2 * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014, 2015, 2016 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
17 #ifndef OPENFLOW_NICIRA_EXT_H
18 #define OPENFLOW_NICIRA_EXT_H 1
20 #include <openflow/openflow.h>
21 #include <openvswitch/types.h>
23 /* The following vendor extensions, proposed by Nicira, are not yet
24 * standardized, so they are not included in openflow.h. Some of them may be
25 * suitable for standardization; others we never expect to standardize. */
28 /* Nicira vendor-specific error messages extension.
30 * OpenFlow 1.0 has a set of predefined error types (OFPET_*) and codes (which
31 * are specific to each type). It does not have any provision for
32 * vendor-specific error codes, and it does not even provide "generic" error
33 * codes that can apply to problems not anticipated by the OpenFlow
34 * specification authors.
36 * This extension attempts to address the problem by adding a generic "error
37 * vendor extension". The extension works as follows: use NXET_VENDOR as type
38 * and NXVC_VENDOR_ERROR as code, followed by struct nx_vendor_error with
39 * vendor-specific details, followed by at least 64 bytes of the failed
42 * It would be better to have a type-specific vendor extension, e.g. so that
43 * OFPET_BAD_ACTION could be used with vendor-specific code values. But
44 * OFPET_BAD_ACTION and most other standardized types already specify that
45 * their 'data' values are (the start of) the OpenFlow message being replied
46 * to, so there is no room to insert a vendor ID.
48 * Currently this extension is only implemented by Open vSwitch, but it seems
49 * like a reasonable candidate for future standardization.
52 /* This is a random number to avoid accidental collision with any other
53 * vendor's extension. */
54 #define NXET_VENDOR 0xb0c2
56 /* ofp_error msg 'code' values for NXET_VENDOR. */
58 NXVC_VENDOR_ERROR /* 'data' contains struct nx_vendor_error. */
61 /* 'data' for 'type' == NXET_VENDOR, 'code' == NXVC_VENDOR_ERROR. */
62 struct nx_vendor_error {
63 ovs_be32 vendor; /* Vendor ID as in struct ofp_vendor_header. */
64 ovs_be16 type; /* Vendor-defined type. */
65 ovs_be16 code; /* Vendor-defined subtype. */
66 /* Followed by at least the first 64 bytes of the failed request. */
69 /* Nicira vendor requests and replies. */
71 /* Fields to use when hashing flows. */
73 /* Ethernet source address (NXM_OF_ETH_SRC) only. */
74 NX_HASH_FIELDS_ETH_SRC,
76 /* L2 through L4, symmetric across src/dst. Specifically, each of the
77 * following fields, if present, is hashed (slashes separate symmetric
80 * - NXM_OF_ETH_DST / NXM_OF_ETH_SRC
82 * - The VID bits from NXM_OF_VLAN_TCI, ignoring PCP and CFI.
84 * - NXM_OF_IP_SRC / NXM_OF_IP_DST
85 * - NXM_OF_TCP_SRC / NXM_OF_TCP_DST
87 NX_HASH_FIELDS_SYMMETRIC_L4,
89 /* L3+L4 only, including the following fields:
92 * - NXM_OF_IP_SRC / NXM_OF_IP_DST
93 * - NXM_OF_SCTP_SRC / NXM_OF_SCTP_DST
94 * - NXM_OF_TCP_SRC / NXM_OF_TCP_DST
96 NX_HASH_FIELDS_SYMMETRIC_L3L4,
98 /* L3+L4 only with UDP ports, including the following fields:
101 * - NXM_OF_IP_SRC / NXM_OF_IP_DST
102 * - NXM_OF_SCTP_SRC / NXM_OF_SCTP_DST
103 * - NXM_OF_TCP_SRC / NXM_OF_TCP_DST
104 * - NXM_OF_UDP_SRC / NXM_OF_UDP_DST
106 NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP
111 /* This command enables or disables an Open vSwitch extension that allows a
112 * controller to specify the OpenFlow table to which a flow should be added,
113 * instead of having the switch decide which table is most appropriate as
114 * required by OpenFlow 1.0. Because NXM was designed as an extension to
115 * OpenFlow 1.0, the extension applies equally to ofp10_flow_mod and
116 * nx_flow_mod. By default, the extension is disabled.
118 * When this feature is enabled, Open vSwitch treats struct ofp10_flow_mod's
119 * and struct nx_flow_mod's 16-bit 'command' member as two separate fields.
120 * The upper 8 bits are used as the table ID, the lower 8 bits specify the
121 * command as usual. A table ID of 0xff is treated like a wildcarded table ID.
123 * The specific treatment of the table ID depends on the type of flow mod:
125 * - OFPFC_ADD: Given a specific table ID, the flow is always placed in that
126 * table. If an identical flow already exists in that table only, then it
127 * is replaced. If the flow cannot be placed in the specified table,
128 * either because the table is full or because the table cannot support
129 * flows of the given type, the switch replies with an OFPFMFC_TABLE_FULL
130 * error. (A controller can distinguish these cases by comparing the
131 * current and maximum number of entries reported in ofp_table_stats.)
133 * If the table ID is wildcarded, the switch picks an appropriate table
134 * itself. If an identical flow already exist in the selected flow table,
135 * then it is replaced. The choice of table might depend on the flows
136 * that are already in the switch; for example, if one table fills up then
137 * the switch might fall back to another one.
139 * - OFPFC_MODIFY, OFPFC_DELETE: Given a specific table ID, only flows
140 * within that table are matched and modified or deleted. If the table ID
141 * is wildcarded, flows within any table may be matched and modified or
144 * - OFPFC_MODIFY_STRICT, OFPFC_DELETE_STRICT: Given a specific table ID,
145 * only a flow within that table may be matched and modified or deleted.
146 * If the table ID is wildcarded and exactly one flow within any table
147 * matches, then it is modified or deleted; if flows in more than one
148 * table match, then none is modified or deleted.
