mptcp: only admit explicitly supported sockopt
authorPaolo Abeni <pabeni@redhat.com>
Thu, 15 Apr 2021 23:44:52 +0000 (16:44 -0700)
committerDavid S. Miller <davem@davemloft.net>
Fri, 16 Apr 2021 22:23:09 +0000 (15:23 -0700)
Unrolling mcast state at msk dismantel time is bug prone, as
syzkaller reported:

======================================================
WARNING: possible circular locking dependency detected
5.11.0-syzkaller #0 Not tainted
------------------------------------------------------
syz-executor905/8822 is trying to acquire lock:
ffffffff8d678fe8 (rtnl_mutex){+.+.}-{3:3}, at: ipv6_sock_mc_close+0xd7/0x110 net/ipv6/mcast.c:323

but task is already holding lock:
ffff888024390120 (sk_lock-AF_INET6){+.+.}-{0:0}, at: lock_sock include/net/sock.h:1600 [inline]
ffff888024390120 (sk_lock-AF_INET6){+.+.}-{0:0}, at: mptcp6_release+0x57/0x130 net/mptcp/protocol.c:3507

which lock already depends on the new lock.

Instead we can simply forbid any mcast-related setsockopt.
Let's do the same with all other non supported sockopts.

Fixes: 717e79c867ca5 ("mptcp: Add setsockopt()/getsockopt() socket operations")
Co-developed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/mptcp/sockopt.c

index 479f756..fb98fab 100644 (file)
@@ -82,6 +82,219 @@ static int mptcp_setsockopt_v6(struct mptcp_sock *msk, int optname,
        return ret;
 }
 
