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 /* Header for Nicira vendor stats request and reply messages in OpenFlow
73 struct nicira10_stats_msg {
74 struct ofp10_vendor_stats_msg vsm; /* Vendor NX_VENDOR_ID. */
75 ovs_be32 subtype; /* One of NXST_* below. */
76 uint8_t pad[4]; /* Align to 64-bits. */
78 OFP_ASSERT(sizeof(struct nicira10_stats_msg) == 24);
80 /* Fields to use when hashing flows. */
82 /* Ethernet source address (NXM_OF_ETH_SRC) only. */
83 NX_HASH_FIELDS_ETH_SRC,
85 /* L2 through L4, symmetric across src/dst. Specifically, each of the
86 * following fields, if present, is hashed (slashes separate symmetric
89 * - NXM_OF_ETH_DST / NXM_OF_ETH_SRC
91 * - The VID bits from NXM_OF_VLAN_TCI, ignoring PCP and CFI.
93 * - NXM_OF_IP_SRC / NXM_OF_IP_DST
94 * - NXM_OF_TCP_SRC / NXM_OF_TCP_DST
96 NX_HASH_FIELDS_SYMMETRIC_L4,
98 /* L3+L4 only, 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
105 NX_HASH_FIELDS_SYMMETRIC_L3L4,
107 /* L3+L4 only with UDP ports, including the following fields:
110 * - NXM_OF_IP_SRC / NXM_OF_IP_DST
111 * - NXM_OF_SCTP_SRC / NXM_OF_SCTP_DST
112 * - NXM_OF_TCP_SRC / NXM_OF_TCP_DST
113 * - NXM_OF_UDP_SRC / NXM_OF_UDP_DST
115 NX_HASH_FIELDS_SYMMETRIC_L3L4_UDP
120 /* This command enables or disables an Open vSwitch extension that allows a
121 * controller to specify the OpenFlow table to which a flow should be added,
122 * instead of having the switch decide which table is most appropriate as
123 * required by OpenFlow 1.0. Because NXM was designed as an extension to
124 * OpenFlow 1.0, the extension applies equally to ofp10_flow_mod and
125 * nx_flow_mod. By default, the extension is disabled.
127 * When this feature is enabled, Open vSwitch treats struct ofp10_flow_mod's
128 * and struct nx_flow_mod's 16-bit 'command' member as two separate fields.
129 * The upper 8 bits are used as the table ID, the lower 8 bits specify the
130 * command as usual. A table ID of 0xff is treated like a wildcarded table ID.
132 * The specific treatment of the table ID depends on the type of flow mod:
134 * - OFPFC_ADD: Given a specific table ID, the flow is always placed in that
135 * table. If an identical flow already exists in that table only, then it
136 * is replaced. If the flow cannot be placed in the specified table,
137 * either because the table is full or because the table cannot support
138 * flows of the given type, the switch replies with an OFPFMFC_TABLE_FULL
139 * error. (A controller can distinguish these cases by comparing the
140 * current and maximum number of entries reported in ofp_table_stats.)
142 * If the table ID is wildcarded, the switch picks an appropriate table
143 * itself. If an identical flow already exist in the selected flow table,
144 * then it is replaced. The choice of table might depend on the flows
145 * that are already in the switch; for example, if one table fills up then
146 * the switch might fall back to another one.
148 * - OFPFC_MODIFY, OFPFC_DELETE: Given a specific table ID, only flows
149 * within that table are matched and modified or deleted. If the table ID
150 * is wildcarded, flows within any table may be matched and modified or
153 * - OFPFC_MODIFY_STRICT, OFPFC_DELETE_STRICT: Given a specific table ID,
154 * only a flow within that table may be matched and modified or deleted.
155 * If the table ID is wildcarded and exactly one flow within any table
156 * matches, then it is modified or deleted; if flows in more than one
157 * table match, then none is modified or deleted.
159 struct nx_flow_mod_table_id {
160 uint8_t set; /* Nonzero to enable, zero to disable. */
163 OFP_ASSERT(sizeof(struct nx_flow_mod_table_id) == 8);
165 enum nx_packet_in_format {
166 NXPIF_OPENFLOW10 = 0, /* Standard OpenFlow 1.0 compatible. */
167 NXPIF_NXM = 1 /* Nicira Extended. */
170 /* NXT_SET_PACKET_IN_FORMAT request. */
171 struct nx_set_packet_in_format {
172 ovs_be32 format; /* One of NXPIF_*. */
174 OFP_ASSERT(sizeof(struct nx_set_packet_in_format) == 4);
176 /* NXT_PACKET_IN (analogous to OFPT_PACKET_IN).
178 * NXT_PACKET_IN is similar to the OpenFlow 1.2 OFPT_PACKET_IN. The
181 * - NXT_PACKET_IN includes the cookie of the rule that triggered the
182 * message. (OpenFlow 1.3 OFPT_PACKET_IN also includes the cookie.)
184 * - The metadata fields use NXM (instead of OXM) field numbers.
