Skip to contentSkip to conversation

staging: ks7010: Add null pointer check for skb

PATCH In review Aymen Qader 2 participants 6 messages Sep 27, 2018 – Sep 28, 2018 Staging

lore
Share & tools
lore
#0 lore raw

PATCH v3 1/1 staging: ks7010: Add null pointer check for skb

Add a null pointer check for the socket buffer in ks_hostif.c to avoid a possible null pointer deference, and remove a later now-redundant null pointer check.

---
v2: Remove redundant pointer check
v3: Style fix

 drivers/staging/ks7010/ks_hostif.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
drivers/staging/ks7010/ks_hostif.c +5 -1
index 0e554e3359b5..95b6c7557e84 100644
@@ -1011,6 +1011,11 @@ int hostif_data_request(struct ks_wlan_private *priv, struct sk_buff *skb)
 	size_t size;
 	int ret;
 
+	if (!skb) {
+		ret = -ENOMEM;
+		goto err_kfree;
+	}
+
 	skb_len = skb->len;
 	if (skb_len > ETH_FRAME_LEN) {
 		netdev_err(priv->net_dev, "bad length skb_len=%d\n", skb_len);
@@ -1023,7 +1028,6 @@ int hostif_data_request(struct ks_wlan_private *priv, struct sk_buff *skb)
 	    priv->wpa.mic_failure.stop) {
 		if (netif_queue_stopped(priv->net_dev))
 			netif_wake_queue(priv->net_dev);
-		if (skb)
 			dev_kfree_skb(skb);
 
 		return 0;
Signature · 1 line
2.17.1
#1 lore raw

Retraction: in hindsight I see that with the current usage of this function, there is already a check for the socket buffer so this check is unnecessary. However, I'm not sure if it's considered good practice to keep this check anyway--in any case, ENOMEM isn't the right error to return.

Quoted 39 lines · Aymen Qader

Add a null pointer check for the socket buffer in ks_hostif.c to avoid a possible null pointer deference, and remove a later now-redundant null pointer check.

---
v2: Remove redundant pointer check
v3: Style fix

 drivers/staging/ks7010/ks_hostif.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
drivers/staging/ks7010/ks_hostif.c +5 -1
index 0e554e3359b5..95b6c7557e84 100644
@@ -1011,6 +1011,11 @@ int hostif_data_request(struct ks_wlan_private *priv, struct sk_buff *skb)
 	size_t size;
 	int ret;
 
+	if (!skb) {
+		ret = -ENOMEM;
+		goto err_kfree;
+	}
+
 	skb_len = skb->len;
 	if (skb_len > ETH_FRAME_LEN) {
 		netdev_err(priv->net_dev, "bad length skb_len=%d\n", skb_len);
@@ -1023,7 +1028,6 @@ int hostif_data_request(struct ks_wlan_private *priv, struct sk_buff *skb)
 	    priv->wpa.mic_failure.stop) {
 		if (netif_queue_stopped(priv->net_dev))
 			netif_wake_queue(priv->net_dev);
-		if (skb)
 			dev_kfree_skb(skb);
 
 		return 0;
Signature · 1 line
2.17.1
#2 lore raw
Quoted 5 lines · Aymen Qader

Retraction: in hindsight I see that with the current usage of this function, there is already a check for the socket buffer so this check is unnecessary. However, I'm not sure if it's considered good practice to keep this check anyway--in any case, ENOMEM isn't the right error to return.

When we find inconsistent NULL checks then we fix it to make sense. Generally, we prefer a minimal style, with no extra code for future proofing. (The future seldom goes the way you expect and those extra NULL checks would be easy to add back).

So, yes, do remove the NULL check but also fix the indenting while you're at it.

Take your time to write patches. I write them, then I sit on them over night then I send them in the morning. It means that sometimes other people have already sent it but that's fine. If you have to redo a patch then don't send the v2 patch on the same day. v2 patches are stressful and you imagine that everyone is waiting for you to send it or something. We are not waiting for you. We don't care if you wait until next week to send these...

