[PATCH rdma-next 0/3] Add support for multicast in RoCEv2

[PATCH rdma-next 0/3] Add support for multicast in RoCEv2

[Leon Romanovsky]

This series from Noa enables multicast traffic over RoCE v2.

To allow this, several changes were required.

First of all, there is a need to validate MLID when link layer is
Infiniband. This check is mandatory according to Infiniband spec
but the MLID should be ignored for RoCE according to RoCEv2 Annex
(Annex 17, A17.5.6).

Second, there is need to cange the mapping of IPv4 multicast
addresses to MGIDs to be according to RoCEv2 Annex (A17.4.1.1 L3
Addresses): GID = ::ffff:<IPv4 Address>

And third, multicast addresses need to be mapped to multicast MAC
instead of searching for it in the GID table.

Thanks

Noa Osherovich (3):
  IB/core: Fix the validations of a multicast LID in attach or detach
    operations
  IB/core: Set RoCEv2 MGID according to spec
  IB/core: Add support for RoCEv2 multicast

 drivers/infiniband/core/cma.c   | 13 +++++----
 drivers/infiniband/core/verbs.c | 62 +++++++++++++++++++++++++++++++++++++----
 include/rdma/ib_addr.h          |  8 +++++-
 3 files changed, 70 insertions(+), 13 deletions(-)

--
2.12.2

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH rdma-next 1/3] IB/core: Fix the validations of a multicast LID in attach or detach operations

[Leon Romanovsky]

From: Noa Osherovich <noaos-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>

RoCE Annex (A16.9.10/11) declares that during attach (detach) QP to a
multicast group, if the QP is associated with a RoCE port, the
multicast group MLID is unused and is ignored.

During attach or detach multicast, when the QP is associated with a
port, it is enough to check the port's link layer and validate the
LID only if it is Infiniband. Otherwise, avoid validating the
multicast LID.

Fixes: 8561eae60ff9 ("IB/core: For multicast functions, verify that LIDs are multicast LIDs")
Signed-off-by: Noa Osherovich <noaos-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Reviewed-by: Moni Shoua <monis-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Signed-off-by: Leon Romanovsky <leon-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org>
---
 drivers/infiniband/core/verbs.c | 44 +++++++++++++++++++++++++++++++++++++----
 1 file changed, 40 insertions(+), 4 deletions(-)

diff --git a/drivers/infiniband/core/verbs.c b/drivers/infiniband/core/verbs.c
index 4792f5209ac2..38f6bec08d56 100644
--- a/drivers/infiniband/core/verbs.c
+++ b/drivers/infiniband/core/verbs.c
@@ -1519,6 +1519,44 @@ EXPORT_SYMBOL(ib_dealloc_fmr);

 /* Multicast groups */

+static bool is_valid_mcast_lid(struct ib_qp *qp, u16 lid)
+{
+	struct ib_qp_init_attr init_attr = {};
+	struct ib_qp_attr attr = {};
+	int num_eth_ports = 0;
+	int port;
+
+	/* If QP state >= init, it is assigned to a port and we can check this
+	 * port only.
+	 */
+	if (!ib_query_qp(qp, &attr, IB_QP_STATE | IB_QP_PORT, &init_attr)) {
+		if (attr.qp_state >= IB_QPS_INIT) {
+			if (qp->device->get_link_layer(qp->device, attr.port_num) !=
+			    IB_LINK_LAYER_INFINIBAND)
+				return true;
+			goto lid_check;
+		}
+	}
+
+	/* Can't get a quick answer, iterate over all ports */
+	for (port = 0; port < qp->device->phys_port_cnt; port++)
+		if (qp->device->get_link_layer(qp->device, port) !=
+		    IB_LINK_LAYER_INFINIBAND)
+			num_eth_ports++;
+
+	/* If we have at lease one Ethernet port, RoCE annex declares that
+	 * multicast LID should be ignored. We can't tell at this step if the
+	 * QP belongs to an IB or Ethernet port.
+	 */
+	if (num_eth_ports)
+		return true;
+
+	/* If all the ports are IB, we can check according to IB spec. */
+lid_check:
+	return !(lid < be16_to_cpu(IB_MULTICAST_LID_BASE) ||
+		 lid == be16_to_cpu(IB_LID_PERMISSIVE));
+}
+
 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid)
 {
 	int ret;
@@ -1526,8 +1564,7 @@ int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid)
 	if (!qp->device->attach_mcast)
 		return -ENOSYS;
 	if (gid->raw[0] != 0xff || qp->qp_type != IB_QPT_UD ||
-	    lid < be16_to_cpu(IB_MULTICAST_LID_BASE) ||
-	    lid == be16_to_cpu(IB_LID_PERMISSIVE))
+	    !is_valid_mcast_lid(qp, lid))
 		return -EINVAL;

