From: Håkon Bugge haakon.bugge@oracle.com Sent: 12 August 2025 07:08 PM
If rdma_dev_init_net() has an early return because the supplied net is the default init_net, we need to call rdma_nl_net_exit() before returning.
Not really. We still need to create the netlink socket so that rdma commands can be operational in init_net.
However, there is a bug in incorrectly cleaning up the init_net during ib_core driver unload flow.
I reviewed a fix internally from Mark Bloch for it. It should be posted anytime soon from Leon's queue.
Mark, Can you please comment on plan to post the fix to rdma-rc?
Fixes: 4e0f7b907072 ("RDMA/core: Implement compat device/sysfs tree in net namespace") Cc: stable@vger.kernel.org Signed-off-by: Håkon Bugge haakon.bugge@oracle.com
drivers/infiniband/core/device.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/drivers/infiniband/core/device.c b/drivers/infiniband/core/device.c index 3145cb34a1d20..ec5642e70c5db 100644 --- a/drivers/infiniband/core/device.c +++ b/drivers/infiniband/core/device.c @@ -1203,8 +1203,10 @@ static __net_init int rdma_dev_init_net(struct net *net) return ret;
/* No need to create any compat devices in default init_net. */
- if (net_eq(net, &init_net))
if (net_eq(net, &init_net)) {
rdma_nl_net_exit(rnet);return 0;
}
ret = xa_alloc(&rdma_nets, &rnet->id, rnet, xa_limit_32b,
GFP_KERNEL); if (ret) { -- 2.43.5