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);
240 * NXT_PACKET_IN2 is conceptually similar to OFPT_PACKET_IN but it is expressed
241 * as an extensible set of properties instead of using a fixed structure.
243 * Added in Open vSwitch 2.6. */
244 enum nx_packet_in2_prop_type {
246 NXPINT_PACKET, /* Raw packet data. */
247 NXPINT_FULL_LEN, /* ovs_be32: Full packet len, if truncated. */
248 NXPINT_BUFFER_ID, /* ovs_be32: Buffer ID, if buffered. */
250 /* Information about the flow that triggered the packet-in. */
251 NXPINT_TABLE_ID, /* uint8_t: Table ID. */
252 NXPINT_COOKIE, /* ovs_be64: Flow cookie. */
255 NXPINT_REASON, /* uint8_t, one of OFPR_*. */
256 NXPINT_METADATA, /* NXM or OXM for metadata fields. */
259 /* Configures the "role" of the sending controller. The default role is:
261 * - Other (NX_ROLE_OTHER), which allows the controller access to all
264 * The other possible roles are a related pair:
266 * - Master (NX_ROLE_MASTER) is equivalent to Other, except that there may
267 * be at most one Master controller at a time: when a controller
268 * configures itself as Master, any existing Master is demoted to the
271 * - Slave (NX_ROLE_SLAVE) allows the controller read-only access to
272 * OpenFlow features. In particular attempts to modify the flow table
273 * will be rejected with an OFPBRC_EPERM error.
275 * Slave controllers do not receive OFPT_PACKET_IN or OFPT_FLOW_REMOVED
276 * messages, but they do receive OFPT_PORT_STATUS messages.
278 struct nx_role_request {
279 ovs_be32 role; /* One of NX_ROLE_*. */
281 OFP_ASSERT(sizeof(struct nx_role_request) == 4);
284 NX_ROLE_OTHER, /* Default role, full access. */
285 NX_ROLE_MASTER, /* Full access, at most one. */
286 NX_ROLE_SLAVE /* Read-only access. */
289 /* NXT_SET_ASYNC_CONFIG.
291 * Sent by a controller, this message configures the asynchronous messages that
292 * the controller wants to receive. Element 0 in each array specifies messages
293 * of interest when the controller has an "other" or "master" role; element 1,
294 * when the controller has a "slave" role.
296 * Each array element is a bitmask in which a 0-bit disables receiving a
297 * particular message and a 1-bit enables receiving it. Each bit controls the
298 * message whose 'reason' corresponds to the bit index. For example, the bit
299 * with value 1<<2 == 4 in port_status_mask[1] determines whether the
300 * controller will receive OFPT_PORT_STATUS messages with reason OFPPR_MODIFY
301 * (value 2) when the controller has a "slave" role.
303 * As a side effect, for service controllers, this message changes the
304 * miss_send_len from default of zero to OFP_DEFAULT_MISS_SEND_LEN (128).
306 struct nx_async_config {
307 ovs_be32 packet_in_mask[2]; /* Bitmasks of OFPR_* values. */
308 ovs_be32 port_status_mask[2]; /* Bitmasks of OFPRR_* values. */
309 ovs_be32 flow_removed_mask[2]; /* Bitmasks of OFPPR_* values. */
311 OFP_ASSERT(sizeof(struct nx_async_config) == 24);
313 /* Flexible flow specifications (aka NXM = Nicira Extended Match).
315 * OpenFlow 1.0 has "struct ofp10_match" for specifying flow matches. This
316 * structure is fixed-length and hence difficult to extend. This section
317 * describes a more flexible, variable-length flow match, called "nx_match" for
318 * short, that is also supported by Open vSwitch. This section also defines a
319 * replacement for each OpenFlow message that includes struct ofp10_match.
321 * OpenFlow 1.2+ introduced OpenFlow Extensible Match (OXM), adapting
322 * the design of NXM. The format of NXM and OXM are compatible.
328 * An nx_match is a sequence of zero or more "nxm_entry"s, which are
329 * type-length-value (TLV) entries, each 5 to 259 (inclusive) bytes long.
