From: Jiayuan Chen Similar to commit 950803f72547 ("bonding: fix type confusion in bond_setup_by_slave()") team has the same class of header_ops type confusion. For non-Ethernet ports, team_setup_by_port() copies port_dev->header_ops directly. When the team device later calls dev_hard_header() or dev_parse_header(), these callbacks can run with the team net_device instead of the real lower device, so netdev_priv(dev) is interpreted as the wrong private type and can crash. Fix this by introducing team header_ops wrappers for create/parse, selecting a team port under RCU, and calling the lower device callbacks with port->dev, so each callback always sees the correct net_device context. Also pass the selected lower device to the lower parse callback, so recursion is bounded in stacked non-Ethernet topologies and parse callbacks always run with the correct device context. Fixes: 1d76efe1577b ("team: add support for non-ethernet devices") Reported-by: syzbot+3d8bc31c45e11450f24c@syzkaller.appspotmail.com Closes: https://lore.kernel.org/all/69b46af7.050a0220.36eb34.000e.GAE@google.com/T/ Cc: Jiayuan Chen Signed-off-by: Jiayuan Chen --- drivers/net/team/team_core.c | 60 +++++++++++++++++++++++++++++++++++- 1 file changed, 59 insertions(+), 1 deletion(-) diff --git a/drivers/net/team/team_core.c b/drivers/net/team/team_core.c index b7282f5c9632..b3f659ed06ca 100644 --- a/drivers/net/team/team_core.c +++ b/drivers/net/team/team_core.c @@ -2058,6 +2058,64 @@ static const struct ethtool_ops team_ethtool_ops = { * rt netlink interface ***********************/ +/* For tx path we need a linkup && enabled port; for parse any port suffices. */ +static struct team_port *team_header_port_get_rcu(struct team *team, bool txable) +{ + struct team_port *port; + + list_for_each_entry_rcu(port, &team->port_list, list) { + if (!txable || team_port_txable(port)) + return port; + } + + return NULL; +} + +static int team_header_create(struct sk_buff *skb, struct net_device *team_dev, + unsigned short type, const void *daddr, + const void *saddr, unsigned int len) +{ + struct team *team = netdev_priv(team_dev); + const struct header_ops *port_ops; + struct team_port *port; + int ret = 0; + + rcu_read_lock(); + port = team_header_port_get_rcu(team, true); + if (port) { + port_ops = READ_ONCE(port->dev->header_ops); + if (port_ops && port_ops->create) + ret = port_ops->create(skb, port->dev, + type, daddr, saddr, len); + } + rcu_read_unlock(); + return ret; +} + +static int team_header_parse(const struct sk_buff *skb, const struct net_device *dev, + unsigned char *haddr) +{ + struct team *team = netdev_priv(dev); + const struct header_ops *port_ops; + struct team_port *port; + int ret = 0; + + rcu_read_lock(); + port = team_header_port_get_rcu(team, false); + if (port) { + port_ops = READ_ONCE(port->dev->header_ops); + if (port_ops && port_ops->parse) + ret = port_ops->parse(skb, port->dev, haddr); + } + rcu_read_unlock(); + return ret; +} + +static const struct header_ops team_header_ops = { + .create = team_header_create, + .parse = team_header_parse, +}; + static void team_setup_by_port(struct net_device *dev, struct net_device *port_dev) { @@ -2066,7 +2124,7 @@ static void team_setup_by_port(struct net_device *dev, if (port_dev->type == ARPHRD_ETHER) dev->header_ops = team->header_ops_cache; else - dev->header_ops = port_dev->header_ops; + dev->header_ops = port_dev->header_ops ? &team_header_ops : NULL; dev->type = port_dev->type; dev->hard_header_len = port_dev->hard_header_len; dev->needed_headroom = port_dev->needed_headroom; -- 2.43.0 From: Jiayuan Chen Add a team selftest that sets up: g0 (gre) -> b0 (bond) -> t0 (team) and triggers IPv6 traffic on t0. This reproduces the non-Ethernet header_ops confusion scenario and protects against regressions in stacked team/bond/gre configurations. Using this script, the panic reported by syzkaller can be reproduced [1]. After the fix: # ./non_ether_header_ops.sh PASS: non-Ethernet header_ops stacking did not crash [1] https://syzkaller.appspot.com/bug?extid=3d8bc31c45e11450f24c Cc: Jiayuan Chen Signed-off-by: Jiayuan Chen --- .../selftests/drivers/net/team/Makefile | 1 + .../testing/selftests/drivers/net/team/config | 2 + .../drivers/net/team/non_ether_header_ops.sh | 40 +++++++++++++++++++ 3 files changed, 43 insertions(+) create mode 100755 tools/testing/selftests/drivers/net/team/non_ether_header_ops.sh diff --git a/tools/testing/selftests/drivers/net/team/Makefile b/tools/testing/selftests/drivers/net/team/Makefile index 45a3e7ad3dcb..02d6f51d5a06 100644 --- a/tools/testing/selftests/drivers/net/team/Makefile +++ b/tools/testing/selftests/drivers/net/team/Makefile @@ -3,6 +3,7 @@ TEST_PROGS := \ dev_addr_lists.sh \ + non_ether_header_ops.sh \ options.sh \ propagation.sh \ refleak.sh \ diff --git a/tools/testing/selftests/drivers/net/team/config b/tools/testing/selftests/drivers/net/team/config index 558e1d0cf565..40fdd5fbcbdb 100644 --- a/tools/testing/selftests/drivers/net/team/config +++ b/tools/testing/selftests/drivers/net/team/config @@ -1,5 +1,7 @@ +CONFIG_BONDING=y CONFIG_DUMMY=y CONFIG_IPV6=y +CONFIG_NET_IPGRE=y CONFIG_MACVLAN=y CONFIG_NETDEVSIM=m CONFIG_NET_TEAM=y diff --git a/tools/testing/selftests/drivers/net/team/non_ether_header_ops.sh b/tools/testing/selftests/drivers/net/team/non_ether_header_ops.sh new file mode 100755 index 000000000000..19bd57f40688 --- /dev/null +++ b/tools/testing/selftests/drivers/net/team/non_ether_header_ops.sh @@ -0,0 +1,40 @@ +#!/bin/bash +# SPDX-License-Identifier: GPL-2.0 +# +# Reproduce the non-Ethernet header_ops confusion scenario with: +# g0 (gre) -> b0 (bond) -> t0 (team) +# +# Before the fix, direct header_ops inheritance in this stack could call +# callbacks with the wrong net_device context and crash. + +lib_dir=$(dirname "$0") +source "$lib_dir"/../../../net/lib.sh + +trap cleanup_all_ns EXIT + +setup_ns ns1 + +ip -n "$ns1" link add d0 type dummy +ip -n "$ns1" addr add 10.10.10.1/24 dev d0 +ip -n "$ns1" link set d0 up + +ip -n "$ns1" link add g0 type gre local 10.10.10.1 +ip -n "$ns1" link add b0 type bond mode active-backup +ip -n "$ns1" link add t0 type team + +ip -n "$ns1" link set g0 master b0 +ip -n "$ns1" link set b0 master t0 + +ip -n "$ns1" link set g0 up +ip -n "$ns1" link set b0 up +ip -n "$ns1" link set t0 up + +# GRE makes team inherit IFF_POINTOPOINT|IFF_NOARP, so IPv4 skips +# dev_hard_header() entirely... use IPv6 NDP which still calls it. +ip -n "$ns1" -6 addr add 2001:db8:1::1/64 dev t0 nodad +for i in $(seq 1 20); do + ip netns exec "$ns1" ping -6 -I t0 ff02::1 -c1 -W1 &>/dev/null || true +done + +echo "PASS: non-Ethernet header_ops stacking did not crash" +exit "$EXIT_STATUS" -- 2.43.0