1 Design Decisions In Open vSwitch
2 ================================
4 This document describes design decisions that went into implementing
5 Open vSwitch. While we believe these to be reasonable decisions, it is
6 impossible to predict how Open vSwitch will be used in all environments.
7 Understanding assumptions made by Open vSwitch is critical to a
8 successful deployment. The end of this document contains contact
9 information that can be used to let us know how we can make Open vSwitch
10 more generally useful.
15 The OpenFlow 1.0 specification requires the output port of the OFPAT_ENQUEUE
16 action to "refer to a valid physical port (i.e. < OFPP_MAX) or OFPP_IN_PORT".
17 Although OFPP_LOCAL is not less than OFPP_MAX, it is an 'internal' port which
18 can have QoS applied to it in Linux. Since we allow the OFPAT_ENQUEUE to apply
19 to 'internal' ports whose port numbers are less than OFPP_MAX, we interpret
20 OFPP_LOCAL as a physical port and support OFPAT_ENQUEUE on it as well.
26 The OpenFlow 1.0 specification for the behavior of OFPT_FLOW_MOD is
27 confusing. The following table summarizes the Open vSwitch
28 implementation of its behavior in the following categories:
30 - "match on priority": Whether the flow_mod acts only on flows
31 whose priority matches that included in the flow_mod message.
33 - "match on out_port": Whether the flow_mod acts only on flows
34 that output to the out_port included in the flow_mod message (if
35 out_port is not OFPP_NONE).
37 - "updates flow_cookie": Whether the flow_mod changes the
38 flow_cookie of the flow or flows that it matches to the
39 flow_cookie included in the flow_mod message.
41 - "updates OFPFF_ flags": Whether the flow_mod changes the
42 OFPFF_SEND_FLOW_REM flag of the flow or flows that it matches to
43 the setting included in the flags of the flow_mod message.
45 - "honors OFPFF_CHECK_OVERLAP": Whether the OFPFF_CHECK_OVERLAP
46 flag in the flow_mod is significant.
48 - "updates idle_timeout" and "updates hard_timeout": Whether the
49 idle_timeout and hard_timeout in the flow_mod, respectively,
50 have an effect on the flow or flows matched by the flow_mod.
52 - "updates idle timer": Whether the flow_mod resets the per-flow
53 timer that measures how long a flow has been idle.
55 - "updates hard timer": Whether the flow_mod resets the per-flow
56 timer that measures how long it has been since a flow was
59 - "zeros counters": Whether the flow_mod resets per-flow packet
60 and byte counters to zero.
62 - "sends flow_removed message": Whether the flow_mod generates a
63 flow_removed message for the flow or flows that it affects.
65 An entry labeled "yes" means that the flow mod type does have the
66 indicated behavior, "---" means that it does not, an empty cell means
67 that the property is not applicable, and other values are explained
71 ADD MODIFY STRICT DELETE STRICT
72 === ====== ====== ====== ======
73 match on priority --- --- yes --- yes
74 match on out_port --- --- --- yes yes
75 updates flow_cookie yes yes yes
76 updates OFPFF_SEND_FLOW_REM yes + +
77 honors OFPFF_CHECK_OVERLAP yes + +
78 updates idle_timeout yes + +
79 updates hard_timeout yes + +
80 resets idle timer yes + +
81 resets hard timer yes yes yes
82 zeros counters yes + +
83 sends flow_removed message --- --- --- % %
85 (+) "modify" and "modify-strict" only take these actions when they
86 create a new flow, not when they update an existing flow.
88 (%) "delete" and "delete_strict" generates a flow_removed message if
89 the deleted flow or flows have the OFPFF_SEND_FLOW_REM flag set.
90 (Each controller can separately control whether it wants to
91 receive the generated messages.)
94 Multiple Table Support
95 ======================
97 OpenFlow 1.0 has only rudimentary support for multiple flow tables.
98 Notably, OpenFlow 1.0 does not allow the controller to specify the
99 flow table to which a flow is to be added. Open vSwitch adds an
100 extension for this purpose, which is enabled on a per-OpenFlow
101 connection basis using the NXT_FLOW_MOD_TABLE_ID message. When the
102 extension is enabled, the upper 8 bits of the 'command' member in an
103 OFPT_FLOW_MOD or NXT_FLOW_MOD message designates the table to which a
106 The Open vSwitch software switch implementation offers 255 flow
107 tables. On packet ingress, only the first flow table (table 0) is
108 searched, and the contents of the remaining tables are not considered
109 in any way. Tables other than table 0 only come into play when an
110 NXAST_RESUBMIT_TABLE action specifies another table to search.
112 Tables 128 and above are reserved for use by the switch itself.
113 Controllers should use only tables 0 through 127.
119 Open vSwitch supports stateless handling of IPv6 packets. Flows can be
120 written to support matching TCP, UDP, and ICMPv6 headers within an IPv6
121 packet. Deeper matching of some Neighbor Discovery messages is also
124 IPv6 was not designed to interact well with middle-boxes. This,
125 combined with Open vSwitch's stateless nature, have affected the
126 processing of IPv6 traffic, which is detailed below.
