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.
50
message and reason code other slave
51
---------------------------------------- ------- -----
52
OFPT_PACKET_IN / NXT_PACKET_IN
55
OFPR_INVALID_TTL --- ---
57
OFPT_FLOW_REMOVED / NXT_FLOW_REMOVED
58
OFPRR_IDLE_TIMEOUT yes ---
59
OFPRR_HARD_TIMEOUT yes ---
67
The NXT_SET_ASYNC_CONFIG message directly sets all of the values in
68
this table for the current connection. The
69
OFPC_INVALID_TTL_TO_CONTROLLER bit in the OFPT_SET_CONFIG message
70
controls the setting for OFPR_INVALID_TTL for the "master" role.
76
The OpenFlow 1.0 specification requires the output port of the OFPAT_ENQUEUE
77
action to "refer to a valid physical port (i.e. < OFPP_MAX) or OFPP_IN_PORT".
78
Although OFPP_LOCAL is not less than OFPP_MAX, it is an 'internal' port which
79
can have QoS applied to it in Linux. Since we allow the OFPAT_ENQUEUE to apply
80
to 'internal' ports whose port numbers are less than OFPP_MAX, we interpret
81
OFPP_LOCAL as a physical port and support OFPAT_ENQUEUE on it as well.
87
The OpenFlow specification for the behavior of OFPT_FLOW_MOD is
88
confusing. The following tables summarize the Open vSwitch
89
implementation of its behavior in the following categories:
91
- "match on priority": Whether the flow_mod acts only on flows
92
whose priority matches that included in the flow_mod message.
94
- "match on out_port": Whether the flow_mod acts only on flows
95
that output to the out_port included in the flow_mod message (if
96
out_port is not OFPP_NONE). OpenFlow 1.1 and later have a
97
similar feature (not listed separately here) for out_group.
99
- "match on flow_cookie": Whether the flow_mod acts only on flows
100
whose flow_cookie matches an optional controller-specified value
103
- "updates flow_cookie": Whether the flow_mod changes the
104
flow_cookie of the flow or flows that it matches to the
105
flow_cookie included in the flow_mod message.
107
- "updates OFPFF_ flags": Whether the flow_mod changes the
108
OFPFF_SEND_FLOW_REM flag of the flow or flows that it matches to
109
the setting included in the flags of the flow_mod message.
111
- "honors OFPFF_CHECK_OVERLAP": Whether the OFPFF_CHECK_OVERLAP
112
flag in the flow_mod is significant.
114
- "updates idle_timeout" and "updates hard_timeout": Whether the
115
idle_timeout and hard_timeout in the flow_mod, respectively,
116
have an effect on the flow or flows matched by the flow_mod.
118
- "updates idle timer": Whether the flow_mod resets the per-flow
119
timer that measures how long a flow has been idle.
121
- "updates hard timer": Whether the flow_mod resets the per-flow
122
timer that measures how long it has been since a flow was
125
- "zeros counters": Whether the flow_mod resets per-flow packet
126
and byte counters to zero.
128
- "may add a new flow": Whether the flow_mod may add a new flow to
129
the flow table. (Obviously this is always true for "add"
130
commands but in some OpenFlow versions "modify" and
131
"modify-strict" can also add new flows.)
133
- "sends flow_removed message": Whether the flow_mod generates a
134
flow_removed message for the flow or flows that it affects.
136
An entry labeled "yes" means that the flow mod type does have the
137
indicated behavior, "---" means that it does not, an empty cell means
138
that the property is not applicable, and other values are explained
145
ADD MODIFY STRICT DELETE STRICT
146
=== ====== ====== ====== ======
147
match on priority yes --- yes --- yes
148
match on out_port --- --- --- yes yes
149
match on flow_cookie --- --- --- --- ---
150
match on table_id --- --- --- --- ---
151
controller chooses table_id --- --- ---
152
updates flow_cookie yes yes yes
153
updates OFPFF_SEND_FLOW_REM yes + +
154
honors OFPFF_CHECK_OVERLAP yes + +
155
updates idle_timeout yes + +
156
updates hard_timeout yes + +
157
resets idle timer yes + +
158
resets hard timer yes yes yes
159
zeros counters yes + +
160
may add a new flow yes yes yes
161
sends flow_removed message --- --- --- % %
163
(+) "modify" and "modify-strict" only take these actions when they
164
create a new flow, not when they update an existing flow.