 	ret = qp->device->attach_mcast(qp, gid, lid);
@@ -1544,8 +1581,7 @@ int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid)
 	if (!qp->device->detach_mcast)
 		return -ENOSYS;
 	if (gid->raw[0] != 0xff || qp->qp_type != IB_QPT_UD ||
-	    lid < be16_to_cpu(IB_MULTICAST_LID_BASE) ||
-	    lid == be16_to_cpu(IB_LID_PERMISSIVE))
+	    !is_valid_mcast_lid(qp, lid))
 		return -EINVAL;

 	ret = qp->device->detach_mcast(qp, gid, lid);
--
2.12.2

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH rdma-next 2/3] IB/core: Set RoCEv2 MGID according to spec

[Leon Romanovsky]

From: Noa Osherovich <noaos-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>

RoCEv2 Annex states that for RoCEv2 over IPv4, the corresponding
IPv4 address is encoded into the GID according to the following rule:
GID= :ffff:<IPv4 address>

Remove the 0xff0e prefix for RoCEv2 packets with IPv4 and leave it
zeroed and change rdma_is_multicast_addr() to consider the new logic.

Signed-off-by: Noa Osherovich <noaos-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Reviewed-by: Moni Shoua <monis-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Signed-off-by: Leon Romanovsky <leon-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org>
---
 drivers/infiniband/core/cma.c   | 13 +++++++------
 drivers/infiniband/core/verbs.c | 10 ++++++----
 include/rdma/ib_addr.h          |  8 +++++++-
 3 files changed, 20 insertions(+), 11 deletions(-)

diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c
index 31bb82d8ecd7..303a8207513a 100644
--- a/drivers/infiniband/core/cma.c
+++ b/drivers/infiniband/core/cma.c
@@ -4022,7 +4022,8 @@ static void iboe_mcast_work_handler(struct work_struct *work)
 	kfree(mw);
 }

-static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid)
+static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid,
+			      enum ib_gid_type gid_type)
 {
 	struct sockaddr_in *sin = (struct sockaddr_in *)addr;
 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
@@ -4032,8 +4033,8 @@ static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid)
 	} else if (addr->sa_family == AF_INET6) {
 		memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
 	} else {
-		mgid->raw[0] = 0xff;
-		mgid->raw[1] = 0x0e;
+		mgid->raw[0] = (gid_type == IB_GID_TYPE_IB) ? 0xff : 0;
+		mgid->raw[1] = (gid_type == IB_GID_TYPE_IB) ? 0x0e : 0;
 		mgid->raw[2] = 0;
 		mgid->raw[3] = 0;
 		mgid->raw[4] = 0;
@@ -4074,7 +4075,9 @@ static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
 		goto out1;
 	}

-	cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid);
+	gid_type = id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
+		   rdma_start_port(id_priv->cma_dev->device)];
+	cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid, gid_type);

 	mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
 	if (id_priv->id.ps == RDMA_PS_UDP)
@@ -4090,8 +4093,6 @@ static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
 	mc->multicast.ib->rec.hop_limit = 1;
 	mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);

