2 * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014, 2015 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_OPENFLOW10 = 0, /* Standard OpenFlow 1.0 compatible. */
158 NXPIF_NXM = 1 /* Nicira Extended. */
161 /* NXT_SET_PACKET_IN_FORMAT request. */
162 struct nx_set_packet_in_format {
163 ovs_be32 format; /* One of NXPIF_*. */
165 OFP_ASSERT(sizeof(struct nx_set_packet_in_format) == 4);
167 /* NXT_PACKET_IN (analogous to OFPT_PACKET_IN).
169 * NXT_PACKET_IN is similar to the OpenFlow 1.2 OFPT_PACKET_IN. The
172 * - NXT_PACKET_IN includes the cookie of the rule that triggered the
173 * message. (OpenFlow 1.3 OFPT_PACKET_IN also includes the cookie.)
175 * - The metadata fields use NXM (instead of OXM) field numbers.
177 * Open vSwitch 1.9.0 and later omits metadata fields that are zero (as allowed
178 * by OpenFlow 1.2). Earlier versions included all implemented metadata
181 * Open vSwitch does not include non-metadata in the nx_match, because by
182 * definition that information can be found in the packet itself. The format
183 * and the standards allow this, however, so controllers should be prepared to
184 * tolerate future changes.
186 * The NXM format is convenient for reporting metadata values, but it is
187 * important not to interpret the format as matching against a flow, because it
188 * does not. Nothing is being matched; arbitrary metadata masks would not be
191 * Whereas in most cases a controller can expect to only get back NXM fields
192 * that it set up itself (e.g. flow dumps will ordinarily report only NXM
193 * fields from flows that the controller added), NXT_PACKET_IN messages might
194 * contain fields that the controller does not understand, because the switch
195 * might support fields (new registers, new protocols, etc.) that the
196 * controller does not. The controller must prepared to tolerate these.
198 * The 'cookie' field has no meaning when 'reason' is OFPR_NO_MATCH. In this
199 * case it should be UINT64_MAX. */
200 struct nx_packet_in {
201 ovs_be32 buffer_id; /* ID assigned by datapath. */
202 ovs_be16 total_len; /* Full length of frame. */
203 uint8_t reason; /* Reason packet is sent (one of OFPR_*). */
204 uint8_t table_id; /* ID of the table that was looked up. */
205 ovs_be64 cookie; /* Cookie of the rule that was looked up. */
206 ovs_be16 match_len; /* Size of nx_match. */
207 uint8_t pad[6]; /* Align to 64-bits. */
209 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
210 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
211 * all-zero bytes, then
212 * - Exactly 2 all-zero padding bytes, then
213 * - An Ethernet frame whose length is inferred from nxh.header.length.
215 * The padding bytes preceding the Ethernet frame ensure that the IP
216 * header (if any) following the Ethernet header is 32-bit aligned. */
218 /* uint8_t nxm_fields[...]; */ /* NXM headers. */
219 /* uint8_t pad[2]; */ /* Align to 64 bit + 16 bit. */
220 /* uint8_t data[0]; */ /* Ethernet frame. */
222 OFP_ASSERT(sizeof(struct nx_packet_in) == 24);
224 /* Configures the "role" of the sending controller. The default role is:
226 * - Other (NX_ROLE_OTHER), which allows the controller access to all
229 * The other possible roles are a related pair:
231 * - Master (NX_ROLE_MASTER) is equivalent to Other, except that there may
232 * be at most one Master controller at a time: when a controller
233 * configures itself as Master, any existing Master is demoted to the
236 * - Slave (NX_ROLE_SLAVE) allows the controller read-only access to
237 * OpenFlow features. In particular attempts to modify the flow table
238 * will be rejected with an OFPBRC_EPERM error.
240 * Slave controllers do not receive OFPT_PACKET_IN or OFPT_FLOW_REMOVED
241 * messages, but they do receive OFPT_PORT_STATUS messages.
243 struct nx_role_request {
244 ovs_be32 role; /* One of NX_ROLE_*. */
246 OFP_ASSERT(sizeof(struct nx_role_request) == 4);
249 NX_ROLE_OTHER, /* Default role, full access. */
250 NX_ROLE_MASTER, /* Full access, at most one. */
251 NX_ROLE_SLAVE /* Read-only access. */
254 /* NXT_SET_ASYNC_CONFIG.
256 * Sent by a controller, this message configures the asynchronous messages that
257 * the controller wants to receive. Element 0 in each array specifies messages
258 * of interest when the controller has an "other" or "master" role; element 1,
259 * when the controller has a "slave" role.
