Skip to contentSkip to conversation

staging: r8188eu: fix a potential integer underflow bug

PATCH In review Dan Carpenter 4 participants 6 messages Feb 23, 2023 – Mar 9, 2023

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 staging: r8188eu: fix a potential integer underflow bug

Here the code is testing to see if skb->len meets a minimum size requirement. However if skb->len is very small then the ETH_HLEN subtraction will result in a negative which is then type promoted to an unsigned int and the condition will be true.

Generally, when you have an untrusted variable like skb->len, you should move all the math to the other side of the comparison.

Fixes 15865124feed ("staging: r8188eu: introduce new core dir for RTL8188eu driver")

---
Compile tested only. This is basic algebra of moving parts of the equation from one side to the other and I am surprisingly bad at something that I was supposed to have learned in 9th grade.

 drivers/staging/r8188eu/core/rtw_br_ext.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/staging/r8188eu/core/rtw_br_ext.c +1 -1
index a7c67014dde0..f49e32c33372 100644
@@ -538,7 +538,7 @@ int nat25_db_handle(struct adapter *priv, struct sk_buff *skb, int method)
 		/*------------------------------------------------*/
 		struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + ETH_HLEN);
 
-		if (sizeof(*iph) >= (skb->len - ETH_HLEN))
+		if (skb->len <= sizeof(*iph) + ETH_HLEN)
 			return -1;
 
 		switch (method) {
Signature · 1 line
2.39.1
Tested-by on v1 1/1
#1 lore raw
Quoted 23 lines · Dan Carpenter

Here the code is testing to see if skb->len meets a minimum size requirement. However if skb->len is very small then the ETH_HLEN subtraction will result in a negative which is then type promoted to an unsigned int and the condition will be true.

Generally, when you have an untrusted variable like skb->len, you should move all the math to the other side of the comparison.

Fixes 15865124feed ("staging: r8188eu: introduce new core dir for RTL8188eu driver")

---
Compile tested only. This is basic algebra of moving parts of the equation from one side to the other and I am surprisingly bad at something that I was supposed to have learned in 9th grade.

  drivers/staging/r8188eu/core/rtw_br_ext.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
drivers/staging/r8188eu/core/rtw_br_ext.c +0 -0
index a7c67014dde0..f49e32c33372 100644

drivers/staging/r8188eu/core/rtw_br_ext.c int nat25_db_handle(struct adapter *priv, struct sk_buff *skb, int method) · line 544

Quoted 8 lines
drivers/staging/r8188eu/core/rtw_br_ext.c +1 -1
@@ -538,7 +538,7 @@ int nat25_db_handle(struct adapter *priv, struct sk_buff *skb, int method)
  		/*------------------------------------------------*/
  		struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + ETH_HLEN);
 
-		if (sizeof(*iph) >= (skb->len - ETH_HLEN))
+		if (skb->len <= sizeof(*iph) + ETH_HLEN)
  			return -1;
 
  		switch (method) {
#2 lore raw
In reply to Dan Carpenter

Hi Dan,

Dan Carpenter <[email protected]> says:

Quoted 23 lines

Here the code is testing to see if skb->len meets a minimum size requirement. However if skb->len is very small then the ETH_HLEN subtraction will result in a negative which is then type promoted to an unsigned int and the condition will be true.

Generally, when you have an untrusted variable like skb->len, you should move all the math to the other side of the comparison.

Fixes 15865124feed ("staging: r8188eu: introduce new core dir for RTL8188eu driver")

---
Compile tested only. This is basic algebra of moving parts of the equation from one side to the other and I am surprisingly bad at something that I was supposed to have learned in 9th grade.

  drivers/staging/r8188eu/core/rtw_br_ext.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
drivers/staging/r8188eu/core/rtw_br_ext.c +0 -0
index a7c67014dde0..f49e32c33372 100644

drivers/staging/r8188eu/core/rtw_br_ext.c int nat25_db_handle(struct adapter *priv, struct sk_buff *skb, int method) · line 542

Quoted 6 lines
drivers/staging/r8188eu/core/rtw_br_ext.c +1 -1
@@ -538,7 +538,7 @@ int nat25_db_handle(struct adapter *priv, struct sk_buff *skb, int method)
  		/*------------------------------------------------*/
  		struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + ETH_HLEN);
 
-		if (sizeof(*iph) >= (skb->len - ETH_HLEN))
+		if (skb->len <= sizeof(*iph) + ETH_HLEN)
  			return -1;

Thanks for the patch!

I am wondering, if it make sense to use generic skb APIs which will do error handling for us?

