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 Over time, Open vSwitch has added many knobs that control whether a
16 given controller receives OpenFlow asynchronous messages. This
17 section describes how all of these features interact.
19 First, a service controller never receives any asynchronous messages
20 unless it changes its miss_send_len from the service controller
21 default of zero in one of the following ways:
23 - Sending an OFPT_SET_CONFIG message with nonzero miss_send_len.
25 - Sending any NXT_SET_ASYNC_CONFIG message: as a side effect, this
26 message changes the miss_send_len to
27 OFP_DEFAULT_MISS_SEND_LEN (128) for service controllers.
29 Second, OFPT_FLOW_REMOVED and NXT_FLOW_REMOVED messages are generated
30 only if the flow that was removed had the OFPFF_SEND_FLOW_REM flag
33 Third, OFPT_PACKET_IN and NXT_PACKET_IN messages are sent only to
34 OpenFlow controller connections that have the correct connection ID
35 (see "struct nx_controller_id" and "struct nx_action_controller"):
37 - For packet-in messages generated by a NXAST_CONTROLLER action,
38 the controller ID specified in the action.
40 - For other packet-in messages, controller ID zero. (This is the
41 default ID when an OpenFlow controller does not configure one.)
43 Finally, Open vSwitch consults a per-connection table indexed by the
44 message type, reason code, and current role. The following table
45 shows how this table is initialized by default when an OpenFlow
46 connection is made. An entry labeled "yes" means that the message is
47 sent, an entry labeled "---" means that the message is suppressed.
51 message and reason code other slave
52 ---------------------------------------- ------- -----
53 OFPT_PACKET_IN / NXT_PACKET_IN
56 OFPR_INVALID_TTL --- ---
57 OFPR_GROUP (OF1.4+) yes ---
59 OFPT_FLOW_REMOVED / NXT_FLOW_REMOVED
60 OFPRR_IDLE_TIMEOUT yes ---
61 OFPRR_HARD_TIMEOUT yes ---
70 The NXT_SET_ASYNC_CONFIG message directly sets all of the values in
71 this table for the current connection. The
72 OFPC_INVALID_TTL_TO_CONTROLLER bit in the OFPT_SET_CONFIG message
73 controls the setting for OFPR_INVALID_TTL for the "master" role.
79 The OpenFlow 1.0 specification requires the output port of the OFPAT_ENQUEUE
80 action to "refer to a valid physical port (i.e. < OFPP_MAX) or OFPP_IN_PORT".
81 Although OFPP_LOCAL is not less than OFPP_MAX, it is an 'internal' port which
82 can have QoS applied to it in Linux. Since we allow the OFPAT_ENQUEUE to apply
83 to 'internal' ports whose port numbers are less than OFPP_MAX, we interpret
84 OFPP_LOCAL as a physical port and support OFPAT_ENQUEUE on it as well.
90 The OpenFlow specification for the behavior of OFPT_FLOW_MOD is
91 confusing. The following tables summarize the Open vSwitch
92 implementation of its behavior in the following categories:
94 - "match on priority": Whether the flow_mod acts only on flows
95 whose priority matches that included in the flow_mod message.
97 - "match on out_port": Whether the flow_mod acts only on flows
98 that output to the out_port included in the flow_mod message (if
99 out_port is not OFPP_NONE). OpenFlow 1.1 and later have a
100 similar feature (not listed separately here) for out_group.
102 - "match on flow_cookie": Whether the flow_mod acts only on flows
103 whose flow_cookie matches an optional controller-specified value
106 - "updates flow_cookie": Whether the flow_mod changes the
107 flow_cookie of the flow or flows that it matches to the
108 flow_cookie included in the flow_mod message.
110 - "updates OFPFF_ flags": Whether the flow_mod changes the
111 OFPFF_SEND_FLOW_REM flag of the flow or flows that it matches to
112 the setting included in the flags of the flow_mod message.
114 - "honors OFPFF_CHECK_OVERLAP": Whether the OFPFF_CHECK_OVERLAP
115 flag in the flow_mod is significant.
117 - "updates idle_timeout" and "updates hard_timeout": Whether the
118 idle_timeout and hard_timeout in the flow_mod, respectively,
119 have an effect on the flow or flows matched by the flow_mod.
121 - "updates idle timer": Whether the flow_mod resets the per-flow
122 timer that measures how long a flow has been idle.
124 - "updates hard timer": Whether the flow_mod resets the per-flow
125 timer that measures how long it has been since a flow was
128 - "zeros counters": Whether the flow_mod resets per-flow packet
129 and byte counters to zero.