261 * Each array element is a bitmask in which a 0-bit disables receiving a
262 * particular message and a 1-bit enables receiving it. Each bit controls the
263 * message whose 'reason' corresponds to the bit index. For example, the bit
264 * with value 1<<2 == 4 in port_status_mask[1] determines whether the
265 * controller will receive OFPT_PORT_STATUS messages with reason OFPPR_MODIFY
266 * (value 2) when the controller has a "slave" role.
268 * As a side effect, for service controllers, this message changes the
269 * miss_send_len from default of zero to OFP_DEFAULT_MISS_SEND_LEN (128).
271 struct nx_async_config {
272 ovs_be32 packet_in_mask[2]; /* Bitmasks of OFPR_* values. */
273 ovs_be32 port_status_mask[2]; /* Bitmasks of OFPRR_* values. */
274 ovs_be32 flow_removed_mask[2]; /* Bitmasks of OFPPR_* values. */
276 OFP_ASSERT(sizeof(struct nx_async_config) == 24);
278 /* Flexible flow specifications (aka NXM = Nicira Extended Match).
280 * OpenFlow 1.0 has "struct ofp10_match" for specifying flow matches. This
281 * structure is fixed-length and hence difficult to extend. This section
282 * describes a more flexible, variable-length flow match, called "nx_match" for
283 * short, that is also supported by Open vSwitch. This section also defines a
284 * replacement for each OpenFlow message that includes struct ofp10_match.
286 * OpenFlow 1.2+ introduced OpenFlow Extensible Match (OXM), adapting
287 * the design of NXM. The format of NXM and OXM are compatible.
293 * An nx_match is a sequence of zero or more "nxm_entry"s, which are
294 * type-length-value (TLV) entries, each 5 to 259 (inclusive) bytes long.
295 * "nxm_entry"s are not aligned on or padded to any multibyte boundary. The
296 * first 4 bytes of an nxm_entry are its "header", followed by the entry's
299 * An nxm_entry's header is interpreted as a 32-bit word in network byte order:
301 * |<-------------------- nxm_type ------------------>|
304 * +----------------------------------+---------------+--+------------------+
305 * | nxm_vendor | nxm_field |hm| nxm_length |
306 * +----------------------------------+---------------+--+------------------+
308 * The most-significant 23 bits of the header are collectively "nxm_type".
309 * Bits 16...31 are "nxm_vendor", one of OFPXMC12_* values. In case of
310 * NXM, it's either OFPXMC12_NXM_0 or OFPXMC12_NXM_1.
311 * Bits 9...15 are "nxm_field", which is a vendor-specific value. nxm_type
312 * normally designates a protocol header, such as the Ethernet type, but it
313 * can also refer to packet metadata, such as the switch port on which a packet
316 * Bit 8 is "nxm_hasmask" (labeled "hm" above for space reasons). The meaning
317 * of this bit is explained later.
319 * The least-significant 8 bits are "nxm_length", a positive integer. The
320 * length of the nxm_entry, including the header, is exactly 4 + nxm_length
323 * For a given nxm_vendor, nxm_field, and nxm_hasmask value, nxm_length is a
324 * constant. It is included only to allow software to minimally parse
325 * "nxm_entry"s of unknown types. (Similarly, for a given nxm_vendor,
326 * nxm_field, and nxm_length, nxm_hasmask is a constant.)
332 * A zero-length nx_match (one with no "nxm_entry"s) matches every packet.
334 * An nxm_entry places a constraint on the packets matched by the nx_match:
336 * - If nxm_hasmask is 0, the nxm_entry's body contains a value for the
337 * field, called "nxm_value". The nx_match matches only packets in which
338 * the field equals nxm_value.
340 * - If nxm_hasmask is 1, then the nxm_entry's body contains a value for the
341 * field (nxm_value), followed by a bitmask of the same length as the
342 * value, called "nxm_mask". For each 1-bit in position J in nxm_mask, the
343 * nx_match matches only packets for which bit J in the given field's value
344 * matches bit J in nxm_value. A 0-bit in nxm_mask causes the
345 * corresponding bit in nxm_value is ignored (it should be 0; Open vSwitch
346 * may enforce this someday), as is the corresponding bit in the field's
347 * value. (The sense of the nxm_mask bits is the opposite of that used by
348 * the "wildcards" member of struct ofp10_match.)