186 * Open vSwitch 1.9.0 and later omits metadata fields that are zero (as allowed
187 * by OpenFlow 1.2). Earlier versions included all implemented metadata
190 * Open vSwitch does not include non-metadata in the nx_match, because by
191 * definition that information can be found in the packet itself. The format
192 * and the standards allow this, however, so controllers should be prepared to
193 * tolerate future changes.
195 * The NXM format is convenient for reporting metadata values, but it is
196 * important not to interpret the format as matching against a flow, because it
197 * does not. Nothing is being matched; arbitrary metadata masks would not be
200 * Whereas in most cases a controller can expect to only get back NXM fields
201 * that it set up itself (e.g. flow dumps will ordinarily report only NXM
202 * fields from flows that the controller added), NXT_PACKET_IN messages might
203 * contain fields that the controller does not understand, because the switch
204 * might support fields (new registers, new protocols, etc.) that the
205 * controller does not. The controller must prepared to tolerate these.
207 * The 'cookie' field has no meaning when 'reason' is OFPR_NO_MATCH. In this
208 * case it should be UINT64_MAX. */
209 struct nx_packet_in {
210 ovs_be32 buffer_id; /* ID assigned by datapath. */
211 ovs_be16 total_len; /* Full length of frame. */
212 uint8_t reason; /* Reason packet is sent (one of OFPR_*). */
213 uint8_t table_id; /* ID of the table that was looked up. */
214 ovs_be64 cookie; /* Cookie of the rule that was looked up. */
215 ovs_be16 match_len; /* Size of nx_match. */
216 uint8_t pad[6]; /* Align to 64-bits. */
218 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
219 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
220 * all-zero bytes, then
221 * - Exactly 2 all-zero padding bytes, then
222 * - An Ethernet frame whose length is inferred from nxh.header.length.
224 * The padding bytes preceding the Ethernet frame ensure that the IP
225 * header (if any) following the Ethernet header is 32-bit aligned. */
227 /* uint8_t nxm_fields[...]; */ /* NXM headers. */
228 /* uint8_t pad[2]; */ /* Align to 64 bit + 16 bit. */
229 /* uint8_t data[0]; */ /* Ethernet frame. */
231 OFP_ASSERT(sizeof(struct nx_packet_in) == 24);
233 /* Configures the "role" of the sending controller. The default role is:
235 * - Other (NX_ROLE_OTHER), which allows the controller access to all
238 * The other possible roles are a related pair:
240 * - Master (NX_ROLE_MASTER) is equivalent to Other, except that there may
241 * be at most one Master controller at a time: when a controller
242 * configures itself as Master, any existing Master is demoted to the
245 * - Slave (NX_ROLE_SLAVE) allows the controller read-only access to
246 * OpenFlow features. In particular attempts to modify the flow table
247 * will be rejected with an OFPBRC_EPERM error.
249 * Slave controllers do not receive OFPT_PACKET_IN or OFPT_FLOW_REMOVED
250 * messages, but they do receive OFPT_PORT_STATUS messages.
252 struct nx_role_request {
253 ovs_be32 role; /* One of NX_ROLE_*. */
255 OFP_ASSERT(sizeof(struct nx_role_request) == 4);
258 NX_ROLE_OTHER, /* Default role, full access. */
259 NX_ROLE_MASTER, /* Full access, at most one. */
260 NX_ROLE_SLAVE /* Read-only access. */
263 /* NXT_SET_ASYNC_CONFIG.
265 * Sent by a controller, this message configures the asynchronous messages that
266 * the controller wants to receive. Element 0 in each array specifies messages
267 * of interest when the controller has an "other" or "master" role; element 1,
268 * when the controller has a "slave" role.
270 * Each array element is a bitmask in which a 0-bit disables receiving a
271 * particular message and a 1-bit enables receiving it. Each bit controls the
272 * message whose 'reason' corresponds to the bit index. For example, the bit
273 * with value 1<<2 == 4 in port_status_mask[1] determines whether the
274 * controller will receive OFPT_PORT_STATUS messages with reason OFPPR_MODIFY
275 * (value 2) when the controller has a "slave" role.
277 * As a side effect, for service controllers, this message changes the
278 * miss_send_len from default of zero to OFP_DEFAULT_MISS_SEND_LEN (128).
280 struct nx_async_config {
281 ovs_be32 packet_in_mask[2]; /* Bitmasks of OFPR_* values. */
282 ovs_be32 port_status_mask[2]; /* Bitmasks of OFPRR_* values. */
283 ovs_be32 flow_removed_mask[2]; /* Bitmasks of OFPPR_* values. */
285 OFP_ASSERT(sizeof(struct nx_async_config) == 24);
287 /* Flexible flow specifications (aka NXM = Nicira Extended Match).
289 * OpenFlow 1.0 has "struct ofp10_match" for specifying flow matches. This
290 * structure is fixed-length and hence difficult to extend. This section
291 * describes a more flexible, variable-length flow match, called "nx_match" for
292 * short, that is also supported by Open vSwitch. This section also defines a
293 * replacement for each OpenFlow message that includes struct ofp10_match.