330 * "nxm_entry"s are not aligned on or padded to any multibyte boundary. The
331 * first 4 bytes of an nxm_entry are its "header", followed by the entry's
334 * An nxm_entry's header is interpreted as a 32-bit word in network byte order:
336 * |<-------------------- nxm_type ------------------>|
339 * +----------------------------------+---------------+--+------------------+
340 * | nxm_vendor | nxm_field |hm| nxm_length |
341 * +----------------------------------+---------------+--+------------------+
343 * The most-significant 23 bits of the header are collectively "nxm_type".
344 * Bits 16...31 are "nxm_vendor", one of OFPXMC12_* values. In case of
345 * NXM, it's either OFPXMC12_NXM_0 or OFPXMC12_NXM_1.
346 * Bits 9...15 are "nxm_field", which is a vendor-specific value. nxm_type
347 * normally designates a protocol header, such as the Ethernet type, but it
348 * can also refer to packet metadata, such as the switch port on which a packet
351 * Bit 8 is "nxm_hasmask" (labeled "hm" above for space reasons). The meaning
352 * of this bit is explained later.
354 * The least-significant 8 bits are "nxm_length", a positive integer. The
355 * length of the nxm_entry, including the header, is exactly 4 + nxm_length
358 * For a given nxm_vendor, nxm_field, and nxm_hasmask value, nxm_length is a
359 * constant. It is included only to allow software to minimally parse
360 * "nxm_entry"s of unknown types. (Similarly, for a given nxm_vendor,
361 * nxm_field, and nxm_length, nxm_hasmask is a constant.)
367 * A zero-length nx_match (one with no "nxm_entry"s) matches every packet.
369 * An nxm_entry places a constraint on the packets matched by the nx_match:
371 * - If nxm_hasmask is 0, the nxm_entry's body contains a value for the
372 * field, called "nxm_value". The nx_match matches only packets in which
373 * the field equals nxm_value.
375 * - If nxm_hasmask is 1, then the nxm_entry's body contains a value for the
376 * field (nxm_value), followed by a bitmask of the same length as the
377 * value, called "nxm_mask". For each 1-bit in position J in nxm_mask, the
378 * nx_match matches only packets for which bit J in the given field's value
379 * matches bit J in nxm_value. A 0-bit in nxm_mask causes the
380 * corresponding bit in nxm_value is ignored (it should be 0; Open vSwitch
381 * may enforce this someday), as is the corresponding bit in the field's
382 * value. (The sense of the nxm_mask bits is the opposite of that used by
383 * the "wildcards" member of struct ofp10_match.)
385 * When nxm_hasmask is 1, nxm_length is always even.
387 * An all-zero-bits nxm_mask is equivalent to omitting the nxm_entry
388 * entirely. An all-one-bits nxm_mask is equivalent to specifying 0 for
391 * When there are multiple "nxm_entry"s, all of the constraints must be met.
397 * Masks may be restricted:
399 * - Some nxm_types may not support masked wildcards, that is, nxm_hasmask
400 * must always be 0 when these fields are specified. For example, the
401 * field that identifies the port on which a packet was received may not be
404 * - Some nxm_types that do support masked wildcards may only support certain
405 * nxm_mask patterns. For example, fields that have IPv4 address values
406 * may be restricted to CIDR masks.
408 * These restrictions should be noted in specifications for individual fields.
409 * A switch may accept an nxm_hasmask or nxm_mask value that the specification
410 * disallows, if the switch correctly implements support for that nxm_hasmask
411 * or nxm_mask value. A switch must reject an attempt to set up a flow that
412 * contains a nxm_hasmask or nxm_mask value that it does not support.
415 * Prerequisite Restrictions
416 * =========================
418 * The presence of an nxm_entry with a given nxm_type may be restricted based
419 * on the presence of or values of other "nxm_entry"s. For example:
421 * - An nxm_entry for nxm_type=NXM_OF_IP_TOS is allowed only if it is
422 * preceded by another entry with nxm_type=NXM_OF_ETH_TYPE, nxm_hasmask=0,
423 * and nxm_value=0x0800. That is, matching on the IP source address is
424 * allowed only if the Ethernet type is explicitly set to IP.