350 * When nxm_hasmask is 1, nxm_length is always even.
352 * An all-zero-bits nxm_mask is equivalent to omitting the nxm_entry
353 * entirely. An all-one-bits nxm_mask is equivalent to specifying 0 for
356 * When there are multiple "nxm_entry"s, all of the constraints must be met.
362 * Masks may be restricted:
364 * - Some nxm_types may not support masked wildcards, that is, nxm_hasmask
365 * must always be 0 when these fields are specified. For example, the
366 * field that identifies the port on which a packet was received may not be
369 * - Some nxm_types that do support masked wildcards may only support certain
370 * nxm_mask patterns. For example, fields that have IPv4 address values
371 * may be restricted to CIDR masks.
373 * These restrictions should be noted in specifications for individual fields.
374 * A switch may accept an nxm_hasmask or nxm_mask value that the specification
375 * disallows, if the switch correctly implements support for that nxm_hasmask
376 * or nxm_mask value. A switch must reject an attempt to set up a flow that
377 * contains a nxm_hasmask or nxm_mask value that it does not support.
380 * Prerequisite Restrictions
381 * =========================
383 * The presence of an nxm_entry with a given nxm_type may be restricted based
384 * on the presence of or values of other "nxm_entry"s. For example:
386 * - An nxm_entry for nxm_type=NXM_OF_IP_TOS is allowed only if it is
387 * preceded by another entry with nxm_type=NXM_OF_ETH_TYPE, nxm_hasmask=0,
388 * and nxm_value=0x0800. That is, matching on the IP source address is
389 * allowed only if the Ethernet type is explicitly set to IP.
391 * - An nxm_entry for nxm_type=NXM_OF_TCP_SRC is allowed only if it is
392 * preceded by an entry with nxm_type=NXM_OF_ETH_TYPE, nxm_hasmask=0, and
393 * nxm_value either 0x0800 or 0x86dd, and another with
394 * nxm_type=NXM_OF_IP_PROTO, nxm_hasmask=0, nxm_value=6, in that order.
395 * That is, matching on the TCP source port is allowed only if the Ethernet
396 * type is IP or IPv6 and the IP protocol is TCP.
398 * These restrictions should be noted in specifications for individual fields.
399 * A switch may implement relaxed versions of these restrictions. A switch
400 * must reject an attempt to set up a flow that violates its restrictions.
403 * Ordering Restrictions
404 * =====================
406 * An nxm_entry that has prerequisite restrictions must appear after the
407 * "nxm_entry"s for its prerequisites. Ordering of "nxm_entry"s within an
408 * nx_match is not otherwise constrained.
410 * Any given nxm_type may appear in an nx_match at most once.
416 * These examples show the format of a single nxm_entry with particular
417 * nxm_hasmask and nxm_length values. The diagrams are labeled with field
418 * numbers and byte indexes.
421 * 8-bit nxm_value, nxm_hasmask=1, nxm_length=2:
424 * +------------+---+---+
426 * +------------+---+---+
429 * 16-bit nxm_value, nxm_hasmask=0, nxm_length=2:
432 * +------------+------+
434 * +------------+------+
437 * 32-bit nxm_value, nxm_hasmask=0, nxm_length=4:
440 * +------------+-------------+
441 * | header | nxm_value |
442 * +------------+-------------+
445 * 48-bit nxm_value, nxm_hasmask=0, nxm_length=6:
448 * +------------+------------------+
449 * | header | nxm_value |
450 * +------------+------------------+
453 * 48-bit nxm_value, nxm_hasmask=1, nxm_length=12:
456 * +------------+------------------+------------------+
457 * | header | nxm_value | nxm_mask |
458 * +------------+------------------+------------------+
464 * A switch should report an error in an nx_match using error type
465 * OFPET_BAD_REQUEST and one of the NXBRC_NXM_* codes. Ideally the switch
466 * should report a specific error code, if one is assigned for the particular
467 * problem, but NXBRC_NXM_INVALID is also available to report a generic
471 /* Number of registers allocated NXM field IDs. */
472 #define NXM_NX_MAX_REGS 16
474 /* Bits in the value of NXM_NX_IP_FRAG. */
475 #define NX_IP_FRAG_ANY (1 << 0) /* Is this a fragment? */
476 #define NX_IP_FRAG_LATER (1 << 1) /* Is this a fragment with nonzero offset? */
478 /* Bits in the value of NXM_NX_TUN_FLAGS. */
479 #define NX_TUN_FLAG_OAM (1 << 0) /* Is this an OAM packet? */
481 /* ## --------------------- ## */
482 /* ## Requests and replies. ## */
483 /* ## --------------------- ## */
485 enum nx_flow_format {
486 NXFF_OPENFLOW10 = 0, /* Standard OpenFlow 1.0 compatible. */
487 NXFF_NXM = 2 /* Nicira extended match. */
490 /* NXT_SET_FLOW_FORMAT request. */
491 struct nx_set_flow_format {
492 ovs_be32 format; /* One of NXFF_*. */
494 OFP_ASSERT(sizeof(struct nx_set_flow_format) == 4);
496 /* NXT_FLOW_MOD (analogous to OFPT_FLOW_MOD).