295 * OpenFlow 1.2+ introduced OpenFlow Extensible Match (OXM), adapting
296 * the design of NXM. The format of NXM and OXM are compatible.
302 * An nx_match is a sequence of zero or more "nxm_entry"s, which are
303 * type-length-value (TLV) entries, each 5 to 259 (inclusive) bytes long.
304 * "nxm_entry"s are not aligned on or padded to any multibyte boundary. The
305 * first 4 bytes of an nxm_entry are its "header", followed by the entry's
308 * An nxm_entry's header is interpreted as a 32-bit word in network byte order:
310 * |<-------------------- nxm_type ------------------>|
313 * +----------------------------------+---------------+--+------------------+
314 * | nxm_vendor | nxm_field |hm| nxm_length |
315 * +----------------------------------+---------------+--+------------------+
317 * The most-significant 23 bits of the header are collectively "nxm_type".
318 * Bits 16...31 are "nxm_vendor", one of OFPXMC12_* values. In case of
319 * NXM, it's either OFPXMC12_NXM_0 or OFPXMC12_NXM_1.
320 * Bits 9...15 are "nxm_field", which is a vendor-specific value. nxm_type
321 * normally designates a protocol header, such as the Ethernet type, but it
322 * can also refer to packet metadata, such as the switch port on which a packet
325 * Bit 8 is "nxm_hasmask" (labeled "hm" above for space reasons). The meaning
326 * of this bit is explained later.
328 * The least-significant 8 bits are "nxm_length", a positive integer. The
329 * length of the nxm_entry, including the header, is exactly 4 + nxm_length
332 * For a given nxm_vendor, nxm_field, and nxm_hasmask value, nxm_length is a
333 * constant. It is included only to allow software to minimally parse
334 * "nxm_entry"s of unknown types. (Similarly, for a given nxm_vendor,
335 * nxm_field, and nxm_length, nxm_hasmask is a constant.)
341 * A zero-length nx_match (one with no "nxm_entry"s) matches every packet.
343 * An nxm_entry places a constraint on the packets matched by the nx_match:
345 * - If nxm_hasmask is 0, the nxm_entry's body contains a value for the
346 * field, called "nxm_value". The nx_match matches only packets in which
347 * the field equals nxm_value.
349 * - If nxm_hasmask is 1, then the nxm_entry's body contains a value for the
350 * field (nxm_value), followed by a bitmask of the same length as the
351 * value, called "nxm_mask". For each 1-bit in position J in nxm_mask, the
352 * nx_match matches only packets for which bit J in the given field's value
353 * matches bit J in nxm_value. A 0-bit in nxm_mask causes the
354 * corresponding bit in nxm_value is ignored (it should be 0; Open vSwitch
355 * may enforce this someday), as is the corresponding bit in the field's
356 * value. (The sense of the nxm_mask bits is the opposite of that used by
357 * the "wildcards" member of struct ofp10_match.)
359 * When nxm_hasmask is 1, nxm_length is always even.
361 * An all-zero-bits nxm_mask is equivalent to omitting the nxm_entry
362 * entirely. An all-one-bits nxm_mask is equivalent to specifying 0 for
365 * When there are multiple "nxm_entry"s, all of the constraints must be met.
371 * Masks may be restricted:
373 * - Some nxm_types may not support masked wildcards, that is, nxm_hasmask
374 * must always be 0 when these fields are specified. For example, the
375 * field that identifies the port on which a packet was received may not be
378 * - Some nxm_types that do support masked wildcards may only support certain
379 * nxm_mask patterns. For example, fields that have IPv4 address values
380 * may be restricted to CIDR masks.
382 * These restrictions should be noted in specifications for individual fields.
383 * A switch may accept an nxm_hasmask or nxm_mask value that the specification
384 * disallows, if the switch correctly implements support for that nxm_hasmask
385 * or nxm_mask value. A switch must reject an attempt to set up a flow that
386 * contains a nxm_hasmask or nxm_mask value that it does not support.
389 * Prerequisite Restrictions
390 * =========================
392 * The presence of an nxm_entry with a given nxm_type may be restricted based
393 * on the presence of or values of other "nxm_entry"s. For example:
395 * - An nxm_entry for nxm_type=NXM_OF_IP_TOS is allowed only if it is
396 * preceded by another entry with nxm_type=NXM_OF_ETH_TYPE, nxm_hasmask=0,
397 * and nxm_value=0x0800. That is, matching on the IP source address is
398 * allowed only if the Ethernet type is explicitly set to IP.
400 * - An nxm_entry for nxm_type=NXM_OF_TCP_SRC is allowed only if it is
401 * preceded by an entry with nxm_type=NXM_OF_ETH_TYPE, nxm_hasmask=0, and
402 * nxm_value either 0x0800 or 0x86dd, and another with
403 * nxm_type=NXM_OF_IP_PROTO, nxm_hasmask=0, nxm_value=6, in that order.
