Skip to contentSkip to conversation

staging: wlan-ng: properly check endpoint types

PATCH In review Rustam Kovhaev 2 participants 3 messages Jul 18, 2020 – Jul 19, 2020 Staging

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 staging: wlan-ng: properly check endpoint types

As syzkaller detected, wlan-ng driver submits bulk urb without checking that the endpoint type is actually bulk, add usb_urb_ep_type_check()

---
 drivers/staging/wlan-ng/hfa384x_usb.c | 14 ++++++++++++++
 1 file changed, 14 insertions(+)
drivers/staging/wlan-ng/hfa384x_usb.c +14 -0
index fa1bf8b069fd..7cde60ea68a2 100644
@@ -339,6 +339,12 @@ static int submit_rx_urb(struct hfa384x *hw, gfp_t memflags)
 
 	hw->rx_urb_skb = skb;
 
+	result = usb_urb_ep_type_check(&hw->rx_urb);
+	if (result) {
+	       netdev_warn(hw->wlandev->netdev, "invalid rx endpoint");
+	       goto cleanup;
+	}
+
 	result = -ENOLINK;
 	if (!hw->wlandev->hwremoved &&
 	    !test_bit(WORK_RX_HALT, &hw->usb_flags)) {
@@ -354,6 +360,7 @@ static int submit_rx_urb(struct hfa384x *hw, gfp_t memflags)
 		}
 	}
 
+cleanup:
 	/* Don't leak memory if anything should go wrong */
 	if (result != 0) {
 		dev_kfree_skb(skb);
@@ -388,6 +395,12 @@ static int submit_tx_urb(struct hfa384x *hw, struct urb *tx_urb, gfp_t memflags)
 	struct net_device *netdev = hw->wlandev->netdev;
 	int result;
 
+	result = usb_urb_ep_type_check(&hw->tx_urb);
+	if (result) {
+	       netdev_warn(hw->wlandev->netdev, "invalid tx endpoint");
+	       goto done;
+	}
+
 	result = -ENOLINK;
 	if (netif_running(netdev)) {
 		if (!hw->wlandev->hwremoved &&
@@ -407,6 +420,7 @@ static int submit_tx_urb(struct hfa384x *hw, struct urb *tx_urb, gfp_t memflags)
 		}
 	}
 
+done:
 	return result;
 }
 
Signature · 1 line
2.27.0
maintainer
#1 lore raw
Quoted 15 lines · Rustam Kovhaev

As syzkaller detected, wlan-ng driver submits bulk urb without checking that the endpoint type is actually bulk, add usb_urb_ep_type_check()

---
 drivers/staging/wlan-ng/hfa384x_usb.c | 14 ++++++++++++++
 1 file changed, 14 insertions(+)
drivers/staging/wlan-ng/hfa384x_usb.c +0 -0
index fa1bf8b069fd..7cde60ea68a2 100644

drivers/staging/wlan-ng/hfa384x_usb.c static int submit_rx_urb(struct hfa384x *hw, gfp_t memflags) · line 346

Quoted 8 lines
drivers/staging/wlan-ng/hfa384x_usb.c +5 -0
@@ -339,6 +339,12 @@ static int submit_rx_urb(struct hfa384x *hw, gfp_t memflags)
 
 	hw->rx_urb_skb = skb;
 
+	result = usb_urb_ep_type_check(&hw->rx_urb);
+	if (result) {
+	       netdev_warn(hw->wlandev->netdev, "invalid rx endpoint");
+	       goto cleanup;
+	}

In looking at this again, can you just make these checks in the probe function, and abort binding the driver to the device at that point in time? This feels really late in the init sequence.

The real problem here is in the hfa384x_create() function, where it blindly takes the 1 and 2 endpoints and assumes that those are the "correct type". How about checking the types there, and if they are incorrect, returning an error from that function and have the caller return the error as well.

That should keep anything else in the driver from being initialized and set up, and stop bad devices from being bound to the driver at a much earlier point in time.

Note, just checking for the valid type/direction of those endpoints should be sufficient.

thanks,

greg k-h

#2 lore raw
Quoted 35 lines · Greg KH
Quoted 15 lines · Rustam Kovhaev

As syzkaller detected, wlan-ng driver submits bulk urb without checking that the endpoint type is actually bulk, add usb_urb_ep_type_check()

---
 drivers/staging/wlan-ng/hfa384x_usb.c | 14 ++++++++++++++
 1 file changed, 14 insertions(+)
drivers/staging/wlan-ng/hfa384x_usb.c +0 -0
index fa1bf8b069fd..7cde60ea68a2 100644

drivers/staging/wlan-ng/hfa384x_usb.c static int submit_rx_urb(struct hfa384x *hw, gfp_t memflags) · line 346

Quoted 8 lines
drivers/staging/wlan-ng/hfa384x_usb.c +5 -0
@@ -339,6 +339,12 @@ static int submit_rx_urb(struct hfa384x *hw, gfp_t memflags)
 
 	hw->rx_urb_skb = skb;
 
+	result = usb_urb_ep_type_check(&hw->rx_urb);
+	if (result) {
+	       netdev_warn(hw->wlandev->netdev, "invalid rx endpoint");
+	       goto cleanup;
+	}

In looking at this again, can you just make these checks in the probe function, and abort binding the driver to the device at that point in time? This feels really late in the init sequence.

The real problem here is in the hfa384x_create() function, where it blindly takes the 1 and 2 endpoints and assumes that those are the "correct type". How about checking the types there, and if they are incorrect, returning an error from that function and have the caller return the error as well.

Quoted 7 lines

That should keep anything else in the driver from being initialized and set up, and stop bad devices from being bound to the driver at a much earlier point in time.

Note, just checking for the valid type/direction of those endpoints should be sufficient.

tyvm for the feedback! makes perfect sense, i'll send a new patch