498 * It is possible to limit flow deletions and modifications to certain
499 * cookies by using the NXM_NX_COOKIE(_W) matches. The "cookie" field
500 * is used only to add or modify flow cookies.
503 ovs_be64 cookie; /* Opaque controller-issued identifier. */
504 ovs_be16 command; /* OFPFC_* + possibly a table ID (see comment
505 * on struct nx_flow_mod_table_id). */
506 ovs_be16 idle_timeout; /* Idle time before discarding (seconds). */
507 ovs_be16 hard_timeout; /* Max time before discarding (seconds). */
508 ovs_be16 priority; /* Priority level of flow entry. */
509 ovs_be32 buffer_id; /* Buffered packet to apply to (or -1).
510 Not meaningful for OFPFC_DELETE*. */
511 ovs_be16 out_port; /* For OFPFC_DELETE* commands, require
512 matching entries to include this as an
513 output port. A value of OFPP_NONE
514 indicates no restriction. */
515 ovs_be16 flags; /* One of OFPFF_*. */
516 ovs_be16 match_len; /* Size of nx_match. */
517 uint8_t pad[6]; /* Align to 64-bits. */
519 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
520 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
521 * all-zero bytes, then
522 * - Actions to fill out the remainder of the message length (always a
526 OFP_ASSERT(sizeof(struct nx_flow_mod) == 32);
528 /* NXT_FLOW_REMOVED (analogous to OFPT_FLOW_REMOVED).
530 * 'table_id' is present only in Open vSwitch 1.11 and later. In earlier
531 * versions of Open vSwitch, this is a padding byte that is always zeroed.
532 * Therefore, a 'table_id' value of 0 indicates that the table ID is not known,
533 * and other values may be interpreted as one more than the flow's former table
535 struct nx_flow_removed {
536 ovs_be64 cookie; /* Opaque controller-issued identifier. */
537 ovs_be16 priority; /* Priority level of flow entry. */
538 uint8_t reason; /* One of OFPRR_*. */
539 uint8_t table_id; /* Flow's former table ID, plus one. */
540 ovs_be32 duration_sec; /* Time flow was alive in seconds. */
541 ovs_be32 duration_nsec; /* Time flow was alive in nanoseconds beyond
543 ovs_be16 idle_timeout; /* Idle timeout from original flow mod. */
544 ovs_be16 match_len; /* Size of nx_match. */
545 ovs_be64 packet_count;
548 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
549 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
552 OFP_ASSERT(sizeof(struct nx_flow_removed) == 40);
554 /* Nicira vendor stats request of type NXST_FLOW (analogous to OFPST_FLOW
557 * It is possible to limit matches to certain cookies by using the
558 * NXM_NX_COOKIE and NXM_NX_COOKIE_W matches.
560 struct nx_flow_stats_request {
561 ovs_be16 out_port; /* Require matching entries to include this
562 as an output port. A value of OFPP_NONE
563 indicates no restriction. */
564 ovs_be16 match_len; /* Length of nx_match. */
565 uint8_t table_id; /* ID of table to read (from ofp_table_stats)
566 or 0xff for all tables. */
567 uint8_t pad[3]; /* Align to 64 bits. */
569 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
570 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
571 * all-zero bytes, which must also exactly fill out the length of the
575 OFP_ASSERT(sizeof(struct nx_flow_stats_request) == 8);
577 /* Body for Nicira vendor stats reply of type NXST_FLOW (analogous to
580 * The values of 'idle_age' and 'hard_age' are only meaningful when talking to
581 * a switch that implements the NXT_FLOW_AGE extension. Zero means that the
582 * true value is unknown, perhaps because hardware does not track the value.
