Re: [PATCH 2/2] iscsi_tcp: Check if tcp_conn is valid in

[Wenchao Hao]

On 2022/3/3 22:59, Mike Christie wrote:
> On 3/3/22 8:56 PM, Wenchao Hao wrote:
> > iscsi_create_conn() would add newly alloced iscsi_cls_conn to connlist,
> > it means when userspace sends ISCSI_UEVENT_SET_PARAM, iscsi_conn_lookup()
> > would found this iscsi_cls_conn and call the set_param callback which is
> > iscsi_sw_tcp_conn_set_param(). While the iscsi_conn's dd_data might not
> > been initialized.
> >
> > Signed-off-by: Wenchao Hao <haowenchao@xxxxxxxxxx>
> > Signed-off-by: Wu Bo <wubo40@xxxxxxxxxx>
> > ---
> >   drivers/scsi/iscsi_tcp.c | 6 +++++-
> >   1 file changed, 5 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/scsi/iscsi_tcp.c b/drivers/scsi/iscsi_tcp.c
> > index 14db224486be..a42449df6156 100644
> > --- a/drivers/scsi/iscsi_tcp.c
> > +++ b/drivers/scsi/iscsi_tcp.c
> > @@ -716,13 +716,17 @@ static int iscsi_sw_tcp_conn_set_param(struct iscsi_cls_conn *cls_conn,
> >   {
> >   	struct iscsi_conn *conn = cls_conn->dd_data;
> >   	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
> > -	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
> > +	struct iscsi_sw_tcp_conn *tcp_sw_conn;
> >   
> >   	switch(param) {
> >   	case ISCSI_PARAM_HDRDGST_EN:
> >   		iscsi_set_param(cls_conn, param, buf, buflen);
> >   		break;
> >   	case ISCSI_PARAM_DATADGST_EN:
> > +		if (!tcp_conn || !tcp_conn->dd_data)
> > +			return -ENOTCONN;
> > +
> > +		tcp_sw_conn = tcp_conn->dd_data;
> >   		iscsi_set_param(cls_conn, param, buf, buflen);
> >   		tcp_sw_conn->sendpage = conn->datadgst_en ?
> >   			sock_no_sendpage : tcp_sw_conn->sock->ops->sendpage;
>
> Is this something you hit or from code review?

It's from code review. I reviewed the code because the panic mentioned 
in my first patch. The issue seems would not happen, so just ignore it.

> We have those state checks:
>
> if ((conn->state == ISCSI_CONN_BOUND) ||
>      (conn->state == ISCSI_CONN_UP)) {
> 	err = transport->set_param(conn, ev->u.set_param.param,
>
> so we don't call set_param until after we have bound the
> connection which will be after ISCSI_UEVENT_CREATE_CONN has returned.
>
> Also for this specific bug iscsi_if_recv_msg is called with the
> rx_queue_mutex, so set_param can only be called after the
> ISCSI_UEVENT_CREATE_CONN cmd has returned.
> .