166
(%) "delete" and "delete_strict" generates a flow_removed message if
167
the deleted flow or flows have the OFPFF_SEND_FLOW_REM flag set.
168
(Each controller can separately control whether it wants to
169
receive the generated messages.)
174
OpenFlow 1.1 makes these changes:
176
- The controller now must specify the table_id of the flow match
177
searched and into which a flow may be inserted. Behavior for a
178
table_id of 255 is undefined.
180
- A flow_mod, except an "add", can now match on the flow_cookie.
182
- When a flow_mod matches on the flow_cookie, "modify" and
183
"modify-strict" never insert a new flow.
186
ADD MODIFY STRICT DELETE STRICT
187
=== ====== ====== ====== ======
188
match on priority yes --- yes --- yes
189
match on out_port --- --- --- yes yes
190
match on flow_cookie --- yes yes yes yes
191
match on table_id yes yes yes yes yes
192
controller chooses table_id yes yes yes
193
updates flow_cookie yes --- ---
194
updates OFPFF_SEND_FLOW_REM yes + +
195
honors OFPFF_CHECK_OVERLAP yes + +
196
updates idle_timeout yes + +
197
updates hard_timeout yes + +
198
resets idle timer yes + +
199
resets hard timer yes yes yes
200
zeros counters yes + +
201
may add a new flow yes # #
202
sends flow_removed message --- --- --- % %
204
(+) "modify" and "modify-strict" only take these actions when they
205
create a new flow, not when they update an existing flow.
207
(%) "delete" and "delete_strict" generates a flow_removed message if
208
the deleted flow or flows have the OFPFF_SEND_FLOW_REM flag set.
209
(Each controller can separately control whether it wants to
210
receive the generated messages.)
212
(#) "modify" and "modify-strict" only add a new flow if the flow_mod
213
does not match on any bits of the flow cookie
218
OpenFlow 1.2 makes these changes:
220
- Only "add" commands ever add flows, "modify" and "modify-strict"
223
- A new flag OFPFF_RESET_COUNTS now controls whether "modify" and
224
"modify-strict" reset counters, whereas previously they never
225
reset counters (except when they inserted a new flow).
228
ADD MODIFY STRICT DELETE STRICT
229
=== ====== ====== ====== ======
230
match on priority yes --- yes --- yes
231
match on out_port --- --- --- yes yes
232
match on flow_cookie --- yes yes yes yes
233
match on table_id yes yes yes yes yes
234
controller chooses table_id yes yes yes
235
updates flow_cookie yes --- ---
236
updates OFPFF_SEND_FLOW_REM yes --- ---
237
honors OFPFF_CHECK_OVERLAP yes --- ---
238
updates idle_timeout yes --- ---
239
updates hard_timeout yes --- ---
240
resets idle timer yes --- ---
241
resets hard timer yes yes yes
242
zeros counters yes & &
243
may add a new flow yes --- ---
244
sends flow_removed message --- --- --- % %
246
(%) "delete" and "delete_strict" generates a flow_removed message if
247
the deleted flow or flows have the OFPFF_SEND_FLOW_REM flag set.
248
(Each controller can separately control whether it wants to
249
receive the generated messages.)
251
(&) "modify" and "modify-strict" reset counters if the
252
OFPFF_RESET_COUNTS flag is specified.
257
OpenFlow 1.3 makes these changes:
259
- Behavior for a table_id of 255 is now defined, for "delete" and
260
"delete-strict" commands, as meaning to delete from all tables.
261
A table_id of 255 is now explicitly invalid for other commands.
263
- New flags OFPFF_NO_PKT_COUNTS and OFPFF_NO_BYT_COUNTS for "add"
266
The table for 1.3 is the same as the one shown above for 1.2.
272
OpenFlow 1.4 does not change flow_mod semantics.
