]> git.ipfire.org Git - thirdparty/kernel/stable-queue.git/blob - queue-6.6/udp-do-not-accept-non-tunnel-gso-skbs-landing-in-a-tunnel.patch
Linux 6.1.85
[thirdparty/kernel/stable-queue.git] / queue-6.6 / udp-do-not-accept-non-tunnel-gso-skbs-landing-in-a-tunnel.patch
1 From 3d010c8031e39f5fa1e8b13ada77e0321091011f Mon Sep 17 00:00:00 2001
2 From: Antoine Tenart <atenart@kernel.org>
3 Date: Tue, 26 Mar 2024 12:33:58 +0100
4 Subject: udp: do not accept non-tunnel GSO skbs landing in a tunnel
5
6 From: Antoine Tenart <atenart@kernel.org>
7
8 commit 3d010c8031e39f5fa1e8b13ada77e0321091011f upstream.
9
10 When rx-udp-gro-forwarding is enabled UDP packets might be GROed when
11 being forwarded. If such packets might land in a tunnel this can cause
12 various issues and udp_gro_receive makes sure this isn't the case by
13 looking for a matching socket. This is performed in
14 udp4/6_gro_lookup_skb but only in the current netns. This is an issue
15 with tunneled packets when the endpoint is in another netns. In such
16 cases the packets will be GROed at the UDP level, which leads to various
17 issues later on. The same thing can happen with rx-gro-list.
18
19 We saw this with geneve packets being GROed at the UDP level. In such
20 case gso_size is set; later the packet goes through the geneve rx path,
21 the geneve header is pulled, the offset are adjusted and frag_list skbs
22 are not adjusted with regard to geneve. When those skbs hit
23 skb_fragment, it will misbehave. Different outcomes are possible
24 depending on what the GROed skbs look like; from corrupted packets to
25 kernel crashes.
26
27 One example is a BUG_ON[1] triggered in skb_segment while processing the
28 frag_list. Because gso_size is wrong (geneve header was pulled)
29 skb_segment thinks there is "geneve header size" of data in frag_list,
30 although it's in fact the next packet. The BUG_ON itself has nothing to
31 do with the issue. This is only one of the potential issues.
32
33 Looking up for a matching socket in udp_gro_receive is fragile: the
34 lookup could be extended to all netns (not speaking about performances)
35 but nothing prevents those packets from being modified in between and we
36 could still not find a matching socket. It's OK to keep the current
37 logic there as it should cover most cases but we also need to make sure
38 we handle tunnel packets being GROed too early.
39
40 This is done by extending the checks in udp_unexpected_gso: GSO packets
41 lacking the SKB_GSO_UDP_TUNNEL/_CSUM bits and landing in a tunnel must
42 be segmented.
43
44 [1] kernel BUG at net/core/skbuff.c:4408!
45 RIP: 0010:skb_segment+0xd2a/0xf70
46 __udp_gso_segment+0xaa/0x560
47
48 Fixes: 9fd1ff5d2ac7 ("udp: Support UDP fraglist GRO/GSO.")
49 Fixes: 36707061d6ba ("udp: allow forwarding of plain (non-fraglisted) UDP GRO packets")
50 Signed-off-by: Antoine Tenart <atenart@kernel.org>
51 Reviewed-by: Willem de Bruijn <willemb@google.com>
52 Signed-off-by: David S. Miller <davem@davemloft.net>
53 Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
54 ---
55 include/linux/udp.h | 28 ++++++++++++++++++++++++++++
56 net/ipv4/udp.c | 7 +++++++
57 net/ipv4/udp_offload.c | 6 ++++--
58 net/ipv6/udp.c | 2 +-
59 4 files changed, 40 insertions(+), 3 deletions(-)
60
61 --- a/include/linux/udp.h
62 +++ b/include/linux/udp.h
63 @@ -140,6 +140,24 @@ static inline void udp_cmsg_recv(struct
64 }
65 }
66
67 +DECLARE_STATIC_KEY_FALSE(udp_encap_needed_key);
68 +#if IS_ENABLED(CONFIG_IPV6)
69 +DECLARE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
70 +#endif
71 +
72 +static inline bool udp_encap_needed(void)
73 +{
74 + if (static_branch_unlikely(&udp_encap_needed_key))
75 + return true;
76 +
77 +#if IS_ENABLED(CONFIG_IPV6)
78 + if (static_branch_unlikely(&udpv6_encap_needed_key))
79 + return true;
80 +#endif
81 +
82 + return false;
83 +}
84 +
85 static inline bool udp_unexpected_gso(struct sock *sk, struct sk_buff *skb)
86 {
87 if (!skb_is_gso(skb))
88 @@ -153,6 +171,16 @@ static inline bool udp_unexpected_gso(st
89 !udp_test_bit(ACCEPT_FRAGLIST, sk))
90 return true;
91
92 + /* GSO packets lacking the SKB_GSO_UDP_TUNNEL/_CSUM bits might still
93 + * land in a tunnel as the socket check in udp_gro_receive cannot be
94 + * foolproof.
95 + */
96 + if (udp_encap_needed() &&
97 + READ_ONCE(udp_sk(sk)->encap_rcv) &&
98 + !(skb_shinfo(skb)->gso_type &
99 + (SKB_GSO_UDP_TUNNEL | SKB_GSO_UDP_TUNNEL_CSUM)))
100 + return true;
101 +
102 return false;
103 }
104
105 --- a/net/ipv4/udp.c
106 +++ b/net/ipv4/udp.c
107 @@ -584,6 +584,13 @@ static inline bool __udp_is_mcast_sock(s
108 }
109
110 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
111 +EXPORT_SYMBOL(udp_encap_needed_key);
112 +
113 +#if IS_ENABLED(CONFIG_IPV6)
114 +DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
115 +EXPORT_SYMBOL(udpv6_encap_needed_key);
116 +#endif
117 +
118 void udp_encap_enable(void)
119 {
120 static_branch_inc(&udp_encap_needed_key);
121 --- a/net/ipv4/udp_offload.c
122 +++ b/net/ipv4/udp_offload.c
123 @@ -552,8 +552,10 @@ struct sk_buff *udp_gro_receive(struct l
124 unsigned int off = skb_gro_offset(skb);
125 int flush = 1;
126
127 - /* we can do L4 aggregation only if the packet can't land in a tunnel
128 - * otherwise we could corrupt the inner stream
129 + /* We can do L4 aggregation only if the packet can't land in a tunnel
130 + * otherwise we could corrupt the inner stream. Detecting such packets
131 + * cannot be foolproof and the aggregation might still happen in some
132 + * cases. Such packets should be caught in udp_unexpected_gso later.
133 */
134 NAPI_GRO_CB(skb)->is_flist = 0;
135 if (!sk || !udp_sk(sk)->gro_receive) {
136 --- a/net/ipv6/udp.c
137 +++ b/net/ipv6/udp.c
138 @@ -450,7 +450,7 @@ csum_copy_err:
139 goto try_again;
140 }
141
142 -DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
143 +DECLARE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
144 void udpv6_encap_enable(void)
145 {
146 static_branch_inc(&udpv6_encap_needed_key);