150 struct nx_flow_mod_table_id {
151 uint8_t set; /* Nonzero to enable, zero to disable. */
154 OFP_ASSERT(sizeof(struct nx_flow_mod_table_id) == 8);
156 enum nx_packet_in_format {
157 NXPIF_STANDARD = 0, /* OFPT_PACKET_IN for this OpenFlow version. */
158 NXPIF_NXT_PACKET_IN = 1, /* NXT_PACKET_IN (since OVS v1.1). */
159 NXPIF_NXT_PACKET_IN2 = 2, /* NXT_PACKET_IN2 (since OVS v2.6). */
162 /* NXT_SET_PACKET_IN_FORMAT request.
164 * For any given OpenFlow version, Open vSwitch supports multiple formats for
165 * "packet-in" messages. The default is always the standard format for the
166 * OpenFlow version in question, but NXT_SET_PACKET_IN_FORMAT can be used to
167 * set an alternative format.
169 * From OVS v1.1 to OVS v2.5, this request was only honored for OpenFlow 1.0.
170 * Requests to set format NXPIF_NXT_PACKET_IN were accepted for OF1.1+ but they
171 * had no effect. (Requests to set formats other than NXPIF_STANDARD or
172 * NXPIF_NXT_PACKET_IN were rejected with OFPBRC_EPERM.)
174 * From OVS v2.6 onward, this request is honored for all OpenFlow versions.
176 struct nx_set_packet_in_format {
177 ovs_be32 format; /* One of NXPIF_*. */
179 OFP_ASSERT(sizeof(struct nx_set_packet_in_format) == 4);
181 /* NXT_PACKET_IN (analogous to OFPT_PACKET_IN).
183 * NXT_PACKET_IN is similar to the OpenFlow 1.2 OFPT_PACKET_IN. The
186 * - NXT_PACKET_IN includes the cookie of the rule that triggered the
187 * message. (OpenFlow 1.3 OFPT_PACKET_IN also includes the cookie.)
189 * - The metadata fields use NXM (instead of OXM) field numbers.
191 * Open vSwitch 1.9.0 and later omits metadata fields that are zero (as allowed
192 * by OpenFlow 1.2). Earlier versions included all implemented metadata
195 * Open vSwitch does not include non-metadata in the nx_match, because by
196 * definition that information can be found in the packet itself. The format
197 * and the standards allow this, however, so controllers should be prepared to
198 * tolerate future changes.
200 * The NXM format is convenient for reporting metadata values, but it is
201 * important not to interpret the format as matching against a flow, because it
202 * does not. Nothing is being matched; arbitrary metadata masks would not be
205 * Whereas in most cases a controller can expect to only get back NXM fields
206 * that it set up itself (e.g. flow dumps will ordinarily report only NXM
207 * fields from flows that the controller added), NXT_PACKET_IN messages might
208 * contain fields that the controller does not understand, because the switch
209 * might support fields (new registers, new protocols, etc.) that the
210 * controller does not. The controller must prepared to tolerate these.
212 * The 'cookie' field has no meaning when 'reason' is OFPR_NO_MATCH. In this
213 * case it should be UINT64_MAX. */
214 struct nx_packet_in {
215 ovs_be32 buffer_id; /* ID assigned by datapath. */
216 ovs_be16 total_len; /* Full length of frame. */
217 uint8_t reason; /* Reason packet is sent (one of OFPR_*). */
218 uint8_t table_id; /* ID of the table that was looked up. */
219 ovs_be64 cookie; /* Cookie of the rule that was looked up. */
220 ovs_be16 match_len; /* Size of nx_match. */
221 uint8_t pad[6]; /* Align to 64-bits. */
223 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
224 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
225 * all-zero bytes, then
226 * - Exactly 2 all-zero padding bytes, then
227 * - An Ethernet frame whose length is inferred from nxh.header.length.
229 * The padding bytes preceding the Ethernet frame ensure that the IP
230 * header (if any) following the Ethernet header is 32-bit aligned. */
232 /* uint8_t nxm_fields[...]; */ /* NXM headers. */
233 /* uint8_t pad[2]; */ /* Align to 64 bit + 16 bit. */
234 /* uint8_t data[0]; */ /* Ethernet frame. */
236 OFP_ASSERT(sizeof(struct nx_packet_in) == 24);
241 * NXT_PACKET_IN2 is conceptually similar to OFPT_PACKET_IN but it is expressed
242 * as an extensible set of properties instead of using a fixed structure.
244 * Added in Open vSwitch 2.6
250 * When a "controller" action specifies the "pause" flag, the controller action
251 * freezes the packet's trip through Open vSwitch flow tables and serializes
252 * that state into the packet-in message as a "continuation". The controller
253 * can later send the continuation back to the switch, which will restart the
254 * packet's traversal from the point where it was interrupted. This permits an
255 * OpenFlow controller to interpose on a packet midway through processing in
258 * Continuations fit into packet processing this way:
260 * 1. A packet ingresses into Open vSwitch, which runs it through the OpenFlow
263 * 2. An OpenFlow flow executes a "controller" action that includes the "pause"
264 * flag. Open vSwitch serializes the packet processing state and sends it,
265 * as an NXT_PACKET_IN2 that includes an additional NXPINT_CONTINUATION
266 * property (the continuation), to the OpenFlow controller.
268 * (The controller must use NXAST_CONTROLLER2 to generate the packet-in,
269 * because only this form of the "controller" action has a "pause" flag.
270 * Similarly, the controller must use NXT_SET_PACKET_IN_FORMAT to select
271 * NXT_PACKET_IN2 as the packet-in format, because this is the only format
272 * that supports continuation passing.)
274 * 3. The controller receives the NXT_PACKET_IN2 and processes it. The
275 * controller can interpret and, if desired, modify some of the contents of
276 * the packet-in, such as the packet and the metadata being processed.
278 * 4. The controller sends the continuation back to the switch, using an
279 * NXT_RESUME message. Packet processing resumes where it left off.