278
The OpenFlow 1.1 specification for OFPT_PACKET_IN is confusing. The
279
definition in OF1.1 openflow.h is[*]:
281
/* Packet received on port (datapath -> controller). */
282
struct ofp_packet_in {
283
struct ofp_header header;
284
uint32_t buffer_id; /* ID assigned by datapath. */
285
uint32_t in_port; /* Port on which frame was received. */
286
uint32_t in_phy_port; /* Physical Port on which frame was received. */
287
uint16_t total_len; /* Full length of frame. */
288
uint8_t reason; /* Reason packet is being sent (one of OFPR_*) */
289
uint8_t table_id; /* ID of the table that was looked up */
290
uint8_t data[0]; /* Ethernet frame, halfway through 32-bit word,
291
so the IP header is 32-bit aligned. The
292
amount of data is inferred from the length
293
field in the header. Because of padding,
294
offsetof(struct ofp_packet_in, data) ==
295
sizeof(struct ofp_packet_in) - 2. */
297
OFP_ASSERT(sizeof(struct ofp_packet_in) == 24);
299
The confusing part is the comment on the data[] member. This comment
300
is a leftover from OF1.0 openflow.h, in which the comment was correct:
301
sizeof(struct ofp_packet_in) is 20 in OF1.0 and offsetof(struct
302
ofp_packet_in, data) is 18. When OF1.1 was written, the structure
303
members were changed but the comment was carelessly not updated, and
304
the comment became wrong: sizeof(struct ofp_packet_in) and
305
offsetof(struct ofp_packet_in, data) are both 24 in OF1.1.
307
That leaves the question of how to implement ofp_packet_in in OF1.1.
308
The OpenFlow reference implementation for OF1.1 does not include any
309
padding, that is, the first byte of the encapsulated frame immediately
310
follows the 'table_id' member without a gap. Open vSwitch therefore
311
implements it the same way for compatibility.
313
For an earlier discussion, please see the thread archived at:
314
https://mailman.stanford.edu/pipermail/openflow-discuss/2011-August/002604.html
316
[*] The quoted definition is directly from OF1.1. Definitions used
317
inside OVS omit the 8-byte ofp_header members, so the sizes in
318
this discussion are 8 bytes larger than those declared in OVS
325
The 802.1Q VLAN header causes more trouble than any other 4 bytes in
326
networking. More specifically, three versions of OpenFlow and Open
327
vSwitch have among them four different ways to match the contents and
328
presence of the VLAN header. The following table describes how each
331
Match NXM OF1.0 OF1.1 OF1.2
332
----- --------- ----------- ----------- ------------
333
[1] 0000/0000 ????/1,??/? ????/1,??/? 0000/0000,--
334
[2] 0000/ffff ffff/0,??/? ffff/0,??/? 0000/ffff,--
335
[3] 1xxx/1fff 0xxx/0,??/1 0xxx/0,??/1 1xxx/ffff,--
336
[4] z000/f000 ????/1,0y/0 fffe/0,0y/0 1000/1000,0y
337
[5] zxxx/ffff 0xxx/0,0y/0 0xxx/0,0y/0 1xxx/ffff,0y
338
[6] 0000/0fff <none> <none> <none>
339
[7] 0000/f000 <none> <none> <none>
340
[8] 0000/efff <none> <none> <none>
341
[9] 1001/1001 <none> <none> 1001/1001,--
342
[10] 3000/3000 <none> <none> <none>
344
Each column is interpreted as follows.
346
- Match: See the list below.
348
- NXM: xxxx/yyyy means NXM_OF_VLAN_TCI_W with value xxxx and mask
349
yyyy. A mask of 0000 is equivalent to omitting
350
NXM_OF_VLAN_TCI(_W), a mask of ffff is equivalent to
353
- OF1.0 and OF1.1: wwww/x,yy/z means dl_vlan wwww, OFPFW_DL_VLAN
354
x, dl_vlan_pcp yy, and OFPFW_DL_VLAN_PCP z. ? means that the
355
given nibble is ignored (and conventionally 0 for wwww or yy,
356
conventionally 1 for x or z). <none> means that the given match
359
- OF1.2: xxxx/yyyy,zz means OXM_OF_VLAN_VID_W with value xxxx and
360
mask yyyy, and OXM_OF_VLAN_PCP (which is not maskable) with
361
value zz. A mask of 0000 is equivalent to omitting
362
OXM_OF_VLAN_VID(_W), a mask of ffff is equivalent to
363
OXM_OF_VLAN_VID. -- means that OXM_OF_VLAN_PCP is omitted.
364
<none> means that the given match is not supported.
368
[1] Matches any packet, that is, one without an 802.1Q header or with
369
an 802.1Q header with any TCI value.
371
[2] Matches only packets without an 802.1Q header.
373
NXM: Any match with (vlan_tci == 0) and (vlan_tci_mask & 0x1000)
374
!= 0 is equivalent to the one listed in the table.