Like following (not even build-tested tho)

With regards,
Pavel Skripkin

drivers/staging/r8188eu/core/rtw_br_ext.c +12 -9
index a7c67014dde0..8f5f2ef26056 100644
@@ -536,26 +536,29 @@ int nat25_db_handle(struct adapter *priv, struct sk_buff *skb, int method)
 		/*------------------------------------------------*/
 		/*         Handle IPV6 frame			  */
 		/*------------------------------------------------*/
-		struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + ETH_HLEN);
+		u8 header *h = skb->data;
+		struct ipv6hdr *iph = skb_pull(skb, ETH_HLEN);
 
-		if (sizeof(*iph) >= (skb->len - ETH_HLEN))
+		if (!iph)
 			return -1;
 
 		switch (method) {
 		case NAT25_CHECK:
-			if (skb->data[0] & 1)
+			if (h[0] & 1)
 				return 0;
 			return -1;
 		case NAT25_INSERT:
 			if (memcmp(&iph->saddr, "\x0\x0\x0\x0\x0\x0\x0\x0\x0\x0\x0\x0\x0\x0\x0\x0", 16)) {
 				__nat25_generate_ipv6_network_addr(addr, (unsigned int *)&iph->saddr);
-				__nat25_db_network_insert(priv, skb->data + ETH_ALEN, addr);
+				__nat25_db_network_insert(priv, (void *)iph, addr);
+
+				if (iph->nexthdr == IPPROTO_ICMPV6) {
+					struct ipv6hdr *hdr = skb_pull(skb, sizeof(*iph));
+
+					if (!iph)
+						return 0;
 
-				if (iph->nexthdr == IPPROTO_ICMPV6 &&
-						skb->len > (ETH_HLEN +  sizeof(*iph) + 4)) {
-					if (update_nd_link_layer_addr(skb->data + ETH_HLEN + sizeof(*iph),
-								      skb->len - ETH_HLEN - sizeof(*iph), GET_MY_HWADDR(priv))) {
-						struct icmp6hdr  *hdr = (struct icmp6hdr *)(skb->data + ETH_HLEN + sizeof(*iph));
+					if (update_nd_link_layer_addr(hdr, skb_len(skb), GET_MY_HWADDR(priv))) {
 						hdr->icmp6_cksum = 0;
 						hdr->icmp6_cksum = csum_ipv6_magic(&iph->saddr, &iph->daddr,
 										be16_to_cpu(iph->payload_len),
#3 lore raw
Quoted 40 lines · Pavel Skripkin

drivers/staging/r8188eu/core/rtw_br_ext.c int nat25_db_handle(struct adapter *priv, struct sk_buff *skb, int method) · line 541

Quoted 10 lines
drivers/staging/r8188eu/core/rtw_br_ext.c +1 -1
index a7c67014dde0..f49e32c33372 100644
@@ -538,7 +538,7 @@ int nat25_db_handle(struct adapter *priv, struct sk_buff *skb, int method)
  		/*------------------------------------------------*/
  		struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + ETH_HLEN);
-		if (sizeof(*iph) >= (skb->len - ETH_HLEN))
+		if (skb->len <= sizeof(*iph) + ETH_HLEN)
  			return -1;

Thanks for the patch!

I am wondering, if it make sense to use generic skb APIs which will do error handling for us?

Like following (not even build-tested tho)

With regards,
Pavel Skripkin

drivers/staging/r8188eu/core/rtw_br_ext.c +3 -2
index a7c67014dde0..8f5f2ef26056 100644
@@ -536,26 +536,29 @@ int nat25_db_handle(struct adapter *priv, struct sk_buff *skb, int method)
 		/*------------------------------------------------*/
 		/*         Handle IPV6 frame			  */
 		/*------------------------------------------------*/
-		struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + ETH_HLEN);
+		u8 header *h = skb->data;
+		struct ipv6hdr *iph = skb_pull(skb, ETH_HLEN);
 
-		if (sizeof(*iph) >= (skb->len - ETH_HLEN))
+		if (!iph)
 			return -1;
 
 		switch (method) {

drivers/staging/r8188eu/core/rtw_br_ext.c int nat25_db_handle(struct adapter *priv, struct sk_buff *skb, int method) · line 553

Quoted 10 lines
drivers/staging/r8188eu/core/rtw_br_ext.c +2 -2
 		case NAT25_CHECK:
-			if (skb->data[0] & 1)
+			if (h[0] & 1)
 				return 0;
 			return -1;
 		case NAT25_INSERT:
 			if (memcmp(&iph->saddr, "\x0\x0\x0\x0\x0\x0\x0\x0\x0\x0\x0\x0\x0\x0\x0\x0", 16)) {
 				__nat25_generate_ipv6_network_addr(addr, (unsigned int *)&iph->saddr);
-				__nat25_db_network_insert(priv, skb->data + ETH_ALEN, addr);
+				__nat25_db_network_insert(priv, (void *)iph, addr);

Doing it this way introduces a read overflow because there is no guarantee that iph is large enough. You would still need a check for if (skb->len < sizeof(*iph)). The existing check is <= instead of <, but that was in the original code. I should have investigated why it is <= and if it should be change to <. Sorry, this patch was not good.

