[PATCH net] bareudp: Fix invalid read beyond skb's linear data

Data beyond the UDP header might not be part of the skb's linear data.
Use skb_copy_bits() instead of direct access to skb->data+X, so that
we read the correct bytes even on a fragmented skb.

Fixes: 4b5f67232d95 ("net: Special handling for IP & MPLS.")
Signed-off-by: Guillaume Nault <redacted>
---
 drivers/net/bareudp.c | 16 +++++++++++-----
 1 file changed, 11 insertions(+), 5 deletions(-)

diff --git a/drivers/net/bareudp.c b/drivers/net/bareudp.c
index a7ee0af1af90..54e321a695ce 100644
--- a/drivers/net/bareudp.c
+++ b/drivers/net/bareudp.c

@@ -71,12 +71,18 @@ static int bareudp_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 		family = AF_INET6;
 
 	if (bareudp->ethertype == htons(ETH_P_IP)) {
-		struct iphdr *iphdr;
+		__u8 ipversion;
 
-		iphdr = (struct iphdr *)(skb->data + BAREUDP_BASE_HLEN);
-		if (iphdr->version == 4) {
-			proto = bareudp->ethertype;
-		} else if (bareudp->multi_proto_mode && (iphdr->version == 6)) {
+		if (skb_copy_bits(skb, BAREUDP_BASE_HLEN, &ipversion,
+				  sizeof(ipversion))) {
+			bareudp->dev->stats.rx_dropped++;
+			goto drop;
+		}
+		ipversion >>= 4;
+
+		if (ipversion == 4) {
+			proto = htons(ETH_P_IP);
+		} else if (ipversion == 6 && bareudp->multi_proto_mode) {
 			proto = htons(ETH_P_IPV6);
 		} else {
 			bareudp->dev->stats.rx_dropped++;

-- 
2.21.3

On Fri, 6 Aug 2021 17:52:06 +0200 Guillaume Nault wrote:

No preference just curious - could skb_header_pointer() be better suited?

+			bareudp->dev->stats.rx_dropped++;
+			goto drop;
+		}
+		ipversion >>= 4;
+
+		if (ipversion == 4) {
+			proto = htons(ETH_P_IP);
+		} else if (ipversion == 6 && bareudp->multi_proto_mode) {
 			proto = htons(ETH_P_IPV6);
 		} else {
 			bareudp->dev->stats.rx_dropped++;

On Fri, Aug 06, 2021 at 04:22:34PM -0700, Jakub Kicinski wrote:

On Fri, 6 Aug 2021 17:52:06 +0200 Guillaume Nault wrote:

quoted

Data beyond the UDP header might not be part of the skb's linear data.
Use skb_copy_bits() instead of direct access to skb->data+X, so that
we read the correct bytes even on a fragmented skb.

Fixes: 4b5f67232d95 ("net: Special handling for IP & MPLS.")
Signed-off-by: Guillaume Nault <redacted>
---
 drivers/net/bareudp.c | 16 +++++++++++-----
 1 file changed, 11 insertions(+), 5 deletions(-)

diff --git a/drivers/net/bareudp.c b/drivers/net/bareudp.c
index a7ee0af1af90..54e321a695ce 100644
--- a/drivers/net/bareudp.c
+++ b/drivers/net/bareudp.c

@@ -71,12 +71,18 @@ static int bareudp_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 		family = AF_INET6;
 
 	if (bareudp->ethertype == htons(ETH_P_IP)) {
-		struct iphdr *iphdr;
+		__u8 ipversion;
 
-		iphdr = (struct iphdr *)(skb->data + BAREUDP_BASE_HLEN);
-		if (iphdr->version == 4) {
-			proto = bareudp->ethertype;
-		} else if (bareudp->multi_proto_mode && (iphdr->version == 6)) {
+		if (skb_copy_bits(skb, BAREUDP_BASE_HLEN, &ipversion,
+				  sizeof(ipversion))) {

No preference just curious - could skb_header_pointer() be better suited?

I have no preference either. I just used skb_copy_bits() because it
didn't seem useful to get a pointer to the buffer (just to read one
byte of data).

But I don't mind reposting with skb_header_pointer() if anyone prefers
that solution.

On Sun, 8 Aug 2021 18:16:25 +0200 Guillaume Nault wrote:

On Fri, Aug 06, 2021 at 04:22:34PM -0700, Jakub Kicinski wrote:

quoted

On Fri, 6 Aug 2021 17:52:06 +0200 Guillaume Nault wrote:  

quoted

Data beyond the UDP header might not be part of the skb's linear data.
Use skb_copy_bits() instead of direct access to skb->data+X, so that
we read the correct bytes even on a fragmented skb.