376
OF1.0: The spec doesn't define behavior if dl_vlan is set to
377
0xffff and OFPFW_DL_VLAN_PCP is not set.
379
OF1.1: The spec says explicitly to ignore dl_vlan_pcp when
380
dl_vlan is set to 0xffff.
382
OF1.2: The spec doesn't say what should happen if (vlan_vid == 0)
383
and (vlan_vid_mask & 0x1000) != 0 but (vlan_vid_mask != 0x1000),
384
but it would be straightforward to also interpret as [2].
386
[3] Matches only packets that have an 802.1Q header with VID xxx (and
389
[4] Matches only packets that have an 802.1Q header with PCP y (and
392
NXM: z is ((y << 1) | 1).
394
OF1.0: The spec isn't very clear, but OVS implements it this way.
396
OF1.2: Presumably other masks such that (vlan_vid_mask & 0x1fff)
397
== 0x1000 would also work, but the spec doesn't define their
400
[5] Matches only packets that have an 802.1Q header with VID xxx and
403
NXM: z is ((y << 1) | 1).
405
OF1.2: Presumably other masks such that (vlan_vid_mask & 0x1fff)
406
== 0x1fff would also work.
408
[6] Matches packets with no 802.1Q header or with an 802.1Q header
409
with a VID of 0. Only possible with NXM.
411
[7] Matches packets with no 802.1Q header or with an 802.1Q header
412
with a PCP of 0. Only possible with NXM.
414
[8] Matches packets with no 802.1Q header or with an 802.1Q header
415
with both VID and PCP of 0. Only possible with NXM.
417
[9] Matches only packets that have an 802.1Q header with an
418
odd-numbered VID (and any PCP). Only possible with NXM and
419
OF1.2. (This is just an example; one can match on any desired
422
[10] Matches only packets that have an 802.1Q header with an
423
odd-numbered PCP (and any VID). Only possible with NXM. (This
424
is just an example; one can match on any desired VID bit
429
- OF1.2: The top three bits of OXM_OF_VLAN_VID are fixed to zero,
430
so bits 13, 14, and 15 in the masks listed in the table may be
431
set to arbitrary values, as long as the corresponding value bits
432
are also zero. The suggested ffff mask for [2], [3], and [5]
433
allows a shorter OXM representation (the mask is omitted) than
434
the minimal 1fff mask.
440
OpenFlow 1.0 and later versions have the concept of a "flow cookie",
441
which is a 64-bit integer value attached to each flow. The treatment
442
of the flow cookie has varied greatly across OpenFlow versions,
447
- OFPFC_ADD set the cookie in the flow that it added.
449
- OFPFC_MODIFY and OFPFC_MODIFY_STRICT updated the cookie for
450
the flow or flows that it modified.
452
- OFPST_FLOW messages included the flow cookie.
454
- OFPT_FLOW_REMOVED messages reported the cookie of the flow
457
OpenFlow 1.1 made the following changes:
459
- Flow mod operations OFPFC_MODIFY, OFPFC_MODIFY_STRICT,
460
OFPFC_DELETE, and OFPFC_DELETE_STRICT, plus flow stats
461
requests and aggregate stats requests, gained the ability to
462
match on flow cookies with an arbitrary mask.
464
- OFPFC_MODIFY and OFPFC_MODIFY_STRICT were changed to add a
465
new flow, in the case of no match, only if the flow table
466
modification operation did not match on the cookie field.
467
(In OpenFlow 1.0, modify operations always added a new flow
468
when there was no match.)
470
- OFPFC_MODIFY and OFPFC_MODIFY_STRICT no longer updated flow
473
OpenFlow 1.2 made the following changes:
475
- OFPC_MODIFY and OFPFC_MODIFY_STRICT were changed to never
476
add a new flow, regardless of whether the flow cookie was
479
Open vSwitch support for OpenFlow 1.0 implements the OpenFlow 1.0
480
behavior with the following extensions:
482
- An NXM extension field NXM_NX_COOKIE(_W) allows the NXM
483
versions of OFPFC_MODIFY, OFPFC_MODIFY_STRICT, OFPFC_DELETE,
484
and OFPFC_DELETE_STRICT flow_mods, plus flow stats requests
485
and aggregate stats requests, to match on flow cookies with
486
arbitrary masks. This is much like the equivalent OpenFlow
489
- Like OpenFlow 1.1, OFPC_MODIFY and OFPFC_MODIFY_STRICT add a
490
new flow if there is no match and the mask is zero (or not
493
- The "cookie" field in OFPT_FLOW_MOD and NXT_FLOW_MOD messages
494
is used as the cookie value for OFPFC_ADD commands, as
495
described in OpenFlow 1.0. For OFPFC_MODIFY and
496
OFPFC_MODIFY_STRICT commands, the "cookie" field is used as a
497
new cookie for flows that match unless it is UINT64_MAX, in
498
which case the flow's cookie is not updated.