281 * The controller might change the pipeline configuration concurrently with
282 * steps 2 through 4. For example, it might add or remove OpenFlow flows. If
283 * that happens, then the packet will experience a mix of processing from the
284 * two configurations, that is, the initial processing (before
285 * NXAST_CONTROLLER2) uses the initial flow table, and the later processing
286 * (after NXT_RESUME) uses the later flow table. This means that the
287 * controller needs to take care to avoid incompatible pipeline changes while
288 * processing continuations.
290 * External side effects (e.g. "output") of OpenFlow actions processed before
291 * NXAST_CONTROLLER2 is encountered might be executed during step 2 or step 4,
292 * and the details may vary among Open vSwitch features and versions. Thus, a
293 * controller that wants to make sure that side effects are executed must pass
294 * the continuation back to the switch, that is, must not skip step 4.
296 * Architecturally, continuations may be "stateful" or "stateless", that is,
297 * they may or may not refer to buffered state maintained in Open vSwitch.
298 * This means that a controller should not attempt to resume a given
299 * continuations more than once (because the switch might have discarded the
300 * buffered state after the first use). For the same reason, continuations
301 * might become "stale" if the controller takes too long to resume them
302 * (because the switch might have discarded old buffered state). Taken
303 * together with the previous note, this means that a controller should resume
304 * each continuation exactly once (and promptly).
306 * Without the information in NXPINT_CONTINUATION, the controller can (with
307 * careful design, and help from the flow cookie) determine where the packet is
308 * in the pipeline, but in the general case it can't determine what nested
309 * "resubmit"s that may be in progress, or what data is on the stack maintained
310 * by NXAST_STACK_PUSH and NXAST_STACK_POP actions, what is in the OpenFlow
313 * Continuations are expensive because they require a round trip between the
314 * switch and the controller. Thus, they should not be used to implement
315 * processing that needs to happen at "line rate".
317 * The contents of NXPINT_CONTINUATION are private to the switch, may change
318 * unpredictably from one version of Open vSwitch to another, and are not
319 * documented here. The contents are also tied to a given Open vSwitch process
320 * and bridge, so that restarting Open vSwitch or deleting and recreating a
321 * bridge will cause the corresponding NXT_RESUME to be rejected.
323 * In the current implementation, Open vSwitch forks the packet processing
324 * pipeline across patch ports. Suppose, for example, that the pipeline for
325 * br0 outputs to a patch port whose peer belongs to br1, and that the pipeline
326 * for br1 executes a controller action with the "pause" flag. This only
327 * pauses processing within br1, and processing in br0 continues and possibly
328 * completes with visible side effects, such as outputting to ports, before
329 * br1's controller receives or processes the continuation. This
330 * implementation maintains the independence of separate bridges and, since
331 * processing in br1 cannot affect the behavior of br0 anyway, should not cause
332 * visible behavioral changes.
334 * A stateless implementation of continuations may ignore the "controller"
335 * action max_len, always sending the whole packet, because the full packet is
336 * required to continue traversal.
338 enum nx_packet_in2_prop_type {
340 NXPINT_PACKET, /* Raw packet data. */
341 NXPINT_FULL_LEN, /* ovs_be32: Full packet len, if truncated. */
342 NXPINT_BUFFER_ID, /* ovs_be32: Buffer ID, if buffered. */
344 /* Information about the flow that triggered the packet-in. */
345 NXPINT_TABLE_ID, /* uint8_t: Table ID. */
346 NXPINT_COOKIE, /* ovs_be64: Flow cookie. */
349 NXPINT_REASON, /* uint8_t, one of OFPR_*. */
350 NXPINT_METADATA, /* NXM or OXM for metadata fields. */
351 NXPINT_USERDATA, /* From NXAST_CONTROLLER2 userdata. */
352 NXPINT_CONTINUATION, /* Private data for continuing processing. */
355 /* Configures the "role" of the sending controller. The default role is:
357 * - Other (NX_ROLE_OTHER), which allows the controller access to all
360 * The other possible roles are a related pair:
362 * - Master (NX_ROLE_MASTER) is equivalent to Other, except that there may
363 * be at most one Master controller at a time: when a controller
364 * configures itself as Master, any existing Master is demoted to the
367 * - Slave (NX_ROLE_SLAVE) allows the controller read-only access to
368 * OpenFlow features. In particular attempts to modify the flow table
369 * will be rejected with an OFPBRC_EPERM error.
371 * Slave controllers do not receive OFPT_PACKET_IN or OFPT_FLOW_REMOVED
372 * messages, but they do receive OFPT_PORT_STATUS messages.
374 struct nx_role_request {
375 ovs_be32 role; /* One of NX_ROLE_*. */
377 OFP_ASSERT(sizeof(struct nx_role_request) == 4);
380 NX_ROLE_OTHER, /* Default role, full access. */
381 NX_ROLE_MASTER, /* Full access, at most one. */
382 NX_ROLE_SLAVE /* Read-only access. */
385 /* NXT_SET_ASYNC_CONFIG.
387 * Sent by a controller, this message configures the asynchronous messages that
388 * the controller wants to receive. Element 0 in each array specifies messages
389 * of interest when the controller has an "other" or "master" role; element 1,
390 * when the controller has a "slave" role.
392 * Each array element is a bitmask in which a 0-bit disables receiving a
393 * particular message and a 1-bit enables receiving it. Each bit controls the
394 * message whose 'reason' corresponds to the bit index. For example, the bit
395 * with value 1<<2 == 4 in port_status_mask[1] determines whether the
396 * controller will receive OFPT_PORT_STATUS messages with reason OFPPR_MODIFY
397 * (value 2) when the controller has a "slave" role.
399 * As a side effect, for service controllers, this message changes the
400 * miss_send_len from default of zero to OFP_DEFAULT_MISS_SEND_LEN (128).
402 struct nx_async_config {
403 ovs_be32 packet_in_mask[2]; /* Bitmasks of OFPR_* values. */
404 ovs_be32 port_status_mask[2]; /* Bitmasks of OFPRR_* values. */
405 ovs_be32 flow_removed_mask[2]; /* Bitmasks of OFPPR_* values. */
407 OFP_ASSERT(sizeof(struct nx_async_config) == 24);
409 /* Flexible flow specifications (aka NXM = Nicira Extended Match).