131 - "may add a new flow": Whether the flow_mod may add a new flow to
132 the flow table. (Obviously this is always true for "add"
133 commands but in some OpenFlow versions "modify" and
134 "modify-strict" can also add new flows.)
136 - "sends flow_removed message": Whether the flow_mod generates a
137 flow_removed message for the flow or flows that it affects.
139 An entry labeled "yes" means that the flow mod type does have the
140 indicated behavior, "---" means that it does not, an empty cell means
141 that the property is not applicable, and other values are explained
149 ADD MODIFY STRICT DELETE STRICT
150 === ====== ====== ====== ======
151 match on priority yes --- yes --- yes
152 match on out_port --- --- --- yes yes
153 match on flow_cookie --- --- --- --- ---
154 match on table_id --- --- --- --- ---
155 controller chooses table_id --- --- ---
156 updates flow_cookie yes yes yes
157 updates OFPFF_SEND_FLOW_REM yes + +
158 honors OFPFF_CHECK_OVERLAP yes + +
159 updates idle_timeout yes + +
160 updates hard_timeout yes + +
161 resets idle timer yes + +
162 resets hard timer yes yes yes
163 zeros counters yes + +
164 may add a new flow yes yes yes
165 sends flow_removed message --- --- --- % %
167 (+) "modify" and "modify-strict" only take these actions when they
168 create a new flow, not when they update an existing flow.
170 (%) "delete" and "delete_strict" generates a flow_removed message if
171 the deleted flow or flows have the OFPFF_SEND_FLOW_REM flag set.
172 (Each controller can separately control whether it wants to
173 receive the generated messages.)
179 OpenFlow 1.1 makes these changes:
181 - The controller now must specify the table_id of the flow match
182 searched and into which a flow may be inserted. Behavior for a
183 table_id of 255 is undefined.
185 - A flow_mod, except an "add", can now match on the flow_cookie.
187 - When a flow_mod matches on the flow_cookie, "modify" and
188 "modify-strict" never insert a new flow.
192 ADD MODIFY STRICT DELETE STRICT
193 === ====== ====== ====== ======
194 match on priority yes --- yes --- yes
195 match on out_port --- --- --- yes yes
196 match on flow_cookie --- yes yes yes yes
197 match on table_id yes yes yes yes yes
198 controller chooses table_id yes yes yes
199 updates flow_cookie yes --- ---
200 updates OFPFF_SEND_FLOW_REM yes + +
201 honors OFPFF_CHECK_OVERLAP yes + +
202 updates idle_timeout yes + +
203 updates hard_timeout yes + +
204 resets idle timer yes + +
205 resets hard timer yes yes yes
206 zeros counters yes + +
207 may add a new flow yes # #
208 sends flow_removed message --- --- --- % %
210 (+) "modify" and "modify-strict" only take these actions when they
211 create a new flow, not when they update an existing flow.
213 (%) "delete" and "delete_strict" generates a flow_removed message if
214 the deleted flow or flows have the OFPFF_SEND_FLOW_REM flag set.
215 (Each controller can separately control whether it wants to
216 receive the generated messages.)
218 (#) "modify" and "modify-strict" only add a new flow if the flow_mod
219 does not match on any bits of the flow cookie
225 OpenFlow 1.2 makes these changes:
227 - Only "add" commands ever add flows, "modify" and "modify-strict"
230 - A new flag OFPFF_RESET_COUNTS now controls whether "modify" and
231 "modify-strict" reset counters, whereas previously they never
232 reset counters (except when they inserted a new flow).
236 ADD MODIFY STRICT DELETE STRICT
237 === ====== ====== ====== ======
238 match on priority yes --- yes --- yes
239 match on out_port --- --- --- yes yes
240 match on flow_cookie --- yes yes yes yes
241 match on table_id yes yes yes yes yes
242 controller chooses table_id yes yes yes
243 updates flow_cookie yes --- ---
244 updates OFPFF_SEND_FLOW_REM yes --- ---
245 honors OFPFF_CHECK_OVERLAP yes --- ---
246 updates idle_timeout yes --- ---
247 updates hard_timeout yes --- ---
248 resets idle timer yes --- ---
249 resets hard timer yes yes yes
250 zeros counters yes & &
251 may add a new flow yes --- ---
252 sends flow_removed message --- --- --- % %
254 (%) "delete" and "delete_strict" generates a flow_removed message if
255 the deleted flow or flows have the OFPFF_SEND_FLOW_REM flag set.
256 (Each controller can separately control whether it wants to
257 receive the generated messages.)
