netif_dbg(dev, rx_err, dev->net,
+				  "size err header=0x%08x\n", header);
+			return 0;
+		}
+
 		if (unlikely(header & RX_STS_ES_)) {
 			netif_dbg(dev, rx_err, dev->net,
 				  "Error header=0x%08x\n", header);
-- 
2.40.0

Re: [PATCH v2] net: usb: smsc95xx: Limit packet length to skb->len

[Jakub Kicinski]

On Thu, 16 Mar 2023 11:19:54 +0100 Szymon Heidrich wrote:
> Packet length retrieved from descriptor may be larger than
> the actual socket buffer length. In such case the cloned
> skb passed up the network stack will leak kernel memory contents.
> 
> Fixes: 2f7ca802bdae ("net: Add SMSC LAN9500 USB2.0 10/100 ethernet adapter driver")
> Signed-off-by: Szymon Heidrich <szymon.heidrich@gmail.com>

Reviewed-by: Jakub Kicinski <kuba@kernel.org>

Re: [PATCH v2] net: usb: smsc95xx: Limit packet length to skb->len

[patchwork-bot+netdevbpf]

Hello:

This patch was applied to netdev/net.git (main)
by Jakub Kicinski <kuba@kernel.org>:

On Thu, 16 Mar 2023 11:19:54 +0100 you wrote:
> Packet length retrieved from descriptor may be larger than
> the actual socket buffer length. In such case the cloned
> skb passed up the network stack will leak kernel memory contents.
> 
> Fixes: 2f7ca802bdae ("net: Add SMSC LAN9500 USB2.0 10/100 ethernet adapter driver")
> Signed-off-by: Szymon Heidrich <szymon.heidrich@gmail.com>
> 
> [...]

Here is the summary with links:
  - [v2] net: usb: smsc95xx: Limit packet length to skb->len
    https://git.kernel.org/netdev/net/c/ff821092cf02

You are awesome, thank you!
-- 
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html

Re: [PATCH] net: usb: smsc95xx: Limit packet length to skb->len

[Szymon Heidrich]

On 16/03/2023 05:24, Jakub Kicinski wrote:
> On Mon, 13 Mar 2023 23:01:24 +0100 Szymon Heidrich wrote:
>> Packet length retrieved from skb data may be larger than
> 
> nit: s/skb data/descriptor/ may be better in terms of terminology
> 
>> the actual socket buffer length (up to 1526 bytes). In such
> 
> nit: the "up to 1526 bytes" is a bit confusing, I'd remove it.
> 
>> case the cloned skb passed up the network stack will leak
>> kernel memory contents.
> 
> 
> 
>> Fixes: 2f7ca802bdae ("net: Add SMSC LAN9500 USB2.0 10/100 ethernet adapter driver")
>> Signed-off-by: Szymon Heidrich <szymon.heidrich@gmail.com>
>> ---
>>  drivers/net/usb/smsc95xx.c | 3 ++-
>>  1 file changed, 2 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/net/usb/smsc95xx.c b/drivers/net/usb/smsc95xx.c
>> index 32d2c60d3..ba766bdb2 100644
>> --- a/drivers/net/usb/smsc95xx.c
>> +++ b/drivers/net/usb/smsc95xx.c
>> @@ -1851,7 +1851,8 @@ static int smsc95xx_rx_fixup(struct usbnet *dev, struct sk_buff *skb)
>>  			}
>>  		} else {
>>  			/* ETH_FRAME_LEN + 4(CRC) + 2(COE) + 4(Vlan) */
>> -			if (unlikely(size > (ETH_FRAME_LEN + 12))) {
>> +			if (unlikely(size > (ETH_FRAME_LEN + 12) ||
>> +				     size > skb->len)) {
> 
> We need this check on both sides of the big if {} statement.
> 
> In case the error bit is set and we drop the packet we still
> end up in skb_pull() which if size > skb->len will panic the
> kernel.
> 
> So let's do this check right after size and align are calculated?
> The patch for smsc75xx has sadly already been applied so you'll
> need to prepare a fix to the fix :(
> 
>>  				netif_dbg(dev, rx_err, dev->net,
>>  					  "size err header=0x%08x\n", header);
>>  				return 0;
> 

Thank you very much for you suggestions Kuba, I provided an updated patch for
smsc95xx and a new patch addressing smsc75xx. Please let me know in case additional
amendments are required.