131 The base IPv6 header is incredibly simple with the intention of only
132 containing information relevant for routing packets between two
133 endpoints. IPv6 relies heavily on the use of extension headers to
134 provide any other functionality. Unfortunately, the extension headers
135 were designed in such a way that it is impossible to move to the next
136 header (including the layer-4 payload) unless the current header is
139 Open vSwitch will process the following extension headers and continue
142 * Fragment (see the next section)
143 * AH (Authentication Header)
146 * Destination Options
148 When a header is encountered that is not in that list, it is considered
149 "terminal". A terminal header's IPv6 protocol value is stored in
150 "nw_proto" for matching purposes. If a terminal header is TCP, UDP, or
151 ICMPv6, the packet will be further processed in an attempt to extract
157 IPv6 requires that every link in the internet have an MTU of 1280 octets
158 or greater (RFC 2460). As such, a terminal header (as described above in
159 "Extension Headers") in the first fragment should generally be
160 reachable. In this case, the terminal header's IPv6 protocol type is
161 stored in the "nw_proto" field for matching purposes. If a terminal
162 header cannot be found in the first fragment (one with a fragment offset
163 of zero), the "nw_proto" field is set to 0. Subsequent fragments (those
164 with a non-zero fragment offset) have the "nw_proto" field set to the
165 IPv6 protocol type for fragments (44).
170 An IPv6 jumbogram (RFC 2675) is a packet containing a payload longer
171 than 65,535 octets. A jumbogram is only relevant in subnets with a link
172 MTU greater than 65,575 octets, and are not required to be supported on
173 nodes that do not connect to link with such large MTUs. Currently, Open
174 vSwitch doesn't process jumbograms.
180 In-band control allows a single network to be used for OpenFlow traffic and
181 other data traffic. See ovs-vswitchd.conf.db(5) for a description of
182 configuring in-band control.
184 This comment is an attempt to describe how in-band control works at a
185 wire- and implementation-level. Correctly implementing in-band
186 control has proven difficult due to its many subtleties, and has thus
187 gone through many iterations. Please read through and understand the
188 reasoning behind the chosen rules before making modifications.
190 In Open vSwitch, in-band control is implemented as "hidden" flows (in that
191 they are not visible through OpenFlow) and at a higher priority than
192 wildcarded flows can be set up by through OpenFlow. This is done so that
193 the OpenFlow controller cannot interfere with them and possibly break
194 connectivity with its switches. It is possible to see all flows, including
195 in-band ones, with the ovs-appctl "bridge/dump-flows" command.
197 The Open vSwitch implementation of in-band control can hide traffic to
198 arbitrary "remotes", where each remote is one TCP port on one IP address.
199 Currently the remotes are automatically configured as the in-band OpenFlow
200 controllers plus the OVSDB managers, if any. (The latter is a requirement
201 because OVSDB managers are responsible for configuring OpenFlow controllers,
202 so if the manager cannot be reached then OpenFlow cannot be reconfigured.)
204 The following rules (with the OFPP_NORMAL action) are set up on any bridge
205 that has any remotes:
207 (a) DHCP requests sent from the local port.
208 (b) ARP replies to the local port's MAC address.
209 (c) ARP requests from the local port's MAC address.
211 In-band also sets up the following rules for each unique next-hop MAC
212 address for the remotes' IPs (the "next hop" is either the remote
213 itself, if it is on a local subnet, or the gateway to reach the remote):
215 (d) ARP replies to the next hop's MAC address.
216 (e) ARP requests from the next hop's MAC address.
218 In-band also sets up the following rules for each unique remote IP address:
220 (f) ARP replies containing the remote's IP address as a target.
221 (g) ARP requests containing the remote's IP address as a source.
223 In-band also sets up the following rules for each unique remote (IP,port)
226 (h) TCP traffic to the remote's IP and port.
227 (i) TCP traffic from the remote's IP and port.
229 The goal of these rules is to be as narrow as possible to allow a
230 switch to join a network and be able to communicate with the
231 remotes. As mentioned earlier, these rules have higher priority
232 than the controller's rules, so if they are too broad, they may
233 prevent the controller from implementing its policy. As such,
234 in-band actively monitors some aspects of flow and packet processing
235 so that the rules can be made more precise.
