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_ACTION_SET (OF1.4+) yes ---
58 OFPR_GROUP (OF1.4+) yes ---
60 OFPT_FLOW_REMOVED / NXT_FLOW_REMOVED
61 OFPRR_IDLE_TIMEOUT yes ---
62 OFPRR_HARD_TIMEOUT yes ---
71 The NXT_SET_ASYNC_CONFIG message directly sets all of the values in
72 this table for the current connection. The
73 OFPC_INVALID_TTL_TO_CONTROLLER bit in the OFPT_SET_CONFIG message
74 controls the setting for OFPR_INVALID_TTL for the "master" role.
80 The OpenFlow 1.0 specification requires the output port of the OFPAT_ENQUEUE
81 action to "refer to a valid physical port (i.e. < OFPP_MAX) or OFPP_IN_PORT".
82 Although OFPP_LOCAL is not less than OFPP_MAX, it is an 'internal' port which
83 can have QoS applied to it in Linux. Since we allow the OFPAT_ENQUEUE to apply
84 to 'internal' ports whose port numbers are less than OFPP_MAX, we interpret
85 OFPP_LOCAL as a physical port and support OFPAT_ENQUEUE on it as well.
91 The OpenFlow specification for the behavior of OFPT_FLOW_MOD is
92 confusing. The following tables summarize the Open vSwitch
93 implementation of its behavior in the following categories:
95 - "match on priority": Whether the flow_mod acts only on flows
96 whose priority matches that included in the flow_mod message.
98 - "match on out_port": Whether the flow_mod acts only on flows
99 that output to the out_port included in the flow_mod message (if
100 out_port is not OFPP_NONE). OpenFlow 1.1 and later have a
101 similar feature (not listed separately here) for out_group.
103 - "match on flow_cookie": Whether the flow_mod acts only on flows
104 whose flow_cookie matches an optional controller-specified value
107 - "updates flow_cookie": Whether the flow_mod changes the
108 flow_cookie of the flow or flows that it matches to the
109 flow_cookie included in the flow_mod message.
111 - "updates OFPFF_ flags": Whether the flow_mod changes the
112 OFPFF_SEND_FLOW_REM flag of the flow or flows that it matches to
113 the setting included in the flags of the flow_mod message.
115 - "honors OFPFF_CHECK_OVERLAP": Whether the OFPFF_CHECK_OVERLAP
116 flag in the flow_mod is significant.
118 - "updates idle_timeout" and "updates hard_timeout": Whether the
119 idle_timeout and hard_timeout in the flow_mod, respectively,
120 have an effect on the flow or flows matched by the flow_mod.
122 - "updates idle timer": Whether the flow_mod resets the per-flow
123 timer that measures how long a flow has been idle.
125 - "updates hard timer": Whether the flow_mod resets the per-flow
126 timer that measures how long it has been since a flow was
129 - "zeros counters": Whether the flow_mod resets per-flow packet
130 and byte counters to zero.
132 - "may add a new flow": Whether the flow_mod may add a new flow to
133 the flow table. (Obviously this is always true for "add"
134 commands but in some OpenFlow versions "modify" and
135 "modify-strict" can also add new flows.)
137 - "sends flow_removed message": Whether the flow_mod generates a
138 flow_removed message for the flow or flows that it affects.
140 An entry labeled "yes" means that the flow mod type does have the
141 indicated behavior, "---" means that it does not, an empty cell means
142 that the property is not applicable, and other values are explained
150 ADD MODIFY STRICT DELETE STRICT
151 === ====== ====== ====== ======
152 match on priority yes --- yes --- yes
153 match on out_port --- --- --- yes yes
154 match on flow_cookie --- --- --- --- ---
155 match on table_id --- --- --- --- ---
156 controller chooses table_id --- --- ---
157 updates flow_cookie yes yes yes
158 updates OFPFF_SEND_FLOW_REM yes + +
159 honors OFPFF_CHECK_OVERLAP yes + +
160 updates idle_timeout yes + +
161 updates hard_timeout yes + +
162 resets idle timer yes + +
163 resets hard timer yes yes yes
164 zeros counters yes + +
165 may add a new flow yes yes yes
166 sends flow_removed message --- --- --- % %
168 (+) "modify" and "modify-strict" only take these actions when they
169 create a new flow, not when they update an existing flow.
171 (%) "delete" and "delete_strict" generates a flow_removed message if
172 the deleted flow or flows have the OFPFF_SEND_FLOW_REM flag set.