259 (&) "modify" and "modify-strict" reset counters if the
260 OFPFF_RESET_COUNTS flag is specified.
266 OpenFlow 1.3 makes these changes:
268 - Behavior for a table_id of 255 is now defined, for "delete" and
269 "delete-strict" commands, as meaning to delete from all tables.
270 A table_id of 255 is now explicitly invalid for other commands.
272 - New flags OFPFF_NO_PKT_COUNTS and OFPFF_NO_BYT_COUNTS for "add"
275 The table for 1.3 is the same as the one shown above for 1.2.
281 OpenFlow 1.4 adds the "importance" field to flow_mods, but it does not
282 explicitly specify which kinds of flow_mods set the importance.For
283 consistency, Open vSwitch uses the same rule for importance as for
284 idle_timeout and hard_timeout, that is, only an "ADD" flow_mod sets
285 the importance. (This issue has been filed with the ONF as EXT-496.)
290 The OpenFlow 1.1 specification for OFPT_PACKET_IN is confusing. The
291 definition in OF1.1 openflow.h is[*]:
294 /* Packet received on port (datapath -> controller). */
295 struct ofp_packet_in {
296 struct ofp_header header;
297 uint32_t buffer_id; /* ID assigned by datapath. */
298 uint32_t in_port; /* Port on which frame was received. */
299 uint32_t in_phy_port; /* Physical Port on which frame was received. */
300 uint16_t total_len; /* Full length of frame. */
301 uint8_t reason; /* Reason packet is being sent (one of OFPR_*) */
302 uint8_t table_id; /* ID of the table that was looked up */
303 uint8_t data[0]; /* Ethernet frame, halfway through 32-bit word,
304 so the IP header is 32-bit aligned. The
305 amount of data is inferred from the length
306 field in the header. Because of padding,
307 offsetof(struct ofp_packet_in, data) ==
308 sizeof(struct ofp_packet_in) - 2. */
310 OFP_ASSERT(sizeof(struct ofp_packet_in) == 24);
313 The confusing part is the comment on the data[] member. This comment
314 is a leftover from OF1.0 openflow.h, in which the comment was correct:
315 sizeof(struct ofp_packet_in) is 20 in OF1.0 and offsetof(struct
316 ofp_packet_in, data) is 18. When OF1.1 was written, the structure
317 members were changed but the comment was carelessly not updated, and
318 the comment became wrong: sizeof(struct ofp_packet_in) and
319 offsetof(struct ofp_packet_in, data) are both 24 in OF1.1.
321 That leaves the question of how to implement ofp_packet_in in OF1.1.
322 The OpenFlow reference implementation for OF1.1 does not include any
323 padding, that is, the first byte of the encapsulated frame immediately
324 follows the 'table_id' member without a gap. Open vSwitch therefore
325 implements it the same way for compatibility.
327 For an earlier discussion, please see the thread archived at:
328 https://mailman.stanford.edu/pipermail/openflow-discuss/2011-August/002604.html
330 [*] The quoted definition is directly from OF1.1. Definitions used
331 inside OVS omit the 8-byte ofp_header members, so the sizes in
332 this discussion are 8 bytes larger than those declared in OVS
339 The 802.1Q VLAN header causes more trouble than any other 4 bytes in
340 networking. More specifically, three versions of OpenFlow and Open
341 vSwitch have among them four different ways to match the contents and
342 presence of the VLAN header. The following table describes how each
345 Match NXM OF1.0 OF1.1 OF1.2
346 ----- --------- ----------- ----------- ------------
347 [1] 0000/0000 ????/1,??/? ????/1,??/? 0000/0000,--
348 [2] 0000/ffff ffff/0,??/? ffff/0,??/? 0000/ffff,--
349 [3] 1xxx/1fff 0xxx/0,??/1 0xxx/0,??/1 1xxx/ffff,--
350 [4] z000/f000 ????/1,0y/0 fffe/0,0y/0 1000/1000,0y
351 [5] zxxx/ffff 0xxx/0,0y/0 0xxx/0,0y/0 1xxx/ffff,0y
352 [6] 0000/0fff <none> <none> <none>
353 [7] 0000/f000 <none> <none> <none>
354 [8] 0000/efff <none> <none> <none>
355 [9] 1001/1001 <none> <none> 1001/1001,--
356 [10] 3000/3000 <none> <none> <none>
358 Each column is interpreted as follows.
360 - Match: See the list below.
