VLAN filtering design document

doc/content/design/vlan-filtering.md

@@ -0,0 +1,188 @@
+---
+title: VLAN filtering support
+layout: default
+design_doc: true
+revision: 1
+status: proposed
+---
+
+<!--toc:start-->
+- [Overview](#overview)
+- [Use Cases](#use-cases)
+- [Changes](#changes)
+  - [Database schema](#database-schema)
+  - [API](#api)
+  - [Behavior change](#behavior-change)
+- [Possible designs](#possible-designs)
+  - [Using VIF](#using-vif)
+  - [Using Network](#using-network)
+  - [Using PIF](#using-pif)
+  - [Using new object](#using-new-object)
+  - [Conclusion](#conclusion)
+- [Glossary](#glossary)
+<!--toc:end-->
+
+## Overview
+
+VLAN filtering is a Layer 2 network segmentation mechanism that controls how Ethernet frames are forwarded across switch ports based on VLAN membership.
+It enables a single physical switching infrastructure to support multiple isolated broadcast domains while maintaining traffic separation and policy enforcement.
+
+When VLAN filtering is enabled, the switch evaluates incoming frames against configured VLAN rules before allowing traffic to traverse the network.
+Frames are sent either tagged (IEEE 802.1Q) or untagged depending on port configuration requirements.
+
+VLAN filtering is commonly used to isolate network segments, and provides only partial view of a trunk link to VM.
+
+## Use Cases
+
+The general use case is providing trunk to some VM in multi-tenants configuration.
+For example, having a VM firewall with VLAN trunking but seeing only a subset of the VLANs present on the trunk.
+
+As examples, the following configurations could be considered,
+with one host having a PIF interface with the following VLANs \[10,11,12,13\] on it :
+
+* VM-A with VIF as trunk port in xenbr0 bridge, `trunks=[]` : VM-A sees all tagged packets so from VLANs 10 to 13 (what we are able to do currently)
+* VM-B with VIF as access port in xenbr0 bridge, `tag=10` : VM-B sees untagged packets from VLAN 10 (what we are able to do currently)
+* VM-C with VIF as trunk port in xenbr0 bridge, `trunks=[10,11]` : VM-C sees tagged packets from VLANs 10 and 11
+* VM-D with VIF as trunk port in xenbr0 bridge, `trunks=[11,12]` : VM-D sees tagged packets from VLANs 11 and 12
+* VM-E with VIF as trunk port in xenbr0 bridge, `trunks=[10]` : VM-E sees tagged packets from VLAN 10
+
+## Changes
+
+### Database schema
+
+The *VIF* class would be extended with a new attribute:
+
+* "trunks" (set int, default to empty): the 802.1Q VLANs that this port trunks (if available) ; if it is empty, then the port trunks all VLANs.
+
+### API
+
+This is a new API introduced to manage `trunks` attribute.
+
+* VIF.add_trunks
+  * self (ref VIF): reference to a valid VIF;
+  * value (int): The 802.1Q VLAN which will be associated with the VIF.
+
+* VIF.remove_trunks
+  * self (ref VIF): reference to a valid VIF;
+  * value (int): the 802.1Q VLAN which will be removed from the VIF.
+
+* VIF.set_trunks
+  * self (ref VIF): reference to a valid VIF;
+  * value (set int): The 802.1Q VLANs which will be associated with the VIF.
+
+### Behavior change
+
+When a VIF is created, *trunks* attribute on VIF is synchronized to `trunks` attribute on `Port` table in OpenvSwitch.
+As the empty list is the current default in OpenvSwitch,
+it doesn't introduce changes from current behaviour when default value is used.
+
+The `trunks` attribute on `Port` table is kept synchronized with *trunks* attribute on VIF during all the lifecycle of the port.
+
+From <https://www.openvswitch.org/support/dist-docs/ovs-vswitchd.conf.db.5.html>:
+
+> **trunks**: set of up to 4,096 integers, in range 0 to 4,095
+> For a trunk, native-tagged, or native-untagged port, the 802.1Q VLAN or VLANs that this port trunks;
+> if it is empty, then the port trunks all VLANs.
+> Must be empty if this is an access port.
+
+The type of the port is defined by `vlan_mode` column on the `Port` table.
+
+As in XAPI we don't set it, we are using the default mode defined as following:
+
+> * If tag contains a value, the port is an access port. The trunks column should be empty.
+> * Otherwise, the port is a trunk port. The trunks column value is honored if it is present.
+
+The `tag` in OpenvSwitch is derived from the **PIF's VLAN tag** on the VIF's Network.
+For consistency with OpenvSwitch, the `trunks` attribute is so expected to be empty if tag is also set.
+In XAPI term it means that a VIF with not empty *trunks* attribute could only be associated to not VLAN Network (Network with PIF with VLAN = -1).