173 (Each controller can separately control whether it wants to
174 receive the generated messages.)
180 OpenFlow 1.1 makes these changes:
182 - The controller now must specify the table_id of the flow match
183 searched and into which a flow may be inserted. Behavior for a
184 table_id of 255 is undefined.
186 - A flow_mod, except an "add", can now match on the flow_cookie.
188 - When a flow_mod matches on the flow_cookie, "modify" and
189 "modify-strict" never insert a new flow.
193 ADD MODIFY STRICT DELETE STRICT
194 === ====== ====== ====== ======
195 match on priority yes --- yes --- yes
196 match on out_port --- --- --- yes yes
197 match on flow_cookie --- yes yes yes yes
198 match on table_id yes yes yes yes yes
199 controller chooses table_id yes yes yes
200 updates flow_cookie yes --- ---
201 updates OFPFF_SEND_FLOW_REM yes + +
202 honors OFPFF_CHECK_OVERLAP yes + +
203 updates idle_timeout yes + +
204 updates hard_timeout yes + +
205 resets idle timer yes + +
206 resets hard timer yes yes yes
207 zeros counters yes + +
208 may add a new flow yes # #
209 sends flow_removed message --- --- --- % %
211 (+) "modify" and "modify-strict" only take these actions when they
212 create a new flow, not when they update an existing flow.
214 (%) "delete" and "delete_strict" generates a flow_removed message if
215 the deleted flow or flows have the OFPFF_SEND_FLOW_REM flag set.
216 (Each controller can separately control whether it wants to
217 receive the generated messages.)
219 (#) "modify" and "modify-strict" only add a new flow if the flow_mod
220 does not match on any bits of the flow cookie
226 OpenFlow 1.2 makes these changes:
228 - Only "add" commands ever add flows, "modify" and "modify-strict"
231 - A new flag OFPFF_RESET_COUNTS now controls whether "modify" and
232 "modify-strict" reset counters, whereas previously they never
233 reset counters (except when they inserted a new flow).
237 ADD MODIFY STRICT DELETE STRICT
238 === ====== ====== ====== ======
239 match on priority yes --- yes --- yes
240 match on out_port --- --- --- yes yes
241 match on flow_cookie --- yes yes yes yes
242 match on table_id yes yes yes yes yes
243 controller chooses table_id yes yes yes
244 updates flow_cookie yes --- ---
245 updates OFPFF_SEND_FLOW_REM yes --- ---
246 honors OFPFF_CHECK_OVERLAP yes --- ---
247 updates idle_timeout yes --- ---
248 updates hard_timeout yes --- ---
249 resets idle timer yes --- ---
250 resets hard timer yes yes yes
251 zeros counters yes & &
252 may add a new flow yes --- ---
253 sends flow_removed message --- --- --- % %
255 (%) "delete" and "delete_strict" generates a flow_removed message if
256 the deleted flow or flows have the OFPFF_SEND_FLOW_REM flag set.
257 (Each controller can separately control whether it wants to
258 receive the generated messages.)
260 (&) "modify" and "modify-strict" reset counters if the
261 OFPFF_RESET_COUNTS flag is specified.
267 OpenFlow 1.3 makes these changes:
269 - Behavior for a table_id of 255 is now defined, for "delete" and
270 "delete-strict" commands, as meaning to delete from all tables.
271 A table_id of 255 is now explicitly invalid for other commands.
273 - New flags OFPFF_NO_PKT_COUNTS and OFPFF_NO_BYT_COUNTS for "add"
276 The table for 1.3 is the same as the one shown above for 1.2.
282 OpenFlow 1.4 adds the "importance" field to flow_mods, but it does not
283 explicitly specify which kinds of flow_mods set the importance.For
284 consistency, Open vSwitch uses the same rule for importance as for
285 idle_timeout and hard_timeout, that is, only an "ADD" flow_mod sets
286 the importance. (This issue has been filed with the ONF as EXT-496.)