426 * - An nxm_entry for nxm_type=NXM_OF_TCP_SRC is allowed only if it is
427 * preceded by an entry with nxm_type=NXM_OF_ETH_TYPE, nxm_hasmask=0, and
428 * nxm_value either 0x0800 or 0x86dd, and another with
429 * nxm_type=NXM_OF_IP_PROTO, nxm_hasmask=0, nxm_value=6, in that order.
430 * That is, matching on the TCP source port is allowed only if the Ethernet
431 * type is IP or IPv6 and the IP protocol is TCP.
433 * These restrictions should be noted in specifications for individual fields.
434 * A switch may implement relaxed versions of these restrictions. A switch
435 * must reject an attempt to set up a flow that violates its restrictions.
438 * Ordering Restrictions
439 * =====================
441 * An nxm_entry that has prerequisite restrictions must appear after the
442 * "nxm_entry"s for its prerequisites. Ordering of "nxm_entry"s within an
443 * nx_match is not otherwise constrained.
445 * Any given nxm_type may appear in an nx_match at most once.
451 * These examples show the format of a single nxm_entry with particular
452 * nxm_hasmask and nxm_length values. The diagrams are labeled with field
453 * numbers and byte indexes.
456 * 8-bit nxm_value, nxm_hasmask=1, nxm_length=2:
459 * +------------+---+---+
461 * +------------+---+---+
464 * 16-bit nxm_value, nxm_hasmask=0, nxm_length=2:
467 * +------------+------+
469 * +------------+------+
472 * 32-bit nxm_value, nxm_hasmask=0, nxm_length=4:
475 * +------------+-------------+
476 * | header | nxm_value |
477 * +------------+-------------+
480 * 48-bit nxm_value, nxm_hasmask=0, nxm_length=6:
483 * +------------+------------------+
484 * | header | nxm_value |
485 * +------------+------------------+
488 * 48-bit nxm_value, nxm_hasmask=1, nxm_length=12:
491 * +------------+------------------+------------------+
492 * | header | nxm_value | nxm_mask |
493 * +------------+------------------+------------------+
499 * A switch should report an error in an nx_match using error type
500 * OFPET_BAD_REQUEST and one of the NXBRC_NXM_* codes. Ideally the switch
501 * should report a specific error code, if one is assigned for the particular
502 * problem, but NXBRC_NXM_INVALID is also available to report a generic
506 /* Number of registers allocated NXM field IDs. */
507 #define NXM_NX_MAX_REGS 16
509 /* Bits in the value of NXM_NX_IP_FRAG. */
510 #define NX_IP_FRAG_ANY (1 << 0) /* Is this a fragment? */
511 #define NX_IP_FRAG_LATER (1 << 1) /* Is this a fragment with nonzero offset? */
513 /* Bits in the value of NXM_NX_TUN_FLAGS. */
514 #define NX_TUN_FLAG_OAM (1 << 0) /* Is this an OAM packet? */
516 /* ## --------------------- ## */
517 /* ## Requests and replies. ## */
518 /* ## --------------------- ## */
520 enum nx_flow_format {
521 NXFF_OPENFLOW10 = 0, /* Standard OpenFlow 1.0 compatible. */
522 NXFF_NXM = 2 /* Nicira extended match. */
525 /* NXT_SET_FLOW_FORMAT request. */
526 struct nx_set_flow_format {
527 ovs_be32 format; /* One of NXFF_*. */
529 OFP_ASSERT(sizeof(struct nx_set_flow_format) == 4);
531 /* NXT_FLOW_MOD (analogous to OFPT_FLOW_MOD).
533 * It is possible to limit flow deletions and modifications to certain
534 * cookies by using the NXM_NX_COOKIE(_W) matches. The "cookie" field
535 * is used only to add or modify flow cookies.
538 ovs_be64 cookie; /* Opaque controller-issued identifier. */
539 ovs_be16 command; /* OFPFC_* + possibly a table ID (see comment
540 * on struct nx_flow_mod_table_id). */
541 ovs_be16 idle_timeout; /* Idle time before discarding (seconds). */
542 ovs_be16 hard_timeout; /* Max time before discarding (seconds). */
543 ovs_be16 priority; /* Priority level of flow entry. */
544 ovs_be32 buffer_id; /* Buffered packet to apply to (or -1).