404 * That is, matching on the TCP source port is allowed only if the Ethernet
405 * type is IP or IPv6 and the IP protocol is TCP.
407 * These restrictions should be noted in specifications for individual fields.
408 * A switch may implement relaxed versions of these restrictions. A switch
409 * must reject an attempt to set up a flow that violates its restrictions.
412 * Ordering Restrictions
413 * =====================
415 * An nxm_entry that has prerequisite restrictions must appear after the
416 * "nxm_entry"s for its prerequisites. Ordering of "nxm_entry"s within an
417 * nx_match is not otherwise constrained.
419 * Any given nxm_type may appear in an nx_match at most once.
425 * These examples show the format of a single nxm_entry with particular
426 * nxm_hasmask and nxm_length values. The diagrams are labeled with field
427 * numbers and byte indexes.
430 * 8-bit nxm_value, nxm_hasmask=1, nxm_length=2:
433 * +------------+---+---+
435 * +------------+---+---+
438 * 16-bit nxm_value, nxm_hasmask=0, nxm_length=2:
441 * +------------+------+
443 * +------------+------+
446 * 32-bit nxm_value, nxm_hasmask=0, nxm_length=4:
449 * +------------+-------------+
450 * | header | nxm_value |
451 * +------------+-------------+
454 * 48-bit nxm_value, nxm_hasmask=0, nxm_length=6:
457 * +------------+------------------+
458 * | header | nxm_value |
459 * +------------+------------------+
462 * 48-bit nxm_value, nxm_hasmask=1, nxm_length=12:
465 * +------------+------------------+------------------+
466 * | header | nxm_value | nxm_mask |
467 * +------------+------------------+------------------+
473 * A switch should report an error in an nx_match using error type
474 * OFPET_BAD_REQUEST and one of the NXBRC_NXM_* codes. Ideally the switch
475 * should report a specific error code, if one is assigned for the particular
476 * problem, but NXBRC_NXM_INVALID is also available to report a generic
480 /* Number of registers allocated NXM field IDs. */
481 #define NXM_NX_MAX_REGS 16
483 /* Bits in the value of NXM_NX_IP_FRAG. */
484 #define NX_IP_FRAG_ANY (1 << 0) /* Is this a fragment? */
485 #define NX_IP_FRAG_LATER (1 << 1) /* Is this a fragment with nonzero offset? */
487 /* Bits in the value of NXM_NX_TUN_FLAGS. */
488 #define NX_TUN_FLAG_OAM (1 << 0) /* Is this an OAM packet? */
490 /* ## --------------------- ## */
491 /* ## Requests and replies. ## */
492 /* ## --------------------- ## */
494 enum nx_flow_format {
495 NXFF_OPENFLOW10 = 0, /* Standard OpenFlow 1.0 compatible. */
496 NXFF_NXM = 2 /* Nicira extended match. */
499 /* NXT_SET_FLOW_FORMAT request. */
500 struct nx_set_flow_format {
501 ovs_be32 format; /* One of NXFF_*. */
503 OFP_ASSERT(sizeof(struct nx_set_flow_format) == 4);
505 /* NXT_FLOW_MOD (analogous to OFPT_FLOW_MOD).
507 * It is possible to limit flow deletions and modifications to certain
508 * cookies by using the NXM_NX_COOKIE(_W) matches. The "cookie" field
509 * is used only to add or modify flow cookies.
512 ovs_be64 cookie; /* Opaque controller-issued identifier. */
513 ovs_be16 command; /* OFPFC_* + possibly a table ID (see comment
514 * on struct nx_flow_mod_table_id). */
515 ovs_be16 idle_timeout; /* Idle time before discarding (seconds). */
516 ovs_be16 hard_timeout; /* Max time before discarding (seconds). */
517 ovs_be16 priority; /* Priority level of flow entry. */
518 ovs_be32 buffer_id; /* Buffered packet to apply to (or -1).
519 Not meaningful for OFPFC_DELETE*. */
520 ovs_be16 out_port; /* For OFPFC_DELETE* commands, require
521 matching entries to include this as an
522 output port. A value of OFPP_NONE
523 indicates no restriction. */
524 ovs_be16 flags; /* One of OFPFF_*. */
525 ovs_be16 match_len; /* Size of nx_match. */
526 uint8_t pad[6]; /* Align to 64-bits. */
528 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
529 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
530 * all-zero bytes, then
531 * - Actions to fill out the remainder of the message length (always a
535 OFP_ASSERT(sizeof(struct nx_flow_mod) == 32);
537 /* NXT_FLOW_REMOVED (analogous to OFPT_FLOW_REMOVED).
539 * 'table_id' is present only in Open vSwitch 1.11 and later. In earlier
540 * versions of Open vSwitch, this is a padding byte that is always zeroed.