583 * (Zero is also the value that one should ordinarily expect to see talking to
584 * a switch that does not implement NXT_FLOW_AGE, since those switches zero the
585 * padding bytes that these fields replaced.) A nonzero value X represents X-1
586 * seconds. A value of 65535 represents 65534 or more seconds.
588 * 'idle_age' is the number of seconds that the flow has been idle, that is,
589 * the number of seconds since a packet passed through the flow. 'hard_age' is
590 * the number of seconds since the flow was last modified (e.g. OFPFC_MODIFY or
591 * OFPFC_MODIFY_STRICT). (The 'duration_*' fields are the elapsed time since
592 * the flow was added, regardless of subsequent modifications.)
594 * For a flow with an idle or hard timeout, 'idle_age' or 'hard_age',
595 * respectively, will ordinarily be smaller than the timeout, but flow
596 * expiration times are only approximate and so one must be prepared to
597 * tolerate expirations that occur somewhat early or late.
599 struct nx_flow_stats {
600 ovs_be16 length; /* Length of this entry. */
601 uint8_t table_id; /* ID of table flow came from. */
603 ovs_be32 duration_sec; /* Time flow has been alive in seconds. */
604 ovs_be32 duration_nsec; /* Time flow has been alive in nanoseconds
605 beyond duration_sec. */
606 ovs_be16 priority; /* Priority of the entry. */
607 ovs_be16 idle_timeout; /* Number of seconds idle before expiration. */
608 ovs_be16 hard_timeout; /* Number of seconds before expiration. */
609 ovs_be16 match_len; /* Length of nx_match. */
610 ovs_be16 idle_age; /* Seconds since last packet, plus one. */
611 ovs_be16 hard_age; /* Seconds since last modification, plus one. */
612 ovs_be64 cookie; /* Opaque controller-issued identifier. */
613 ovs_be64 packet_count; /* Number of packets, UINT64_MAX if unknown. */
614 ovs_be64 byte_count; /* Number of bytes, UINT64_MAX if unknown. */
616 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
617 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
618 * all-zero bytes, then
619 * - Actions to fill out the remainder 'length' bytes (always a multiple
623 OFP_ASSERT(sizeof(struct nx_flow_stats) == 48);
625 /* Nicira vendor stats request of type NXST_AGGREGATE (analogous to
626 * OFPST_AGGREGATE request).
628 * The reply format is identical to the reply format for OFPST_AGGREGATE,
629 * except for the header. */
630 struct nx_aggregate_stats_request {
631 ovs_be16 out_port; /* Require matching entries to include this
632 as an output port. A value of OFPP_NONE
633 indicates no restriction. */
634 ovs_be16 match_len; /* Length of nx_match. */
635 uint8_t table_id; /* ID of table to read (from ofp_table_stats)
636 or 0xff for all tables. */
637 uint8_t pad[3]; /* Align to 64 bits. */
639 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
640 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
641 * all-zero bytes, which must also exactly fill out the length of the
645 OFP_ASSERT(sizeof(struct nx_aggregate_stats_request) == 8);
647 /* NXT_SET_CONTROLLER_ID.
649 * Each OpenFlow controller connection has a 16-bit identifier that is
650 * initially 0. This message changes the connection's ID to 'id'.
652 * Controller connection IDs need not be unique.
654 * The NXAST_CONTROLLER action is the only current user of controller
656 struct nx_controller_id {
657 uint8_t zero[6]; /* Must be zero. */
658 ovs_be16 controller_id; /* New controller connection ID. */
660 OFP_ASSERT(sizeof(struct nx_controller_id) == 8);
662 /* Flow Table Monitoring
663 * =====================
665 * NXST_FLOW_MONITOR allows a controller to keep track of changes to OpenFlow
666 * flow table(s) or subsets of them, with the following workflow:
668 * 1. The controller sends an NXST_FLOW_MONITOR request to begin monitoring
669 * flows. The 'id' in the request must be unique among all monitors that
670 * the controller has started and not yet canceled on this OpenFlow
673 * 2. The switch responds with an NXST_FLOW_MONITOR reply. If the request's
674 * 'flags' included NXFMF_INITIAL, the reply includes all the flows that
675 * matched the request at the time of the request (with event NXFME_ADDED).
676 * If 'flags' did not include NXFMF_INITIAL, the reply is empty.
678 * The reply uses the xid of the request (as do all replies to OpenFlow
681 * 3. Whenever a change to a flow table entry matches some outstanding monitor
682 * request's criteria and flags, the switch sends a notification to the
683 * controller as an additional NXST_FLOW_MONITOR reply with xid 0.