291 The OpenFlow 1.1 specification for OFPT_PACKET_IN is confusing. The
292 definition in OF1.1 openflow.h is[*]:
295 /* Packet received on port (datapath -> controller). */
296 struct ofp_packet_in {
297 struct ofp_header header;
298 uint32_t buffer_id; /* ID assigned by datapath. */
299 uint32_t in_port; /* Port on which frame was received. */
300 uint32_t in_phy_port; /* Physical Port on which frame was received. */
301 uint16_t total_len; /* Full length of frame. */
302 uint8_t reason; /* Reason packet is being sent (one of OFPR_*) */
303 uint8_t table_id; /* ID of the table that was looked up */
304 uint8_t data[0]; /* Ethernet frame, halfway through 32-bit word,
305 so the IP header is 32-bit aligned. The
306 amount of data is inferred from the length
307 field in the header. Because of padding,
308 offsetof(struct ofp_packet_in, data) ==
309 sizeof(struct ofp_packet_in) - 2. */
311 OFP_ASSERT(sizeof(struct ofp_packet_in) == 24);
314 The confusing part is the comment on the data[] member. This comment
315 is a leftover from OF1.0 openflow.h, in which the comment was correct:
316 sizeof(struct ofp_packet_in) is 20 in OF1.0 and offsetof(struct
317 ofp_packet_in, data) is 18. When OF1.1 was written, the structure
318 members were changed but the comment was carelessly not updated, and
319 the comment became wrong: sizeof(struct ofp_packet_in) and
320 offsetof(struct ofp_packet_in, data) are both 24 in OF1.1.
322 That leaves the question of how to implement ofp_packet_in in OF1.1.
323 The OpenFlow reference implementation for OF1.1 does not include any
324 padding, that is, the first byte of the encapsulated frame immediately
325 follows the 'table_id' member without a gap. Open vSwitch therefore
326 implements it the same way for compatibility.
328 For an earlier discussion, please see the thread archived at:
329 https://mailman.stanford.edu/pipermail/openflow-discuss/2011-August/002604.html
331 [*] The quoted definition is directly from OF1.1. Definitions used
332 inside OVS omit the 8-byte ofp_header members, so the sizes in
333 this discussion are 8 bytes larger than those declared in OVS
340 The 802.1Q VLAN header causes more trouble than any other 4 bytes in
341 networking. More specifically, three versions of OpenFlow and Open
342 vSwitch have among them four different ways to match the contents and
343 presence of the VLAN header. The following table describes how each
346 Match NXM OF1.0 OF1.1 OF1.2
347 ----- --------- ----------- ----------- ------------
348 [1] 0000/0000 ????/1,??/? ????/1,??/? 0000/0000,--
349 [2] 0000/ffff ffff/0,??/? ffff/0,??/? 0000/ffff,--
350 [3] 1xxx/1fff 0xxx/0,??/1 0xxx/0,??/1 1xxx/ffff,--
351 [4] z000/f000 ????/1,0y/0 fffe/0,0y/0 1000/1000,0y
352 [5] zxxx/ffff 0xxx/0,0y/0 0xxx/0,0y/0 1xxx/ffff,0y
353 [6] 0000/0fff <none> <none> <none>
354 [7] 0000/f000 <none> <none> <none>
355 [8] 0000/efff <none> <none> <none>
356 [9] 1001/1001 <none> <none> 1001/1001,--
357 [10] 3000/3000 <none> <none> <none>
359 Each column is interpreted as follows.
361 - Match: See the list below.
363 - NXM: xxxx/yyyy means NXM_OF_VLAN_TCI_W with value xxxx and mask
364 yyyy. A mask of 0000 is equivalent to omitting
365 NXM_OF_VLAN_TCI(_W), a mask of ffff is equivalent to
368 - OF1.0 and OF1.1: wwww/x,yy/z means dl_vlan wwww, OFPFW_DL_VLAN
369 x, dl_vlan_pcp yy, and OFPFW_DL_VLAN_PCP z. ? means that the
370 given nibble is ignored (and conventionally 0 for wwww or yy,
371 conventionally 1 for x or z). <none> means that the given match
374 - OF1.2: xxxx/yyyy,zz means OXM_OF_VLAN_VID_W with value xxxx and
375 mask yyyy, and OXM_OF_VLAN_PCP (which is not maskable) with
376 value zz. A mask of 0000 is equivalent to omitting
377 OXM_OF_VLAN_VID(_W), a mask of ffff is equivalent to
378 OXM_OF_VLAN_VID. -- means that OXM_OF_VLAN_PCP is omitted.
379 <none> means that the given match is not supported.
383 [1] Matches any packet, that is, one without an 802.1Q header or with
384 an 802.1Q header with any TCI value.
386 [2] Matches only packets without an 802.1Q header.
