if CONFIG_LOCKDEP is enabled, the following
kernel warning message is generated because
rdma_accept() checks whehter the handler_mutex
is held by lockdep_assert_held. CM(Connection
Manager) holds the mutex before CM handler
callback is called.

[   63.211405 ] WARNING: CPU: 1 PID: 345 at drivers/infiniband/core/cma.c:4405 rdma_accept+0x17a/0x350
[   63.212080 ] RIP: 0010:rdma_accept+0x17a/0x350
...
[   63.214036 ] Call Trace:
[   63.214098 ]  <TASK>
[   63.214185 ]  smb_direct_accept_client+0xb4/0x170 [ksmbd]
[   63.214412 ]  smb_direct_prepare+0x322/0x8c0 [ksmbd]
[   63.214555 ]  ? rcu_read_lock_sched_held+0x3a/0x70
[   63.214700 ]  ksmbd_conn_handler_loop+0x63/0x270 [ksmbd]
[   63.214826 ]  ? ksmbd_conn_alive+0x80/0x80 [ksmbd]
[   63.214952 ]  kthread+0x171/0x1a0
[   63.215039 ]  ? set_kthread_struct+0x40/0x40
[   63.215128 ]  ret_from_fork+0x22/0x30

To avoid this, move creating a queue pair and accepting
a client from transport_ops->prepare() to
smb_direct_handle_connect_request().
Acked-by: Namjae Jeon <linkinjeon@kernel.org>
Signed-off-by: Hyunchul Lee <hyc.lee@gmail.com>
Signed-off-by: Steve French <stfrench@microsoft.com>