411 * OpenFlow 1.0 has "struct ofp10_match" for specifying flow matches. This
412 * structure is fixed-length and hence difficult to extend. This section
413 * describes a more flexible, variable-length flow match, called "nx_match" for
414 * short, that is also supported by Open vSwitch. This section also defines a
415 * replacement for each OpenFlow message that includes struct ofp10_match.
417 * OpenFlow 1.2+ introduced OpenFlow Extensible Match (OXM), adapting
418 * the design of NXM. The format of NXM and OXM are compatible.
424 * An nx_match is a sequence of zero or more "nxm_entry"s, which are
425 * type-length-value (TLV) entries, each 5 to 259 (inclusive) bytes long.
426 * "nxm_entry"s are not aligned on or padded to any multibyte boundary. The
427 * first 4 bytes of an nxm_entry are its "header", followed by the entry's
430 * An nxm_entry's header is interpreted as a 32-bit word in network byte order:
432 * |<-------------------- nxm_type ------------------>|
435 * +----------------------------------+---------------+--+------------------+
436 * | nxm_vendor | nxm_field |hm| nxm_length |
437 * +----------------------------------+---------------+--+------------------+
439 * The most-significant 23 bits of the header are collectively "nxm_type".
440 * Bits 16...31 are "nxm_vendor", one of OFPXMC12_* values. In case of
441 * NXM, it's either OFPXMC12_NXM_0 or OFPXMC12_NXM_1.
442 * Bits 9...15 are "nxm_field", which is a vendor-specific value. nxm_type
443 * normally designates a protocol header, such as the Ethernet type, but it
444 * can also refer to packet metadata, such as the switch port on which a packet
447 * Bit 8 is "nxm_hasmask" (labeled "hm" above for space reasons). The meaning
448 * of this bit is explained later.
450 * The least-significant 8 bits are "nxm_length", a positive integer. The
451 * length of the nxm_entry, including the header, is exactly 4 + nxm_length
454 * For a given nxm_vendor, nxm_field, and nxm_hasmask value, nxm_length is a
455 * constant. It is included only to allow software to minimally parse
456 * "nxm_entry"s of unknown types. (Similarly, for a given nxm_vendor,
457 * nxm_field, and nxm_length, nxm_hasmask is a constant.)
463 * A zero-length nx_match (one with no "nxm_entry"s) matches every packet.
465 * An nxm_entry places a constraint on the packets matched by the nx_match:
467 * - If nxm_hasmask is 0, the nxm_entry's body contains a value for the
468 * field, called "nxm_value". The nx_match matches only packets in which
469 * the field equals nxm_value.
471 * - If nxm_hasmask is 1, then the nxm_entry's body contains a value for the
472 * field (nxm_value), followed by a bitmask of the same length as the
473 * value, called "nxm_mask". For each 1-bit in position J in nxm_mask, the
474 * nx_match matches only packets for which bit J in the given field's value
475 * matches bit J in nxm_value. A 0-bit in nxm_mask causes the
476 * corresponding bit in nxm_value is ignored (it should be 0; Open vSwitch
477 * may enforce this someday), as is the corresponding bit in the field's
478 * value. (The sense of the nxm_mask bits is the opposite of that used by
479 * the "wildcards" member of struct ofp10_match.)
481 * When nxm_hasmask is 1, nxm_length is always even.
483 * An all-zero-bits nxm_mask is equivalent to omitting the nxm_entry
484 * entirely. An all-one-bits nxm_mask is equivalent to specifying 0 for
487 * When there are multiple "nxm_entry"s, all of the constraints must be met.
493 * Masks may be restricted:
495 * - Some nxm_types may not support masked wildcards, that is, nxm_hasmask
496 * must always be 0 when these fields are specified. For example, the
497 * field that identifies the port on which a packet was received may not be
500 * - Some nxm_types that do support masked wildcards may only support certain
501 * nxm_mask patterns. For example, fields that have IPv4 address values
502 * may be restricted to CIDR masks.
504 * These restrictions should be noted in specifications for individual fields.
505 * A switch may accept an nxm_hasmask or nxm_mask value that the specification
506 * disallows, if the switch correctly implements support for that nxm_hasmask
507 * or nxm_mask value. A switch must reject an attempt to set up a flow that
508 * contains a nxm_hasmask or nxm_mask value that it does not support.
511 * Prerequisite Restrictions
512 * =========================
514 * The presence of an nxm_entry with a given nxm_type may be restricted based
515 * on the presence of or values of other "nxm_entry"s. For example:
517 * - An nxm_entry for nxm_type=NXM_OF_IP_TOS is allowed only if it is
518 * preceded by another entry with nxm_type=NXM_OF_ETH_TYPE, nxm_hasmask=0,
519 * and nxm_value=0x0800. That is, matching on the IP source address is
520 * allowed only if the Ethernet type is explicitly set to IP.
522 * - An nxm_entry for nxm_type=NXM_OF_TCP_SRC is allowed only if it is
523 * preceded by an entry with nxm_type=NXM_OF_ETH_TYPE, nxm_hasmask=0, and
524 * nxm_value either 0x0800 or 0x86dd, and another with
525 * nxm_type=NXM_OF_IP_PROTO, nxm_hasmask=0, nxm_value=6, in that order.
526 * That is, matching on the TCP source port is allowed only if the Ethernet
527 * type is IP or IPv6 and the IP protocol is TCP.
529 * These restrictions should be noted in specifications for individual fields.
530 * A switch may implement relaxed versions of these restrictions. A switch
531 * must reject an attempt to set up a flow that violates its restrictions.
534 * Ordering Restrictions
535 * =====================
537 * An nxm_entry that has prerequisite restrictions must appear after the
538 * "nxm_entry"s for its prerequisites. Ordering of "nxm_entry"s within an
539 * nx_match is not otherwise constrained.
541 * Any given nxm_type may appear in an nx_match at most once.