388 NXM: Any match with (vlan_tci == 0) and (vlan_tci_mask & 0x1000)
389 != 0 is equivalent to the one listed in the table.
391 OF1.0: The spec doesn't define behavior if dl_vlan is set to
392 0xffff and OFPFW_DL_VLAN_PCP is not set.
394 OF1.1: The spec says explicitly to ignore dl_vlan_pcp when
395 dl_vlan is set to 0xffff.
397 OF1.2: The spec doesn't say what should happen if (vlan_vid == 0)
398 and (vlan_vid_mask & 0x1000) != 0 but (vlan_vid_mask != 0x1000),
399 but it would be straightforward to also interpret as [2].
401 [3] Matches only packets that have an 802.1Q header with VID xxx (and
404 [4] Matches only packets that have an 802.1Q header with PCP y (and
407 NXM: z is ((y << 1) | 1).
409 OF1.0: The spec isn't very clear, but OVS implements it this way.
411 OF1.2: Presumably other masks such that (vlan_vid_mask & 0x1fff)
412 == 0x1000 would also work, but the spec doesn't define their
415 [5] Matches only packets that have an 802.1Q header with VID xxx and
418 NXM: z is ((y << 1) | 1).
420 OF1.2: Presumably other masks such that (vlan_vid_mask & 0x1fff)
421 == 0x1fff would also work.
423 [6] Matches packets with no 802.1Q header or with an 802.1Q header
424 with a VID of 0. Only possible with NXM.
426 [7] Matches packets with no 802.1Q header or with an 802.1Q header
427 with a PCP of 0. Only possible with NXM.
429 [8] Matches packets with no 802.1Q header or with an 802.1Q header
430 with both VID and PCP of 0. Only possible with NXM.
432 [9] Matches only packets that have an 802.1Q header with an
433 odd-numbered VID (and any PCP). Only possible with NXM and
434 OF1.2. (This is just an example; one can match on any desired
437 [10] Matches only packets that have an 802.1Q header with an
438 odd-numbered PCP (and any VID). Only possible with NXM. (This
439 is just an example; one can match on any desired VID bit
444 - OF1.2: The top three bits of OXM_OF_VLAN_VID are fixed to zero,
445 so bits 13, 14, and 15 in the masks listed in the table may be
446 set to arbitrary values, as long as the corresponding value bits
447 are also zero. The suggested ffff mask for [2], [3], and [5]
448 allows a shorter OXM representation (the mask is omitted) than
449 the minimal 1fff mask.
455 OpenFlow 1.0 and later versions have the concept of a "flow cookie",
456 which is a 64-bit integer value attached to each flow. The treatment
457 of the flow cookie has varied greatly across OpenFlow versions,
462 - OFPFC_ADD set the cookie in the flow that it added.
464 - OFPFC_MODIFY and OFPFC_MODIFY_STRICT updated the cookie for
465 the flow or flows that it modified.
467 - OFPST_FLOW messages included the flow cookie.
469 - OFPT_FLOW_REMOVED messages reported the cookie of the flow
472 OpenFlow 1.1 made the following changes:
474 - Flow mod operations OFPFC_MODIFY, OFPFC_MODIFY_STRICT,
475 OFPFC_DELETE, and OFPFC_DELETE_STRICT, plus flow stats
476 requests and aggregate stats requests, gained the ability to
477 match on flow cookies with an arbitrary mask.
479 - OFPFC_MODIFY and OFPFC_MODIFY_STRICT were changed to add a
480 new flow, in the case of no match, only if the flow table
481 modification operation did not match on the cookie field.
482 (In OpenFlow 1.0, modify operations always added a new flow
483 when there was no match.)
485 - OFPFC_MODIFY and OFPFC_MODIFY_STRICT no longer updated flow
488 OpenFlow 1.2 made the following changes:
490 - OFPC_MODIFY and OFPFC_MODIFY_STRICT were changed to never
491 add a new flow, regardless of whether the flow cookie was
494 Open vSwitch support for OpenFlow 1.0 implements the OpenFlow 1.0
495 behavior with the following extensions:
497 - An NXM extension field NXM_NX_COOKIE(_W) allows the NXM
498 versions of OFPFC_MODIFY, OFPFC_MODIFY_STRICT, OFPFC_DELETE,
499 and OFPFC_DELETE_STRICT flow_mods, plus flow stats requests
500 and aggregate stats requests, to match on flow cookies with
501 arbitrary masks. This is much like the equivalent OpenFlow
504 - Like OpenFlow 1.1, OFPC_MODIFY and OFPFC_MODIFY_STRICT add a
505 new flow if there is no match and the mask is zero (or not
508 - The "cookie" field in OFPT_FLOW_MOD and NXT_FLOW_MOD messages
509 is used as the cookie value for OFPFC_ADD commands, as
510 described in OpenFlow 1.0. For OFPFC_MODIFY and
511 OFPFC_MODIFY_STRICT commands, the "cookie" field is used as a
512 new cookie for flows that match unless it is UINT64_MAX, in
513 which case the flow's cookie is not updated.