-	gid_type = id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
-		   rdma_start_port(id_priv->cma_dev->device)];
 	if (addr->sa_family == AF_INET) {
 		if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
 			mc->multicast.ib->rec.hop_limit = IPV6_DEFAULT_HOPLIMIT;
diff --git a/drivers/infiniband/core/verbs.c b/drivers/infiniband/core/verbs.c
index 38f6bec08d56..0e8e6d1b5b5b 100644
--- a/drivers/infiniband/core/verbs.c
+++ b/drivers/infiniband/core/verbs.c
@@ -1563,8 +1563,9 @@ int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid)

 	if (!qp->device->attach_mcast)
 		return -ENOSYS;
-	if (gid->raw[0] != 0xff || qp->qp_type != IB_QPT_UD ||
-	    !is_valid_mcast_lid(qp, lid))
+
+	if (!rdma_is_multicast_addr((struct in6_addr *)gid->raw) ||
+	    qp->qp_type != IB_QPT_UD || !is_valid_mcast_lid(qp, lid))
 		return -EINVAL;

 	ret = qp->device->attach_mcast(qp, gid, lid);
@@ -1580,8 +1581,9 @@ int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid)

 	if (!qp->device->detach_mcast)
 		return -ENOSYS;
-	if (gid->raw[0] != 0xff || qp->qp_type != IB_QPT_UD ||
-	    !is_valid_mcast_lid(qp, lid))
+
+	if (!rdma_is_multicast_addr((struct in6_addr *)gid->raw) ||
+	    qp->qp_type != IB_QPT_UD || !is_valid_mcast_lid(qp, lid))
 		return -EINVAL;

 	ret = qp->device->detach_mcast(qp, gid, lid);
diff --git a/include/rdma/ib_addr.h b/include/rdma/ib_addr.h
index 4b34c51f859e..503a02f7c3da 100644
--- a/include/rdma/ib_addr.h
+++ b/include/rdma/ib_addr.h
@@ -302,7 +302,13 @@ static inline void rdma_get_ll_mac(struct in6_addr *addr, u8 *mac)

 static inline int rdma_is_multicast_addr(struct in6_addr *addr)
 {
-	return addr->s6_addr[0] == 0xff;
+	u32 ipv4_addr;
+
+	if (addr->s6_addr[0] == 0xff)
+		return 1;
+
+	memcpy(&ipv4_addr, addr->s6_addr + 12, 4);
+	return (ipv6_addr_v4mapped(addr) && ipv4_is_multicast(ipv4_addr));
 }

 static inline void rdma_get_mcast_mac(struct in6_addr *addr, u8 *mac)
--
2.12.2

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH rdma-next 3/3] IB/core: Add support for RoCEv2 multicast

[Leon Romanovsky]

From: Noa Osherovich <noaos-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>

When creating address handle from multicast GID, set MAC according to
the appropriate formula instead of searching for it in the GID table:
- For IPv4 multicast GID use ip_eth_mc_map().
- For IPv6 multicast GID use ipv6_eth_mc_map().

Signed-off-by: Noa Osherovich <noaos-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Reviewed-by: Moni Shoua <monis-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Signed-off-by: Leon Romanovsky <leon-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org>
---
 drivers/infiniband/core/verbs.c | 12 ++++++++++++
 1 file changed, 12 insertions(+)

diff --git a/drivers/infiniband/core/verbs.c b/drivers/infiniband/core/verbs.c
index 0e8e6d1b5b5b..c24be769aab9 100644
--- a/drivers/infiniband/core/verbs.c
+++ b/drivers/infiniband/core/verbs.c
@@ -1220,6 +1220,18 @@ int ib_resolve_eth_dmac(struct ib_device *device,
 	if (rdma_link_local_addr((struct in6_addr *)grh->dgid.raw)) {
 		rdma_get_ll_mac((struct in6_addr *)grh->dgid.raw,
 				ah_attr->roce.dmac);
+		return 0;
+	}
+	if (rdma_is_multicast_addr((struct in6_addr *)ah_attr->grh.dgid.raw)) {
+		if (ipv6_addr_v4mapped((struct in6_addr *)ah_attr->grh.dgid.raw)) {
+			__be32 addr = 0;
+
+			memcpy(&addr, ah_attr->grh.dgid.raw + 12, 4);
+			ip_eth_mc_map(addr, (char *)ah_attr->roce.dmac);
+		} else {
+			ipv6_eth_mc_map((struct in6_addr *)ah_attr->grh.dgid.raw,
+					(char *)ah_attr->roce.dmac);
+		}
 	} else {
 		union ib_gid		sgid;
 		struct ib_gid_attr	sgid_attr;
--
2.12.2

