Hostif Generic Netlink interface #936

padmanarayana · 2019-04-04T00:19:14Z

A netdev tunnel channel is proposed where an application can create a virtual tunnel interface (over a loopback) to receive trapped packets/buffers from the SAI driver. This can be used as an alternative to callback/FD based mechanisms.

atitjain · 2019-04-04T11:00:38Z

inc/saihostif.h

+ */
+typedef enum _sai_hostif_channel_encap_type_t
+{
+    /** Receive packets via GRE/ERSPAN Type 2*/


Can we just call it SAI_HOSTIF_TABLE_CHANNEL_ENCAP_TYPE_GRE and add one for IPinIP, then it covers all tunnels:

SAI_TUNNEL_TYPE_IPINIP, SAI_TUNNEL_TYPE_IPINIP_GRE, SAI_TUNNEL_TYPE_VXLAN,

This is a tunnel to lift packets from the SAI driver to applications. We do not plan to leverage the SAI_TUNNEL infra for this..

atitjain · 2019-04-04T11:02:00Z

inc/saihostif.h

+
+    /** End of custom range base */
+    SAI_HOSTIF_TUNNEL_ATTR_CUSTOM_RANGE_END
+} sai_hostif_tunnel_attr_t;


Where is this used? What is the use of this attribute? Cant we just use the tunnel id? Are you planning to extend:

* @type sai_object_id_t * @flags MANDATORY_ON_CREATE | CREATE_ONLY * @objects SAI_OBJECT_TYPE_PORT, SAI_OBJECT_TYPE_LAG, SAI_OBJECT_TYPE_ROUTER_INTERFACE * @condition SAI_HOSTIF_TABLE_ENTRY_ATTR_TYPE == SAI_HOSTIF_TABLE_ENTRY_TYPE_PORT or SAI_HOSTIF_TABLE_ENTRY_ATTR_TYPE == SAI_HOSTIF_TABLE_ENTRY_TYPE_VLAN or SAI_HOSTIF_TABLE_ENTRY_ATTR_TYPE == SAI_HOSTIF_TABLE_ENTRY_TYPE_LAG */

The primary use case for this will be sFlow.

Why not leverage the sFlow proposal for this. In this looks like you will do the sFlow from the CPU path? I think better is to just enable sFlow, like we do from Port, from tunnel id itself? If the idea is to sample the packets from the transit tunnel, you can apply sFlow action using ACL. If the Scaling is concern, we might need a tunnel table entry that doesnt terminate but use action sFlow.

Refer:
saisamplepacket.h
saiport.h

/** * @brief Enable/Disable Samplepacket session * * Enable ingress sampling by assigning samplepacket object id Disable * ingress sampling by assigning #SAI_NULL_OBJECT_ID as attribute value. * * @type sai_object_id_t * @flags CREATE_AND_SET * @objects SAI_OBJECT_TYPE_SAMPLEPACKET * @allownull true * @default SAI_NULL_OBJECT_ID */ SAI_PORT_ATTR_INGRESS_SAMPLEPACKET_ENABLE, /** * @brief Enable/Disable Samplepacket session * * Enable egress sampling by assigning samplepacket object id Disable * egress sampling by assigning #SAI_NULL_OBJECT_ID as attribute value. * * @type sai_object_id_t * @flags CREATE_AND_SET * @objects SAI_OBJECT_TYPE_SAMPLEPACKET * @allownull true * @default SAI_NULL_OBJECT_ID */ SAI_PORT_ATTR_EGRESS_SAMPLEPACKET_ENABLE,

saiacl.h

/** Set ingress packet sampling */ SAI_ACL_ACTION_TYPE_INGRESS_SAMPLEPACKET_ENABLE, /** Set egress packet sampling */ SAI_ACL_ACTION_TYPE_EGRESS_SAMPLEPACKET_ENABLE,

saitunnel.h
sai_tunnel_term_table_entry_attr_t

I removed the use of tunnel since it's being misunderstood with the tunnel infra..

atitjain · 2019-04-04T11:08:04Z

A netdev tunnel channel is proposed where an application can create a virtual tunnel interface (over a loopback) to receive trapped packets/buffers from the SAI driver. This can be used as an alternative to callback/FD based mechanisms.