515 - NXT_PACKET_IN (the Nicira extended version of
516 OFPT_PACKET_IN) reports the cookie of the rule that
517 generated the packet, or all-1-bits if no rule generated the
518 packet. (Older versions of OVS used all-0-bits instead of
521 The following table shows the handling of different protocols when
522 receiving OFPFC_MODIFY and OFPFC_MODIFY_STRICT messages. A mask of 0
523 indicates either an explicit mask of zero or an implicit one by not
524 specifying the NXM_NX_COOKIE(_W) field.
527 Match Update Add on miss Add on miss
528 cookie cookie mask!=0 mask==0
529 ====== ====== =========== ===========
530 OpenFlow 1.0 no yes <always add on miss>
531 OpenFlow 1.1 yes no no yes
532 OpenFlow 1.2 yes no no no
535 * Updates the flow's cookie unless the "cookie" field is UINT64_MAX.
538 Multiple Table Support
539 ======================
541 OpenFlow 1.0 has only rudimentary support for multiple flow tables.
542 Notably, OpenFlow 1.0 does not allow the controller to specify the
543 flow table to which a flow is to be added. Open vSwitch adds an
544 extension for this purpose, which is enabled on a per-OpenFlow
545 connection basis using the NXT_FLOW_MOD_TABLE_ID message. When the
546 extension is enabled, the upper 8 bits of the 'command' member in an
547 OFPT_FLOW_MOD or NXT_FLOW_MOD message designates the table to which a
550 The Open vSwitch software switch implementation offers 255 flow
551 tables. On packet ingress, only the first flow table (table 0) is
552 searched, and the contents of the remaining tables are not considered
553 in any way. Tables other than table 0 only come into play when an
554 NXAST_RESUBMIT_TABLE action specifies another table to search.
556 Tables 128 and above are reserved for use by the switch itself.
557 Controllers should use only tables 0 through 127.
563 Open vSwitch supports stateless handling of IPv6 packets. Flows can be
564 written to support matching TCP, UDP, and ICMPv6 headers within an IPv6
565 packet. Deeper matching of some Neighbor Discovery messages is also
568 IPv6 was not designed to interact well with middle-boxes. This,
569 combined with Open vSwitch's stateless nature, have affected the
570 processing of IPv6 traffic, which is detailed below.
575 The base IPv6 header is incredibly simple with the intention of only
576 containing information relevant for routing packets between two
577 endpoints. IPv6 relies heavily on the use of extension headers to
578 provide any other functionality. Unfortunately, the extension headers
579 were designed in such a way that it is impossible to move to the next
580 header (including the layer-4 payload) unless the current header is
583 Open vSwitch will process the following extension headers and continue
586 * Fragment (see the next section)
587 * AH (Authentication Header)
590 * Destination Options
592 When a header is encountered that is not in that list, it is considered
593 "terminal". A terminal header's IPv6 protocol value is stored in
594 "nw_proto" for matching purposes. If a terminal header is TCP, UDP, or
595 ICMPv6, the packet will be further processed in an attempt to extract
601 IPv6 requires that every link in the internet have an MTU of 1280 octets
602 or greater (RFC 2460). As such, a terminal header (as described above in
603 "Extension Headers") in the first fragment should generally be
604 reachable. In this case, the terminal header's IPv6 protocol type is
605 stored in the "nw_proto" field for matching purposes. If a terminal
606 header cannot be found in the first fragment (one with a fragment offset
607 of zero), the "nw_proto" field is set to 0. Subsequent fragments (those
608 with a non-zero fragment offset) have the "nw_proto" field set to the
609 IPv6 protocol type for fragments (44).
614 An IPv6 jumbogram (RFC 2675) is a packet containing a payload longer
615 than 65,535 octets. A jumbogram is only relevant in subnets with a link
616 MTU greater than 65,575 octets, and are not required to be supported on
617 nodes that do not connect to link with such large MTUs. Currently, Open
618 vSwitch doesn't process jumbograms.