541 * Therefore, a 'table_id' value of 0 indicates that the table ID is not known,
542 * and other values may be interpreted as one more than the flow's former table
544 struct nx_flow_removed {
545 ovs_be64 cookie; /* Opaque controller-issued identifier. */
546 ovs_be16 priority; /* Priority level of flow entry. */
547 uint8_t reason; /* One of OFPRR_*. */
548 uint8_t table_id; /* Flow's former table ID, plus one. */
549 ovs_be32 duration_sec; /* Time flow was alive in seconds. */
550 ovs_be32 duration_nsec; /* Time flow was alive in nanoseconds beyond
552 ovs_be16 idle_timeout; /* Idle timeout from original flow mod. */
553 ovs_be16 match_len; /* Size of nx_match. */
554 ovs_be64 packet_count;
557 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
558 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
561 OFP_ASSERT(sizeof(struct nx_flow_removed) == 40);
563 /* Nicira vendor stats request of type NXST_FLOW (analogous to OFPST_FLOW
566 * It is possible to limit matches to certain cookies by using the
567 * NXM_NX_COOKIE and NXM_NX_COOKIE_W matches.
569 struct nx_flow_stats_request {
570 ovs_be16 out_port; /* Require matching entries to include this
571 as an output port. A value of OFPP_NONE
572 indicates no restriction. */
573 ovs_be16 match_len; /* Length of nx_match. */
574 uint8_t table_id; /* ID of table to read (from ofp_table_stats)
575 or 0xff for all tables. */
576 uint8_t pad[3]; /* Align to 64 bits. */
578 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
579 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
580 * all-zero bytes, which must also exactly fill out the length of the
584 OFP_ASSERT(sizeof(struct nx_flow_stats_request) == 8);
586 /* Body for Nicira vendor stats reply of type NXST_FLOW (analogous to
589 * The values of 'idle_age' and 'hard_age' are only meaningful when talking to
590 * a switch that implements the NXT_FLOW_AGE extension. Zero means that the
591 * true value is unknown, perhaps because hardware does not track the value.
592 * (Zero is also the value that one should ordinarily expect to see talking to
593 * a switch that does not implement NXT_FLOW_AGE, since those switches zero the
594 * padding bytes that these fields replaced.) A nonzero value X represents X-1
595 * seconds. A value of 65535 represents 65534 or more seconds.
597 * 'idle_age' is the number of seconds that the flow has been idle, that is,
598 * the number of seconds since a packet passed through the flow. 'hard_age' is
599 * the number of seconds since the flow was last modified (e.g. OFPFC_MODIFY or
600 * OFPFC_MODIFY_STRICT). (The 'duration_*' fields are the elapsed time since
601 * the flow was added, regardless of subsequent modifications.)
603 * For a flow with an idle or hard timeout, 'idle_age' or 'hard_age',
604 * respectively, will ordinarily be smaller than the timeout, but flow
605 * expiration times are only approximate and so one must be prepared to
606 * tolerate expirations that occur somewhat early or late.
608 struct nx_flow_stats {
609 ovs_be16 length; /* Length of this entry. */
610 uint8_t table_id; /* ID of table flow came from. */
612 ovs_be32 duration_sec; /* Time flow has been alive in seconds. */
613 ovs_be32 duration_nsec; /* Time flow has been alive in nanoseconds
614 beyond duration_sec. */
615 ovs_be16 priority; /* Priority of the entry. */
616 ovs_be16 idle_timeout; /* Number of seconds idle before expiration. */
617 ovs_be16 hard_timeout; /* Number of seconds before expiration. */
618 ovs_be16 match_len; /* Length of nx_match. */
619 ovs_be16 idle_age; /* Seconds since last packet, plus one. */
620 ovs_be16 hard_age; /* Seconds since last modification, plus one. */
621 ovs_be64 cookie; /* Opaque controller-issued identifier. */
622 ovs_be64 packet_count; /* Number of packets, UINT64_MAX if unknown. */
623 ovs_be64 byte_count; /* Number of bytes, UINT64_MAX if unknown. */
625 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
626 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
627 * all-zero bytes, then
628 * - Actions to fill out the remainder 'length' bytes (always a multiple
632 OFP_ASSERT(sizeof(struct nx_flow_stats) == 48);
634 /* Nicira vendor stats request of type NXST_AGGREGATE (analogous to
635 * OFPST_AGGREGATE request).
637 * The reply format is identical to the reply format for OFPST_AGGREGATE,
638 * except for the header. */
639 struct nx_aggregate_stats_request {
640 ovs_be16 out_port; /* Require matching entries to include this
641 as an output port. A value of OFPP_NONE
642 indicates no restriction. */
643 ovs_be16 match_len; /* Length of nx_match. */
644 uint8_t table_id; /* ID of table to read (from ofp_table_stats)
645 or 0xff for all tables. */
646 uint8_t pad[3]; /* Align to 64 bits. */
648 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
649 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
650 * all-zero bytes, which must also exactly fill out the length of the
654 OFP_ASSERT(sizeof(struct nx_aggregate_stats_request) == 8);
656 /* NXT_SET_CONTROLLER_ID.