545 Not meaningful for OFPFC_DELETE*. */
546 ovs_be16 out_port; /* For OFPFC_DELETE* commands, require
547 matching entries to include this as an
548 output port. A value of OFPP_NONE
549 indicates no restriction. */
550 ovs_be16 flags; /* One of OFPFF_*. */
551 ovs_be16 match_len; /* Size of nx_match. */
552 uint8_t pad[6]; /* Align to 64-bits. */
554 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
555 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
556 * all-zero bytes, then
557 * - Actions to fill out the remainder of the message length (always a
561 OFP_ASSERT(sizeof(struct nx_flow_mod) == 32);
563 /* NXT_FLOW_REMOVED (analogous to OFPT_FLOW_REMOVED).
565 * 'table_id' is present only in Open vSwitch 1.11 and later. In earlier
566 * versions of Open vSwitch, this is a padding byte that is always zeroed.
567 * Therefore, a 'table_id' value of 0 indicates that the table ID is not known,
568 * and other values may be interpreted as one more than the flow's former table
570 struct nx_flow_removed {
571 ovs_be64 cookie; /* Opaque controller-issued identifier. */
572 ovs_be16 priority; /* Priority level of flow entry. */
573 uint8_t reason; /* One of OFPRR_*. */
574 uint8_t table_id; /* Flow's former table ID, plus one. */
575 ovs_be32 duration_sec; /* Time flow was alive in seconds. */
576 ovs_be32 duration_nsec; /* Time flow was alive in nanoseconds beyond
578 ovs_be16 idle_timeout; /* Idle timeout from original flow mod. */
579 ovs_be16 match_len; /* Size of nx_match. */
580 ovs_be64 packet_count;
583 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
584 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
587 OFP_ASSERT(sizeof(struct nx_flow_removed) == 40);
589 /* Nicira vendor stats request of type NXST_FLOW (analogous to OFPST_FLOW
592 * It is possible to limit matches to certain cookies by using the
593 * NXM_NX_COOKIE and NXM_NX_COOKIE_W matches.
595 struct nx_flow_stats_request {
596 ovs_be16 out_port; /* Require matching entries to include this
597 as an output port. A value of OFPP_NONE
598 indicates no restriction. */
599 ovs_be16 match_len; /* Length of nx_match. */
600 uint8_t table_id; /* ID of table to read (from ofp_table_stats)
601 or 0xff for all tables. */
602 uint8_t pad[3]; /* Align to 64 bits. */
604 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
605 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
606 * all-zero bytes, which must also exactly fill out the length of the
610 OFP_ASSERT(sizeof(struct nx_flow_stats_request) == 8);
612 /* Body for Nicira vendor stats reply of type NXST_FLOW (analogous to
615 * The values of 'idle_age' and 'hard_age' are only meaningful when talking to
616 * a switch that implements the NXT_FLOW_AGE extension. Zero means that the
617 * true value is unknown, perhaps because hardware does not track the value.
618 * (Zero is also the value that one should ordinarily expect to see talking to
619 * a switch that does not implement NXT_FLOW_AGE, since those switches zero the
620 * padding bytes that these fields replaced.) A nonzero value X represents X-1
621 * seconds. A value of 65535 represents 65534 or more seconds.
623 * 'idle_age' is the number of seconds that the flow has been idle, that is,
624 * the number of seconds since a packet passed through the flow. 'hard_age' is
625 * the number of seconds since the flow was last modified (e.g. OFPFC_MODIFY or
626 * OFPFC_MODIFY_STRICT). (The 'duration_*' fields are the elapsed time since
627 * the flow was added, regardless of subsequent modifications.)
629 * For a flow with an idle or hard timeout, 'idle_age' or 'hard_age',
630 * respectively, will ordinarily be smaller than the timeout, but flow
631 * expiration times are only approximate and so one must be prepared to
632 * tolerate expirations that occur somewhat early or late.
