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.
23 Multiple Table Support
24 ======================
26 OpenFlow 1.0 has only rudimentary support for multiple flow tables.
27 Notably, OpenFlow 1.0 does not allow the controller to specify the
28 flow table to which a flow is to be added. Open vSwitch adds an
29 extension for this purpose, which is enabled on a per-OpenFlow
30 connection basis using the NXT_FLOW_MOD_TABLE_ID message. When the
31 extension is enabled, the upper 8 bits of the 'command' member in an
32 OFPT_FLOW_MOD or NXT_FLOW_MOD message designates the table to which a
35 The Open vSwitch software switch implementation offers 255 flow
36 tables. On packet ingress, only the first flow table (table 0) is
37 searched, and the contents of the remaining tables are not considered
38 in any way. Tables other than table 0 only come into play when an
39 NXAST_RESUBMIT_TABLE action specifies another table to search.
41 Tables 128 and above are reserved for use by the switch itself.
42 Controllers should use only tables 0 through 127.
48 Open vSwitch supports stateless handling of IPv6 packets. Flows can be
49 written to support matching TCP, UDP, and ICMPv6 headers within an IPv6
50 packet. Deeper matching of some Neighbor Discovery messages is also
53 IPv6 was not designed to interact well with middle-boxes. This,
54 combined with Open vSwitch's stateless nature, have affected the
55 processing of IPv6 traffic, which is detailed below.
60 The base IPv6 header is incredibly simple with the intention of only
61 containing information relevant for routing packets between two
62 endpoints. IPv6 relies heavily on the use of extension headers to
63 provide any other functionality. Unfortunately, the extension headers
64 were designed in such a way that it is impossible to move to the next
65 header (including the layer-4 payload) unless the current header is
68 Open vSwitch will process the following extension headers and continue
71 * Fragment (see the next section)
72 * AH (Authentication Header)
77 When a header is encountered that is not in that list, it is considered
78 "terminal". A terminal header's IPv6 protocol value is stored in
79 "nw_proto" for matching purposes. If a terminal header is TCP, UDP, or
80 ICMPv6, the packet will be further processed in an attempt to extract
86 IPv6 requires that every link in the internet have an MTU of 1280 octets
87 or greater (RFC 2460). As such, a terminal header (as described above in
88 "Extension Headers") in the first fragment should generally be
89 reachable. In this case, the terminal header's IPv6 protocol type is
90 stored in the "nw_proto" field for matching purposes. If a terminal
91 header cannot be found in the first fragment (one with a fragment offset
92 of zero), the "nw_proto" field is set to 0. Subsequent fragments (those
93 with a non-zero fragment offset) have the "nw_proto" field set to the
94 IPv6 protocol type for fragments (44).
99 An IPv6 jumbogram (RFC 2675) is a packet containing a payload longer
100 than 65,535 octets. A jumbogram is only relevant in subnets with a link
101 MTU greater than 65,575 octets, and are not required to be supported on
102 nodes that do not connect to link with such large MTUs. Currently, Open
103 vSwitch doesn't process jumbograms.
109 In-band control allows a single network to be used for OpenFlow traffic and
110 other data traffic. See ovs-vswitchd.conf.db(5) for a description of
111 configuring in-band control.
113 This comment is an attempt to describe how in-band control works at a
114 wire- and implementation-level. Correctly implementing in-band
115 control has proven difficult due to its many subtleties, and has thus
116 gone through many iterations. Please read through and understand the
117 reasoning behind the chosen rules before making modifications.
119 In Open vSwitch, in-band control is implemented as "hidden" flows (in that
120 they are not visible through OpenFlow) and at a higher priority than
121 wildcarded flows can be set up by through OpenFlow. This is done so that
122 the OpenFlow controller cannot interfere with them and possibly break
123 connectivity with its switches. It is possible to see all flows, including
124 in-band ones, with the ovs-appctl "bridge/dump-flows" command.
126 The Open vSwitch implementation of in-band control can hide traffic to
127 arbitrary "remotes", where each remote is one TCP port on one IP address.
128 Currently the remotes are automatically configured as the in-band OpenFlow
129 controllers plus the OVSDB managers, if any. (The latter is a requirement
130 because OVSDB managers are responsible for configuring OpenFlow controllers,
131 so if the manager cannot be reached then OpenFlow cannot be reconfigured.)
133 The following rules (with the OFPP_NORMAL action) are set up on any bridge
134 that has any remotes:
136 (a) DHCP requests sent from the local port.
137 (b) ARP replies to the local port's MAC address.
138 (c) ARP requests from the local port's MAC address.
140 In-band also sets up the following rules for each unique next-hop MAC
141 address for the remotes' IPs (the "next hop" is either the remote
142 itself, if it is on a local subnet, or the gateway to reach the remote):
144 (d) ARP replies to the next hop's MAC address.
145 (e) ARP requests from the next hop's MAC address.
147 In-band also sets up the following rules for each unique remote IP address:
149 (f) ARP replies containing the remote's IP address as a target.
150 (g) ARP requests containing the remote's IP address as a source.
152 In-band also sets up the following rules for each unique remote (IP,port)
155 (h) TCP traffic to the remote's IP and port.