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

RE: [PATCH rdma-next 1/3] IB/core: Fix the validations of a multicast LID in attach or detach operations

[Hefty, Sean]

> +static bool is_valid_mcast_lid(struct ib_qp *qp, u16 lid)
> +{
> +	struct ib_qp_init_attr init_attr = {};
> +	struct ib_qp_attr attr = {};
> +	int num_eth_ports = 0;
> +	int port;
> +
> +	/* If QP state >= init, it is assigned to a port and we can
> check this
> +	 * port only.
> +	 */
> +	if (!ib_query_qp(qp, &attr, IB_QP_STATE | IB_QP_PORT,
> &init_attr)) {
> +		if (attr.qp_state >= IB_QPS_INIT) {
> +			if (qp->device->get_link_layer(qp->device,
> attr.port_num) !=
> +			    IB_LINK_LAYER_INFINIBAND)
> +				return true;
> +			goto lid_check;
> +		}
> +	}
> +
> +	/* Can't get a quick answer, iterate over all ports */
> +	for (port = 0; port < qp->device->phys_port_cnt; port++)
> +		if (qp->device->get_link_layer(qp->device, port) !=
> +		    IB_LINK_LAYER_INFINIBAND)
> +			num_eth_ports++;
> +
> +	/* If we have at lease one Ethernet port, RoCE annex declares
> that
> +	 * multicast LID should be ignored. We can't tell at this step
> if the
> +	 * QP belongs to an IB or Ethernet port.
> +	 */
> +	if (num_eth_ports)
> +		return true;

Reporting that a lid *might* be valid is a really weak check.  Can we require that RoCE QPs be in the INIT state prior to joining, or use the mgid, or something to make the validation check work?  Do we even know if the qp is a roce qp?


> +
> +	/* If all the ports are IB, we can check according to IB spec.
> */
> +lid_check:
> +	return !(lid < be16_to_cpu(IB_MULTICAST_LID_BASE) ||
> +		 lid == be16_to_cpu(IB_LID_PERMISSIVE));
> +}
> +
>  int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid)
>  {
>  	int ret;
> @@ -1526,8 +1564,7 @@ int ib_attach_mcast(struct ib_qp *qp, union
> ib_gid *gid, u16 lid)
>  	if (!qp->device->attach_mcast)
>  		return -ENOSYS;
>  	if (gid->raw[0] != 0xff || qp->qp_type != IB_QPT_UD ||
> -	    lid < be16_to_cpu(IB_MULTICAST_LID_BASE) ||
> -	    lid == be16_to_cpu(IB_LID_PERMISSIVE))
> +	    !is_valid_mcast_lid(qp, lid))
>  		return -EINVAL;
> 
>  	ret = qp->device->attach_mcast(qp, gid, lid);
> @@ -1544,8 +1581,7 @@ int ib_detach_mcast(struct ib_qp *qp, union
> ib_gid *gid, u16 lid)
>  	if (!qp->device->detach_mcast)
>  		return -ENOSYS;
>  	if (gid->raw[0] != 0xff || qp->qp_type != IB_QPT_UD ||
> -	    lid < be16_to_cpu(IB_MULTICAST_LID_BASE) ||
> -	    lid == be16_to_cpu(IB_LID_PERMISSIVE))
> +	    !is_valid_mcast_lid(qp, lid))
>  		return -EINVAL;
> 
>  	ret = qp->device->detach_mcast(qp, gid, lid);
> --
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH rdma-next 1/3] IB/core: Fix the validations of a multicast LID in attach or detach operations

