From: Hangbin Liu <redacted>
When bond_enslave() succeeds up to the XDP setup stage, slave_cnt is
already incremented. If XDP setup subsequently fails, the error paths
jump directly to err_sysfs_del, bypassing the slave_cnt decrement.
This causes slave_cnt to drift upward on each failed enslaving attempt,
which would lead to unbalanced traffic distribution with round-robin
mode.
Add an error out to make sure slave_cnt gets decreased correctly.
Fixes: 9e2ee5c7e7c3 ("net, bonding: Add XDP support to the bonding driver")
Reviewed-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: Hangbin Liu <redacted>
---
This issue is found during code review. The reproducer is from
Eric's patch
ip link add bond0 type bond mode balance-rr
ip link set bond0 xdpdrv obj xdp_dummy.bpf.o sec xdp
for i in 0 1 2; do
ip link add d$i type dummy
ip link set d$i master bond0 # fails: Slave does not support XDP
done
num_slaves read via SIOCBONDINFOQUERY is 3 before this patch, 0 after.
For Sashiko:
This patch only focuses on fixing the bond_slave_cnt leak issue.
1. `bond_xdp_xmit()` may call the slave's `ndo_xdp_xmit` before checking
whether `slave->netdev_ops->ndo_xdp_xmit` exists. This is a separate issue
and I will address it in another patch.
2. `bond_slave_ns_maddrs_add()` and device promiscuous/allmulti error
handling are also separate issues, which I will fix separately.
PS: I'm working on a restruct of bond_enslave to fix all these issues.
---
Changes in v5:
- Re-post, no code change.
- Link to v4: https://lore.kernel.org/r/20260914-bond_slave_cnt-v4-1-d14864a39e60@kylinos.cn (local)
Changes in v4:
- Revert to v1 version as the bond_select_active_slave() just under
slave_cnt increase also calls bond_update_slave_arr() (sashiko)
- re-run all bonding selftests with debug kernel, all passed.
- Link to v3: https://lore.kernel.org/r/20260907-bond_slave_cnt-v3-1-57df3b3cf2cb@kylinos.cn (local)
Changes in v3:
- move the slave_cnt increasement before bond_update_slave_arr() (selftest)
- run all bonding selftests on debug kernel to make sure no regression (Jakub)
- Link to v2: https://lore.kernel.org/r/20260903-bond_slave_cnt-v2-1-02e27304ca36@kylinos.cn (local)
Changes in v2:
- move the slave_cnt increasement after XDP setup (Nikolay Aleksandrov)
- balance-xor mode is not affected, not mention it (Nikolay Aleksandrov)
- Link to v1: https://lore.kernel.org/r/20260902-bond_slave_cnt-v1-1-36e95bf4a6ff@kylinos.cn (local)
---
drivers/net/bonding/bond_main.c | 9 ++++++---
1 file changed, 6 insertions(+), 3 deletions(-)
diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/bond_main.c
index a1c156c145b2..34f1b8083fbb 100644
--- a/drivers/net/bonding/bond_main.c
+++ b/drivers/net/bonding/bond_main.c
@@ -2305,7 +2305,7 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
SLAVE_NL_ERR(bond_dev, slave_dev, extack,
"Slave does not support XDP");
res = -EOPNOTSUPP;
- goto err_sysfs_del;
+ goto err_slave_cnt;
}
} else if (bond->xdp_prog) {
struct netdev_bpf xdp = {@@ -2319,14 +2319,14 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
SLAVE_NL_ERR(bond_dev, slave_dev, extack,
"Slave has XDP program loaded, please unload before enslaving");
res = -EOPNOTSUPP;
- goto err_sysfs_del;
+ goto err_slave_cnt;
}
res = dev_xdp_propagate(slave_dev, &xdp);
if (res < 0) {
/* ndo_bpf() sets extack error message */
slave_dbg(bond_dev, slave_dev, "Error %d calling ndo_bpf\n", res);
- goto err_sysfs_del;
+ goto err_slave_cnt;
}
if (bond->xdp_prog)
bpf_prog_inc(bond->xdp_prog);@@ -2348,6 +2348,9 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
return 0;
/* Undo stages on error */
+err_slave_cnt:
+ WRITE_ONCE(bond->slave_cnt, bond->slave_cnt - 1);
+
err_sysfs_del:
bond_sysfs_slave_del(new_slave);
---
base-commit: fafbc4051123d5459ea6950899cec08ed946e278
change-id: 20260807-bond_slave_cnt-88e78c5f0ab9
Best regards,
--
Hangbin Liu [off-list ref]