627 An OpenFlow switch must establish and maintain a TCP network
628 connection to its controller. There are two basic ways to categorize
629 the network that this connection traverses: either it is completely
630 separate from the one that the switch is otherwise controlling, or its
631 path may overlap the network that the switch controls. We call the
632 former case "out-of-band control", the latter case "in-band control".
634 Out-of-band control has the following benefits:
636 - Simplicity: Out-of-band control slightly simplifies the switch
639 - Reliability: Excessive switch traffic volume cannot interfere
640 with control traffic.
642 - Integrity: Machines not on the control network cannot
643 impersonate a switch or a controller.
645 - Confidentiality: Machines not on the control network cannot
646 snoop on control traffic.
648 In-band control, on the other hand, has the following advantages:
650 - No dedicated port: There is no need to dedicate a physical
651 switch port to control, which is important on switches that have
652 few ports (e.g. wireless routers, low-end embedded platforms).
654 - No dedicated network: There is no need to build and maintain a
655 separate control network. This is important in many
656 environments because it reduces proliferation of switches and
659 Open vSwitch supports both out-of-band and in-band control. This
660 section describes the principles behind in-band control. See the
661 description of the Controller table in ovs-vswitchd.conf.db(5) to
662 configure OVS for in-band control.
667 The fundamental principle of in-band control is that an OpenFlow
668 switch must recognize and switch control traffic without involving the
669 OpenFlow controller. All the details of implementing in-band control
670 are special cases of this principle.
672 The rationale for this principle is simple. If the switch does not
673 handle in-band control traffic itself, then it will be caught in a
674 contradiction: it must contact the controller, but it cannot, because
675 only the controller can set up the flows that are needed to contact
678 The following points describe important special cases of this
681 - In-band control must be implemented regardless of whether the
684 It is tempting to implement the in-band control rules only when
685 the switch is not connected to the controller, using the
686 reasoning that the controller should have complete control once
687 it has established a connection with the switch.
689 This does not work in practice. Consider the case where the
690 switch is connected to the controller. Occasionally it can
691 happen that the controller forgets or otherwise needs to obtain
692 the MAC address of the switch. To do so, the controller sends a
693 broadcast ARP request. A switch that implements the in-band
694 control rules only when it is disconnected will then send an
695 OFPT_PACKET_IN message up to the controller. The controller will
696 be unable to respond, because it does not know the MAC address of
697 the switch. This is a deadlock situation that can only be
698 resolved by the switch noticing that its connection to the
699 controller has hung and reconnecting.
701 - In-band control must override flows set up by the controller.
703 It is reasonable to assume that flows set up by the OpenFlow
704 controller should take precedence over in-band control, on the
705 basis that the controller should be in charge of the switch.
707 Again, this does not work in practice. Reasonable controller
708 implementations may set up a "last resort" fallback rule that
709 wildcards every field and, e.g., sends it up to the controller or
710 discards it. If a controller does that, then it will isolate
711 itself from the switch.
713 - The switch must recognize all control traffic.
715 The fundamental principle of in-band control states, in part,
716 that a switch must recognize control traffic without involving
717 the OpenFlow controller. More specifically, the switch must
718 recognize *all* control traffic. "False negatives", that is,
719 packets that constitute control traffic but that the switch does
720 not recognize as control traffic, lead to control traffic storms.
722 Consider an OpenFlow switch that only recognizes control packets
723 sent to or from that switch. Now suppose that two switches of
724 this type, named A and B, are connected to ports on an Ethernet
725 hub (not a switch) and that an OpenFlow controller is connected
726 to a third hub port. In this setup, control traffic sent by
727 switch A will be seen by switch B, which will send it to the
728 controller as part of an OFPT_PACKET_IN message. Switch A will
729 then see the OFPT_PACKET_IN message's packet, re-encapsulate it
730 in another OFPT_PACKET_IN, and send it to the controller. Switch
731 B will then see that OFPT_PACKET_IN, and so on in an infinite
734 Incidentally, the consequences of "false positives", where
735 packets that are not control traffic are nevertheless recognized
736 as control traffic, are much less severe. The controller will
737 not be able to control their behavior, but the network will
738 remain in working order. False positives do constitute a
741 - The switch should use echo-requests to detect disconnection.