[Leon Romanovsky]

[-- Attachment #1: Type: text/plain, Size: 2753 bytes --]

On Mon, Jun 12, 2017 at 04:15:11PM +0000, Hefty, Sean wrote:
> > +static bool is_valid_mcast_lid(struct ib_qp *qp, u16 lid)
> > +{
> > +	struct ib_qp_init_attr init_attr = {};
> > +	struct ib_qp_attr attr = {};
> > +	int num_eth_ports = 0;
> > +	int port;
> > +
> > +	/* If QP state >= init, it is assigned to a port and we can
> > check this
> > +	 * port only.
> > +	 */
> > +	if (!ib_query_qp(qp, &attr, IB_QP_STATE | IB_QP_PORT,
> > &init_attr)) {
> > +		if (attr.qp_state >= IB_QPS_INIT) {
> > +			if (qp->device->get_link_layer(qp->device,
> > attr.port_num) !=
> > +			    IB_LINK_LAYER_INFINIBAND)
> > +				return true;
> > +			goto lid_check;
> > +		}
> > +	}
> > +
> > +	/* Can't get a quick answer, iterate over all ports */
> > +	for (port = 0; port < qp->device->phys_port_cnt; port++)
> > +		if (qp->device->get_link_layer(qp->device, port) !=
> > +		    IB_LINK_LAYER_INFINIBAND)
> > +			num_eth_ports++;
> > +
> > +	/* If we have at lease one Ethernet port, RoCE annex declares
> > that
> > +	 * multicast LID should be ignored. We can't tell at this step
> > if the
> > +	 * QP belongs to an IB or Ethernet port.
> > +	 */
> > +	if (num_eth_ports)
> > +		return true;
>
> Reporting that a lid *might* be valid is a really weak check.  Can we require that RoCE QPs be in the INIT state prior to joining, or use the mgid, or something to make the validation check work?  Do we even know if the qp is a roce qp?

It would be possible if ah existed (ib_qp_attr->rdma_ah_attr->rdma_ah_attr_type), but it isn't filled for IB_QPT_UD in mlx5.

>
>
> > +
> > +	/* If all the ports are IB, we can check according to IB spec.
> > */
> > +lid_check:
> > +	return !(lid < be16_to_cpu(IB_MULTICAST_LID_BASE) ||
> > +		 lid == be16_to_cpu(IB_LID_PERMISSIVE));
> > +}
> > +
> >  int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid)
> >  {
> >  	int ret;
> > @@ -1526,8 +1564,7 @@ int ib_attach_mcast(struct ib_qp *qp, union
> > ib_gid *gid, u16 lid)
> >  	if (!qp->device->attach_mcast)
> >  		return -ENOSYS;
> >  	if (gid->raw[0] != 0xff || qp->qp_type != IB_QPT_UD ||
> > -	    lid < be16_to_cpu(IB_MULTICAST_LID_BASE) ||
> > -	    lid == be16_to_cpu(IB_LID_PERMISSIVE))
> > +	    !is_valid_mcast_lid(qp, lid))
> >  		return -EINVAL;
> >
> >  	ret = qp->device->attach_mcast(qp, gid, lid);
> > @@ -1544,8 +1581,7 @@ int ib_detach_mcast(struct ib_qp *qp, union
> > ib_gid *gid, u16 lid)
> >  	if (!qp->device->detach_mcast)
> >  		return -ENOSYS;
> >  	if (gid->raw[0] != 0xff || qp->qp_type != IB_QPT_UD ||
> > -	    lid < be16_to_cpu(IB_MULTICAST_LID_BASE) ||
> > -	    lid == be16_to_cpu(IB_LID_PERMISSIVE))
> > +	    !is_valid_mcast_lid(qp, lid))
> >  		return -EINVAL;
> >
> >  	ret = qp->device->detach_mcast(qp, gid, lid);
> > --