634 struct nx_flow_stats {
635 ovs_be16 length; /* Length of this entry. */
636 uint8_t table_id; /* ID of table flow came from. */
638 ovs_be32 duration_sec; /* Time flow has been alive in seconds. */
639 ovs_be32 duration_nsec; /* Time flow has been alive in nanoseconds
640 beyond duration_sec. */
641 ovs_be16 priority; /* Priority of the entry. */
642 ovs_be16 idle_timeout; /* Number of seconds idle before expiration. */
643 ovs_be16 hard_timeout; /* Number of seconds before expiration. */
644 ovs_be16 match_len; /* Length of nx_match. */
645 ovs_be16 idle_age; /* Seconds since last packet, plus one. */
646 ovs_be16 hard_age; /* Seconds since last modification, plus one. */
647 ovs_be64 cookie; /* Opaque controller-issued identifier. */
648 ovs_be64 packet_count; /* Number of packets, UINT64_MAX if unknown. */
649 ovs_be64 byte_count; /* Number of bytes, UINT64_MAX if unknown. */
651 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
652 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
653 * all-zero bytes, then
654 * - Actions to fill out the remainder 'length' bytes (always a multiple
658 OFP_ASSERT(sizeof(struct nx_flow_stats) == 48);
660 /* Nicira vendor stats request of type NXST_AGGREGATE (analogous to
661 * OFPST_AGGREGATE request).
663 * The reply format is identical to the reply format for OFPST_AGGREGATE,
664 * except for the header. */
665 struct nx_aggregate_stats_request {
666 ovs_be16 out_port; /* Require matching entries to include this
667 as an output port. A value of OFPP_NONE
668 indicates no restriction. */
669 ovs_be16 match_len; /* Length of nx_match. */
670 uint8_t table_id; /* ID of table to read (from ofp_table_stats)
671 or 0xff for all tables. */
672 uint8_t pad[3]; /* Align to 64 bits. */
674 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
675 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
676 * all-zero bytes, which must also exactly fill out the length of the
680 OFP_ASSERT(sizeof(struct nx_aggregate_stats_request) == 8);
682 /* NXT_SET_CONTROLLER_ID.
684 * Each OpenFlow controller connection has a 16-bit identifier that is
685 * initially 0. This message changes the connection's ID to 'id'.
687 * Controller connection IDs need not be unique.
689 * The NXAST_CONTROLLER action is the only current user of controller
691 struct nx_controller_id {
692 uint8_t zero[6]; /* Must be zero. */
693 ovs_be16 controller_id; /* New controller connection ID. */
695 OFP_ASSERT(sizeof(struct nx_controller_id) == 8);
697 /* Flow Table Monitoring
698 * =====================
700 * NXST_FLOW_MONITOR allows a controller to keep track of changes to OpenFlow
701 * flow table(s) or subsets of them, with the following workflow:
703 * 1. The controller sends an NXST_FLOW_MONITOR request to begin monitoring
704 * flows. The 'id' in the request must be unique among all monitors that
705 * the controller has started and not yet canceled on this OpenFlow
708 * 2. The switch responds with an NXST_FLOW_MONITOR reply. If the request's
709 * 'flags' included NXFMF_INITIAL, the reply includes all the flows that
710 * matched the request at the time of the request (with event NXFME_ADDED).
711 * If 'flags' did not include NXFMF_INITIAL, the reply is empty.
713 * The reply uses the xid of the request (as do all replies to OpenFlow
716 * 3. Whenever a change to a flow table entry matches some outstanding monitor
717 * request's criteria and flags, the switch sends a notification to the
718 * controller as an additional NXST_FLOW_MONITOR reply with xid 0.
720 * When multiple outstanding monitors match a single change, only a single
721 * notification is sent. This merged notification includes the information
722 * requested in any of the individual monitors. That is, if any of the
723 * matching monitors requests actions (NXFMF_ACTIONS), the notification
724 * includes actions, and if any of the monitors request full changes for the
725 * controller's own changes (NXFMF_OWN), the controller's own changes will
726 * be included in full.
728 * 4. The controller may cancel a monitor with NXT_FLOW_MONITOR_CANCEL. No
729 * further notifications will be sent on the basis of the canceled monitor
736 * OpenFlow messages for flow monitor notifications can overflow the buffer
737 * space available to the switch, either temporarily (e.g. due to network
738 * conditions slowing OpenFlow traffic) or more permanently (e.g. the sustained
739 * rate of flow table change exceeds the network bandwidth between switch and
742 * When Open vSwitch's notification buffer space reaches a limiting threshold,
743 * OVS reacts as follows:
745 * 1. OVS sends an NXT_FLOW_MONITOR_PAUSED message to the controller, following
746 * all the already queued notifications. After it receives this message,
747 * the controller knows that its view of the flow table, as represented by
748 * flow monitor notifications, is incomplete.