Fixes: 4b5f67232d95 ("net: Special handling for IP & MPLS.")
Signed-off-by: Guillaume Nault <redacted>
---
 drivers/net/bareudp.c | 16 +++++++++++-----
 1 file changed, 11 insertions(+), 5 deletions(-)

diff --git a/drivers/net/bareudp.c b/drivers/net/bareudp.c
index a7ee0af1af90..54e321a695ce 100644
--- a/drivers/net/bareudp.c
+++ b/drivers/net/bareudp.c

@@ -71,12 +71,18 @@ static int bareudp_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 		family = AF_INET6;
 
 	if (bareudp->ethertype == htons(ETH_P_IP)) {
-		struct iphdr *iphdr;
+		__u8 ipversion;
 
-		iphdr = (struct iphdr *)(skb->data + BAREUDP_BASE_HLEN);
-		if (iphdr->version == 4) {
-			proto = bareudp->ethertype;
-		} else if (bareudp->multi_proto_mode && (iphdr->version == 6)) {
+		if (skb_copy_bits(skb, BAREUDP_BASE_HLEN, &ipversion,
+				  sizeof(ipversion))) {  

No preference just curious - could skb_header_pointer() be better suited?  

I have no preference either. I just used skb_copy_bits() because it
didn't seem useful to get a pointer to the buffer (just to read one
byte of data).

Right, the advantage would be in the "fast" case of skb_header_pointer()
being inlined.

But I don't mind reposting with skb_header_pointer() if anyone prefers
that solution.

Hello:

This patch was applied to netdev/net.git (refs/heads/master):

On Fri, 6 Aug 2021 17:52:06 +0200 you wrote:

Data beyond the UDP header might not be part of the skb's linear data.
Use skb_copy_bits() instead of direct access to skb->data+X, so that
we read the correct bytes even on a fragmented skb.

Fixes: 4b5f67232d95 ("net: Special handling for IP & MPLS.")
Signed-off-by: Guillaume Nault <redacted>

[...]

Here is the summary with links:
  - [net] bareudp: Fix invalid read beyond skb's linear data
    https://git.kernel.org/netdev/net/c/143a8526ab5f

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

On Mon, Aug 09, 2021 at 09:19:18AM -0700, Jakub Kicinski wrote:

On Sun, 8 Aug 2021 18:16:25 +0200 Guillaume Nault wrote:

quoted

On Fri, Aug 06, 2021 at 04:22:34PM -0700, Jakub Kicinski wrote:

quoted

On Fri, 6 Aug 2021 17:52:06 +0200 Guillaume Nault wrote:  

quoted

Data beyond the UDP header might not be part of the skb's linear data.
Use skb_copy_bits() instead of direct access to skb->data+X, so that
we read the correct bytes even on a fragmented skb.

Fixes: 4b5f67232d95 ("net: Special handling for IP & MPLS.")
Signed-off-by: Guillaume Nault <redacted>
---
 drivers/net/bareudp.c | 16 +++++++++++-----
 1 file changed, 11 insertions(+), 5 deletions(-)

diff --git a/drivers/net/bareudp.c b/drivers/net/bareudp.c
index a7ee0af1af90..54e321a695ce 100644
--- a/drivers/net/bareudp.c
+++ b/drivers/net/bareudp.c

@@ -71,12 +71,18 @@ static int bareudp_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 		family = AF_INET6;
 
 	if (bareudp->ethertype == htons(ETH_P_IP)) {
-		struct iphdr *iphdr;
+		__u8 ipversion;
 
-		iphdr = (struct iphdr *)(skb->data + BAREUDP_BASE_HLEN);
-		if (iphdr->version == 4) {
-			proto = bareudp->ethertype;
-		} else if (bareudp->multi_proto_mode && (iphdr->version == 6)) {
+		if (skb_copy_bits(skb, BAREUDP_BASE_HLEN, &ipversion,
+				  sizeof(ipversion))) {  

No preference just curious - could skb_header_pointer() be better suited?  

I have no preference either. I just used skb_copy_bits() because it
didn't seem useful to get a pointer to the buffer (just to read one
byte of data).

Right, the advantage would be in the "fast" case of skb_header_pointer()
being inlined.

Yes indeed. The problem was found because of some automated functionnal
tests, not because of any practical use cases. So I didn't consider the
possible performance differences.

I see that you've applied the patch as is already. I can switch to
skb_header_pointer() in the future, if anyone sees any practical
benefit for it.

Thanks for the review.