From: Jiayuan Chen <[email protected]> This series has gone through several rounds of discussion and the maintainers hold different views on where the fix should live (in the generic xdp_master_redirect() path vs. inside bonding). I respect all of the suggestions, but I would like to get the crash fixed first, so this version takes the approach of checking whether the master device is up in xdp_master_redirect(), as suggested by Daniel Borkmann. If a different shape is preferred later it can be done as a follow-up, but the null-ptr-deref should not linger.
syzkaller reported a kernel panic, full decoded trace here: https://syzkaller.appspot.com/bug?extid=80e046b8da2820b6ba73 Problem Description bond_rr_gen_slave_id() dereferences bond->rr_tx_counter without a NULL check. rr_tx_counter is a per-CPU counter that bonding only allocates in bond_open() when the mode is round-robin. If the bond device was never brought up, rr_tx_counter stays NULL. The XDP redirect path can still reach that code on a bond that was never opened: bpf_master_redirect_enabled_key is a global static key, so as soon as any bond device has native XDP attached, the XDP_TX -> xdp_master_redirect() interception is enabled for every slave system-wide. The path xdp_master_redirect() -> bond_xdp_get_xmit_slave() -> bond_xdp_xmit_roundrobin_slave_get() -> bond_rr_gen_slave_id() then runs against a bond that has no rr_tx_counter and crashes. Solution Patch 1: Fix this in the generic xdp_master_redirect() by skipping master interception when the master device is not running. Returning XDP_TX keeps the original XDP_TX behaviour on the receiving slave, and avoids calling into any master ->ndo_xdp_get_xmit_slave() on a device that has not fully initialized its XDP state. This is not specific to bonding: any current or future master that defers XDP state allocation to ->ndo_open() is protected. Patch 2: Add a selftest that reproduces the above scenario. Changes since v5: https://lore.kernel.org/netdev/[email protected]/ - Moved the fix back into xdp_master_redirect() and use netif_running() on the master device to decide whether to intercept (Suggested by Daniel Borkmann, seconded by Paolo Abeni and Eric Dumazet) Changes since v4: https://lore.kernel.org/netdev/[email protected]/ - Reverted unconditional alloc in bond_init(); instead add a NULL check with unlikely()/READ_ONCE() in bond_rr_gen_slave_id() and WRITE_ONCE() in bond_open(), avoiding memory waste for non-RR modes (Suggested by Nikolay Aleksandrov, patch by Jay Vosburgh) Changes since v3: https://lore.kernel.org/netdev/[email protected]/T/#t - Added code comment and commit log explaining why rr_tx_counter is allocated unconditionally for all modes (Suggested by Jay Vosburgh) Changes since v2: https://lore.kernel.org/netdev/[email protected]/T/#t - Moved allocation from bond_create_init() helper into bond_init() (ndo_init), which is the natural single point covering both creation paths and also handles post-creation mode changes to round-robin Changes since v1: https://lore.kernel.org/netdev/[email protected]/T/#t - Moved the guard for NULL rr_tx_counter from xdp_master_redirect() into the bonding subsystem itself (Suggested by Sebastian Andrzej Siewior <[email protected]>) [1] https://syzkaller.appspot.com/bug?extid=80e046b8da2820b6ba73 Jiayuan Chen (2): net, bpf: fix null-ptr-deref in xdp_master_redirect() for down master selftests/bpf: add test for xdp_master_redirect with bond not up net/core/filter.c | 2 + .../selftests/bpf/prog_tests/xdp_bonding.c | 101 +++++++++++++++++- 2 files changed, 101 insertions(+), 2 deletions(-) -- 2.43.0
