5 ** OVN_Northbound schema
7 *** Needs to support extra routes
9 Currently a router port has a single route associated with it, but
10 presumably we should support multiple routes. For connections from
11 one router to another, this doesn't seem to matter (just put more than
12 one connection between them), but for connections between a router and
13 a switch it might matter because a switch has only one router port.
17 *** Allow output to ingress port
19 Sometimes when a packet ingresses into a router, it has to egress the
20 same port. One example is a "one-armed" router that has multiple
21 routes on a single port (or in which a host is (mis)configured to send
22 every IP packet to the router, e.g. due to a bad netmask). Another is
23 when a router needs to send an ICMP reply to an ingressing packet.
25 To some degree this problem is layered, because there are two
26 different notions of "ingress port". The first is the OpenFlow
27 ingress port, essentially a physical port identifier. This is
28 implemented as part of ovs-vswitchd's OpenFlow implementation. It
29 prevents a reply from being sent across the tunnel on which it
30 arrived. It is questionable whether this OpenFlow feature is useful
31 to OVN. (OVN already has to override it to allow a packet from one
32 nested container to be forwarded to a different nested container.)
33 OVS make it possible to disable this feature of OpenFlow by setting
34 the OpenFlow input port field to 0. (If one does this too early, of
35 course, it means that there's no way to actually match on the input
36 port in the OpenFlow flow tables, but one can work around that by
37 instead setting the input port just before the output action, possibly
38 wrapping these actions in push/pop pairs to preserve the input port
41 The second is the OVN logical ingress port, which is implemented in
42 ovn-controller as part of the logical abstraction, using an OVS
43 register. Dropping packets directed to the logical ingress port is
44 implemented through an OpenFlow table not directly visible to the
45 logical flow table. Currently this behavior can't be disabled, but
46 various ways to ensure it could be implemented, e.g. the same as for
47 OpenFlow by allowing the logical inport to be zeroed, or by
48 introducing a new action that ignores the inport.
52 *** What flows should it generate?
54 See description in ovn-northd(8).
56 ** New OVN logical actions
60 Generates an ARP packet based on the current IPv4 packet and allows it
61 to be processed as part of the current pipeline (and then pop back to
62 processing the original IPv4 packet).
64 TCP/IP stacks typically limit the rate at which ARPs are sent, e.g. to
65 one per second for a given target. We might need to do this too.
67 We probably need to buffer the packet that generated the ARP. I don't
68 know where to do that.
70 *** icmp4 { action... }
72 Generates an ICMPv4 packet based on the current IPv4 packet and
73 processes it according to each nested action (and then pops back to
74 processing the original IPv4 packet). The intended use case is for
75 generating "time exceeded" and "destination unreachable" errors.
77 ovn-sb.xml includes a tentative specification for this action.
79 Tentatively, the icmp4 action sets a default icmp_type and icmp_code
80 and lets the nested actions override it. This means that we'd have to
81 make icmp_type and icmp_code writable. Because changing icmp_type and
82 icmp_code can change the interpretation of the rest of the data in the
83 ICMP packet, we would want to think this through carefully. If it
84 seems like a bad idea then we could instead make the type and code a
85 parameter to the action: icmp4(type, code) { action... }
87 It is worth considering what should be considered the ingress port for
88 the ICMPv4 packet. It's quite likely that the ICMPv4 packet is going
89 to go back out the ingress port. Maybe the icmp4 action, therefore,
90 should clear the inport, so that output to the original inport won't
95 Transforms the current TCP packet into a RST reply.
97 ovn-sb.xml includes a tentative specification for this action.
99 *** Other actions for IPv6.
101 IPv6 will probably need an action or actions for ND that is similar to
102 the "arp" action, and an action for generating
104 *** ovn-controller translation to OpenFlow
106 The following two translation strategies come to mind. Some of the
107 new actions we might want to implement one way, some of them the
108 other, depending on the details.
110 *** Implementation strategies
112 One way to do this is to define new actions as Open vSwitch extensions
113 to OpenFlow, emit those actions in ovn-controller, and implement them
114 in ovs-vswitchd (possibly pushing the implementations into the Linux
115 and DPDK datapaths as well). This is the only acceptable way for
116 actions that need high performance. None of these actions obviously
117 need high performance, but it might be necessary to have fairness in
118 handling e.g. a flood of incoming packets that require these actions.
119 The main disadvantage of this approach is that it ties ovs-vswitchd
120 (and the Linux kernel module) to supporting these actions essentially
121 forever, which means that we'd want to make sure that they are
122 general-purpose, well designed, maintainable, and supportable.