685 * When multiple outstanding monitors match a single change, only a single
686 * notification is sent. This merged notification includes the information
687 * requested in any of the individual monitors. That is, if any of the
688 * matching monitors requests actions (NXFMF_ACTIONS), the notification
689 * includes actions, and if any of the monitors request full changes for the
690 * controller's own changes (NXFMF_OWN), the controller's own changes will
691 * be included in full.
693 * 4. The controller may cancel a monitor with NXT_FLOW_MONITOR_CANCEL. No
694 * further notifications will be sent on the basis of the canceled monitor
701 * OpenFlow messages for flow monitor notifications can overflow the buffer
702 * space available to the switch, either temporarily (e.g. due to network
703 * conditions slowing OpenFlow traffic) or more permanently (e.g. the sustained
704 * rate of flow table change exceeds the network bandwidth between switch and
707 * When Open vSwitch's notification buffer space reaches a limiting threshold,
708 * OVS reacts as follows:
710 * 1. OVS sends an NXT_FLOW_MONITOR_PAUSED message to the controller, following
711 * all the already queued notifications. After it receives this message,
712 * the controller knows that its view of the flow table, as represented by
713 * flow monitor notifications, is incomplete.
715 * 2. As long as the notification buffer is not empty:
717 * - NXMFE_ADD and NXFME_MODIFIED notifications will not be sent.
719 * - NXFME_DELETED notifications will still be sent, but only for flows
720 * that existed before OVS sent NXT_FLOW_MONITOR_PAUSED.
722 * - NXFME_ABBREV notifications will not be sent. They are treated as
723 * the expanded version (and therefore only the NXFME_DELETED
724 * components, if any, are sent).
726 * 3. When the notification buffer empties, OVS sends NXFME_ADD notifications
727 * for flows added since the buffer reached its limit and NXFME_MODIFIED
728 * notifications for flows that existed before the limit was reached and
729 * changed after the limit was reached.
731 * 4. OVS sends an NXT_FLOW_MONITOR_RESUMED message to the controller. After
732 * it receives this message, the controller knows that its view of the flow
733 * table, as represented by flow monitor notifications, is again complete.
735 * This allows the maximum buffer space requirement for notifications to be
736 * bounded by the limit plus the maximum number of supported flows.
739 * "Flow Removed" messages
740 * =======================
742 * The flow monitor mechanism is independent of OFPT_FLOW_REMOVED and
743 * NXT_FLOW_REMOVED. Flow monitor updates for deletion are sent if
744 * NXFMF_DELETE is set on a monitor, regardless of whether the
745 * OFPFF_SEND_FLOW_REM flag was set when the flow was added. */
747 /* NXST_FLOW_MONITOR request.
749 * The NXST_FLOW_MONITOR request's body consists of an array of zero or more
750 * instances of this structure. The request arranges to monitor the flows
751 * that match the specified criteria, which are interpreted in the same way as
754 * 'id' identifies a particular monitor for the purpose of allowing it to be
755 * canceled later with NXT_FLOW_MONITOR_CANCEL. 'id' must be unique among
756 * existing monitors that have not already been canceled.
758 * The reply includes the initial flow matches for monitors that have the
759 * NXFMF_INITIAL flag set. No single flow will be included in the reply more
760 * than once, even if more than one requested monitor matches that flow. The
761 * reply will be empty if none of the monitors has NXFMF_INITIAL set or if none
762 * of the monitors initially matches any flows.
764 * For NXFMF_ADD, an event will be reported if 'out_port' matches against the
765 * actions of the flow being added or, for a flow that is replacing an existing
766 * flow, if 'out_port' matches against the actions of the flow being replaced.
767 * For NXFMF_DELETE, 'out_port' matches against the actions of a flow being
768 * deleted. For NXFMF_MODIFY, an event will be reported if 'out_port' matches
769 * either the old or the new actions. */
770 struct nx_flow_monitor_request {
771 ovs_be32 id; /* Controller-assigned ID for this monitor. */
772 ovs_be16 flags; /* NXFMF_*. */
773 ovs_be16 out_port; /* Required output port, if not OFPP_NONE. */
774 ovs_be16 match_len; /* Length of nx_match. */
775 uint8_t table_id; /* One table's ID or 0xff for all tables. */
776 uint8_t zeros[5]; /* Align to 64 bits (must be zero). */
778 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
779 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
782 OFP_ASSERT(sizeof(struct nx_flow_monitor_request) == 16);
784 /* 'flags' bits in struct nx_flow_monitor_request. */
785 enum nx_flow_monitor_flags {
786 /* When to send updates. */
787 NXFMF_INITIAL = 1 << 0, /* Initially matching flows. */
788 NXFMF_ADD = 1 << 1, /* New matching flows as they are added. */
789 NXFMF_DELETE = 1 << 2, /* Old matching flows as they are removed. */
790 NXFMF_MODIFY = 1 << 3, /* Matching flows as they are changed. */
792 /* What to include in updates. */
793 NXFMF_ACTIONS = 1 << 4, /* If set, actions are included. */
794 NXFMF_OWN = 1 << 5, /* If set, include own changes in full. */
797 /* NXST_FLOW_MONITOR reply header.