547 * These examples show the format of a single nxm_entry with particular
548 * nxm_hasmask and nxm_length values. The diagrams are labeled with field
549 * numbers and byte indexes.
552 * 8-bit nxm_value, nxm_hasmask=1, nxm_length=2:
555 * +------------+---+---+
557 * +------------+---+---+
560 * 16-bit nxm_value, nxm_hasmask=0, nxm_length=2:
563 * +------------+------+
565 * +------------+------+
568 * 32-bit nxm_value, nxm_hasmask=0, nxm_length=4:
571 * +------------+-------------+
572 * | header | nxm_value |
573 * +------------+-------------+
576 * 48-bit nxm_value, nxm_hasmask=0, nxm_length=6:
579 * +------------+------------------+
580 * | header | nxm_value |
581 * +------------+------------------+
584 * 48-bit nxm_value, nxm_hasmask=1, nxm_length=12:
587 * +------------+------------------+------------------+
588 * | header | nxm_value | nxm_mask |
589 * +------------+------------------+------------------+
595 * A switch should report an error in an nx_match using error type
596 * OFPET_BAD_REQUEST and one of the NXBRC_NXM_* codes. Ideally the switch
597 * should report a specific error code, if one is assigned for the particular
598 * problem, but NXBRC_NXM_INVALID is also available to report a generic
602 /* Number of registers allocated NXM field IDs. */
603 #define NXM_NX_MAX_REGS 16
605 /* Bits in the value of NXM_NX_IP_FRAG. */
606 #define NX_IP_FRAG_ANY (1 << 0) /* Is this a fragment? */
607 #define NX_IP_FRAG_LATER (1 << 1) /* Is this a fragment with nonzero offset? */
609 /* Bits in the value of NXM_NX_TUN_FLAGS. */
610 #define NX_TUN_FLAG_OAM (1 << 0) /* Is this an OAM packet? */
612 /* ## --------------------- ## */
613 /* ## Requests and replies. ## */
614 /* ## --------------------- ## */
616 enum nx_flow_format {
617 NXFF_OPENFLOW10 = 0, /* Standard OpenFlow 1.0 compatible. */
618 NXFF_NXM = 2 /* Nicira extended match. */
621 /* NXT_SET_FLOW_FORMAT request. */
622 struct nx_set_flow_format {
623 ovs_be32 format; /* One of NXFF_*. */
625 OFP_ASSERT(sizeof(struct nx_set_flow_format) == 4);
627 /* NXT_FLOW_MOD (analogous to OFPT_FLOW_MOD).
629 * It is possible to limit flow deletions and modifications to certain
630 * cookies by using the NXM_NX_COOKIE(_W) matches. The "cookie" field
631 * is used only to add or modify flow cookies.
634 ovs_be64 cookie; /* Opaque controller-issued identifier. */
635 ovs_be16 command; /* OFPFC_* + possibly a table ID (see comment
636 * on struct nx_flow_mod_table_id). */
637 ovs_be16 idle_timeout; /* Idle time before discarding (seconds). */
638 ovs_be16 hard_timeout; /* Max time before discarding (seconds). */
639 ovs_be16 priority; /* Priority level of flow entry. */
640 ovs_be32 buffer_id; /* Buffered packet to apply to (or -1).
641 Not meaningful for OFPFC_DELETE*. */
642 ovs_be16 out_port; /* For OFPFC_DELETE* commands, require
643 matching entries to include this as an
644 output port. A value of OFPP_NONE
645 indicates no restriction. */
646 ovs_be16 flags; /* One of OFPFF_*. */
647 ovs_be16 match_len; /* Size of nx_match. */
648 uint8_t pad[6]; /* Align to 64-bits. */
650 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
651 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
652 * all-zero bytes, then
653 * - Actions to fill out the remainder of the message length (always a
657 OFP_ASSERT(sizeof(struct nx_flow_mod) == 32);
659 /* NXT_FLOW_REMOVED (analogous to OFPT_FLOW_REMOVED).
661 * 'table_id' is present only in Open vSwitch 1.11 and later. In earlier
662 * versions of Open vSwitch, this is a padding byte that is always zeroed.
663 * Therefore, a 'table_id' value of 0 indicates that the table ID is not known,
664 * and other values may be interpreted as one more than the flow's former table
666 struct nx_flow_removed {
667 ovs_be64 cookie; /* Opaque controller-issued identifier. */
668 ovs_be16 priority; /* Priority level of flow entry. */
669 uint8_t reason; /* One of OFPRR_*. */
670 uint8_t table_id; /* Flow's former table ID, plus one. */
671 ovs_be32 duration_sec; /* Time flow was alive in seconds. */
672 ovs_be32 duration_nsec; /* Time flow was alive in nanoseconds beyond
674 ovs_be16 idle_timeout; /* Idle timeout from original flow mod. */
675 ovs_be16 match_len; /* Size of nx_match. */
676 ovs_be64 packet_count;
679 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
680 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
683 OFP_ASSERT(sizeof(struct nx_flow_removed) == 40);
685 /* Nicira vendor stats request of type NXST_FLOW (analogous to OFPST_FLOW
688 * It is possible to limit matches to certain cookies by using the
689 * NXM_NX_COOKIE and NXM_NX_COOKIE_W matches.
691 struct nx_flow_stats_request {
692 ovs_be16 out_port; /* Require matching entries to include this
693 as an output port. A value of OFPP_NONE
694 indicates no restriction. */
695 ovs_be16 match_len; /* Length of nx_match. */
696 uint8_t table_id; /* ID of table to read (from ofp_table_stats)
697 or 0xff for all tables. */
698 uint8_t pad[3]; /* Align to 64 bits. */
700 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
701 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
702 * all-zero bytes, which must also exactly fill out the length of the
706 OFP_ASSERT(sizeof(struct nx_flow_stats_request) == 8);
708 /* Body for Nicira vendor stats reply of type NXST_FLOW (analogous to
711 * The values of 'idle_age' and 'hard_age' are only meaningful when talking to
712 * a switch that implements the NXT_FLOW_AGE extension. Zero means that the
713 * true value is unknown, perhaps because hardware does not track the value.