362 - NXM: xxxx/yyyy means NXM_OF_VLAN_TCI_W with value xxxx and mask
363 yyyy. A mask of 0000 is equivalent to omitting
364 NXM_OF_VLAN_TCI(_W), a mask of ffff is equivalent to
367 - OF1.0 and OF1.1: wwww/x,yy/z means dl_vlan wwww, OFPFW_DL_VLAN
368 x, dl_vlan_pcp yy, and OFPFW_DL_VLAN_PCP z. ? means that the
369 given nibble is ignored (and conventionally 0 for wwww or yy,
370 conventionally 1 for x or z). <none> means that the given match
373 - OF1.2: xxxx/yyyy,zz means OXM_OF_VLAN_VID_W with value xxxx and
374 mask yyyy, and OXM_OF_VLAN_PCP (which is not maskable) with
375 value zz. A mask of 0000 is equivalent to omitting
376 OXM_OF_VLAN_VID(_W), a mask of ffff is equivalent to
377 OXM_OF_VLAN_VID. -- means that OXM_OF_VLAN_PCP is omitted.
378 <none> means that the given match is not supported.
382 [1] Matches any packet, that is, one without an 802.1Q header or with
383 an 802.1Q header with any TCI value.
385 [2] Matches only packets without an 802.1Q header.
387 NXM: Any match with (vlan_tci == 0) and (vlan_tci_mask & 0x1000)
388 != 0 is equivalent to the one listed in the table.
390 OF1.0: The spec doesn't define behavior if dl_vlan is set to
391 0xffff and OFPFW_DL_VLAN_PCP is not set.
393 OF1.1: The spec says explicitly to ignore dl_vlan_pcp when
394 dl_vlan is set to 0xffff.
396 OF1.2: The spec doesn't say what should happen if (vlan_vid == 0)
397 and (vlan_vid_mask & 0x1000) != 0 but (vlan_vid_mask != 0x1000),
398 but it would be straightforward to also interpret as [2].
400 [3] Matches only packets that have an 802.1Q header with VID xxx (and
403 [4] Matches only packets that have an 802.1Q header with PCP y (and
406 NXM: z is ((y << 1) | 1).
408 OF1.0: The spec isn't very clear, but OVS implements it this way.
410 OF1.2: Presumably other masks such that (vlan_vid_mask & 0x1fff)
411 == 0x1000 would also work, but the spec doesn't define their
414 [5] Matches only packets that have an 802.1Q header with VID xxx and
417 NXM: z is ((y << 1) | 1).
419 OF1.2: Presumably other masks such that (vlan_vid_mask & 0x1fff)
420 == 0x1fff would also work.
422 [6] Matches packets with no 802.1Q header or with an 802.1Q header
423 with a VID of 0. Only possible with NXM.
425 [7] Matches packets with no 802.1Q header or with an 802.1Q header
426 with a PCP of 0. Only possible with NXM.
428 [8] Matches packets with no 802.1Q header or with an 802.1Q header
429 with both VID and PCP of 0. Only possible with NXM.
431 [9] Matches only packets that have an 802.1Q header with an
432 odd-numbered VID (and any PCP). Only possible with NXM and
433 OF1.2. (This is just an example; one can match on any desired
436 [10] Matches only packets that have an 802.1Q header with an
437 odd-numbered PCP (and any VID). Only possible with NXM. (This
438 is just an example; one can match on any desired VID bit
443 - OF1.2: The top three bits of OXM_OF_VLAN_VID are fixed to zero,
444 so bits 13, 14, and 15 in the masks listed in the table may be
445 set to arbitrary values, as long as the corresponding value bits
446 are also zero. The suggested ffff mask for [2], [3], and [5]
447 allows a shorter OXM representation (the mask is omitted) than
448 the minimal 1fff mask.
454 OpenFlow 1.0 and later versions have the concept of a "flow cookie",
455 which is a 64-bit integer value attached to each flow. The treatment
456 of the flow cookie has varied greatly across OpenFlow versions,
461 - OFPFC_ADD set the cookie in the flow that it added.
463 - OFPFC_MODIFY and OFPFC_MODIFY_STRICT updated the cookie for
464 the flow or flows that it modified.
466 - OFPST_FLOW messages included the flow cookie.
468 - OFPT_FLOW_REMOVED messages reported the cookie of the flow
471 OpenFlow 1.1 made the following changes:
473 - Flow mod operations OFPFC_MODIFY, OFPFC_MODIFY_STRICT,
474 OFPFC_DELETE, and OFPFC_DELETE_STRICT, plus flow stats
475 requests and aggregate stats requests, gained the ability to
476 match on flow cookies with an arbitrary mask.
478 - OFPFC_MODIFY and OFPFC_MODIFY_STRICT were changed to add a
479 new flow, in the case of no match, only if the flow table
480 modification operation did not match on the cookie field.