500
- NXT_PACKET_IN (the Nicira extended version of
501
OFPT_PACKET_IN) reports the cookie of the rule that
502
generated the packet, or all-1-bits if no rule generated the
503
packet. (Older versions of OVS used all-0-bits instead of
506
The following table shows the handling of different protocols when
507
receiving OFPFC_MODIFY and OFPFC_MODIFY_STRICT messages. A mask of 0
508
indicates either an explicit mask of zero or an implicit one by not
509
specifying the NXM_NX_COOKIE(_W) field.
511
Match Update Add on miss Add on miss
512
cookie cookie mask!=0 mask==0
513
====== ====== =========== ===========
514
OpenFlow 1.0 no yes <always add on miss>
515
OpenFlow 1.1 yes no no yes
516
OpenFlow 1.2 yes no no no
519
* Updates the flow's cookie unless the "cookie" field is UINT64_MAX.
522
Multiple Table Support
523
======================
525
OpenFlow 1.0 has only rudimentary support for multiple flow tables.
526
Notably, OpenFlow 1.0 does not allow the controller to specify the
527
flow table to which a flow is to be added. Open vSwitch adds an
528
extension for this purpose, which is enabled on a per-OpenFlow
529
connection basis using the NXT_FLOW_MOD_TABLE_ID message. When the
530
extension is enabled, the upper 8 bits of the 'command' member in an
531
OFPT_FLOW_MOD or NXT_FLOW_MOD message designates the table to which a
534
The Open vSwitch software switch implementation offers 255 flow
535
tables. On packet ingress, only the first flow table (table 0) is
536
searched, and the contents of the remaining tables are not considered
537
in any way. Tables other than table 0 only come into play when an
538
NXAST_RESUBMIT_TABLE action specifies another table to search.
540
Tables 128 and above are reserved for use by the switch itself.
541
Controllers should use only tables 0 through 127.
547
Open vSwitch supports stateless handling of IPv6 packets. Flows can be
548
written to support matching TCP, UDP, and ICMPv6 headers within an IPv6
549
packet. Deeper matching of some Neighbor Discovery messages is also
552
IPv6 was not designed to interact well with middle-boxes. This,
553
combined with Open vSwitch's stateless nature, have affected the
554
processing of IPv6 traffic, which is detailed below.
559
The base IPv6 header is incredibly simple with the intention of only
560
containing information relevant for routing packets between two
561
endpoints. IPv6 relies heavily on the use of extension headers to
562
provide any other functionality. Unfortunately, the extension headers
563
were designed in such a way that it is impossible to move to the next
564
header (including the layer-4 payload) unless the current header is
567
Open vSwitch will process the following extension headers and continue
570
* Fragment (see the next section)
571
* AH (Authentication Header)
574
* Destination Options
576
When a header is encountered that is not in that list, it is considered
577
"terminal". A terminal header's IPv6 protocol value is stored in
578
"nw_proto" for matching purposes. If a terminal header is TCP, UDP, or
579
ICMPv6, the packet will be further processed in an attempt to extract
585
IPv6 requires that every link in the internet have an MTU of 1280 octets
586
or greater (RFC 2460). As such, a terminal header (as described above in
587
"Extension Headers") in the first fragment should generally be
588
reachable. In this case, the terminal header's IPv6 protocol type is
589
stored in the "nw_proto" field for matching purposes. If a terminal
590
header cannot be found in the first fragment (one with a fragment offset
591
of zero), the "nw_proto" field is set to 0. Subsequent fragments (those
592
with a non-zero fragment offset) have the "nw_proto" field set to the
593
IPv6 protocol type for fragments (44).
598
An IPv6 jumbogram (RFC 2675) is a packet containing a payload longer
599
than 65,535 octets. A jumbogram is only relevant in subnets with a link
600
MTU greater than 65,575 octets, and are not required to be supported on
601
nodes that do not connect to link with such large MTUs. Currently, Open
602
vSwitch doesn't process jumbograms.
