geneve: fix header validation in geneve[6]_xmit_skb
authorEric Dumazet <edumazet@google.com>
Fri, 5 Apr 2024 10:30:34 +0000 (10:30 +0000)
committerDavid S. Miller <davem@davemloft.net>
Mon, 8 Apr 2024 10:51:04 +0000 (11:51 +0100)
syzbot is able to trigger an uninit-value in geneve_xmit() [1]

Problem : While most ip tunnel helpers (like ip_tunnel_get_dsfield())
uses skb_protocol(skb, true), pskb_inet_may_pull() is only using
skb->protocol.

If anything else than ETH_P_IPV6 or ETH_P_IP is found in skb->protocol,
pskb_inet_may_pull() does nothing at all.

If a vlan tag was provided by the caller (af_packet in the syzbot case),
the network header might not point to the correct location, and skb
linear part could be smaller than expected.

Add skb_vlan_inet_prepare() to perform a complete mac validation.

Use this in geneve for the moment, I suspect we need to adopt this
more broadly.

v4 - Jakub reported v3 broke l2_tos_ttl_inherit.sh selftest
   - Only call __vlan_get_protocol() for vlan types.
Link: https://lore.kernel.org/netdev/20240404100035.3270a7d5@kernel.org/
v2,v3 - Addressed Sabrina comments on v1 and v2
Link: https://lore.kernel.org/netdev/Zg1l9L2BNoZWZDZG@hog/
[1]

BUG: KMSAN: uninit-value in geneve_xmit_skb drivers/net/geneve.c:910 [inline]
 BUG: KMSAN: uninit-value in geneve_xmit+0x302d/0x5420 drivers/net/geneve.c:1030
  geneve_xmit_skb drivers/net/geneve.c:910 [inline]
  geneve_xmit+0x302d/0x5420 drivers/net/geneve.c:1030
  __netdev_start_xmit include/linux/netdevice.h:4903 [inline]
  netdev_start_xmit include/linux/netdevice.h:4917 [inline]
  xmit_one net/core/dev.c:3531 [inline]
  dev_hard_start_xmit+0x247/0xa20 net/core/dev.c:3547
  __dev_queue_xmit+0x348d/0x52c0 net/core/dev.c:4335
  dev_queue_xmit include/linux/netdevice.h:3091 [inline]
  packet_xmit+0x9c/0x6c0 net/packet/af_packet.c:276
  packet_snd net/packet/af_packet.c:3081 [inline]
  packet_sendmsg+0x8bb0/0x9ef0 net/packet/af_packet.c:3113
  sock_sendmsg_nosec net/socket.c:730 [inline]
  __sock_sendmsg+0x30f/0x380 net/socket.c:745
  __sys_sendto+0x685/0x830 net/socket.c:2191
  __do_sys_sendto net/socket.c:2203 [inline]
  __se_sys_sendto net/socket.c:2199 [inline]
  __x64_sys_sendto+0x125/0x1d0 net/socket.c:2199
 do_syscall_64+0xd5/0x1f0
 entry_SYSCALL_64_after_hwframe+0x6d/0x75

Uninit was created at:
  slab_post_alloc_hook mm/slub.c:3804 [inline]
  slab_alloc_node mm/slub.c:3845 [inline]
  kmem_cache_alloc_node+0x613/0xc50 mm/slub.c:3888
  kmalloc_reserve+0x13d/0x4a0 net/core/skbuff.c:577
  __alloc_skb+0x35b/0x7a0 net/core/skbuff.c:668
  alloc_skb include/linux/skbuff.h:1318 [inline]
  alloc_skb_with_frags+0xc8/0xbf0 net/core/skbuff.c:6504
  sock_alloc_send_pskb+0xa81/0xbf0 net/core/sock.c:2795
  packet_alloc_skb net/packet/af_packet.c:2930 [inline]
  packet_snd net/packet/af_packet.c:3024 [inline]
  packet_sendmsg+0x722d/0x9ef0 net/packet/af_packet.c:3113
  sock_sendmsg_nosec net/socket.c:730 [inline]
  __sock_sendmsg+0x30f/0x380 net/socket.c:745
  __sys_sendto+0x685/0x830 net/socket.c:2191
  __do_sys_sendto net/socket.c:2203 [inline]
  __se_sys_sendto net/socket.c:2199 [inline]
  __x64_sys_sendto+0x125/0x1d0 net/socket.c:2199
 do_syscall_64+0xd5/0x1f0
 entry_SYSCALL_64_after_hwframe+0x6d/0x75

CPU: 0 PID: 5033 Comm: syz-executor346 Not tainted 6.9.0-rc1-syzkaller-00005-g928a87efa423 #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 02/29/2024

Fixes: d13f048dd40e ("net: geneve: modify IP header check in geneve6_xmit_skb and geneve_xmit_skb")
Reported-by: syzbot+9ee20ec1de7b3168db09@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/netdev/000000000000d19c3a06152f9ee4@google.com/
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Phillip Potter <phil@philpotter.co.uk>
Cc: Sabrina Dubroca <sd@queasysnail.net>
Reviewed-by: Sabrina Dubroca <sd@queasysnail.net>
Reviewed-by: Phillip Potter <phil@philpotter.co.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/geneve.c
include/net/ip_tunnels.h

index 2f6739f..6c28350 100644 (file)
@@ -822,7 +822,7 @@ static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
        __be16 sport;
        int err;
 
-       if (!pskb_inet_may_pull(skb))
+       if (!skb_vlan_inet_prepare(skb))
                return -EINVAL;
 
        if (!gs4)
@@ -929,7 +929,7 @@ static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
        __be16 sport;
        int err;
 
-       if (!pskb_inet_may_pull(skb))
+       if (!skb_vlan_inet_prepare(skb))
                return -EINVAL;
 
        if (!gs6)
index 5cd64bb..c286cc2 100644 (file)
@@ -361,6 +361,39 @@ static inline bool pskb_inet_may_pull(struct sk_buff *skb)
        return pskb_network_may_pull(skb, nhlen);
 }
 
+/* Variant of pskb_inet_may_pull().
+ */
+static inline bool skb_vlan_inet_prepare(struct sk_buff *skb)
+{
+       int nhlen = 0, maclen = ETH_HLEN;
+       __be16 type = skb->protocol;
+
+       /* Essentially this is skb_protocol(skb, true)
+        * And we get MAC len.
+        */
+       if (eth_type_vlan(type))
+               type = __vlan_get_protocol(skb, type, &maclen);
+
+       switch (type) {
+#if IS_ENABLED(CONFIG_IPV6)
+       case htons(ETH_P_IPV6):
+               nhlen = sizeof(struct ipv6hdr);
+               break;
+#endif
+       case htons(ETH_P_IP):
+               nhlen = sizeof(struct iphdr);
+               break;
+       }
+       /* For ETH_P_IPV6/ETH_P_IP we make sure to pull
+        * a base network header in skb->head.
+        */
+       if (!pskb_may_pull(skb, maclen + nhlen))
+               return false;
+
+       skb_set_network_header(skb, maclen);
+       return true;
+}
+
 static inline int ip_encap_hlen(struct ip_tunnel_encap *e)
 {
        const struct ip_tunnel_encap_ops *ops;