481 (In OpenFlow 1.0, modify operations always added a new flow
482 when there was no match.)
484 - OFPFC_MODIFY and OFPFC_MODIFY_STRICT no longer updated flow
487 OpenFlow 1.2 made the following changes:
489 - OFPC_MODIFY and OFPFC_MODIFY_STRICT were changed to never
490 add a new flow, regardless of whether the flow cookie was
493 Open vSwitch support for OpenFlow 1.0 implements the OpenFlow 1.0
494 behavior with the following extensions:
496 - An NXM extension field NXM_NX_COOKIE(_W) allows the NXM
497 versions of OFPFC_MODIFY, OFPFC_MODIFY_STRICT, OFPFC_DELETE,
498 and OFPFC_DELETE_STRICT flow_mods, plus flow stats requests
499 and aggregate stats requests, to match on flow cookies with
500 arbitrary masks. This is much like the equivalent OpenFlow
503 - Like OpenFlow 1.1, OFPC_MODIFY and OFPFC_MODIFY_STRICT add a
504 new flow if there is no match and the mask is zero (or not
507 - The "cookie" field in OFPT_FLOW_MOD and NXT_FLOW_MOD messages
508 is used as the cookie value for OFPFC_ADD commands, as
509 described in OpenFlow 1.0. For OFPFC_MODIFY and
510 OFPFC_MODIFY_STRICT commands, the "cookie" field is used as a
511 new cookie for flows that match unless it is UINT64_MAX, in
512 which case the flow's cookie is not updated.
514 - NXT_PACKET_IN (the Nicira extended version of
515 OFPT_PACKET_IN) reports the cookie of the rule that
516 generated the packet, or all-1-bits if no rule generated the
517 packet. (Older versions of OVS used all-0-bits instead of
520 The following table shows the handling of different protocols when
521 receiving OFPFC_MODIFY and OFPFC_MODIFY_STRICT messages. A mask of 0
522 indicates either an explicit mask of zero or an implicit one by not
523 specifying the NXM_NX_COOKIE(_W) field.
526 Match Update Add on miss Add on miss
527 cookie cookie mask!=0 mask==0
528 ====== ====== =========== ===========
529 OpenFlow 1.0 no yes <always add on miss>
530 OpenFlow 1.1 yes no no yes
531 OpenFlow 1.2 yes no no no
534 * Updates the flow's cookie unless the "cookie" field is UINT64_MAX.
537 Multiple Table Support
538 ======================
540 OpenFlow 1.0 has only rudimentary support for multiple flow tables.
541 Notably, OpenFlow 1.0 does not allow the controller to specify the
542 flow table to which a flow is to be added. Open vSwitch adds an
543 extension for this purpose, which is enabled on a per-OpenFlow
544 connection basis using the NXT_FLOW_MOD_TABLE_ID message. When the
545 extension is enabled, the upper 8 bits of the 'command' member in an
546 OFPT_FLOW_MOD or NXT_FLOW_MOD message designates the table to which a
549 The Open vSwitch software switch implementation offers 255 flow
550 tables. On packet ingress, only the first flow table (table 0) is
551 searched, and the contents of the remaining tables are not considered
552 in any way. Tables other than table 0 only come into play when an
553 NXAST_RESUBMIT_TABLE action specifies another table to search.
555 Tables 128 and above are reserved for use by the switch itself.
556 Controllers should use only tables 0 through 127.
562 Open vSwitch supports stateless handling of IPv6 packets. Flows can be
563 written to support matching TCP, UDP, and ICMPv6 headers within an IPv6
564 packet. Deeper matching of some Neighbor Discovery messages is also
567 IPv6 was not designed to interact well with middle-boxes. This,
568 combined with Open vSwitch's stateless nature, have affected the
569 processing of IPv6 traffic, which is detailed below.
574 The base IPv6 header is incredibly simple with the intention of only
575 containing information relevant for routing packets between two
576 endpoints. IPv6 relies heavily on the use of extension headers to
577 provide any other functionality. Unfortunately, the extension headers
578 were designed in such a way that it is impossible to move to the next
579 header (including the layer-4 payload) unless the current header is
582 Open vSwitch will process the following extension headers and continue
585 * Fragment (see the next section)
586 * AH (Authentication Header)