714 * (Zero is also the value that one should ordinarily expect to see talking to
715 * a switch that does not implement NXT_FLOW_AGE, since those switches zero the
716 * padding bytes that these fields replaced.) A nonzero value X represents X-1
717 * seconds. A value of 65535 represents 65534 or more seconds.
719 * 'idle_age' is the number of seconds that the flow has been idle, that is,
720 * the number of seconds since a packet passed through the flow. 'hard_age' is
721 * the number of seconds since the flow was last modified (e.g. OFPFC_MODIFY or
722 * OFPFC_MODIFY_STRICT). (The 'duration_*' fields are the elapsed time since
723 * the flow was added, regardless of subsequent modifications.)
725 * For a flow with an idle or hard timeout, 'idle_age' or 'hard_age',
726 * respectively, will ordinarily be smaller than the timeout, but flow
727 * expiration times are only approximate and so one must be prepared to
728 * tolerate expirations that occur somewhat early or late.
730 struct nx_flow_stats {
731 ovs_be16 length; /* Length of this entry. */
732 uint8_t table_id; /* ID of table flow came from. */
734 ovs_be32 duration_sec; /* Time flow has been alive in seconds. */
735 ovs_be32 duration_nsec; /* Time flow has been alive in nanoseconds
736 beyond duration_sec. */
737 ovs_be16 priority; /* Priority of the entry. */
738 ovs_be16 idle_timeout; /* Number of seconds idle before expiration. */
739 ovs_be16 hard_timeout; /* Number of seconds before expiration. */
740 ovs_be16 match_len; /* Length of nx_match. */
741 ovs_be16 idle_age; /* Seconds since last packet, plus one. */
742 ovs_be16 hard_age; /* Seconds since last modification, plus one. */
743 ovs_be64 cookie; /* Opaque controller-issued identifier. */
744 ovs_be64 packet_count; /* Number of packets, UINT64_MAX if unknown. */
745 ovs_be64 byte_count; /* Number of bytes, UINT64_MAX if unknown. */
747 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
748 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
749 * all-zero bytes, then
750 * - Actions to fill out the remainder 'length' bytes (always a multiple
754 OFP_ASSERT(sizeof(struct nx_flow_stats) == 48);
756 /* Nicira vendor stats request of type NXST_AGGREGATE (analogous to
757 * OFPST_AGGREGATE request).
759 * The reply format is identical to the reply format for OFPST_AGGREGATE,
760 * except for the header. */
761 struct nx_aggregate_stats_request {
762 ovs_be16 out_port; /* Require matching entries to include this
763 as an output port. A value of OFPP_NONE
764 indicates no restriction. */
765 ovs_be16 match_len; /* Length of nx_match. */
766 uint8_t table_id; /* ID of table to read (from ofp_table_stats)
767 or 0xff for all tables. */
768 uint8_t pad[3]; /* Align to 64 bits. */
770 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
771 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
772 * all-zero bytes, which must also exactly fill out the length of the
776 OFP_ASSERT(sizeof(struct nx_aggregate_stats_request) == 8);
778 /* NXT_SET_CONTROLLER_ID.
780 * Each OpenFlow controller connection has a 16-bit identifier that is
781 * initially 0. This message changes the connection's ID to 'id'.
783 * Controller connection IDs need not be unique.
785 * The NXAST_CONTROLLER action is the only current user of controller
787 struct nx_controller_id {
788 uint8_t zero[6]; /* Must be zero. */
789 ovs_be16 controller_id; /* New controller connection ID. */
791 OFP_ASSERT(sizeof(struct nx_controller_id) == 8);
793 /* Flow Table Monitoring
794 * =====================
796 * NXST_FLOW_MONITOR allows a controller to keep track of changes to OpenFlow
797 * flow table(s) or subsets of them, with the following workflow:
799 * 1. The controller sends an NXST_FLOW_MONITOR request to begin monitoring
800 * flows. The 'id' in the request must be unique among all monitors that
801 * the controller has started and not yet canceled on this OpenFlow
804 * 2. The switch responds with an NXST_FLOW_MONITOR reply. If the request's
805 * 'flags' included NXFMF_INITIAL, the reply includes all the flows that
806 * matched the request at the time of the request (with event NXFME_ADDED).
807 * If 'flags' did not include NXFMF_INITIAL, the reply is empty.
809 * The reply uses the xid of the request (as do all replies to OpenFlow
812 * 3. Whenever a change to a flow table entry matches some outstanding monitor
813 * request's criteria and flags, the switch sends a notification to the
814 * controller as an additional NXST_FLOW_MONITOR reply with xid 0.
816 * When multiple outstanding monitors match a single change, only a single
817 * notification is sent. This merged notification includes the information
818 * requested in any of the individual monitors. That is, if any of the
819 * matching monitors requests actions (NXFMF_ACTIONS), the notification
820 * includes actions, and if any of the monitors request full changes for the
821 * controller's own changes (NXFMF_OWN), the controller's own changes will
822 * be included in full.
824 * 4. The controller may cancel a monitor with NXT_FLOW_MONITOR_CANCEL. No
825 * further notifications will be sent on the basis of the canceled monitor
832 * OpenFlow messages for flow monitor notifications can overflow the buffer
833 * space available to the switch, either temporarily (e.g. due to network
834 * conditions slowing OpenFlow traffic) or more permanently (e.g. the sustained
835 * rate of flow table change exceeds the network bandwidth between switch and
838 * When Open vSwitch's notification buffer space reaches a limiting threshold,
839 * OVS reacts as follows:
841 * 1. OVS sends an NXT_FLOW_MONITOR_PAUSED message to the controller, following
842 * all the already queued notifications. After it receives this message,
843 * the controller knows that its view of the flow table, as represented by
844 * flow monitor notifications, is incomplete.
846 * 2. As long as the notification buffer is not empty:
848 * - NXMFE_ADD and NXFME_MODIFIED notifications will not be sent.
850 * - NXFME_DELETED notifications will still be sent, but only for flows
851 * that existed before OVS sent NXT_FLOW_MONITOR_PAUSED.