658 * Each OpenFlow controller connection has a 16-bit identifier that is
659 * initially 0. This message changes the connection's ID to 'id'.
661 * Controller connection IDs need not be unique.
663 * The NXAST_CONTROLLER action is the only current user of controller
665 struct nx_controller_id {
666 uint8_t zero[6]; /* Must be zero. */
667 ovs_be16 controller_id; /* New controller connection ID. */
669 OFP_ASSERT(sizeof(struct nx_controller_id) == 8);
671 /* Flow Table Monitoring
672 * =====================
674 * NXST_FLOW_MONITOR allows a controller to keep track of changes to OpenFlow
675 * flow table(s) or subsets of them, with the following workflow:
677 * 1. The controller sends an NXST_FLOW_MONITOR request to begin monitoring
678 * flows. The 'id' in the request must be unique among all monitors that
679 * the controller has started and not yet canceled on this OpenFlow
682 * 2. The switch responds with an NXST_FLOW_MONITOR reply. If the request's
683 * 'flags' included NXFMF_INITIAL, the reply includes all the flows that
684 * matched the request at the time of the request (with event NXFME_ADDED).
685 * If 'flags' did not include NXFMF_INITIAL, the reply is empty.
687 * The reply uses the xid of the request (as do all replies to OpenFlow
690 * 3. Whenever a change to a flow table entry matches some outstanding monitor
691 * request's criteria and flags, the switch sends a notification to the
692 * controller as an additional NXST_FLOW_MONITOR reply with xid 0.
694 * When multiple outstanding monitors match a single change, only a single
695 * notification is sent. This merged notification includes the information
696 * requested in any of the individual monitors. That is, if any of the
697 * matching monitors requests actions (NXFMF_ACTIONS), the notification
698 * includes actions, and if any of the monitors request full changes for the
699 * controller's own changes (NXFMF_OWN), the controller's own changes will
700 * be included in full.
702 * 4. The controller may cancel a monitor with NXT_FLOW_MONITOR_CANCEL. No
703 * further notifications will be sent on the basis of the canceled monitor
710 * OpenFlow messages for flow monitor notifications can overflow the buffer
711 * space available to the switch, either temporarily (e.g. due to network
712 * conditions slowing OpenFlow traffic) or more permanently (e.g. the sustained
713 * rate of flow table change exceeds the network bandwidth between switch and
716 * When Open vSwitch's notification buffer space reaches a limiting threshold,
717 * OVS reacts as follows:
719 * 1. OVS sends an NXT_FLOW_MONITOR_PAUSED message to the controller, following
720 * all the already queued notifications. After it receives this message,
721 * the controller knows that its view of the flow table, as represented by
722 * flow monitor notifications, is incomplete.
724 * 2. As long as the notification buffer is not empty:
726 * - NXMFE_ADD and NXFME_MODIFIED notifications will not be sent.
728 * - NXFME_DELETED notifications will still be sent, but only for flows
729 * that existed before OVS sent NXT_FLOW_MONITOR_PAUSED.
731 * - NXFME_ABBREV notifications will not be sent. They are treated as
732 * the expanded version (and therefore only the NXFME_DELETED
733 * components, if any, are sent).
735 * 3. When the notification buffer empties, OVS sends NXFME_ADD notifications
736 * for flows added since the buffer reached its limit and NXFME_MODIFIED
737 * notifications for flows that existed before the limit was reached and
738 * changed after the limit was reached.
740 * 4. OVS sends an NXT_FLOW_MONITOR_RESUMED message to the controller. After
741 * it receives this message, the controller knows that its view of the flow
742 * table, as represented by flow monitor notifications, is again complete.
744 * This allows the maximum buffer space requirement for notifications to be
745 * bounded by the limit plus the maximum number of supported flows.
748 * "Flow Removed" messages
749 * =======================
751 * The flow monitor mechanism is independent of OFPT_FLOW_REMOVED and
752 * NXT_FLOW_REMOVED. Flow monitor updates for deletion are sent if
753 * NXFMF_DELETE is set on a monitor, regardless of whether the
754 * OFPFF_SEND_FLOW_REM flag was set when the flow was added. */
756 /* NXST_FLOW_MONITOR request.
758 * The NXST_FLOW_MONITOR request's body consists of an array of zero or more
759 * instances of this structure. The request arranges to monitor the flows
760 * that match the specified criteria, which are interpreted in the same way as
763 * 'id' identifies a particular monitor for the purpose of allowing it to be
764 * canceled later with NXT_FLOW_MONITOR_CANCEL. 'id' must be unique among
765 * existing monitors that have not already been canceled.
767 * The reply includes the initial flow matches for monitors that have the
768 * NXFMF_INITIAL flag set. No single flow will be included in the reply more
769 * than once, even if more than one requested monitor matches that flow. The
770 * reply will be empty if none of the monitors has NXFMF_INITIAL set or if none
771 * of the monitors initially matches any flows.