589 * Destination Options
591 When a header is encountered that is not in that list, it is considered
592 "terminal". A terminal header's IPv6 protocol value is stored in
593 "nw_proto" for matching purposes. If a terminal header is TCP, UDP, or
594 ICMPv6, the packet will be further processed in an attempt to extract
600 IPv6 requires that every link in the internet have an MTU of 1280 octets
601 or greater (RFC 2460). As such, a terminal header (as described above in
602 "Extension Headers") in the first fragment should generally be
603 reachable. In this case, the terminal header's IPv6 protocol type is
604 stored in the "nw_proto" field for matching purposes. If a terminal
605 header cannot be found in the first fragment (one with a fragment offset
606 of zero), the "nw_proto" field is set to 0. Subsequent fragments (those
607 with a non-zero fragment offset) have the "nw_proto" field set to the
608 IPv6 protocol type for fragments (44).
613 An IPv6 jumbogram (RFC 2675) is a packet containing a payload longer
614 than 65,535 octets. A jumbogram is only relevant in subnets with a link
615 MTU greater than 65,575 octets, and are not required to be supported on
616 nodes that do not connect to link with such large MTUs. Currently, Open
617 vSwitch doesn't process jumbograms.
626 An OpenFlow switch must establish and maintain a TCP network
627 connection to its controller. There are two basic ways to categorize
628 the network that this connection traverses: either it is completely
629 separate from the one that the switch is otherwise controlling, or its
630 path may overlap the network that the switch controls. We call the
631 former case "out-of-band control", the latter case "in-band control".
633 Out-of-band control has the following benefits:
635 - Simplicity: Out-of-band control slightly simplifies the switch
638 - Reliability: Excessive switch traffic volume cannot interfere
639 with control traffic.
641 - Integrity: Machines not on the control network cannot
642 impersonate a switch or a controller.
644 - Confidentiality: Machines not on the control network cannot
645 snoop on control traffic.
647 In-band control, on the other hand, has the following advantages:
649 - No dedicated port: There is no need to dedicate a physical
650 switch port to control, which is important on switches that have
651 few ports (e.g. wireless routers, low-end embedded platforms).
653 - No dedicated network: There is no need to build and maintain a
654 separate control network. This is important in many
655 environments because it reduces proliferation of switches and
658 Open vSwitch supports both out-of-band and in-band control. This
659 section describes the principles behind in-band control. See the
660 description of the Controller table in ovs-vswitchd.conf.db(5) to
661 configure OVS for in-band control.
666 The fundamental principle of in-band control is that an OpenFlow
667 switch must recognize and switch control traffic without involving the
668 OpenFlow controller. All the details of implementing in-band control
669 are special cases of this principle.
671 The rationale for this principle is simple. If the switch does not
672 handle in-band control traffic itself, then it will be caught in a
673 contradiction: it must contact the controller, but it cannot, because
674 only the controller can set up the flows that are needed to contact
677 The following points describe important special cases of this
680 - In-band control must be implemented regardless of whether the
683 It is tempting to implement the in-band control rules only when
684 the switch is not connected to the controller, using the
685 reasoning that the controller should have complete control once
686 it has established a connection with the switch.
688 This does not work in practice. Consider the case where the
689 switch is connected to the controller. Occasionally it can
690 happen that the controller forgets or otherwise needs to obtain
691 the MAC address of the switch. To do so, the controller sends a
692 broadcast ARP request. A switch that implements the in-band
693 control rules only when it is disconnected will then send an
694 OFPT_PACKET_IN message up to the controller. The controller will
695 be unable to respond, because it does not know the MAC address of
696 the switch. This is a deadlock situation that can only be
697 resolved by the switch noticing that its connection to the
698 controller has hung and reconnecting.
700 - In-band control must override flows set up by the controller.
702 It is reasonable to assume that flows set up by the OpenFlow
703 controller should take precedence over in-band control, on the
704 basis that the controller should be in charge of the switch.
706 Again, this does not work in practice. Reasonable controller
707 implementations may set up a "last resort" fallback rule that
708 wildcards every field and, e.g., sends it up to the controller or
709 discards it. If a controller does that, then it will isolate
710 itself from the switch.
712 - The switch must recognize all control traffic.
714 The fundamental principle of in-band control states, in part,
715 that a switch must recognize control traffic without involving
716 the OpenFlow controller. More specifically, the switch must
717 recognize *all* control traffic. "False negatives", that is,
718 packets that constitute control traffic but that the switch does
719 not recognize as control traffic, lead to control traffic storms.