+static bool mptcp_supported_sockopt(int level, int optname)
+{
+       if (level == SOL_SOCKET) {
+               switch (optname) {
+               case SO_DEBUG:
+               case SO_REUSEPORT:
+               case SO_REUSEADDR:
+
+               /* the following ones need a better implementation,
+                * but are quite common we want to preserve them
+                */
+               case SO_BINDTODEVICE:
+               case SO_SNDBUF:
+               case SO_SNDBUFFORCE:
+               case SO_RCVBUF:
+               case SO_RCVBUFFORCE:
+               case SO_KEEPALIVE:
+               case SO_PRIORITY:
+               case SO_LINGER:
+               case SO_TIMESTAMP_OLD:
+               case SO_TIMESTAMP_NEW:
+               case SO_TIMESTAMPNS_OLD:
+               case SO_TIMESTAMPNS_NEW:
+               case SO_TIMESTAMPING_OLD:
+               case SO_TIMESTAMPING_NEW:
+               case SO_RCVLOWAT:
+               case SO_RCVTIMEO_OLD:
+               case SO_RCVTIMEO_NEW:
+               case SO_SNDTIMEO_OLD:
+               case SO_SNDTIMEO_NEW:
+               case SO_MARK:
+               case SO_INCOMING_CPU:
+               case SO_BINDTOIFINDEX:
+               case SO_BUSY_POLL:
+               case SO_PREFER_BUSY_POLL:
+               case SO_BUSY_POLL_BUDGET:
+
+               /* next ones are no-op for plain TCP */
+               case SO_NO_CHECK:
+               case SO_DONTROUTE:
+               case SO_BROADCAST:
+               case SO_BSDCOMPAT:
+               case SO_PASSCRED:
+               case SO_PASSSEC:
+               case SO_RXQ_OVFL:
+               case SO_WIFI_STATUS:
+               case SO_NOFCS:
+               case SO_SELECT_ERR_QUEUE:
+                       return true;
+               }
+
+               /* SO_OOBINLINE is not supported, let's avoid the related mess */
+               /* SO_ATTACH_FILTER, SO_ATTACH_BPF, SO_ATTACH_REUSEPORT_CBPF,
+                * SO_DETACH_REUSEPORT_BPF, SO_DETACH_FILTER, SO_LOCK_FILTER,
+                * we must be careful with subflows
+                */
+               /* SO_ATTACH_REUSEPORT_EBPF is not supported, at it checks
+                * explicitly the sk_protocol field
+                */
+               /* SO_PEEK_OFF is unsupported, as it is for plain TCP */
+               /* SO_MAX_PACING_RATE is unsupported, we must be careful with subflows */
+               /* SO_CNX_ADVICE is currently unsupported, could possibly be relevant,
+                * but likely needs careful design
+                */
+               /* SO_ZEROCOPY is currently unsupported, TODO in sndmsg */
+               /* SO_TXTIME is currently unsupported */
+               return false;
+       }
+       if (level == SOL_IP) {
+               switch (optname) {
+               /* should work fine */
+               case IP_FREEBIND:
+               case IP_TRANSPARENT:
+
+               /* the following are control cmsg related */
+               case IP_PKTINFO:
+               case IP_RECVTTL:
+               case IP_RECVTOS:
+               case IP_RECVOPTS:
+               case IP_RETOPTS:
+               case IP_PASSSEC:
+               case IP_RECVORIGDSTADDR:
+               case IP_CHECKSUM:
+               case IP_RECVFRAGSIZE:
+
+               /* common stuff that need some love */
+               case IP_TOS:
+               case IP_TTL:
+               case IP_BIND_ADDRESS_NO_PORT:
+               case IP_MTU_DISCOVER:
+               case IP_RECVERR:
+
+               /* possibly less common may deserve some love */
+               case IP_MINTTL:
+
+               /* the following is apparently a no-op for plain TCP */
+               case IP_RECVERR_RFC4884:
+                       return true;
+               }
+
+               /* IP_OPTIONS is not supported, needs subflow care */
+               /* IP_HDRINCL, IP_NODEFRAG are not supported, RAW specific */
+               /* IP_MULTICAST_TTL, IP_MULTICAST_LOOP, IP_UNICAST_IF,
+                * IP_ADD_MEMBERSHIP, IP_ADD_SOURCE_MEMBERSHIP, IP_DROP_MEMBERSHIP,
+                * IP_DROP_SOURCE_MEMBERSHIP, IP_BLOCK_SOURCE, IP_UNBLOCK_SOURCE,
+                * MCAST_JOIN_GROUP, MCAST_LEAVE_GROUP MCAST_JOIN_SOURCE_GROUP,
+                * MCAST_LEAVE_SOURCE_GROUP, MCAST_BLOCK_SOURCE, MCAST_UNBLOCK_SOURCE,
+                * MCAST_MSFILTER, IP_MULTICAST_ALL are not supported, better not deal
+                * with mcast stuff
+                */
+               /* IP_IPSEC_POLICY, IP_XFRM_POLICY are nut supported, unrelated here */