799 * The body of an NXST_FLOW_MONITOR reply is an array of variable-length
800 * structures, each of which begins with this header. The 'length' member may
801 * be used to traverse the array, and the 'event' member may be used to
802 * determine the particular structure.
804 * Every instance is a multiple of 8 bytes long. */
805 struct nx_flow_update_header {
806 ovs_be16 length; /* Length of this entry. */
807 ovs_be16 event; /* One of NXFME_*. */
808 /* ...other data depending on 'event'... */
810 OFP_ASSERT(sizeof(struct nx_flow_update_header) == 4);
812 /* 'event' values in struct nx_flow_update_header. */
813 enum nx_flow_update_event {
814 /* struct nx_flow_update_full. */
815 NXFME_ADDED = 0, /* Flow was added. */
816 NXFME_DELETED = 1, /* Flow was deleted. */
817 NXFME_MODIFIED = 2, /* Flow (generally its actions) was changed. */
819 /* struct nx_flow_update_abbrev. */
820 NXFME_ABBREV = 3, /* Abbreviated reply. */
823 /* NXST_FLOW_MONITOR reply for NXFME_ADDED, NXFME_DELETED, and
825 struct nx_flow_update_full {
826 ovs_be16 length; /* Length is 24. */
827 ovs_be16 event; /* One of NXFME_*. */
828 ovs_be16 reason; /* OFPRR_* for NXFME_DELETED, else zero. */
829 ovs_be16 priority; /* Priority of the entry. */
830 ovs_be16 idle_timeout; /* Number of seconds idle before expiration. */
831 ovs_be16 hard_timeout; /* Number of seconds before expiration. */
832 ovs_be16 match_len; /* Length of nx_match. */
833 uint8_t table_id; /* ID of flow's table. */
834 uint8_t pad; /* Reserved, currently zeroed. */
835 ovs_be64 cookie; /* Opaque controller-issued identifier. */
837 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
838 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
839 * all-zero bytes, then
840 * - Actions to fill out the remainder 'length' bytes (always a multiple
841 * of 8). If NXFMF_ACTIONS was not specified, or 'event' is
842 * NXFME_DELETED, no actions are included.
845 OFP_ASSERT(sizeof(struct nx_flow_update_full) == 24);
847 /* NXST_FLOW_MONITOR reply for NXFME_ABBREV.
849 * When the controller does not specify NXFMF_OWN in a monitor request, any
850 * flow tables changes due to the controller's own requests (on the same
851 * OpenFlow channel) will be abbreviated, when possible, to this form, which
852 * simply specifies the 'xid' of the OpenFlow request (e.g. an OFPT_FLOW_MOD or
853 * NXT_FLOW_MOD) that caused the change.
855 * Some changes cannot be abbreviated and will be sent in full:
857 * - Changes that only partially succeed. This can happen if, for example,
858 * a flow_mod with type OFPFC_MODIFY affects multiple flows, but only some
859 * of those modifications succeed (e.g. due to hardware limitations).
861 * This cannot occur with the Open vSwitch software datapath. This also
862 * cannot occur in Open vSwitch 2.4 and later, because these versions only
863 * execute any flow modifications if all of them will succeed.
865 * - Changes that race with conflicting changes made by other controllers or
866 * other flow_mods (not separated by barriers) by the same controller.
868 * This cannot occur with the current Open vSwitch implementation
869 * (regardless of datapath) because Open vSwitch internally serializes
870 * potentially conflicting changes.
872 * - Changes that occur when flow notification is paused (see "Buffer
873 * Management" above).
875 * A flow_mod that does not change the flow table will not trigger any
876 * notification, even an abbreviated one. For example, a "modify" or "delete"
877 * flow_mod that does not match any flows will not trigger a notification.