611
An OpenFlow switch must establish and maintain a TCP network
612
connection to its controller. There are two basic ways to categorize
613
the network that this connection traverses: either it is completely
614
separate from the one that the switch is otherwise controlling, or its
615
path may overlap the network that the switch controls. We call the
616
former case "out-of-band control", the latter case "in-band control".
618
Out-of-band control has the following benefits:
620
- Simplicity: Out-of-band control slightly simplifies the switch
623
- Reliability: Excessive switch traffic volume cannot interfere
624
with control traffic.
626
- Integrity: Machines not on the control network cannot
627
impersonate a switch or a controller.
629
- Confidentiality: Machines not on the control network cannot
630
snoop on control traffic.
632
In-band control, on the other hand, has the following advantages:
634
- No dedicated port: There is no need to dedicate a physical
635
switch port to control, which is important on switches that have
636
few ports (e.g. wireless routers, low-end embedded platforms).
638
- No dedicated network: There is no need to build and maintain a
639
separate control network. This is important in many
640
environments because it reduces proliferation of switches and
643
Open vSwitch supports both out-of-band and in-band control. This
644
section describes the principles behind in-band control. See the
645
description of the Controller table in ovs-vswitchd.conf.db(5) to
646
configure OVS for in-band control.
651
The fundamental principle of in-band control is that an OpenFlow
652
switch must recognize and switch control traffic without involving the
653
OpenFlow controller. All the details of implementing in-band control
654
are special cases of this principle.
656
The rationale for this principle is simple. If the switch does not
657
handle in-band control traffic itself, then it will be caught in a
658
contradiction: it must contact the controller, but it cannot, because
659
only the controller can set up the flows that are needed to contact
662
The following points describe important special cases of this
665
- In-band control must be implemented regardless of whether the
668
It is tempting to implement the in-band control rules only when
669
the switch is not connected to the controller, using the
670
reasoning that the controller should have complete control once
671
it has established a connection with the switch.
673
This does not work in practice. Consider the case where the
674
switch is connected to the controller. Occasionally it can
675
happen that the controller forgets or otherwise needs to obtain
676
the MAC address of the switch. To do so, the controller sends a
677
broadcast ARP request. A switch that implements the in-band
678
control rules only when it is disconnected will then send an
679
OFPT_PACKET_IN message up to the controller. The controller will
680
be unable to respond, because it does not know the MAC address of
681
the switch. This is a deadlock situation that can only be
682
resolved by the switch noticing that its connection to the
683
controller has hung and reconnecting.
685
- In-band control must override flows set up by the controller.
687
It is reasonable to assume that flows set up by the OpenFlow
688
controller should take precedence over in-band control, on the
689
basis that the controller should be in charge of the switch.
691
Again, this does not work in practice. Reasonable controller
692
implementations may set up a "last resort" fallback rule that
693
wildcards every field and, e.g., sends it up to the controller or
694
discards it. If a controller does that, then it will isolate
695
itself from the switch.
697
- The switch must recognize all control traffic.
699
The fundamental principle of in-band control states, in part,
700
that a switch must recognize control traffic without involving
701
the OpenFlow controller. More specifically, the switch must
702
recognize *all* control traffic. "False negatives", that is,
703
packets that constitute control traffic but that the switch does
704
not recognize as control traffic, lead to control traffic storms.
706
Consider an OpenFlow switch that only recognizes control packets
707
sent to or from that switch. Now suppose that two switches of
708
this type, named A and B, are connected to ports on an Ethernet
709
hub (not a switch) and that an OpenFlow controller is connected
710
to a third hub port. In this setup, control traffic sent by
711
switch A will be seen by switch B, which will send it to the
712
controller as part of an OFPT_PACKET_IN message. Switch A will
713
then see the OFPT_PACKET_IN message's packet, re-encapsulate it
714
in another OFPT_PACKET_IN, and send it to the controller. Switch
715
B will then see that OFPT_PACKET_IN, and so on in an infinite
718
Incidentally, the consequences of "false positives", where
719
packets that are not control traffic are nevertheless recognized
720
as control traffic, are much less severe. The controller will
721
not be able to control their behavior, but the network will
722
remain in working order. False positives do constitute a
725
- The switch should use echo-requests to detect disconnection.