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH rdma-next 1/3] IB/core: Fix the validations of a multicast LID in attach or detach operations

[Moni Shoua]

On Mon, Jun 12, 2017 at 7:15 PM, Hefty, Sean <sean.hefty-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org> wrote:
>> +static bool is_valid_mcast_lid(struct ib_qp *qp, u16 lid)
>> +{
>> +     struct ib_qp_init_attr init_attr = {};
>> +     struct ib_qp_attr attr = {};
>> +     int num_eth_ports = 0;
>> +     int port;
>> +
>> +     /* If QP state >= init, it is assigned to a port and we can
>> check this
>> +      * port only.
>> +      */
>> +     if (!ib_query_qp(qp, &attr, IB_QP_STATE | IB_QP_PORT,
>> &init_attr)) {
>> +             if (attr.qp_state >= IB_QPS_INIT) {
>> +                     if (qp->device->get_link_layer(qp->device,
>> attr.port_num) !=
>> +                         IB_LINK_LAYER_INFINIBAND)
>> +                             return true;
>> +                     goto lid_check;
>> +             }
>> +     }
>> +
>> +     /* Can't get a quick answer, iterate over all ports */
>> +     for (port = 0; port < qp->device->phys_port_cnt; port++)
>> +             if (qp->device->get_link_layer(qp->device, port) !=
>> +                 IB_LINK_LAYER_INFINIBAND)
>> +                     num_eth_ports++;
>> +
>> +     /* If we have at lease one Ethernet port, RoCE annex declares
>> that
>> +      * multicast LID should be ignored. We can't tell at this step
>> if the
>> +      * QP belongs to an IB or Ethernet port.
>> +      */
>> +     if (num_eth_ports)
>> +             return true;
>
> Reporting that a lid *might* be valid is a really weak check.  Can we require that RoCE QPs be in the INIT state prior to joining, or use the mgid, or something to make the validation check work?  Do we even know if the qp is a roce qp?
>
>
This is not a join flow but attach flow. AFAIK, attaching a QP to a
mcast group can be when QP is in RESET state. This patch does the best
it can to give a valid answer but when QP is in RESET state and at
least one of the ports is Eth port then we can't invalidate a MLID.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH rdma-next 0/3] Add support for multicast in RoCEv2

[Doug Ledford]

On Mon, 2017-06-12 at 11:14 +0300, Leon Romanovsky wrote:
> This series from Noa enables multicast traffic over RoCE v2.
> 
> To allow this, several changes were required.
> 
> First of all, there is a need to validate MLID when link layer is
> Infiniband. This check is mandatory according to Infiniband spec
> but the MLID should be ignored for RoCE according to RoCEv2 Annex
> (Annex 17, A17.5.6).
> 
> Second, there is need to cange the mapping of IPv4 multicast
> addresses to MGIDs to be according to RoCEv2 Annex (A17.4.1.1 L3
> Addresses): GID = ::ffff:<IPv4 Address>
> 
> And third, multicast addresses need to be mapped to multicast MAC
> instead of searching for it in the GID table.
> 
> Thanks
> 
> Noa Osherovich (3):
>   IB/core: Fix the validations of a multicast LID in attach or detach
>     operations
>   IB/core: Set RoCEv2 MGID according to spec
>   IB/core: Add support for RoCEv2 multicast
> 
>  drivers/infiniband/core/cma.c   | 13 +++++----
>  drivers/infiniband/core/verbs.c | 62
> +++++++++++++++++++++++++++++++++++++----
>  include/rdma/ib_addr.h          |  8 +++++-
>  3 files changed, 70 insertions(+), 13 deletions(-)

Thanks, series applied.

-- 
Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
    GPG KeyID: B826A3330E572FDD
    Key fingerprint = AE6B 1BDA 122B 23B4 265B  1274 B826 A333 0E57 2FDD

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html