750 * 2. As long as the notification buffer is not empty:
752 * - NXMFE_ADD and NXFME_MODIFIED notifications will not be sent.
754 * - NXFME_DELETED notifications will still be sent, but only for flows
755 * that existed before OVS sent NXT_FLOW_MONITOR_PAUSED.
757 * - NXFME_ABBREV notifications will not be sent. They are treated as
758 * the expanded version (and therefore only the NXFME_DELETED
759 * components, if any, are sent).
761 * 3. When the notification buffer empties, OVS sends NXFME_ADD notifications
762 * for flows added since the buffer reached its limit and NXFME_MODIFIED
763 * notifications for flows that existed before the limit was reached and
764 * changed after the limit was reached.
766 * 4. OVS sends an NXT_FLOW_MONITOR_RESUMED message to the controller. After
767 * it receives this message, the controller knows that its view of the flow
768 * table, as represented by flow monitor notifications, is again complete.
770 * This allows the maximum buffer space requirement for notifications to be
771 * bounded by the limit plus the maximum number of supported flows.
774 * "Flow Removed" messages
775 * =======================
777 * The flow monitor mechanism is independent of OFPT_FLOW_REMOVED and
778 * NXT_FLOW_REMOVED. Flow monitor updates for deletion are sent if
779 * NXFMF_DELETE is set on a monitor, regardless of whether the
780 * OFPFF_SEND_FLOW_REM flag was set when the flow was added. */
782 /* NXST_FLOW_MONITOR request.
784 * The NXST_FLOW_MONITOR request's body consists of an array of zero or more
785 * instances of this structure. The request arranges to monitor the flows
786 * that match the specified criteria, which are interpreted in the same way as
789 * 'id' identifies a particular monitor for the purpose of allowing it to be
790 * canceled later with NXT_FLOW_MONITOR_CANCEL. 'id' must be unique among
791 * existing monitors that have not already been canceled.
793 * The reply includes the initial flow matches for monitors that have the
794 * NXFMF_INITIAL flag set. No single flow will be included in the reply more
795 * than once, even if more than one requested monitor matches that flow. The
796 * reply will be empty if none of the monitors has NXFMF_INITIAL set or if none
797 * of the monitors initially matches any flows.
799 * For NXFMF_ADD, an event will be reported if 'out_port' matches against the
800 * actions of the flow being added or, for a flow that is replacing an existing
801 * flow, if 'out_port' matches against the actions of the flow being replaced.
802 * For NXFMF_DELETE, 'out_port' matches against the actions of a flow being
803 * deleted. For NXFMF_MODIFY, an event will be reported if 'out_port' matches
804 * either the old or the new actions. */
805 struct nx_flow_monitor_request {
806 ovs_be32 id; /* Controller-assigned ID for this monitor. */
807 ovs_be16 flags; /* NXFMF_*. */
808 ovs_be16 out_port; /* Required output port, if not OFPP_NONE. */
809 ovs_be16 match_len; /* Length of nx_match. */
810 uint8_t table_id; /* One table's ID or 0xff for all tables. */
811 uint8_t zeros[5]; /* Align to 64 bits (must be zero). */
813 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
814 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
817 OFP_ASSERT(sizeof(struct nx_flow_monitor_request) == 16);
819 /* 'flags' bits in struct nx_flow_monitor_request. */
820 enum nx_flow_monitor_flags {
821 /* When to send updates. */
822 NXFMF_INITIAL = 1 << 0, /* Initially matching flows. */
823 NXFMF_ADD = 1 << 1, /* New matching flows as they are added. */
824 NXFMF_DELETE = 1 << 2, /* Old matching flows as they are removed. */
825 NXFMF_MODIFY = 1 << 3, /* Matching flows as they are changed. */
827 /* What to include in updates. */
828 NXFMF_ACTIONS = 1 << 4, /* If set, actions are included. */
829 NXFMF_OWN = 1 << 5, /* If set, include own changes in full. */
832 /* NXST_FLOW_MONITOR reply header.