721 Consider an OpenFlow switch that only recognizes control packets
722 sent to or from that switch. Now suppose that two switches of
723 this type, named A and B, are connected to ports on an Ethernet
724 hub (not a switch) and that an OpenFlow controller is connected
725 to a third hub port. In this setup, control traffic sent by
726 switch A will be seen by switch B, which will send it to the
727 controller as part of an OFPT_PACKET_IN message. Switch A will
728 then see the OFPT_PACKET_IN message's packet, re-encapsulate it
729 in another OFPT_PACKET_IN, and send it to the controller. Switch
730 B will then see that OFPT_PACKET_IN, and so on in an infinite
733 Incidentally, the consequences of "false positives", where
734 packets that are not control traffic are nevertheless recognized
735 as control traffic, are much less severe. The controller will
736 not be able to control their behavior, but the network will
737 remain in working order. False positives do constitute a
740 - The switch should use echo-requests to detect disconnection.
742 TCP will notice that a connection has hung, but this can take a
743 considerable amount of time. For example, with default settings
744 the Linux kernel TCP implementation will retransmit for between
745 13 and 30 minutes, depending on the connection's retransmission
746 timeout, according to kernel documentation. This is far too long
747 for a switch to be disconnected, so an OpenFlow switch should
748 implement its own connection timeout. OpenFlow OFPT_ECHO_REQUEST
749 messages are the best way to do this, since they test the
750 OpenFlow connection itself.
755 This section describes how Open vSwitch implements in-band control.
756 Correctly implementing in-band control has proven difficult due to its
757 many subtleties, and has thus gone through many iterations. Please
758 read through and understand the reasoning behind the chosen rules
759 before making modifications.
761 Open vSwitch implements in-band control as "hidden" flows, that is,
762 flows that are not visible through OpenFlow, and at a higher priority
763 than wildcarded flows can be set up through OpenFlow. This is done so
764 that the OpenFlow controller cannot interfere with them and possibly
765 break connectivity with its switches. It is possible to see all
766 flows, including in-band ones, with the ovs-appctl "bridge/dump-flows"
769 The Open vSwitch implementation of in-band control can hide traffic to
770 arbitrary "remotes", where each remote is one TCP port on one IP address.
771 Currently the remotes are automatically configured as the in-band OpenFlow
772 controllers plus the OVSDB managers, if any. (The latter is a requirement
773 because OVSDB managers are responsible for configuring OpenFlow controllers,
774 so if the manager cannot be reached then OpenFlow cannot be reconfigured.)
776 The following rules (with the OFPP_NORMAL action) are set up on any bridge
777 that has any remotes:
779 (a) DHCP requests sent from the local port.
780 (b) ARP replies to the local port's MAC address.
781 (c) ARP requests from the local port's MAC address.
783 In-band also sets up the following rules for each unique next-hop MAC
784 address for the remotes' IPs (the "next hop" is either the remote
785 itself, if it is on a local subnet, or the gateway to reach the remote):
787 (d) ARP replies to the next hop's MAC address.
788 (e) ARP requests from the next hop's MAC address.
790 In-band also sets up the following rules for each unique remote IP address:
792 (f) ARP replies containing the remote's IP address as a target.
793 (g) ARP requests containing the remote's IP address as a source.
795 In-band also sets up the following rules for each unique remote (IP,port)
798 (h) TCP traffic to the remote's IP and port.
799 (i) TCP traffic from the remote's IP and port.
801 The goal of these rules is to be as narrow as possible to allow a
802 switch to join a network and be able to communicate with the
803 remotes. As mentioned earlier, these rules have higher priority
804 than the controller's rules, so if they are too broad, they may
805 prevent the controller from implementing its policy. As such,
806 in-band actively monitors some aspects of flow and packet processing
807 so that the rules can be made more precise.
809 In-band control monitors attempts to add flows into the datapath that
810 could interfere with its duties. The datapath only allows exact
811 match entries, so in-band control is able to be very precise about
812 the flows it prevents. Flows that miss in the datapath are sent to
813 userspace to be processed, so preventing these flows from being
814 cached in the "fast path" does not affect correctness. The only type
815 of flow that is currently prevented is one that would prevent DHCP
816 replies from being seen by the local port. For example, a rule that
817 forwarded all DHCP traffic to the controller would not be allowed,
818 but one that forwarded to all ports (including the local port) would.
820 As mentioned earlier, packets that miss in the datapath are sent to
821 the userspace for processing. The userspace has its own flow table,
822 the "classifier", so in-band checks whether any special processing
823 is needed before the classifier is consulted. If a packet is a DHCP
824 response to a request from the local port, the packet is forwarded to
825 the local port, regardless of the flow table. Note that this requires
826 L7 processing of DHCP replies to determine whether the 'chaddr' field
827 matches the MAC address of the local port.
