On Wed, Feb 08, 2023 at 09:40:50AM -0800, Lee Duncan wrote: > Right now the iscsi_endpoint is only linked to a connection once that > connection has been established. For net namespace filtering of the > sysfs objects, associate an endpoint with the host that it was > allocated for when it is created. > > Signed-off-by: Chris Leech <cleech@xxxxxxxxxx> > Signed-off-by: Lee Duncan <lduncan@xxxxxxxx> > --- > > diff --git a/drivers/infiniband/ulp/iser/iscsi_iser.c b/drivers/infiniband/ulp/iser/iscsi_iser.c > index 6b7603765383..212fa7aa9810 100644 > --- a/drivers/infiniband/ulp/iser/iscsi_iser.c > +++ b/drivers/infiniband/ulp/iser/iscsi_iser.c > @@ -802,7 +802,7 @@ static struct iscsi_endpoint *iscsi_iser_ep_connect(struct Scsi_Host *shost, > struct iser_conn *iser_conn; > struct iscsi_endpoint *ep; > > - ep = iscsi_create_endpoint(0); > + ep = iscsi_create_endpoint(shost, 0); > if (!ep) > return ERR_PTR(-ENOMEM); I started trying[1] to look at iSER, and I think this is a problem. iSER is the only iSCSI driver that uses endpoint objects, but does not require then to be bound to a host. That means that iscsi_iser_ep_connect can be called with a null shost. So this fails, and not in a new namespace. It just breaks iSER entirely. I think we need to preserve support for the iscsi_endpoint device having a virtual device path for iSER. Also, enabling net namespace support for iSER might require the ability to create an endpoint directly in a namespace instead of on a host. Kind of like the create_session discussion for iscsi_tcp. - Chris [1] I say trying, becuase before going and borrowing an RDMA setup I thought I'd give the kernel target and either siw or rxe a try. The isert module seems to have issues with siw, and I think maybe any iWARP, where setting enable_iser on a port will try and re-use the TCP port number and fail due to it being in use. With rxe my host failed, but that's becuase of this create_endpoint issue.