834 * The body of an NXST_FLOW_MONITOR reply is an array of variable-length
835 * structures, each of which begins with this header. The 'length' member may
836 * be used to traverse the array, and the 'event' member may be used to
837 * determine the particular structure.
839 * Every instance is a multiple of 8 bytes long. */
840 struct nx_flow_update_header {
841 ovs_be16 length; /* Length of this entry. */
842 ovs_be16 event; /* One of NXFME_*. */
843 /* ...other data depending on 'event'... */
845 OFP_ASSERT(sizeof(struct nx_flow_update_header) == 4);
847 /* 'event' values in struct nx_flow_update_header. */
848 enum nx_flow_update_event {
849 /* struct nx_flow_update_full. */
850 NXFME_ADDED = 0, /* Flow was added. */
851 NXFME_DELETED = 1, /* Flow was deleted. */
852 NXFME_MODIFIED = 2, /* Flow (generally its actions) was changed. */
854 /* struct nx_flow_update_abbrev. */
855 NXFME_ABBREV = 3, /* Abbreviated reply. */
858 /* NXST_FLOW_MONITOR reply for NXFME_ADDED, NXFME_DELETED, and
860 struct nx_flow_update_full {
861 ovs_be16 length; /* Length is 24. */
862 ovs_be16 event; /* One of NXFME_*. */
863 ovs_be16 reason; /* OFPRR_* for NXFME_DELETED, else zero. */
864 ovs_be16 priority; /* Priority of the entry. */
865 ovs_be16 idle_timeout; /* Number of seconds idle before expiration. */
866 ovs_be16 hard_timeout; /* Number of seconds before expiration. */
867 ovs_be16 match_len; /* Length of nx_match. */
868 uint8_t table_id; /* ID of flow's table. */
869 uint8_t pad; /* Reserved, currently zeroed. */
870 ovs_be64 cookie; /* Opaque controller-issued identifier. */
872 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
873 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
874 * all-zero bytes, then
875 * - Actions to fill out the remainder 'length' bytes (always a multiple
876 * of 8). If NXFMF_ACTIONS was not specified, or 'event' is
877 * NXFME_DELETED, no actions are included.
880 OFP_ASSERT(sizeof(struct nx_flow_update_full) == 24);
882 /* NXST_FLOW_MONITOR reply for NXFME_ABBREV.
884 * When the controller does not specify NXFMF_OWN in a monitor request, any
885 * flow tables changes due to the controller's own requests (on the same
886 * OpenFlow channel) will be abbreviated, when possible, to this form, which
887 * simply specifies the 'xid' of the OpenFlow request (e.g. an OFPT_FLOW_MOD or
888 * NXT_FLOW_MOD) that caused the change.
890 * Some changes cannot be abbreviated and will be sent in full:
892 * - Changes that only partially succeed. This can happen if, for example,
893 * a flow_mod with type OFPFC_MODIFY affects multiple flows, but only some
894 * of those modifications succeed (e.g. due to hardware limitations).
896 * This cannot occur with the Open vSwitch software datapath. This also
897 * cannot occur in Open vSwitch 2.4 and later, because these versions only
898 * execute any flow modifications if all of them will succeed.
900 * - Changes that race with conflicting changes made by other controllers or
901 * other flow_mods (not separated by barriers) by the same controller.
903 * This cannot occur with the current Open vSwitch implementation
904 * (regardless of datapath) because Open vSwitch internally serializes
905 * potentially conflicting changes.
907 * - Changes that occur when flow notification is paused (see "Buffer
908 * Management" above).
910 * A flow_mod that does not change the flow table will not trigger any
911 * notification, even an abbreviated one. For example, a "modify" or "delete"
912 * flow_mod that does not match any flows will not trigger a notification.
913 * Whether an "add" or "modify" that specifies all the same parameters that a
914 * flow already has triggers a notification is unspecified and subject to
915 * change in future versions of Open vSwitch.