156 (i) TCP traffic from the remote's IP and port.
158 The goal of these rules is to be as narrow as possible to allow a
159 switch to join a network and be able to communicate with the
160 remotes. As mentioned earlier, these rules have higher priority
161 than the controller's rules, so if they are too broad, they may
162 prevent the controller from implementing its policy. As such,
163 in-band actively monitors some aspects of flow and packet processing
164 so that the rules can be made more precise.
166 In-band control monitors attempts to add flows into the datapath that
167 could interfere with its duties. The datapath only allows exact
168 match entries, so in-band control is able to be very precise about
169 the flows it prevents. Flows that miss in the datapath are sent to
170 userspace to be processed, so preventing these flows from being
171 cached in the "fast path" does not affect correctness. The only type
172 of flow that is currently prevented is one that would prevent DHCP
173 replies from being seen by the local port. For example, a rule that
174 forwarded all DHCP traffic to the controller would not be allowed,
175 but one that forwarded to all ports (including the local port) would.
177 As mentioned earlier, packets that miss in the datapath are sent to
178 the userspace for processing. The userspace has its own flow table,
179 the "classifier", so in-band checks whether any special processing
180 is needed before the classifier is consulted. If a packet is a DHCP
181 response to a request from the local port, the packet is forwarded to
182 the local port, regardless of the flow table. Note that this requires
183 L7 processing of DHCP replies to determine whether the 'chaddr' field
184 matches the MAC address of the local port.
186 It is interesting to note that for an L3-based in-band control
187 mechanism, the majority of rules are devoted to ARP traffic. At first
188 glance, some of these rules appear redundant. However, each serves an
189 important role. First, in order to determine the MAC address of the
190 remote side (controller or gateway) for other ARP rules, we must allow
191 ARP traffic for our local port with rules (b) and (c). If we are
192 between a switch and its connection to the remote, we have to
193 allow the other switch's ARP traffic to through. This is done with
194 rules (d) and (e), since we do not know the addresses of the other
195 switches a priori, but do know the remote's or gateway's. Finally,
196 if the remote is running in a local guest VM that is not reached
197 through the local port, the switch that is connected to the VM must
198 allow ARP traffic based on the remote's IP address, since it will
199 not know the MAC address of the local port that is sending the traffic
200 or the MAC address of the remote in the guest VM.
202 With a few notable exceptions below, in-band should work in most
203 network setups. The following are considered "supported' in the
204 current implementation:
206 - Locally Connected. The switch and remote are on the same
207 subnet. This uses rules (a), (b), (c), (h), and (i).
209 - Reached through Gateway. The switch and remote are on
210 different subnets and must go through a gateway. This uses
211 rules (a), (b), (c), (h), and (i).
213 - Between Switch and Remote. This switch is between another
214 switch and the remote, and we want to allow the other
215 switch's traffic through. This uses rules (d), (e), (h), and
216 (i). It uses (b) and (c) indirectly in order to know the MAC
217 address for rules (d) and (e). Note that DHCP for the other
218 switch will not work unless an OpenFlow controller explicitly lets this
219 switch pass the traffic.
221 - Between Switch and Gateway. This switch is between another
222 switch and the gateway, and we want to allow the other switch's
223 traffic through. This uses the same rules and logic as the
224 "Between Switch and Remote" configuration described earlier.
226 - Remote on Local VM. The remote is a guest VM on the
227 system running in-band control. This uses rules (a), (b), (c),
230 - Remote on Local VM with Different Networks. The remote
231 is a guest VM on the system running in-band control, but the
232 local port is not used to connect to the remote. For
233 example, an IP address is configured on eth0 of the switch. The
234 remote's VM is connected through eth1 of the switch, but an
235 IP address has not been configured for that port on the switch.
236 As such, the switch will use eth0 to connect to the remote,
237 and eth1's rules about the local port will not work. In the
238 example, the switch attached to eth0 would use rules (a), (b),
239 (c), (h), and (i) on eth0. The switch attached to eth1 would use
240 rules (f), (g), (h), and (i).
242 The following are explicitly *not* supported by in-band control:
244 - Specify Remote by Name. Currently, the remote must be
245 identified by IP address. A naive approach would be to permit
246 all DNS traffic. Unfortunately, this would prevent the
247 controller from defining any policy over DNS. Since switches
248 that are located behind us need to connect to the remote,
249 in-band cannot simply add a rule that allows DNS traffic from
250 the local port. The "correct" way to support this is to parse
251 DNS requests to allow all traffic related to a request for the
252 remote's name through. Due to the potential security
253 problems and amount of processing, we decided to hold off for
256 - Differing Remotes for Switches. All switches must know
257 the L3 addresses for all the remotes that other switches
258 may use, since rules need to be set up to allow traffic related
259 to those remotes through. See rules (f), (g), (h), and (i).
261 - Differing Routes for Switches. In order for the switch to
262 allow other switches to connect to a remote through a
263 gateway, it allows the gateway's traffic through with rules (d)
264 and (e). If the routes to the remote differ for the two
265 switches, we will not know the MAC address of the alternate
272 Suggestions to improve Open vSwitch are welcome at discuss@openvswitch.org.
projects / openvswitch / blob