727
TCP will notice that a connection has hung, but this can take a
728
considerable amount of time. For example, with default settings
729
the Linux kernel TCP implementation will retransmit for between
730
13 and 30 minutes, depending on the connection's retransmission
731
timeout, according to kernel documentation. This is far too long
732
for a switch to be disconnected, so an OpenFlow switch should
733
implement its own connection timeout. OpenFlow OFPT_ECHO_REQUEST
734
messages are the best way to do this, since they test the
735
OpenFlow connection itself.
740
This section describes how Open vSwitch implements in-band control.
741
Correctly implementing in-band control has proven difficult due to its
742
many subtleties, and has thus gone through many iterations. Please
743
read through and understand the reasoning behind the chosen rules
744
before making modifications.
746
Open vSwitch implements in-band control as "hidden" flows, that is,
747
flows that are not visible through OpenFlow, and at a higher priority
748
than wildcarded flows can be set up through OpenFlow. This is done so
749
that the OpenFlow controller cannot interfere with them and possibly
750
break connectivity with its switches. It is possible to see all
751
flows, including in-band ones, with the ovs-appctl "bridge/dump-flows"
754
The Open vSwitch implementation of in-band control can hide traffic to
755
arbitrary "remotes", where each remote is one TCP port on one IP address.
756
Currently the remotes are automatically configured as the in-band OpenFlow
757
controllers plus the OVSDB managers, if any. (The latter is a requirement
758
because OVSDB managers are responsible for configuring OpenFlow controllers,
759
so if the manager cannot be reached then OpenFlow cannot be reconfigured.)
761
The following rules (with the OFPP_NORMAL action) are set up on any bridge
762
that has any remotes:
764
(a) DHCP requests sent from the local port.
765
(b) ARP replies to the local port's MAC address.
766
(c) ARP requests from the local port's MAC address.
768
In-band also sets up the following rules for each unique next-hop MAC
769
address for the remotes' IPs (the "next hop" is either the remote
770
itself, if it is on a local subnet, or the gateway to reach the remote):
772
(d) ARP replies to the next hop's MAC address.
773
(e) ARP requests from the next hop's MAC address.
775
In-band also sets up the following rules for each unique remote IP address:
777
(f) ARP replies containing the remote's IP address as a target.
778
(g) ARP requests containing the remote's IP address as a source.
780
In-band also sets up the following rules for each unique remote (IP,port)
783
(h) TCP traffic to the remote's IP and port.
784
(i) TCP traffic from the remote's IP and port.
786
The goal of these rules is to be as narrow as possible to allow a
787
switch to join a network and be able to communicate with the
788
remotes. As mentioned earlier, these rules have higher priority
789
than the controller's rules, so if they are too broad, they may
790
prevent the controller from implementing its policy. As such,
791
in-band actively monitors some aspects of flow and packet processing
792
so that the rules can be made more precise.
794
In-band control monitors attempts to add flows into the datapath that
795
could interfere with its duties. The datapath only allows exact
796
match entries, so in-band control is able to be very precise about
797
the flows it prevents. Flows that miss in the datapath are sent to
798
userspace to be processed, so preventing these flows from being
799
cached in the "fast path" does not affect correctness. The only type
800
of flow that is currently prevented is one that would prevent DHCP
801
replies from being seen by the local port. For example, a rule that
802
forwarded all DHCP traffic to the controller would not be allowed,
803
but one that forwarded to all ports (including the local port) would.
805
As mentioned earlier, packets that miss in the datapath are sent to
806
the userspace for processing. The userspace has its own flow table,
807
the "classifier", so in-band checks whether any special processing
808
is needed before the classifier is consulted. If a packet is a DHCP
809
response to a request from the local port, the packet is forwarded to
810
the local port, regardless of the flow table. Note that this requires
811
L7 processing of DHCP replies to determine whether the 'chaddr' field
812
matches the MAC address of the local port.
814
It is interesting to note that for an L3-based in-band control
815
mechanism, the majority of rules are devoted to ARP traffic. At first
816
glance, some of these rules appear redundant. However, each serves an
817
important role. First, in order to determine the MAC address of the
818
remote side (controller or gateway) for other ARP rules, we must allow
819
ARP traffic for our local port with rules (b) and (c). If we are
820
between a switch and its connection to the remote, we have to
821
allow the other switch's ARP traffic to through. This is done with
822
rules (d) and (e), since we do not know the addresses of the other
823
switches a priori, but do know the remote's or gateway's. Finally,
824
if the remote is running in a local guest VM that is not reached
825
through the local port, the switch that is connected to the VM must
826
allow ARP traffic based on the remote's IP address, since it will
827
not know the MAC address of the local port that is sending the traffic
828
or the MAC address of the remote in the guest VM.