917 * OVS will always send the notifications for a given flow table change before
918 * the reply to a OFPT_BARRIER_REQUEST request that follows the flow table
919 * change. Thus, if the controller does not receive an abbreviated (or
920 * unabbreviated) notification for a flow_mod before the next
921 * OFPT_BARRIER_REPLY, it will never receive one. */
922 struct nx_flow_update_abbrev {
923 ovs_be16 length; /* Length is 8. */
924 ovs_be16 event; /* NXFME_ABBREV. */
925 ovs_be32 xid; /* Controller-specified xid from flow_mod. */
927 OFP_ASSERT(sizeof(struct nx_flow_update_abbrev) == 8);
929 /* NXT_FLOW_MONITOR_CANCEL.
931 * Used by a controller to cancel an outstanding monitor. */
932 struct nx_flow_monitor_cancel {
933 ovs_be32 id; /* 'id' from nx_flow_monitor_request. */
935 OFP_ASSERT(sizeof(struct nx_flow_monitor_cancel) == 4);
937 /* Variable-length option TLV table maintenance commands.
939 * The option in Type-Length-Value format is widely used in tunnel options,
940 * e.g., the base Geneve header is followed by zero or more options in TLV
941 * format. Each option consists of a four byte option header and a variable
942 * amount of option data interpreted according to the type. The generic TLV
943 * format in tunnel options is as following:
946 * 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
947 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
948 * | Option Class | Type |R|R|R| Length |
949 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
950 * | Variable Option Data |
951 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
953 * In order to work with this variable-length options in TLV format in
954 * tunnel options, we need to maintain a mapping table between an option
955 * TLV (defined by <class, type, length>) and an NXM field that can be
956 * operated on for the purposes of matches, actions, etc. This mapping
957 * must be explicitly specified by the user.
959 * There are two primary groups of OpenFlow messages that are introduced
960 * as Nicira extensions: modification commands (add, delete, clear mappings)
961 * and table status request/reply to dump the current table along with switch
964 * Note that mappings should not be changed while they are in active use by
965 * a flow. The result of doing so is undefined. */
967 /* TLV table commands */
968 enum nx_tlv_table_mod_command {
969 NXTTMC_ADD, /* New mappings (fails if an option is already
971 NXTTMC_DELETE, /* Delete mappings, identified by index
972 * (unmapped options are ignored). */
973 NXTTMC_CLEAR, /* Clear all mappings. Additional information
974 in this command is ignored. */
977 /* Map between an option TLV and an NXM field. */
979 ovs_be16 option_class; /* TLV class. */
980 uint8_t option_type; /* TLV type. */
981 uint8_t option_len; /* TLV length (multiple of 4). */
982 ovs_be16 index; /* NXM_NX_TUN_METADATA<n> index */
985 OFP_ASSERT(sizeof(struct nx_tlv_map) == 8);
987 /* NXT_TLV_TABLE_MOD.
989 * Use to configure a mapping between option TLVs (class, type, length)
990 * and NXM fields (NXM_NX_TUN_METADATA<n> where 'index' is <n>).
992 * This command is atomic: all operations on different options will
993 * either succeed or fail. */
994 struct nx_tlv_table_mod {
995 ovs_be16 command; /* One of NTTTMC_* */
997 /* struct nx_tlv_map[0]; Array of maps between indicies and option
998 TLVs. The number of elements is inferred
999 from the length field in the header. */
1001 OFP_ASSERT(sizeof(struct nx_tlv_table_mod) == 8);
1003 /* NXT_TLV_TABLE_REPLY.
1005 * Issued in reponse to an NXT_TLV_TABLE_REQUEST to give information
1006 * about the current status of the TLV table in the switch. Provides
1007 * both static information about the switch's capabilities as well as
1008 * the configured TLV table. */
1009 struct nx_tlv_table_reply {
1010 ovs_be32 max_option_space; /* Maximum total of option sizes supported. */
1011 ovs_be16 max_fields; /* Maximum number of match fields supported. */
1012 uint8_t reserved[10];
1013 /* struct nx_tlv_map[0]; Array of maps between indicies and option
1014 TLVs. The number of elements is inferred
1015 from the length field in the header. */
1017 OFP_ASSERT(sizeof(struct nx_tlv_table_reply) == 16);
1019 #endif /* openflow/nicira-ext.h */