743 TCP will notice that a connection has hung, but this can take a
744 considerable amount of time. For example, with default settings
745 the Linux kernel TCP implementation will retransmit for between
746 13 and 30 minutes, depending on the connection's retransmission
747 timeout, according to kernel documentation. This is far too long
748 for a switch to be disconnected, so an OpenFlow switch should
749 implement its own connection timeout. OpenFlow OFPT_ECHO_REQUEST
750 messages are the best way to do this, since they test the
751 OpenFlow connection itself.
756 This section describes how Open vSwitch implements in-band control.
757 Correctly implementing in-band control has proven difficult due to its
758 many subtleties, and has thus gone through many iterations. Please
759 read through and understand the reasoning behind the chosen rules
760 before making modifications.
762 Open vSwitch implements in-band control as "hidden" flows, that is,
763 flows that are not visible through OpenFlow, and at a higher priority
764 than wildcarded flows can be set up through OpenFlow. This is done so
765 that the OpenFlow controller cannot interfere with them and possibly
766 break connectivity with its switches. It is possible to see all
767 flows, including in-band ones, with the ovs-appctl "bridge/dump-flows"
770 The Open vSwitch implementation of in-band control can hide traffic to
771 arbitrary "remotes", where each remote is one TCP port on one IP address.
772 Currently the remotes are automatically configured as the in-band OpenFlow
773 controllers plus the OVSDB managers, if any. (The latter is a requirement
774 because OVSDB managers are responsible for configuring OpenFlow controllers,
775 so if the manager cannot be reached then OpenFlow cannot be reconfigured.)
777 The following rules (with the OFPP_NORMAL action) are set up on any bridge
778 that has any remotes:
780 (a) DHCP requests sent from the local port.
781 (b) ARP replies to the local port's MAC address.
782 (c) ARP requests from the local port's MAC address.
784 In-band also sets up the following rules for each unique next-hop MAC
785 address for the remotes' IPs (the "next hop" is either the remote
786 itself, if it is on a local subnet, or the gateway to reach the remote):
788 (d) ARP replies to the next hop's MAC address.
789 (e) ARP requests from the next hop's MAC address.
791 In-band also sets up the following rules for each unique remote IP address:
793 (f) ARP replies containing the remote's IP address as a target.
794 (g) ARP requests containing the remote's IP address as a source.
796 In-band also sets up the following rules for each unique remote (IP,port)
799 (h) TCP traffic to the remote's IP and port.
800 (i) TCP traffic from the remote's IP and port.
802 The goal of these rules is to be as narrow as possible to allow a
803 switch to join a network and be able to communicate with the
804 remotes. As mentioned earlier, these rules have higher priority
805 than the controller's rules, so if they are too broad, they may
806 prevent the controller from implementing its policy. As such,
807 in-band actively monitors some aspects of flow and packet processing
808 so that the rules can be made more precise.
810 In-band control monitors attempts to add flows into the datapath that
811 could interfere with its duties. The datapath only allows exact
812 match entries, so in-band control is able to be very precise about
813 the flows it prevents. Flows that miss in the datapath are sent to
814 userspace to be processed, so preventing these flows from being
815 cached in the "fast path" does not affect correctness. The only type
816 of flow that is currently prevented is one that would prevent DHCP
817 replies from being seen by the local port. For example, a rule that
818 forwarded all DHCP traffic to the controller would not be allowed,
819 but one that forwarded to all ports (including the local port) would.
821 As mentioned earlier, packets that miss in the datapath are sent to
822 the userspace for processing. The userspace has its own flow table,
823 the "classifier", so in-band checks whether any special processing
824 is needed before the classifier is consulted. If a packet is a DHCP
825 response to a request from the local port, the packet is forwarded to
826 the local port, regardless of the flow table. Note that this requires
827 L7 processing of DHCP replies to determine whether the 'chaddr' field
828 matches the MAC address of the local port.
830 It is interesting to note that for an L3-based in-band control
831 mechanism, the majority of rules are devoted to ARP traffic. At first
832 glance, some of these rules appear redundant. However, each serves an
833 important role. First, in order to determine the MAC address of the
834 remote side (controller or gateway) for other ARP rules, we must allow
835 ARP traffic for our local port with rules (b) and (c). If we are
836 between a switch and its connection to the remote, we have to
837 allow the other switch's ARP traffic to through. This is done with
838 rules (d) and (e), since we do not know the addresses of the other
839 switches a priori, but do know the remote's or gateway's. Finally,
840 if the remote is running in a local guest VM that is not reached
841 through the local port, the switch that is connected to the VM must
842 allow ARP traffic based on the remote's IP address, since it will
843 not know the MAC address of the local port that is sending the traffic
844 or the MAC address of the remote in the guest VM.