+
+This introduces a validation constraint preventing incompatible configurations:
+
+* A VIF with non-empty `trunks` cannot be associated with a Network backed by a VLAN-tagged PIF (`PIF.vLAN` ≠ -1).
+* If a VIF is already associated with a Network backed by a VLAN-tagged PIF, its `trunks` attribute must remain empty.
+* A VLAN-tagged PIF (`PIF.vLAN` ≠ -1) cannot be associated with a Network that contains a VIF with non-empty `trunks`.
+
+### Impacts
+
+#### Update from older version
+
+All VIFs will get a new *trunks* attribute which will be the default value (empty set).
+It doesn't introduce any behavior changes.
+
+#### VM start
+
+For each created VIF, the value of *trunks* attribute will be set to `trunks` attribute on OpenvSwitch `Port` table.
+
+#### VM migration
+
+If the VM is migrated, the `trunks` attribute on OpenvSwitch `Port` table on the new host is keep in synchronization with the value of *trunks* attribute.
+
+#### VIF attribute changes
+
+On *trunks* attribute change, the value will be set to `trunks` attribute on OpenvSwitch `Port` table.
+The change is effective immediately and transparently (without replugging the VIF).
+
+
+## Possible designs
+
+The proposed design was chosen after considering the following elements.
+They are taken up here to present the possible options and reasons for choosing to use VIF.
+
+### Using VIF
+
+* **pros**
+  * simplest implementation path : only a new attribute on VIF to synchronize with OpenvSwitch `Port` configuration
+  * per-VIF granularity is a feature here : each VM gets its own filtered view, matching the multi-tenant use case
+  * no need to change existing XAPI invariants (one PIF, one Network, see below)
+  * the attribute lives on the object that actually needs the filter — no scaffolding objects required
+* **cons**
+  * doesn't reuse XAPI's existing VLAN model (VLAN -> tagged PIF -> Network chain), so the concept is somewhat duplicated
+  * configuration is scoped to a single VIF : sharing the same trunk policy across multiple VIFs would require manual duplication
+  * visibility is more limited than a dedicated Network : admins looking at network-level configuration won't immediately see which VLANs are being filtered
+
+### Using Network
+
+* **pros**
+  * natural scoping : the trunk policy would be easily discoverable
+  * consistent with how other per-network settings (MTU, locking mode, etc.) are already attached
+* **cons**
+  * a single PIF would need to be a member of several Networks simultaneously, which is conceptually problematic (a PIF normally belongs to exactly one Network)
+  * would require rethinking or relaxing the "one PIF, one Network" invariant in XAPI
+
+### Using PIF
+
+* **pros**
+  * PIF already has a VLAN-aware model via `vlan-slave-of` (the set of VLANs attached to that PIF)
+  * could express the "filtered view of available VLANs" at the physical interface level
+* **cons**
+  * we would need several PIF backends on the same physical device (one PIF per trunk view), which contradicts the one-backing-device model
+  * does not fit well with VLAN tagging or bonding on top of the PIF ; would need additional restrictions (e.g. blocking VLAN or Bond creation on a PIF that already has trunked views)
+
+### Using new object
+
+* **pros**
+  * clean semantic : a dedicated object expresses "a PIF + a subset of VLANs" without overloading existing entities
+  * could naturally group multiple VIFs under the same trunk policy (like a Network does)
+* **cons**
+  * seems overkill for the use case : adds a new object type, new API surface, and new lifecycle concerns
+  * would force creating a VLAN PIF / Network for each trunk subset, adding scaffolding noise for what is essentially a per-VIF filter
+
+### Conclusion
+
+It was chosen to put `trunks` attribute on VIF as:
+
+* it is simple design solution
+* an installation usage trunks will not have many VIFs sharing the same configuration
+
+The well suited alternative would be to create a new object (like VLAN or Bond) for holding the trunk information, but it seems overkill for the purpose.
+
+Other alternatives would need more changes in current XAPI invariants.
+
+## Glossary
+
+* **VLAN (Virtual LAN)** : A logical subdivision of a physical network that isolates traffic at Layer 2. VLANs are identified by a 12-bit ID (range 0–4095).
+* **VLAN tag** : A 4-byte field inserted into an Ethernet frame by the switch, containing a 12-bit VLAN ID (0–4095).
+* **802.1Q** : IEEE standard defining a system of VLAN tagging for Ethernet frames — adds a VLAN identifier to Ethernet frames to segregate traffic at Layer 2.
+* **Access port** : An OVS port type that carries traffic for a single VLAN. The `trunks` column must be empty; the `tag` column specifies the VLAN. Packets are untagged.
+* **Trunk port** : An OVS port type that carries traffic for multiple VLANs. The `trunks` column specifies which VLANs are allowed; if empty, all VLANs are allowed. Packets are tagged.
+* **Trunk link** : A physical or virtual link that carries traffic for multiple VLANs simultaneously. Packets are tagged.