878 * Whether an "add" or "modify" that specifies all the same parameters that a
879 * flow already has triggers a notification is unspecified and subject to
880 * change in future versions of Open vSwitch.
882 * OVS will always send the notifications for a given flow table change before
883 * the reply to a OFPT_BARRIER_REQUEST request that follows the flow table
884 * change. Thus, if the controller does not receive an abbreviated (or
885 * unabbreviated) notification for a flow_mod before the next
886 * OFPT_BARRIER_REPLY, it will never receive one. */
887 struct nx_flow_update_abbrev {
888 ovs_be16 length; /* Length is 8. */
889 ovs_be16 event; /* NXFME_ABBREV. */
890 ovs_be32 xid; /* Controller-specified xid from flow_mod. */
892 OFP_ASSERT(sizeof(struct nx_flow_update_abbrev) == 8);
894 /* NXT_FLOW_MONITOR_CANCEL.
896 * Used by a controller to cancel an outstanding monitor. */
897 struct nx_flow_monitor_cancel {
898 ovs_be32 id; /* 'id' from nx_flow_monitor_request. */
900 OFP_ASSERT(sizeof(struct nx_flow_monitor_cancel) == 4);
902 /* Variable-length option TLV table maintenance commands.
904 * The option in Type-Length-Value format is widely used in tunnel options,
905 * e.g., the base Geneve header is followed by zero or more options in TLV
906 * format. Each option consists of a four byte option header and a variable
907 * amount of option data interpreted according to the type. The generic TLV
908 * format in tunnel options is as following:
911 * 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
912 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
913 * | Option Class | Type |R|R|R| Length |
914 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
915 * | Variable Option Data |
916 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
918 * In order to work with this variable-length options in TLV format in
919 * tunnel options, we need to maintain a mapping table between an option
920 * TLV (defined by <class, type, length>) and an NXM field that can be
921 * operated on for the purposes of matches, actions, etc. This mapping
922 * must be explicitly specified by the user.
924 * There are two primary groups of OpenFlow messages that are introduced
925 * as Nicira extensions: modification commands (add, delete, clear mappings)
926 * and table status request/reply to dump the current table along with switch
929 * Note that mappings should not be changed while they are in active use by
930 * a flow. The result of doing so is undefined. */
932 /* TLV table commands */
933 enum nx_tlv_table_mod_command {
934 NXTTMC_ADD, /* New mappings (fails if an option is already
936 NXTTMC_DELETE, /* Delete mappings, identified by index
937 * (unmapped options are ignored). */
938 NXTTMC_CLEAR, /* Clear all mappings. Additional information
939 in this command is ignored. */
942 /* Map between an option TLV and an NXM field. */
944 ovs_be16 option_class; /* TLV class. */
945 uint8_t option_type; /* TLV type. */
946 uint8_t option_len; /* TLV length (multiple of 4). */
947 ovs_be16 index; /* NXM_NX_TUN_METADATA<n> index */
950 OFP_ASSERT(sizeof(struct nx_tlv_map) == 8);
952 /* NXT_TLV_TABLE_MOD.
954 * Use to configure a mapping between option TLVs (class, type, length)
955 * and NXM fields (NXM_NX_TUN_METADATA<n> where 'index' is <n>).
957 * This command is atomic: all operations on different options will
958 * either succeed or fail. */
959 struct nx_tlv_table_mod {
960 ovs_be16 command; /* One of NTTTMC_* */
962 /* struct nx_tlv_map[0]; Array of maps between indicies and option
963 TLVs. The number of elements is inferred
964 from the length field in the header. */
966 OFP_ASSERT(sizeof(struct nx_tlv_table_mod) == 8);
968 /* NXT_TLV_TABLE_REPLY.
970 * Issued in reponse to an NXT_TLV_TABLE_REQUEST to give information
971 * about the current status of the TLV table in the switch. Provides
972 * both static information about the switch's capabilities as well as
973 * the configured TLV table. */
974 struct nx_tlv_table_reply {
975 ovs_be32 max_option_space; /* Maximum total of option sizes supported. */
976 ovs_be16 max_fields; /* Maximum number of match fields supported. */
977 uint8_t reserved[10];
978 /* struct nx_tlv_map[0]; Array of maps between indicies and option
979 TLVs. The number of elements is inferred
980 from the length field in the header. */
982 OFP_ASSERT(sizeof(struct nx_tlv_table_reply) == 16);
984 #endif /* openflow/nicira-ext.h */