So when you write a v2 patch wait until the next day to send it. Then you will be calm when you review it.

regards,
dan carpenter

#3 lore raw
Quoted 12 lines · Dan Carpenter

You might want to try running Smatch on your patches. This is the second one where maybe the results would have been interesting.

git clone http://repo.or.cz/w/smatch.git
cd smatch
make
cd ~/kernel/src/
~/smatch/smatch_scripts/kchecker drivers/staging/ks7010/ks_hostif.c

This patch introduced a problem, but the earlier one fixed a potential problem in rtl8723bs:

Quoted 7 lines

drivers/staging/rtl8723bs/core/rtw_mlme_ext.c:1274 OnAssocReq() error: uninitialized symbol 'ie_len'.

Unfortunately, Smatch missed the potential NULL dereference in OnAssocReq()... It wouldn't be that hard to fix...

regards,
dan carpenter

Oh thanks, I've not used Smatch before--that's really helpful, I'm going to use that a lot in the future.

Quoted 10 lines · Dan Carpenter

Btw, if you have the cross function DB built then Smatch says that the NULL check can be removed here no problem.

$ smdb hostif_data_request
[ snip ]
drivers/staging/ks7010/ks_wlan_net.c |   ks_wlan_start_xmit |  hostif_data_request |  PARAM_VALUE |  1 | skb | s64min-(-1),1-s64max

regards,
dan carpenter

Again, thanks very much! That's really cool, Smatch looks like a great tool, I wish I found it before!

Quoted 11 lines · Dan Carpenter
Quoted 5 lines · Aymen Qader

Retraction: in hindsight I see that with the current usage of this function, there is already a check for the socket buffer so this check is unnecessary. However, I'm not sure if it's considered good practice to keep this check anyway--in any case, ENOMEM isn't the right error to return.

When we find inconsistent NULL checks then we fix it to make sense. Generally, we prefer a minimal style, with no extra code for future proofing. (The future seldom goes the way you expect and those extra NULL checks would be easy to add back).

Hm I understand, so if a parameter can currently never be null there's no need to add unnecessary checks for in case that changes in the future. I'll be sure to keep that in mind from now on.

Quoted 12 lines

So, yes, do remove the NULL check but also fix the indenting while you're at it.

Take your time to write patches. I write them, then I sit on them over night then I send them in the morning. It means that sometimes other people have already sent it but that's fine. If you have to redo a patch then don't send the v2 patch on the same day. v2 patches are stressful and you imagine that everyone is waiting for you to send it or something. We are not waiting for you. We don't care if you wait until next week to send these...

Quoted 6 lines

So when you write a v2 patch wait until the next day to send it. Then you will be calm when you review it.

regards,
dan carpenter

Thank you very much for all of these tips, I really appreciate you taking the time. I'm definitely going to try and apply this philosophy in the future, I think I've been too impulsive with my patches so far. I've been thinking "oh no, I made a mistake--I need to quickly send a v2!" and then end up making even more mistakes, so I completely understand what you mean.

Thanks again.

Kind regards,
Aymen Qader

#4 lore raw
In reply to Aymen Qader

You might want to try running Smatch on your patches. This is the second one where maybe the results would have been interesting.

git clone http://repo.or.cz/w/smatch.git
cd smatch
make
cd ~/kernel/src/
~/smatch/smatch_scripts/kchecker drivers/staging/ks7010/ks_hostif.c

This patch introduced a problem, but the earlier one fixed a potential problem in rtl8723bs:

drivers/staging/rtl8723bs/core/rtw_mlme_ext.c:1274 OnAssocReq() error: uninitialized symbol 'ie_len'.

Unfortunately, Smatch missed the potential NULL dereference in OnAssocReq()... It wouldn't be that hard to fix...

regards,
dan carpenter

#5 lore raw

Btw, if you have the cross function DB built then Smatch says that the NULL check can be removed here no problem.

$ smdb hostif_data_request
[ snip ]
drivers/staging/ks7010/ks_wlan_net.c |   ks_wlan_start_xmit |  hostif_data_request |  PARAM_VALUE |  1 | skb | s64min-(-1),1-s64max

regards,
dan carpenter