diff mbox series

[SRU,Bionic,1/3] ipv6: annotate some data-races around sk->sk_prot

Message ID 20221117035621.1091322-10-cascardo@canonical.com
State New
Headers show
Series [SRU,Bionic,1/3] ipv6: annotate some data-races around sk->sk_prot | expand

Commit Message

Thadeu Lima de Souza Cascardo Nov. 17, 2022, 3:56 a.m. UTC 
From: Eric Dumazet <edumazet@google.com>

IPv6 has this hack changing sk->sk_prot when an IPv6 socket
is 'converted' to an IPv4 one with IPV6_ADDRFORM option.

This operation is only performed for TCP and UDP, knowing
their 'struct proto' for the two network families are populated
in the same way, and can not disappear while a reader
might use and dereference sk->sk_prot.

If we think about it all reads of sk->sk_prot while
either socket lock or RTNL is not acquired should be using READ_ONCE().

Also note that other layers like MPTCP, XFRM, CHELSIO_TLS also
write over sk->sk_prot.

BUG: KCSAN: data-race in inet6_recvmsg / ipv6_setsockopt

write to 0xffff8881386f7aa8 of 8 bytes by task 26932 on cpu 0:
 do_ipv6_setsockopt net/ipv6/ipv6_sockglue.c:492 [inline]
 ipv6_setsockopt+0x3758/0x3910 net/ipv6/ipv6_sockglue.c:1019
 udpv6_setsockopt+0x85/0x90 net/ipv6/udp.c:1649
 sock_common_setsockopt+0x5d/0x70 net/core/sock.c:3489
 __sys_setsockopt+0x209/0x2a0 net/socket.c:2180
 __do_sys_setsockopt net/socket.c:2191 [inline]
 __se_sys_setsockopt net/socket.c:2188 [inline]
 __x64_sys_setsockopt+0x62/0x70 net/socket.c:2188
 do_syscall_x64 arch/x86/entry/common.c:50 [inline]
 do_syscall_64+0x44/0xd0 arch/x86/entry/common.c:80
 entry_SYSCALL_64_after_hwframe+0x44/0xae

read to 0xffff8881386f7aa8 of 8 bytes by task 26911 on cpu 1:
 inet6_recvmsg+0x7a/0x210 net/ipv6/af_inet6.c:659
 ____sys_recvmsg+0x16c/0x320
 ___sys_recvmsg net/socket.c:2674 [inline]
 do_recvmmsg+0x3f5/0xae0 net/socket.c:2768
 __sys_recvmmsg net/socket.c:2847 [inline]
 __do_sys_recvmmsg net/socket.c:2870 [inline]
 __se_sys_recvmmsg net/socket.c:2863 [inline]
 __x64_sys_recvmmsg+0xde/0x160 net/socket.c:2863
 do_syscall_x64 arch/x86/entry/common.c:50 [inline]
 do_syscall_64+0x44/0xd0 arch/x86/entry/common.c:80
 entry_SYSCALL_64_after_hwframe+0x44/0xae

value changed: 0xffffffff85e0e980 -> 0xffffffff85e01580

Reported by Kernel Concurrency Sanitizer on:
CPU: 1 PID: 26911 Comm: syz-executor.3 Not tainted 5.17.0-rc2-syzkaller-00316-g0457e5153e0e-dirty #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011

Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
(backported from commit 086d49058cd8471046ae9927524708820f5fd1c7)
[cascardo: context fixup since preempt is still disabled around
 sock_prot_inuse_add, see commit b3cb764aa1d753cf6a58858f9e2097ba71e8100b]
[cascardo: small conflict fixup in inet6_bind]
[cascardo: inet6_sendmsg and inet6_recvmsg are not present, inet6_bind has moved around]
CVE-2022-3567
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@canonical.com>
---
 net/ipv6/af_inet6.c      | 14 ++++++++++----
 net/ipv6/ipv6_sockglue.c |  8 ++++++--
 2 files changed, 16 insertions(+), 6 deletions(-)
diff mbox series

Patch

diff --git a/net/ipv6/af_inet6.c b/net/ipv6/af_inet6.c
index 47830ee2b4e5..dca11e98f937 100644
--- a/net/ipv6/af_inet6.c
+++ b/net/ipv6/af_inet6.c
@@ -279,6 +279,7 @@  int inet6_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 {
 	struct sockaddr_in6 *addr = (struct sockaddr_in6 *)uaddr;
 	struct sock *sk = sock->sk;
+	const struct proto *prot;
 	struct inet_sock *inet = inet_sk(sk);
 	struct ipv6_pinfo *np = inet6_sk(sk);
 	struct net *net = sock_net(sk);
@@ -288,9 +289,11 @@  int inet6_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 	int addr_type = 0;
 	int err = 0;
 
+	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
+	prot = READ_ONCE(sk->sk_prot);
 	/* If the socket has its own bind function then use it. */
-	if (sk->sk_prot->bind)
-		return sk->sk_prot->bind(sk, uaddr, addr_len);
+	if (prot->bind)
+		return prot->bind(sk, uaddr, addr_len);
 
 	if (addr_len < SIN6_LEN_RFC2133)
 		return -EINVAL;
@@ -524,6 +527,7 @@  int inet6_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 {
 	struct sock *sk = sock->sk;
 	struct net *net = sock_net(sk);
+	const struct proto *prot;
 
 	switch (cmd) {
 	case SIOCGSTAMP:
@@ -544,9 +548,11 @@  int inet6_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 	case SIOCSIFDSTADDR:
 		return addrconf_set_dstaddr(net, (void __user *) arg);
 	default:
-		if (!sk->sk_prot->ioctl)
+		/* IPV6_ADDRFORM can change sk->sk_prot under us. */
+		prot = READ_ONCE(sk->sk_prot);
+		if (!prot->ioctl)
 			return -ENOIOCTLCMD;
-		return sk->sk_prot->ioctl(sk, cmd, arg);
+		return prot->ioctl(sk, cmd, arg);
 	}
 	/*NOTREACHED*/
 	return 0;
diff --git a/net/ipv6/ipv6_sockglue.c b/net/ipv6/ipv6_sockglue.c
index ccfba74d40a4..b433adcc0fd1 100644
--- a/net/ipv6/ipv6_sockglue.c
+++ b/net/ipv6/ipv6_sockglue.c
@@ -228,7 +228,9 @@  static int do_ipv6_setsockopt(struct sock *sk, int level, int optname,
 				sock_prot_inuse_add(net, sk->sk_prot, -1);
 				sock_prot_inuse_add(net, &tcp_prot, 1);
 				local_bh_enable();
-				sk->sk_prot = &tcp_prot;
+
+				/* Paired with READ_ONCE(sk->sk_prot) in net/ipv6/af_inet6.c */
+				WRITE_ONCE(sk->sk_prot, &tcp_prot);
 				icsk->icsk_af_ops = &ipv4_specific;
 				sk->sk_socket->ops = &inet_stream_ops;
 				sk->sk_family = PF_INET;
@@ -242,7 +244,9 @@  static int do_ipv6_setsockopt(struct sock *sk, int level, int optname,
 				sock_prot_inuse_add(net, sk->sk_prot, -1);
 				sock_prot_inuse_add(net, prot, 1);
 				local_bh_enable();
-				sk->sk_prot = prot;
+
+				/* Paired with READ_ONCE(sk->sk_prot) in net/ipv6/af_inet6.c */
+				WRITE_ONCE(sk->sk_prot, prot);
 				sk->sk_socket->ops = &inet_dgram_ops;
 				sk->sk_family = PF_INET;
 			}