+               return false;
+       }
+       if (level == SOL_IPV6) {
+               switch (optname) {
+               case IPV6_V6ONLY:
+
+               /* the following are control cmsg related */
+               case IPV6_RECVPKTINFO:
+               case IPV6_2292PKTINFO:
+               case IPV6_RECVHOPLIMIT:
+               case IPV6_2292HOPLIMIT:
+               case IPV6_RECVRTHDR:
+               case IPV6_2292RTHDR:
+               case IPV6_RECVHOPOPTS:
+               case IPV6_2292HOPOPTS:
+               case IPV6_RECVDSTOPTS:
+               case IPV6_2292DSTOPTS:
+               case IPV6_RECVTCLASS:
+               case IPV6_FLOWINFO:
+               case IPV6_RECVPATHMTU:
+               case IPV6_RECVORIGDSTADDR:
+               case IPV6_RECVFRAGSIZE:
+
+               /* the following ones need some love but are quite common */
+               case IPV6_TCLASS:
+               case IPV6_TRANSPARENT:
+               case IPV6_FREEBIND:
+               case IPV6_PKTINFO:
+               case IPV6_2292PKTOPTIONS:
+               case IPV6_UNICAST_HOPS:
+               case IPV6_MTU_DISCOVER:
+               case IPV6_MTU:
+               case IPV6_RECVERR:
+               case IPV6_FLOWINFO_SEND:
+               case IPV6_FLOWLABEL_MGR:
+               case IPV6_MINHOPCOUNT:
+               case IPV6_DONTFRAG:
+               case IPV6_AUTOFLOWLABEL:
+
+               /* the following one is a no-op for plain TCP */
+               case IPV6_RECVERR_RFC4884:
+                       return true;
+               }
+
+               /* IPV6_HOPOPTS, IPV6_RTHDRDSTOPTS, IPV6_RTHDR, IPV6_DSTOPTS are
+                * not supported
+                */
+               /* IPV6_MULTICAST_HOPS, IPV6_MULTICAST_LOOP, IPV6_UNICAST_IF,
+                * IPV6_MULTICAST_IF, IPV6_ADDRFORM,
+                * IPV6_ADD_MEMBERSHIP, IPV6_DROP_MEMBERSHIP, IPV6_JOIN_ANYCAST,
+                * IPV6_LEAVE_ANYCAST, IPV6_MULTICAST_ALL, MCAST_JOIN_GROUP, MCAST_LEAVE_GROUP,
+                * MCAST_JOIN_SOURCE_GROUP, MCAST_LEAVE_SOURCE_GROUP,
+                * MCAST_BLOCK_SOURCE, MCAST_UNBLOCK_SOURCE, MCAST_MSFILTER
+                * are not supported better not deal with mcast
+                */
+               /* IPV6_ROUTER_ALERT, IPV6_ROUTER_ALERT_ISOLATE are not supported, since are evil */
+
+               /* IPV6_IPSEC_POLICY, IPV6_XFRM_POLICY are not supported */
+               /* IPV6_ADDR_PREFERENCES is not supported, we must be careful with subflows */
+               return false;
+       }
+       if (level == SOL_TCP) {
+               switch (optname) {
+               /* the following are no-op or should work just fine */
+               case TCP_THIN_DUPACK:
+               case TCP_DEFER_ACCEPT:
+
+               /* the following need some love */
+               case TCP_MAXSEG:
+               case TCP_NODELAY:
+               case TCP_THIN_LINEAR_TIMEOUTS:
+               case TCP_CONGESTION:
+               case TCP_ULP:
+               case TCP_CORK:
+               case TCP_KEEPIDLE:
+               case TCP_KEEPINTVL:
+               case TCP_KEEPCNT:
+               case TCP_SYNCNT:
+               case TCP_SAVE_SYN:
+               case TCP_LINGER2:
+               case TCP_WINDOW_CLAMP:
+               case TCP_QUICKACK:
+               case TCP_USER_TIMEOUT:
+               case TCP_TIMESTAMP:
+               case TCP_NOTSENT_LOWAT:
+               case TCP_TX_DELAY:
+                       return true;
+               }
+
+               /* TCP_MD5SIG, TCP_MD5SIG_EXT are not supported, MD5 is not compatible with MPTCP */
+
+               /* TCP_REPAIR, TCP_REPAIR_QUEUE, TCP_QUEUE_SEQ, TCP_REPAIR_OPTIONS,
+                * TCP_REPAIR_WINDOW are not supported, better avoid this mess
+                */
+               /* TCP_FASTOPEN_KEY, TCP_FASTOPEN TCP_FASTOPEN_CONNECT, TCP_FASTOPEN_NO_COOKIE,
+                * are not supported fastopen is currently unsupported
+                */
+               /* TCP_INQ is currently unsupported, needs some recvmsg work */
+       }
+       return false;
+}
+
 int mptcp_setsockopt(struct sock *sk, int level, int optname,
                     sockptr_t optval, unsigned int optlen)
 {
@@ -90,6 +303,9 @@ int mptcp_setsockopt(struct sock *sk, int level, int optname,
 
        pr_debug("msk=%p", msk);
 
+       if (!mptcp_supported_sockopt(level, optname))
+               return -ENOPROTOOPT;
+
        if (level == SOL_SOCKET)
                return mptcp_setsockopt_sol_socket(msk, optname, optval, optlen);