830
With a few notable exceptions below, in-band should work in most
831
network setups. The following are considered "supported' in the
832
current implementation:
834
- Locally Connected. The switch and remote are on the same
835
subnet. This uses rules (a), (b), (c), (h), and (i).
837
- Reached through Gateway. The switch and remote are on
838
different subnets and must go through a gateway. This uses
839
rules (a), (b), (c), (h), and (i).
841
- Between Switch and Remote. This switch is between another
842
switch and the remote, and we want to allow the other
843
switch's traffic through. This uses rules (d), (e), (h), and
844
(i). It uses (b) and (c) indirectly in order to know the MAC
845
address for rules (d) and (e). Note that DHCP for the other
846
switch will not work unless an OpenFlow controller explicitly lets this
847
switch pass the traffic.
849
- Between Switch and Gateway. This switch is between another
850
switch and the gateway, and we want to allow the other switch's
851
traffic through. This uses the same rules and logic as the
852
"Between Switch and Remote" configuration described earlier.
854
- Remote on Local VM. The remote is a guest VM on the
855
system running in-band control. This uses rules (a), (b), (c),
858
- Remote on Local VM with Different Networks. The remote
859
is a guest VM on the system running in-band control, but the
860
local port is not used to connect to the remote. For
861
example, an IP address is configured on eth0 of the switch. The
862
remote's VM is connected through eth1 of the switch, but an
863
IP address has not been configured for that port on the switch.
864
As such, the switch will use eth0 to connect to the remote,
865
and eth1's rules about the local port will not work. In the
866
example, the switch attached to eth0 would use rules (a), (b),
867
(c), (h), and (i) on eth0. The switch attached to eth1 would use
868
rules (f), (g), (h), and (i).
870
The following are explicitly *not* supported by in-band control:
872
- Specify Remote by Name. Currently, the remote must be
873
identified by IP address. A naive approach would be to permit
874
all DNS traffic. Unfortunately, this would prevent the
875
controller from defining any policy over DNS. Since switches
876
that are located behind us need to connect to the remote,
877
in-band cannot simply add a rule that allows DNS traffic from
878
the local port. The "correct" way to support this is to parse
879
DNS requests to allow all traffic related to a request for the
880
remote's name through. Due to the potential security
881
problems and amount of processing, we decided to hold off for
884
- Differing Remotes for Switches. All switches must know
885
the L3 addresses for all the remotes that other switches
886
may use, since rules need to be set up to allow traffic related
887
to those remotes through. See rules (f), (g), (h), and (i).
889
- Differing Routes for Switches. In order for the switch to
890
allow other switches to connect to a remote through a
891
gateway, it allows the gateway's traffic through with rules (d)
892
and (e). If the routes to the remote differ for the two
893
switches, we will not know the MAC address of the alternate
900
It seems likely that many controllers, at least at startup, use the
901
OpenFlow "flow statistics" request to obtain existing flows, then
902
compare the flows' actions against the actions that they expect to
903
find. Before version 1.8.0, Open vSwitch always returned exact,
904
byte-for-byte copies of the actions that had been added to the flow
905
table. The current version of Open vSwitch does not always do this in
906
some exceptional cases. This section lists the exceptions that
907
controller authors must keep in mind if they compare actual actions
908
against desired actions in a bytewise fashion:
910
- Open vSwitch zeros padding bytes in action structures,
911
regardless of their values when the flows were added.
913
- Open vSwitch "normalizes" the instructions in OpenFlow 1.1
914
(and later) in the following way:
916
* OVS sorts the instructions into the following order:
917
Apply-Actions, Clear-Actions, Write-Actions,
918
Write-Metadata, Goto-Table.
920
* OVS drops Apply-Actions instructions that have empty
923
* OVS drops Write-Actions instructions that have empty
926
Please report other discrepancies, if you notice any, so that we can
927
fix or document them.
933
Suggestions to improve Open vSwitch are welcome at discuss@openvswitch.org.
added
removed
DESIGN