773 * For NXFMF_ADD, an event will be reported if 'out_port' matches against the
774 * actions of the flow being added or, for a flow that is replacing an existing
775 * flow, if 'out_port' matches against the actions of the flow being replaced.
776 * For NXFMF_DELETE, 'out_port' matches against the actions of a flow being
777 * deleted. For NXFMF_MODIFY, an event will be reported if 'out_port' matches
778 * either the old or the new actions. */
779 struct nx_flow_monitor_request {
780 ovs_be32 id; /* Controller-assigned ID for this monitor. */
781 ovs_be16 flags; /* NXFMF_*. */
782 ovs_be16 out_port; /* Required output port, if not OFPP_NONE. */
783 ovs_be16 match_len; /* Length of nx_match. */
784 uint8_t table_id; /* One table's ID or 0xff for all tables. */
785 uint8_t zeros[5]; /* Align to 64 bits (must be zero). */
787 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
788 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
791 OFP_ASSERT(sizeof(struct nx_flow_monitor_request) == 16);
793 /* 'flags' bits in struct nx_flow_monitor_request. */
794 enum nx_flow_monitor_flags {
795 /* When to send updates. */
796 NXFMF_INITIAL = 1 << 0, /* Initially matching flows. */
797 NXFMF_ADD = 1 << 1, /* New matching flows as they are added. */
798 NXFMF_DELETE = 1 << 2, /* Old matching flows as they are removed. */
799 NXFMF_MODIFY = 1 << 3, /* Matching flows as they are changed. */
801 /* What to include in updates. */
802 NXFMF_ACTIONS = 1 << 4, /* If set, actions are included. */
803 NXFMF_OWN = 1 << 5, /* If set, include own changes in full. */
806 /* NXST_FLOW_MONITOR reply header.
808 * The body of an NXST_FLOW_MONITOR reply is an array of variable-length
809 * structures, each of which begins with this header. The 'length' member may
810 * be used to traverse the array, and the 'event' member may be used to
811 * determine the particular structure.
813 * Every instance is a multiple of 8 bytes long. */
814 struct nx_flow_update_header {
815 ovs_be16 length; /* Length of this entry. */
816 ovs_be16 event; /* One of NXFME_*. */
817 /* ...other data depending on 'event'... */
819 OFP_ASSERT(sizeof(struct nx_flow_update_header) == 4);
821 /* 'event' values in struct nx_flow_update_header. */
822 enum nx_flow_update_event {
823 /* struct nx_flow_update_full. */
824 NXFME_ADDED = 0, /* Flow was added. */
825 NXFME_DELETED = 1, /* Flow was deleted. */
826 NXFME_MODIFIED = 2, /* Flow (generally its actions) was changed. */
828 /* struct nx_flow_update_abbrev. */
829 NXFME_ABBREV = 3, /* Abbreviated reply. */
832 /* NXST_FLOW_MONITOR reply for NXFME_ADDED, NXFME_DELETED, and
834 struct nx_flow_update_full {
835 ovs_be16 length; /* Length is 24. */
836 ovs_be16 event; /* One of NXFME_*. */
837 ovs_be16 reason; /* OFPRR_* for NXFME_DELETED, else zero. */
838 ovs_be16 priority; /* Priority of the entry. */
839 ovs_be16 idle_timeout; /* Number of seconds idle before expiration. */
840 ovs_be16 hard_timeout; /* Number of seconds before expiration. */
841 ovs_be16 match_len; /* Length of nx_match. */
842 uint8_t table_id; /* ID of flow's table. */
843 uint8_t pad; /* Reserved, currently zeroed. */
844 ovs_be64 cookie; /* Opaque controller-issued identifier. */
846 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
847 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
848 * all-zero bytes, then
849 * - Actions to fill out the remainder 'length' bytes (always a multiple
850 * of 8). If NXFMF_ACTIONS was not specified, or 'event' is
851 * NXFME_DELETED, no actions are included.
854 OFP_ASSERT(sizeof(struct nx_flow_update_full) == 24);
856 /* NXST_FLOW_MONITOR reply for NXFME_ABBREV.
858 * When the controller does not specify NXFMF_OWN in a monitor request, any
859 * flow tables changes due to the controller's own requests (on the same
860 * OpenFlow channel) will be abbreviated, when possible, to this form, which
861 * simply specifies the 'xid' of the OpenFlow request (e.g. an OFPT_FLOW_MOD or
862 * NXT_FLOW_MOD) that caused the change.
864 * Some changes cannot be abbreviated and will be sent in full:
866 * - Changes that only partially succeed. This can happen if, for example,
867 * a flow_mod with type OFPFC_MODIFY affects multiple flows, but only some
868 * of those modifications succeed (e.g. due to hardware limitations).
870 * This cannot occur with the Open vSwitch software datapath. This also
871 * cannot occur in Open vSwitch 2.4 and later, because these versions only
872 * execute any flow modifications if all of them will succeed.