853 * - NXFME_ABBREV notifications will not be sent. They are treated as
854 * the expanded version (and therefore only the NXFME_DELETED
855 * components, if any, are sent).
857 * 3. When the notification buffer empties, OVS sends NXFME_ADD notifications
858 * for flows added since the buffer reached its limit and NXFME_MODIFIED
859 * notifications for flows that existed before the limit was reached and
860 * changed after the limit was reached.
862 * 4. OVS sends an NXT_FLOW_MONITOR_RESUMED message to the controller. After
863 * it receives this message, the controller knows that its view of the flow
864 * table, as represented by flow monitor notifications, is again complete.
866 * This allows the maximum buffer space requirement for notifications to be
867 * bounded by the limit plus the maximum number of supported flows.
870 * "Flow Removed" messages
871 * =======================
873 * The flow monitor mechanism is independent of OFPT_FLOW_REMOVED and
874 * NXT_FLOW_REMOVED. Flow monitor updates for deletion are sent if
875 * NXFMF_DELETE is set on a monitor, regardless of whether the
876 * OFPFF_SEND_FLOW_REM flag was set when the flow was added. */
878 /* NXST_FLOW_MONITOR request.
880 * The NXST_FLOW_MONITOR request's body consists of an array of zero or more
881 * instances of this structure. The request arranges to monitor the flows
882 * that match the specified criteria, which are interpreted in the same way as
885 * 'id' identifies a particular monitor for the purpose of allowing it to be
886 * canceled later with NXT_FLOW_MONITOR_CANCEL. 'id' must be unique among
887 * existing monitors that have not already been canceled.
889 * The reply includes the initial flow matches for monitors that have the
890 * NXFMF_INITIAL flag set. No single flow will be included in the reply more
891 * than once, even if more than one requested monitor matches that flow. The
892 * reply will be empty if none of the monitors has NXFMF_INITIAL set or if none
893 * of the monitors initially matches any flows.
895 * For NXFMF_ADD, an event will be reported if 'out_port' matches against the
896 * actions of the flow being added or, for a flow that is replacing an existing
897 * flow, if 'out_port' matches against the actions of the flow being replaced.
898 * For NXFMF_DELETE, 'out_port' matches against the actions of a flow being
899 * deleted. For NXFMF_MODIFY, an event will be reported if 'out_port' matches
900 * either the old or the new actions. */
901 struct nx_flow_monitor_request {
902 ovs_be32 id; /* Controller-assigned ID for this monitor. */
903 ovs_be16 flags; /* NXFMF_*. */
904 ovs_be16 out_port; /* Required output port, if not OFPP_NONE. */
905 ovs_be16 match_len; /* Length of nx_match. */
906 uint8_t table_id; /* One table's ID or 0xff for all tables. */
907 uint8_t zeros[5]; /* Align to 64 bits (must be zero). */
909 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
910 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
913 OFP_ASSERT(sizeof(struct nx_flow_monitor_request) == 16);
915 /* 'flags' bits in struct nx_flow_monitor_request. */
916 enum nx_flow_monitor_flags {
917 /* When to send updates. */
918 NXFMF_INITIAL = 1 << 0, /* Initially matching flows. */
919 NXFMF_ADD = 1 << 1, /* New matching flows as they are added. */
920 NXFMF_DELETE = 1 << 2, /* Old matching flows as they are removed. */
921 NXFMF_MODIFY = 1 << 3, /* Matching flows as they are changed. */
923 /* What to include in updates. */
924 NXFMF_ACTIONS = 1 << 4, /* If set, actions are included. */
925 NXFMF_OWN = 1 << 5, /* If set, include own changes in full. */
928 /* NXST_FLOW_MONITOR reply header.
930 * The body of an NXST_FLOW_MONITOR reply is an array of variable-length
931 * structures, each of which begins with this header. The 'length' member may
932 * be used to traverse the array, and the 'event' member may be used to
933 * determine the particular structure.
935 * Every instance is a multiple of 8 bytes long. */
936 struct nx_flow_update_header {
937 ovs_be16 length; /* Length of this entry. */
938 ovs_be16 event; /* One of NXFME_*. */
939 /* ...other data depending on 'event'... */
941 OFP_ASSERT(sizeof(struct nx_flow_update_header) == 4);
943 /* 'event' values in struct nx_flow_update_header. */
944 enum nx_flow_update_event {
945 /* struct nx_flow_update_full. */
946 NXFME_ADDED = 0, /* Flow was added. */
947 NXFME_DELETED = 1, /* Flow was deleted. */
948 NXFME_MODIFIED = 2, /* Flow (generally its actions) was changed. */
950 /* struct nx_flow_update_abbrev. */
951 NXFME_ABBREV = 3, /* Abbreviated reply. */
954 /* NXST_FLOW_MONITOR reply for NXFME_ADDED, NXFME_DELETED, and
956 struct nx_flow_update_full {
957 ovs_be16 length; /* Length is 24. */
958 ovs_be16 event; /* One of NXFME_*. */
959 ovs_be16 reason; /* OFPRR_* for NXFME_DELETED, else zero. */
960 ovs_be16 priority; /* Priority of the entry. */
961 ovs_be16 idle_timeout; /* Number of seconds idle before expiration. */
962 ovs_be16 hard_timeout; /* Number of seconds before expiration. */
963 ovs_be16 match_len; /* Length of nx_match. */
964 uint8_t table_id; /* ID of flow's table. */
965 uint8_t pad; /* Reserved, currently zeroed. */
966 ovs_be64 cookie; /* Opaque controller-issued identifier. */
968 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
969 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
970 * all-zero bytes, then
971 * - Actions to fill out the remainder 'length' bytes (always a multiple
972 * of 8). If NXFMF_ACTIONS was not specified, or 'event' is
973 * NXFME_DELETED, no actions are included.
976 OFP_ASSERT(sizeof(struct nx_flow_update_full) == 24);
978 /* NXST_FLOW_MONITOR reply for NXFME_ABBREV.