In practice, the tunnel is created by using one of the loopback interfaces only and in that case, it's used to terminate the data traffic itself. So I couldn't quite get the use case you are trying.

kcudnik · 2019-04-09T07:05:37Z

couple issues to fix:

WARNING: coment is ending without space saihostif.h 890: /** Receive packets via GRE/ERSPAN Type 2*/

WARNING: @brief should start with capital letter: saihostif.h 899: * @brief hostif tunnel attributes

WARNING: @brief should start with capital letter: saihostif.h 942: * @brief meta header valid

WARNING: Word 'hostif' is misspelled saihostif.h 899: * @brief hostif tunnel attributes

WARNING: Value SAI_HOSTIF_TABLE_CHANNEL_ENCAP_TYPE_ERSPAN_TYPE_II of sai_hostif_channel_encap_type_t is not prefixed as SAI_HOSTIF_CHANNEL_ENCAP_TYPE_

WARNING: Value SAI_HOSTIF_TABLE_CHANNEL_ENCAP_TYPE_UDP of sai_hostif_channel_encap_type_t is not prefixed as SAI_HOSTIF_CHANNEL_ENCAP_TYPE_

ERROR: enum sai_hostif_table_entry_channel_encap_type_t has no typedef enum sai_hostif_table_entry_channel_encap_type_t

WARNING: Value SAI_HOSTIF_TABLE_CHANNEL_ENCAP_TYPE_ERSPAN_TYPE_II of sai_hostif_channel_encap_type_t is not prefixed as SAI_HOSTIF_CHANNEL_ENCAP_TYPE_

WARNING: Value SAI_HOSTIF_TABLE_CHANNEL_ENCAP_TYPE_UDP of sai_hostif_channel_encap_type_t is not prefixed as SAI_HOSTIF_CHANNEL_ENCAP_TYPE_

lguohan · 2019-04-09T16:51:14Z

I think we need example and scope in the documents.

kcudnik · 2019-04-11T10:01:52Z

you added sai_hostif_tunnel_attr_t structure attributes but there is missing SAI_OBJECT_TYPE_HOSTIF_TUNNEL object type in sai_object_type_t

kcudnik · 2019-04-11T15:18:49Z

you introduced new object type, but there are not api's to create/remove/set/get for this object type in sai_hostif_api_t, you need to add 4 new api's:

create_hostif_tunnel,
remove_hostif_tunnel
set_hostif_tunnel_attribute
get_hostif_tunnel_attribute

you can also try building metadata localy, just go to SAI/meta directory and type "make" to see if everything compiles

marian-pritsak · 2019-04-18T10:41:14Z

Did you consider using the existing psample infrastructure in Linux as an alternative to protobuf? It allows you to carry metadata along with the packet and is already a part of the mainline kernel.

sflow-rt · 2019-04-22T15:55:56Z

+1 for using psample netlink channel to deliver packet samples.

psample is a generic (subsystem independent) method of delivering packet samples and associated metadata. It is also extensible so additional metadata can easily be added in future (queue depth, queuing delay, etc.)

psample is supported by the open source Host sFlow agent.

Add updated Host-Interface proposal Change copyright header and fix UDP port type Change copyright year Fix warnings in saihostif.h Fix warnings in saihostif.h Fix warnings in saihostif.h Fix warnings in saihostif.h Update saitypes.h to include SAI_OBJECT_TYPE_HOSTIF_TUNNEL Add encap parameters to hostif, revert the use of tunnel Fix warning Add default for SAI_HOSTIF_ATTR_ENCAP_TYPE Change conditional to validonly Add defaults for encap attributes fix warnings fix warnings fix warnings fix warnings fix warnings fix warnings fix warnings fix warnings remove use of tunnel in meta header Revert Netdev tunnel approach and use Generic Netlink Fix typo Add netlink to aspell

marian-pritsak · 2019-07-04T09:19:29Z

inc/saihostif.h

     * @type char
     * @flags MANDATORY_ON_CREATE | CREATE_ONLY
-     * @condition SAI_HOSTIF_ATTR_TYPE == SAI_HOSTIF_TYPE_NETDEV
+     * @condition SAI_HOSTIF_ATTR_TYPE == SAI_HOSTIF_TYPE_NETDEV or SAI_HOSTIF_ATTR_TYPE == SAI_HOSTIF_TYPE_GENETLINK


Is it always equal to "NETLINK_GENERIC"?