874 * - Changes that race with conflicting changes made by other controllers or
875 * other flow_mods (not separated by barriers) by the same controller.
877 * This cannot occur with the current Open vSwitch implementation
878 * (regardless of datapath) because Open vSwitch internally serializes
879 * potentially conflicting changes.
881 * - Changes that occur when flow notification is paused (see "Buffer
882 * Management" above).
884 * A flow_mod that does not change the flow table will not trigger any
885 * notification, even an abbreviated one. For example, a "modify" or "delete"
886 * flow_mod that does not match any flows will not trigger a notification.
887 * Whether an "add" or "modify" that specifies all the same parameters that a
888 * flow already has triggers a notification is unspecified and subject to
889 * change in future versions of Open vSwitch.
891 * OVS will always send the notifications for a given flow table change before
892 * the reply to a OFPT_BARRIER_REQUEST request that follows the flow table
893 * change. Thus, if the controller does not receive an abbreviated (or
894 * unabbreviated) notification for a flow_mod before the next
895 * OFPT_BARRIER_REPLY, it will never receive one. */
896 struct nx_flow_update_abbrev {
897 ovs_be16 length; /* Length is 8. */
898 ovs_be16 event; /* NXFME_ABBREV. */
899 ovs_be32 xid; /* Controller-specified xid from flow_mod. */
901 OFP_ASSERT(sizeof(struct nx_flow_update_abbrev) == 8);
903 /* NXT_FLOW_MONITOR_CANCEL.
905 * Used by a controller to cancel an outstanding monitor. */
906 struct nx_flow_monitor_cancel {
907 ovs_be32 id; /* 'id' from nx_flow_monitor_request. */
909 OFP_ASSERT(sizeof(struct nx_flow_monitor_cancel) == 4);
911 /* Variable-length option TLV table maintenance commands.
913 * The option in Type-Length-Value format is widely used in tunnel options,
914 * e.g., the base Geneve header is followed by zero or more options in TLV
915 * format. Each option consists of a four byte option header and a variable
916 * amount of option data interpreted according to the type. The generic TLV
917 * format in tunnel options is as following:
920 * 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
921 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
922 * | Option Class | Type |R|R|R| Length |
923 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
924 * | Variable Option Data |
925 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
927 * In order to work with this variable-length options in TLV format in
928 * tunnel options, we need to maintain a mapping table between an option
929 * TLV (defined by <class, type, length>) and an NXM field that can be
930 * operated on for the purposes of matches, actions, etc. This mapping
931 * must be explicitly specified by the user.
933 * There are two primary groups of OpenFlow messages that are introduced
934 * as Nicira extensions: modification commands (add, delete, clear mappings)
935 * and table status request/reply to dump the current table along with switch
938 * Note that mappings should not be changed while they are in active use by
939 * a flow. The result of doing so is undefined. */
941 /* TLV table commands */
942 enum nx_tlv_table_mod_command {
943 NXTTMC_ADD, /* New mappings (fails if an option is already
945 NXTTMC_DELETE, /* Delete mappings, identified by index
946 * (unmapped options are ignored). */
947 NXTTMC_CLEAR, /* Clear all mappings. Additional information
948 in this command is ignored. */
951 /* Map between an option TLV and an NXM field. */
953 ovs_be16 option_class; /* TLV class. */
954 uint8_t option_type; /* TLV type. */
955 uint8_t option_len; /* TLV length (multiple of 4). */
956 ovs_be16 index; /* NXM_NX_TUN_METADATA<n> index */
959 OFP_ASSERT(sizeof(struct nx_tlv_map) == 8);
961 /* NXT_TLV_TABLE_MOD.
963 * Use to configure a mapping between option TLVs (class, type, length)
964 * and NXM fields (NXM_NX_TUN_METADATA<n> where 'index' is <n>).
966 * This command is atomic: all operations on different options will
967 * either succeed or fail. */
968 struct nx_tlv_table_mod {
969 ovs_be16 command; /* One of NTTTMC_* */
971 /* struct nx_tlv_map[0]; Array of maps between indicies and option
972 TLVs. The number of elements is inferred
973 from the length field in the header. */
975 OFP_ASSERT(sizeof(struct nx_tlv_table_mod) == 8);
977 /* NXT_TLV_TABLE_REPLY.
979 * Issued in reponse to an NXT_TLV_TABLE_REQUEST to give information
980 * about the current status of the TLV table in the switch. Provides
981 * both static information about the switch's capabilities as well as
982 * the configured TLV table. */
983 struct nx_tlv_table_reply {
984 ovs_be32 max_option_space; /* Maximum total of option sizes supported. */
985 ovs_be16 max_fields; /* Maximum number of match fields supported. */
986 uint8_t reserved[10];
987 /* struct nx_tlv_map[0]; Array of maps between indicies and option
988 TLVs. The number of elements is inferred
989 from the length field in the header. */
991 OFP_ASSERT(sizeof(struct nx_tlv_table_reply) == 16);
993 #endif /* openflow/nicira-ext.h */