980 * When the controller does not specify NXFMF_OWN in a monitor request, any
981 * flow tables changes due to the controller's own requests (on the same
982 * OpenFlow channel) will be abbreviated, when possible, to this form, which
983 * simply specifies the 'xid' of the OpenFlow request (e.g. an OFPT_FLOW_MOD or
984 * NXT_FLOW_MOD) that caused the change.
986 * Some changes cannot be abbreviated and will be sent in full:
988 * - Changes that only partially succeed. This can happen if, for example,
989 * a flow_mod with type OFPFC_MODIFY affects multiple flows, but only some
990 * of those modifications succeed (e.g. due to hardware limitations).
992 * This cannot occur with the Open vSwitch software datapath. This also
993 * cannot occur in Open vSwitch 2.4 and later, because these versions only
994 * execute any flow modifications if all of them will succeed.
996 * - Changes that race with conflicting changes made by other controllers or
997 * other flow_mods (not separated by barriers) by the same controller.
999 * This cannot occur with the current Open vSwitch implementation
1000 * (regardless of datapath) because Open vSwitch internally serializes
1001 * potentially conflicting changes.
1003 * - Changes that occur when flow notification is paused (see "Buffer
1004 * Management" above).
1006 * A flow_mod that does not change the flow table will not trigger any
1007 * notification, even an abbreviated one. For example, a "modify" or "delete"
1008 * flow_mod that does not match any flows will not trigger a notification.
1009 * Whether an "add" or "modify" that specifies all the same parameters that a
1010 * flow already has triggers a notification is unspecified and subject to
1011 * change in future versions of Open vSwitch.
1013 * OVS will always send the notifications for a given flow table change before
1014 * the reply to a OFPT_BARRIER_REQUEST request that follows the flow table
1015 * change. Thus, if the controller does not receive an abbreviated (or
1016 * unabbreviated) notification for a flow_mod before the next
1017 * OFPT_BARRIER_REPLY, it will never receive one. */
1018 struct nx_flow_update_abbrev {
1019 ovs_be16 length; /* Length is 8. */
1020 ovs_be16 event; /* NXFME_ABBREV. */
1021 ovs_be32 xid; /* Controller-specified xid from flow_mod. */
1023 OFP_ASSERT(sizeof(struct nx_flow_update_abbrev) == 8);
1025 /* NXT_FLOW_MONITOR_CANCEL.
1027 * Used by a controller to cancel an outstanding monitor. */
1028 struct nx_flow_monitor_cancel {
1029 ovs_be32 id; /* 'id' from nx_flow_monitor_request. */
1031 OFP_ASSERT(sizeof(struct nx_flow_monitor_cancel) == 4);
1033 /* Variable-length option TLV table maintenance commands.
1035 * The option in Type-Length-Value format is widely used in tunnel options,
1036 * e.g., the base Geneve header is followed by zero or more options in TLV
1037 * format. Each option consists of a four byte option header and a variable
1038 * amount of option data interpreted according to the type. The generic TLV
1039 * format in tunnel options is as following:
1042 * 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
1043 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1044 * | Option Class | Type |R|R|R| Length |
1045 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1046 * | Variable Option Data |
1047 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1049 * In order to work with this variable-length options in TLV format in
1050 * tunnel options, we need to maintain a mapping table between an option
1051 * TLV (defined by <class, type, length>) and an NXM field that can be
1052 * operated on for the purposes of matches, actions, etc. This mapping
1053 * must be explicitly specified by the user.
1055 * There are two primary groups of OpenFlow messages that are introduced
1056 * as Nicira extensions: modification commands (add, delete, clear mappings)
1057 * and table status request/reply to dump the current table along with switch
1060 * Note that mappings should not be changed while they are in active use by
1061 * a flow. The result of doing so is undefined. */
1063 /* TLV table commands */
1064 enum nx_tlv_table_mod_command {
1065 NXTTMC_ADD, /* New mappings (fails if an option is already
1067 NXTTMC_DELETE, /* Delete mappings, identified by index
1068 * (unmapped options are ignored). */
1069 NXTTMC_CLEAR, /* Clear all mappings. Additional information
1070 in this command is ignored. */
1073 /* Map between an option TLV and an NXM field. */
1075 ovs_be16 option_class; /* TLV class. */
1076 uint8_t option_type; /* TLV type. */
1077 uint8_t option_len; /* TLV length (multiple of 4). */
1078 ovs_be16 index; /* NXM_NX_TUN_METADATA<n> index */
1081 OFP_ASSERT(sizeof(struct nx_tlv_map) == 8);
1083 /* NXT_TLV_TABLE_MOD.
1085 * Use to configure a mapping between option TLVs (class, type, length)
1086 * and NXM fields (NXM_NX_TUN_METADATA<n> where 'index' is <n>).
1088 * This command is atomic: all operations on different options will
1089 * either succeed or fail. */
1090 struct nx_tlv_table_mod {
1091 ovs_be16 command; /* One of NTTTMC_* */
1093 /* struct nx_tlv_map[0]; Array of maps between indicies and option
1094 TLVs. The number of elements is inferred
1095 from the length field in the header. */
1097 OFP_ASSERT(sizeof(struct nx_tlv_table_mod) == 8);
1099 /* NXT_TLV_TABLE_REPLY.
1101 * Issued in reponse to an NXT_TLV_TABLE_REQUEST to give information
1102 * about the current status of the TLV table in the switch. Provides
1103 * both static information about the switch's capabilities as well as
1104 * the configured TLV table. */
1105 struct nx_tlv_table_reply {
1106 ovs_be32 max_option_space; /* Maximum total of option sizes supported. */
1107 ovs_be16 max_fields; /* Maximum number of match fields supported. */
1108 uint8_t reserved[10];
1109 /* struct nx_tlv_map[0]; Array of maps between indicies and option
1110 TLVs. The number of elements is inferred
1111 from the length field in the header. */
1113 OFP_ASSERT(sizeof(struct nx_tlv_table_reply) == 16);
1115 #endif /* openflow/nicira-ext.h */