829 It is interesting to note that for an L3-based in-band control
830 mechanism, the majority of rules are devoted to ARP traffic. At first
831 glance, some of these rules appear redundant. However, each serves an
832 important role. First, in order to determine the MAC address of the
833 remote side (controller or gateway) for other ARP rules, we must allow
834 ARP traffic for our local port with rules (b) and (c). If we are
835 between a switch and its connection to the remote, we have to
836 allow the other switch's ARP traffic to through. This is done with
837 rules (d) and (e), since we do not know the addresses of the other
838 switches a priori, but do know the remote's or gateway's. Finally,
839 if the remote is running in a local guest VM that is not reached
840 through the local port, the switch that is connected to the VM must
841 allow ARP traffic based on the remote's IP address, since it will
842 not know the MAC address of the local port that is sending the traffic
843 or the MAC address of the remote in the guest VM.
845 With a few notable exceptions below, in-band should work in most
846 network setups. The following are considered "supported' in the
847 current implementation:
849 - Locally Connected. The switch and remote are on the same
850 subnet. This uses rules (a), (b), (c), (h), and (i).
852 - Reached through Gateway. The switch and remote are on
853 different subnets and must go through a gateway. This uses
854 rules (a), (b), (c), (h), and (i).
856 - Between Switch and Remote. This switch is between another
857 switch and the remote, and we want to allow the other
858 switch's traffic through. This uses rules (d), (e), (h), and
859 (i). It uses (b) and (c) indirectly in order to know the MAC
860 address for rules (d) and (e). Note that DHCP for the other
861 switch will not work unless an OpenFlow controller explicitly lets this
862 switch pass the traffic.
864 - Between Switch and Gateway. This switch is between another
865 switch and the gateway, and we want to allow the other switch's
866 traffic through. This uses the same rules and logic as the
867 "Between Switch and Remote" configuration described earlier.
869 - Remote on Local VM. The remote is a guest VM on the
870 system running in-band control. This uses rules (a), (b), (c),
873 - Remote on Local VM with Different Networks. The remote
874 is a guest VM on the system running in-band control, but the
875 local port is not used to connect to the remote. For
876 example, an IP address is configured on eth0 of the switch. The
877 remote's VM is connected through eth1 of the switch, but an
878 IP address has not been configured for that port on the switch.
879 As such, the switch will use eth0 to connect to the remote,
880 and eth1's rules about the local port will not work. In the
881 example, the switch attached to eth0 would use rules (a), (b),
882 (c), (h), and (i) on eth0. The switch attached to eth1 would use
883 rules (f), (g), (h), and (i).
885 The following are explicitly *not* supported by in-band control:
887 - Specify Remote by Name. Currently, the remote must be
888 identified by IP address. A naive approach would be to permit
889 all DNS traffic. Unfortunately, this would prevent the
890 controller from defining any policy over DNS. Since switches
891 that are located behind us need to connect to the remote,
892 in-band cannot simply add a rule that allows DNS traffic from
893 the local port. The "correct" way to support this is to parse
894 DNS requests to allow all traffic related to a request for the
895 remote's name through. Due to the potential security
896 problems and amount of processing, we decided to hold off for
899 - Differing Remotes for Switches. All switches must know
900 the L3 addresses for all the remotes that other switches
901 may use, since rules need to be set up to allow traffic related
902 to those remotes through. See rules (f), (g), (h), and (i).
904 - Differing Routes for Switches. In order for the switch to
905 allow other switches to connect to a remote through a
906 gateway, it allows the gateway's traffic through with rules (d)
907 and (e). If the routes to the remote differ for the two
908 switches, we will not know the MAC address of the alternate
915 It seems likely that many controllers, at least at startup, use the
916 OpenFlow "flow statistics" request to obtain existing flows, then
917 compare the flows' actions against the actions that they expect to
918 find. Before version 1.8.0, Open vSwitch always returned exact,
919 byte-for-byte copies of the actions that had been added to the flow
920 table. The current version of Open vSwitch does not always do this in
921 some exceptional cases. This section lists the exceptions that
922 controller authors must keep in mind if they compare actual actions
923 against desired actions in a bytewise fashion:
925 - Open vSwitch zeros padding bytes in action structures,
926 regardless of their values when the flows were added.
928 - Open vSwitch "normalizes" the instructions in OpenFlow 1.1
929 (and later) in the following way:
931 * OVS sorts the instructions into the following order:
932 Apply-Actions, Clear-Actions, Write-Actions,
933 Write-Metadata, Goto-Table.
935 * OVS drops Apply-Actions instructions that have empty
938 * OVS drops Write-Actions instructions that have empty
941 Please report other discrepancies, if you notice any, so that we can
942 fix or document them.
948 Suggestions to improve Open vSwitch are welcome at discuss@openvswitch.org.