A better way to do this would be to use skb_pull_data()... But it's still buggy because it trusts be16_to_cpu(iph->payload_len). I need to zoom out on this one. Is this NAT25 stuff even required? Do other people do NAT stuff in their wifi drivers?

regards,
dan carpenter

drivers/staging/r8188eu/core/rtw_br_ext.c +9 -10
index f49e32c33372..b84b2d4f23c3 100644
@@ -536,27 +536,26 @@ int nat25_db_handle(struct adapter *priv, struct sk_buff *skb, int method)
 		/*------------------------------------------------*/
 		/*         Handle IPV6 frame			  */
 		/*------------------------------------------------*/
-		struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + ETH_HLEN);
+		u8 *header = skb_pull_data(skb, ETH_HLEN);
+		struct ipv6hdr *iph = skb_pull_data(skb, sizeof(*iph));
 
-		if (skb->len <= sizeof(*iph) + ETH_HLEN)
+		if (!iph)
 			return -1;
 
 		switch (method) {
 		case NAT25_CHECK:
-			if (skb->data[0] & 1)
+			if (header[0] & 1)
 				return 0;
 			return -1;
 		case NAT25_INSERT:
 			if (memcmp(&iph->saddr, "\x0\x0\x0\x0\x0\x0\x0\x0\x0\x0\x0\x0\x0\x0\x0\x0", 16)) {
 				__nat25_generate_ipv6_network_addr(addr, (unsigned int *)&iph->saddr);
-				__nat25_db_network_insert(priv, skb->data + ETH_ALEN, addr);
+				__nat25_db_network_insert(priv, iph, addr);
 
-				if (iph->nexthdr == IPPROTO_ICMPV6 &&
-						skb->len > (ETH_HLEN +  sizeof(*iph) + 4)) {
-					if (update_nd_link_layer_addr(skb->data + ETH_HLEN + sizeof(*iph),
-								      skb->len - ETH_HLEN - sizeof(*iph), GET_MY_HWADDR(priv))) {
-						struct icmp6hdr  *hdr = (struct icmp6hdr *)(skb->data + ETH_HLEN + sizeof(*iph));
-						hdr->icmp6_cksum = 0;
+				if (iph->nexthdr == IPPROTO_ICMPV6 && skb->len > 4) {
+					if (update_nd_link_layer_addr(skb->data, skb->len,
+								      GET_MY_HWADDR(priv))) {
+						struct icmp6hdr  *hdr = (struct icmp6hdr *)skb->data;
 						hdr->icmp6_cksum = csum_ipv6_magic(&iph->saddr, &iph->daddr,
 										be16_to_cpu(iph->payload_len),
 										IPPROTO_ICMPV6,
#4 lore raw
In reply to Dan Carpenter
Quoted 23 lines · Dan Carpenter

Here the code is testing to see if skb->len meets a minimum size requirement. However if skb->len is very small then the ETH_HLEN subtraction will result in a negative which is then type promoted to an unsigned int and the condition will be true.

Generally, when you have an untrusted variable like skb->len, you should move all the math to the other side of the comparison.

Fixes 15865124feed ("staging: r8188eu: introduce new core dir for RTL8188eu driver")

---
Compile tested only. This is basic algebra of moving parts of the equation from one side to the other and I am surprisingly bad at something that I was supposed to have learned in 9th grade.

 drivers/staging/r8188eu/core/rtw_br_ext.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/staging/r8188eu/core/rtw_br_ext.c +0 -0
index a7c67014dde0..f49e32c33372 100644

drivers/staging/r8188eu/core/rtw_br_ext.c int nat25_db_handle(struct adapter *priv, struct sk_buff *skb, int method) · line 542

Quoted 6 lines
drivers/staging/r8188eu/core/rtw_br_ext.c +1 -1
@@ -538,7 +538,7 @@ int nat25_db_handle(struct adapter *priv, struct sk_buff *skb, int method)
 		/*------------------------------------------------*/
 		struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + ETH_HLEN);
 
-		if (sizeof(*iph) >= (skb->len - ETH_HLEN))
+		if (skb->len <= sizeof(*iph) + ETH_HLEN)
 			return -1;

NAK. On reviewing now, if this is a bug, then there is already a read overflow a few lines earlier.

regards,
dan carpenter

#5 lore raw
In reply to Dan Carpenter
Quoted 10 lines · Dan Carpenter

Here the code is testing to see if skb->len meets a minimum size requirement. However if skb->len is very small then the ETH_HLEN subtraction will result in a negative which is then type promoted to an unsigned int and the condition will be true.

Generally, when you have an untrusted variable like skb->len, you should move all the math to the other side of the comparison.

Fixes 15865124feed ("staging: r8188eu: introduce new core dir for RTL8188eu driver")
Quoted 7 lines

---
Compile tested only. This is basic algebra of moving parts of the equation from one side to the other and I am surprisingly bad at something that I was supposed to have learned in 9th grade.

 drivers/staging/r8188eu/core/rtw_br_ext.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

This driver is now deleted, so no need to worry about this anymore.

thanks,

greg k-h