846 With a few notable exceptions below, in-band should work in most
847 network setups. The following are considered "supported' in the
848 current implementation:
850 - Locally Connected. The switch and remote are on the same
851 subnet. This uses rules (a), (b), (c), (h), and (i).
853 - Reached through Gateway. The switch and remote are on
854 different subnets and must go through a gateway. This uses
855 rules (a), (b), (c), (h), and (i).
857 - Between Switch and Remote. This switch is between another
858 switch and the remote, and we want to allow the other
859 switch's traffic through. This uses rules (d), (e), (h), and
860 (i). It uses (b) and (c) indirectly in order to know the MAC
861 address for rules (d) and (e). Note that DHCP for the other
862 switch will not work unless an OpenFlow controller explicitly lets this
863 switch pass the traffic.
865 - Between Switch and Gateway. This switch is between another
866 switch and the gateway, and we want to allow the other switch's
867 traffic through. This uses the same rules and logic as the
868 "Between Switch and Remote" configuration described earlier.
870 - Remote on Local VM. The remote is a guest VM on the
871 system running in-band control. This uses rules (a), (b), (c),
874 - Remote on Local VM with Different Networks. The remote
875 is a guest VM on the system running in-band control, but the
876 local port is not used to connect to the remote. For
877 example, an IP address is configured on eth0 of the switch. The
878 remote's VM is connected through eth1 of the switch, but an
879 IP address has not been configured for that port on the switch.
880 As such, the switch will use eth0 to connect to the remote,
881 and eth1's rules about the local port will not work. In the
882 example, the switch attached to eth0 would use rules (a), (b),
883 (c), (h), and (i) on eth0. The switch attached to eth1 would use
884 rules (f), (g), (h), and (i).
886 The following are explicitly *not* supported by in-band control:
888 - Specify Remote by Name. Currently, the remote must be
889 identified by IP address. A naive approach would be to permit
890 all DNS traffic. Unfortunately, this would prevent the
891 controller from defining any policy over DNS. Since switches
892 that are located behind us need to connect to the remote,
893 in-band cannot simply add a rule that allows DNS traffic from
894 the local port. The "correct" way to support this is to parse
895 DNS requests to allow all traffic related to a request for the
896 remote's name through. Due to the potential security
897 problems and amount of processing, we decided to hold off for
900 - Differing Remotes for Switches. All switches must know
901 the L3 addresses for all the remotes that other switches
902 may use, since rules need to be set up to allow traffic related
903 to those remotes through. See rules (f), (g), (h), and (i).
905 - Differing Routes for Switches. In order for the switch to
906 allow other switches to connect to a remote through a
907 gateway, it allows the gateway's traffic through with rules (d)
908 and (e). If the routes to the remote differ for the two
909 switches, we will not know the MAC address of the alternate
916 It seems likely that many controllers, at least at startup, use the
917 OpenFlow "flow statistics" request to obtain existing flows, then
918 compare the flows' actions against the actions that they expect to
919 find. Before version 1.8.0, Open vSwitch always returned exact,
920 byte-for-byte copies of the actions that had been added to the flow
921 table. The current version of Open vSwitch does not always do this in
922 some exceptional cases. This section lists the exceptions that
923 controller authors must keep in mind if they compare actual actions
924 against desired actions in a bytewise fashion:
926 - Open vSwitch zeros padding bytes in action structures,
927 regardless of their values when the flows were added.
929 - Open vSwitch "normalizes" the instructions in OpenFlow 1.1
930 (and later) in the following way:
932 * OVS sorts the instructions into the following order:
933 Apply-Actions, Clear-Actions, Write-Actions,
934 Write-Metadata, Goto-Table.
936 * OVS drops Apply-Actions instructions that have empty
939 * OVS drops Write-Actions instructions that have empty
942 Please report other discrepancies, if you notice any, so that we can
943 fix or document them.
949 Suggestions to improve Open vSwitch are welcome at discuss@openvswitch.org.
projects / ovs.git / blob