124 The other way to do this is to send the packets across an OpenFlow
125 channel to ovn-controller and have ovn-controller process them. This
126 is acceptable for actions that don't need high performance, and it
127 means that we don't add anything permanently to ovs-vswitchd or the
128 kernel (so we can be more casual about the design). The big
129 disadvantage is that it becomes necessary to add a way to resume the
130 OpenFlow pipeline when it is interrupted in the middle by sending a
131 packet to the controller. This is not as simple as doing a new flow
132 table lookup and resuming from that point. Instead, it is equivalent
133 to the (very complicated) recirculation logic in ofproto-dpif-xlate.c.
134 Much of this logic can be translated into OpenFlow actions (e.g. the
135 call stack and data stack), but some of it is entirely outside
136 OpenFlow (e.g. the state of mirrors). To implement it properly, it
137 seems that we'll have to introduce a new Open vSwitch extension to
138 OpenFlow, a "send-to-controller" action that causes extra data to be
139 sent to the controller, where the extra data packages up the state
140 necessary to resume the pipeline. Maybe the bits of the state that
141 can be represented in OpenFlow can be embedded in this extra data in a
142 controller-readable form, but other bits we might want to be opaque.
143 It's also likely that we'll want to change and extend the form of this
144 opaque data over time, so this should be allowed for, e.g. by
145 including a nonce in the extra data that is newly generated every time
148 *** OpenFlow action definitions
150 Define OpenFlow wire structures for each new OpenFlow action and
151 implement them in lib/ofp-actions.[ch].
153 *** OVS implementation
155 Add code for action translation. Possibly add datapath code for
156 action implementation. However, none of these new actions should
157 require high-bandwidth processing so we could at least start with them
158 implemented in userspace only. (ARP field modification is already
159 userspace-only and no one has complained yet.)
171 Somehow it has to be possible for an L3 logical router to map from an
172 IP address to an Ethernet address. This can happen statically or
173 dynamically. Probably both cases need to be supported eventually.
175 *** Dynamic IP to MAC bindings
177 Some bindings from IP address to MAC will undoubtedly need to be
178 discovered dynamically through ARP requests. It's straightforward
179 enough for a logical L3 router to generate ARP requests and forward
180 them to the appropriate switch.
182 It's more difficult to figure out where the reply should be processed
183 and stored. It might seem at first that a first-cut implementation
184 could just keep track of the binding on the hypervisor that needs to
185 know, but that can't happen easily because the VM that sends the reply
186 might not be on the same HV as the VM that needs the answer (that is,
187 the VM that sent the packet that needs the binding to be resolved) and
188 there isn't an easy way for it to know which HV needs the answer.
190 Thus, the HV that processes the ARP reply (which is unknown when the
191 ARP is sent) has to tell all the HVs the binding. The most obvious
192 place for this in the OVN_Southbound database.
194 Details need to be worked out, including:
196 **** OVN_Southbound schema changes.
198 Possibly bindings could be added to the Port_Binding table by adding
199 or modifying columns. Another possibility is that another table
202 **** Logical_Flow representation
204 It would be really nice to maintain the general-purpose nature of
205 logical flows, but these bindings might have to include some
206 hard-coded special cases, especially when it comes to the relationship
207 with populating the bindings into the OVN_Southbound table.
209 **** Tracking queries
211 It's probably best to only record in the database responses to queries
212 actually issued by an L3 logical router, so somehow they have to be
213 tracked, probably by putting a tentative binding without a MAC address
216 **** Renewal and expiration.
218 Something needs to make sure that bindings remain valid and expire
219 those that become stale.
221 *** MTU handling (fragmentation on output)
227 *** ICMP error generation, TCP reset, UDP unreachable, protocol unreachable, ...
229 As a point of comparison, Linux doesn't ratelimit TCP resets but I
230 think it does everything else.
234 ** ovn-controller parameters and configuration.
236 *** SSL configuration.
238 Can probably get this from Open_vSwitch database.
242 *** Limiting the impact of a compromised chassis.
244 Every instance of ovn-controller has the same full access to the central
245 OVN_Southbound database. This means that a compromised chassis can
246 interfere with the normal operation of the rest of the deployment. Some
247 specific examples include writing to the logical flow table to alter
248 traffic handling or updating the port binding table to claim ports that are
249 actually present on a different chassis. In practice, the compromised host
250 would be fighting against ovn-northd and other instances of ovn-controller
251 that would be trying to restore the correct state. The impact could include
252 at least temporarily redirecting traffic (so the compromised host could
253 receive traffic that it shouldn't) and potentially a more general denial of
256 There are different potential improvements to this area. The first would be
257 to add some sort of ACL scheme to ovsdb-server. A proposal for this should
258 first include an ACL scheme for ovn-controller. An example policy would
259 be to make Logical_Flow read-only. Table-level control is needed, but is
260 not enough. For example, ovn-controller must be able to update the Chassis
261 and Encap tables, but should only be able to modify the rows associated with
262 that chassis and no others.