237 In-band control monitors attempts to add flows into the datapath that
238 could interfere with its duties. The datapath only allows exact
239 match entries, so in-band control is able to be very precise about
240 the flows it prevents. Flows that miss in the datapath are sent to
241 userspace to be processed, so preventing these flows from being
242 cached in the "fast path" does not affect correctness. The only type
243 of flow that is currently prevented is one that would prevent DHCP
244 replies from being seen by the local port. For example, a rule that
245 forwarded all DHCP traffic to the controller would not be allowed,
246 but one that forwarded to all ports (including the local port) would.
248 As mentioned earlier, packets that miss in the datapath are sent to
249 the userspace for processing. The userspace has its own flow table,
250 the "classifier", so in-band checks whether any special processing
251 is needed before the classifier is consulted. If a packet is a DHCP
252 response to a request from the local port, the packet is forwarded to
253 the local port, regardless of the flow table. Note that this requires
254 L7 processing of DHCP replies to determine whether the 'chaddr' field
255 matches the MAC address of the local port.
257 It is interesting to note that for an L3-based in-band control
258 mechanism, the majority of rules are devoted to ARP traffic. At first
259 glance, some of these rules appear redundant. However, each serves an
260 important role. First, in order to determine the MAC address of the
261 remote side (controller or gateway) for other ARP rules, we must allow
262 ARP traffic for our local port with rules (b) and (c). If we are
263 between a switch and its connection to the remote, we have to
264 allow the other switch's ARP traffic to through. This is done with
265 rules (d) and (e), since we do not know the addresses of the other
266 switches a priori, but do know the remote's or gateway's. Finally,
267 if the remote is running in a local guest VM that is not reached
268 through the local port, the switch that is connected to the VM must
269 allow ARP traffic based on the remote's IP address, since it will
270 not know the MAC address of the local port that is sending the traffic
271 or the MAC address of the remote in the guest VM.
273 With a few notable exceptions below, in-band should work in most
274 network setups. The following are considered "supported' in the
275 current implementation:
277 - Locally Connected. The switch and remote are on the same
278 subnet. This uses rules (a), (b), (c), (h), and (i).
280 - Reached through Gateway. The switch and remote are on
281 different subnets and must go through a gateway. This uses
282 rules (a), (b), (c), (h), and (i).
284 - Between Switch and Remote. This switch is between another
285 switch and the remote, and we want to allow the other
286 switch's traffic through. This uses rules (d), (e), (h), and
287 (i). It uses (b) and (c) indirectly in order to know the MAC
288 address for rules (d) and (e). Note that DHCP for the other
289 switch will not work unless an OpenFlow controller explicitly lets this
290 switch pass the traffic.
292 - Between Switch and Gateway. This switch is between another
293 switch and the gateway, and we want to allow the other switch's
294 traffic through. This uses the same rules and logic as the
295 "Between Switch and Remote" configuration described earlier.
297 - Remote on Local VM. The remote is a guest VM on the
298 system running in-band control. This uses rules (a), (b), (c),
301 - Remote on Local VM with Different Networks. The remote
302 is a guest VM on the system running in-band control, but the
303 local port is not used to connect to the remote. For
304 example, an IP address is configured on eth0 of the switch. The
305 remote's VM is connected through eth1 of the switch, but an
306 IP address has not been configured for that port on the switch.
307 As such, the switch will use eth0 to connect to the remote,
308 and eth1's rules about the local port will not work. In the
309 example, the switch attached to eth0 would use rules (a), (b),
310 (c), (h), and (i) on eth0. The switch attached to eth1 would use
311 rules (f), (g), (h), and (i).
313 The following are explicitly *not* supported by in-band control:
315 - Specify Remote by Name. Currently, the remote must be
316 identified by IP address. A naive approach would be to permit
317 all DNS traffic. Unfortunately, this would prevent the
318 controller from defining any policy over DNS. Since switches
319 that are located behind us need to connect to the remote,
320 in-band cannot simply add a rule that allows DNS traffic from
321 the local port. The "correct" way to support this is to parse
322 DNS requests to allow all traffic related to a request for the
323 remote's name through. Due to the potential security
324 problems and amount of processing, we decided to hold off for
327 - Differing Remotes for Switches. All switches must know
328 the L3 addresses for all the remotes that other switches
329 may use, since rules need to be set up to allow traffic related
330 to those remotes through. See rules (f), (g), (h), and (i).
332 - Differing Routes for Switches. In order for the switch to
333 allow other switches to connect to a remote through a
334 gateway, it allows the gateway's traffic through with rules (d)
335 and (e). If the routes to the remote differ for the two
336 switches, we will not know the MAC address of the alternate
343 Suggestions to improve Open vSwitch are welcome at discuss@openvswitch.org.