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