264 A more complex example is the Port_Binding table. Currently, ovn-controller
265 is the source of truth of where a port is located. There seems to be no
266 policy that can prevent malicious behavior of a compromised host with this
269 An alternative scheme for port bindings would be to provide an optional mode
270 where an external entity controls port bindings and make them read-only to
271 ovn-controller. This is actually how OpenStack works today, for example.
272 The part of OpenStack that manages VMs (Nova) tells the networking component
273 (Neutron) where a port will be located, as opposed to the networking
274 component discovering it.
278 ovsdb-server should have adequate features for OVN but it probably
279 needs work for scale and possibly for availability as deployments
280 grow. Here are some thoughts.
282 Andy Zhou is looking at these issues.
284 *** Reducing amount of data sent to clients.
286 Currently, whenever a row monitored by a client changes,
287 ovsdb-server sends the client every monitored column in the row,
288 even if only one column changes. It might be valuable to reduce
289 this only to the columns that changes.
291 Also, whenever a column changes, ovsdb-server sends the entire
292 contents of the column. It might be valuable, for columns that
293 are sets or maps, to send only added or removed values or
296 Currently, clients monitor the entire contents of a table. It
297 might make sense to allow clients to monitor only rows that
298 satisfy specific criteria, e.g. to allow an ovn-controller to
299 receive only Logical_Flow rows for logical networks on its hypervisor.
301 *** Reducing redundant data and code within ovsdb-server.
303 Currently, ovsdb-server separately composes database update
304 information to send to each of its clients. This is fine for a
305 small number of clients, but it wastes time and memory when
306 hundreds of clients all want the same updates (as will be in the
309 (This is somewhat opposed to the idea of letting a client monitor
310 only some rows in a table, since that would increase the diversity
315 If it turns out that other changes don't let ovsdb-server scale
316 adequately, we can multithread ovsdb-server. Initially one might
317 only break protocol handling into separate threads, leaving the
318 actual database work serialized through a lock.
320 ** Increasing availability.
322 Database availability might become an issue. The OVN system
323 shouldn't grind to a halt if the database becomes unavailable, but
324 it would become impossible to bring VIFs up or down, etc.
326 My current thought on how to increase availability is to add
327 clustering to ovsdb-server, probably via the Raft consensus
328 algorithm. As an experiment, I wrote an implementation of Raft
329 for Open vSwitch that you can clone from:
331 https://github.com/blp/ovs-reviews.git raft
333 ** Reducing startup time.
335 As-is, if ovsdb-server restarts, every client will fetch a fresh
336 copy of the part of the database that it cares about. With
337 hundreds of clients, this could cause heavy CPU load on
338 ovsdb-server and use excessive network bandwidth. It would be
339 better to allow incremental updates even across connection loss.
340 One way might be to use "Difference Digests" as described in
341 Epstein et al., "What's the Difference? Efficient Set
342 Reconciliation Without Prior Context". (I'm not yet aware of
343 previous non-academic use of this technique.)
345 ** Support multiple tunnel encapsulations in Chassis.
347 So far, both ovn-controller and ovn-controller-vtep only allow
348 chassis to have one tunnel encapsulation entry. We should extend
349 the implementation to support multiple tunnel encapsulations.
351 ** Update learned MAC addresses from VTEP to OVN
353 The VTEP gateway stores all MAC addresses learned from its
354 physical interfaces in the 'Ucast_Macs_Local' and the
355 'Mcast_Macs_Local' tables. ovn-controller-vtep should be
356 able to update that information back to ovn-sb database,
357 so that other chassis know where to send packets destined
358 to the extended external network instead of broadcasting.
360 ** Translate ovn-sb Multicast_Group table into VTEP config
362 The ovn-controller-vtep daemon should be able to translate
363 the Multicast_Group table entry in ovn-sb database into
364 Mcast_Macs_Remote table configuration in VTEP database.
366 * Use BFD as tunnel monitor.
368 Both ovn-controller and ovn-contorller-vtep should use BFD to
369 monitor the tunnel liveness. Both ovs-vswitchd schema and
370 VTEP schema supports BFD.
376 ** Support reject action.
378 ** Support log option.