This is set to the generic netlink family name whose multicast port id (SAI_HOSTIF_ATTR_GENETLINK_PORT_ID) is used to lift the samples. If the psample driver is used, it should be set to "psample" (refer https://github.com/torvalds/linux/blob/master/include/uapi/linux/psample.h#L33 ).

atitjain reviewed Apr 4, 2019

View reviewed changes

kcudnik approved these changes Apr 18, 2019

View reviewed changes

padmanarayana changed the title ~~Hostif netdev tunnel~~ Hostif Generic Netlink interface May 16, 2019

padmanarayana force-pushed the hostif_netdev_tunnel branch from d4a5170 to f9610a0 Compare May 29, 2019 23:22

Add comment to clarify SAI_HOSTIF_ATTR_NAME while using GENETLINK

1fecd36

prsunny approved these changes May 30, 2019

View reviewed changes

kcudnik approved these changes Jun 4, 2019

View reviewed changes

marian-pritsak suggested changes Jul 4, 2019

View reviewed changes

update comment for SAI_HOSTIF_TABLE_ENTRY_ATTR_HOST_IF

c7afe19

marian-pritsak approved these changes Jul 11, 2019

View reviewed changes

michaelli10 approved these changes Jul 11, 2019

View reviewed changes

lguohan merged commit ae9111b into opencomputeproject:master Jul 11, 2019

dgsudharsan mentioned this pull request Jul 31, 2019

Copp sflow changes sonic-net/sonic-swss#1011

Merged

padmanarayana mentioned this pull request Aug 13, 2019

Modify GENETLINK multicast group attribute from id to name #997

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Hostif Generic Netlink interface #936

Hostif Generic Netlink interface #936

padmanarayana commented Apr 4, 2019

atitjain Apr 4, 2019

padmanarayana Apr 11, 2019

atitjain Apr 4, 2019 •

edited

Loading

padmanarayana Apr 11, 2019

atitjain Apr 15, 2019 •

edited

Loading

padmanarayana Apr 18, 2019

atitjain commented Apr 4, 2019

kcudnik commented Apr 9, 2019

lguohan commented Apr 9, 2019

kcudnik commented Apr 11, 2019

kcudnik commented Apr 11, 2019

marian-pritsak commented Apr 18, 2019

sflow-rt commented Apr 22, 2019

marian-pritsak Jul 4, 2019

padmanarayana Jul 11, 2019

Hostif Generic Netlink interface #936

Hostif Generic Netlink interface #936

Conversation

padmanarayana commented Apr 4, 2019

atitjain Apr 4, 2019

Choose a reason for hiding this comment

padmanarayana Apr 11, 2019

Choose a reason for hiding this comment

atitjain Apr 4, 2019 • edited Loading

Choose a reason for hiding this comment

padmanarayana Apr 11, 2019

Choose a reason for hiding this comment

atitjain Apr 15, 2019 • edited Loading

Choose a reason for hiding this comment

padmanarayana Apr 18, 2019

Choose a reason for hiding this comment

atitjain commented Apr 4, 2019

kcudnik commented Apr 9, 2019

lguohan commented Apr 9, 2019

kcudnik commented Apr 11, 2019

kcudnik commented Apr 11, 2019

marian-pritsak commented Apr 18, 2019

sflow-rt commented Apr 22, 2019

marian-pritsak Jul 4, 2019

Choose a reason for hiding this comment

padmanarayana Jul 11, 2019

Choose a reason for hiding this comment

atitjain Apr 4, 2019 •

edited

Loading

atitjain Apr 15, 2019 •

edited

Loading