2 * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014 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 requests and replies. */
72 struct nicira_header {
73 struct ofp_header header;
74 ovs_be32 vendor; /* NX_VENDOR_ID. */
75 ovs_be32 subtype; /* See the NXT numbers in ofp-msgs.h. */
77 OFP_ASSERT(sizeof(struct nicira_header) == 16);
79 /* Header for Nicira vendor stats request and reply messages in OpenFlow
81 struct nicira10_stats_msg {
82 struct ofp10_vendor_stats_msg vsm; /* Vendor NX_VENDOR_ID. */
83 ovs_be32 subtype; /* One of NXST_* below. */
84 uint8_t pad[4]; /* Align to 64-bits. */
86 OFP_ASSERT(sizeof(struct nicira10_stats_msg) == 24);
88 /* Header for Nicira vendor stats request and reply messages in OpenFlow
90 struct nicira11_stats_msg {
91 struct ofp11_vendor_stats_msg vsm; /* Vendor NX_VENDOR_ID. */
92 ovs_be32 subtype; /* One of NXST_* below. */
94 OFP_ASSERT(sizeof(struct nicira11_stats_msg) == 24);
96 /* Fields to use when hashing flows. */
98 /* Ethernet source address (NXM_OF_ETH_SRC) only. */
99 NX_HASH_FIELDS_ETH_SRC,
101 /* L2 through L4, symmetric across src/dst. Specifically, each of the
102 * following fields, if present, is hashed (slashes separate symmetric
105 * - NXM_OF_ETH_DST / NXM_OF_ETH_SRC
107 * - The VID bits from NXM_OF_VLAN_TCI, ignoring PCP and CFI.
109 * - NXM_OF_IP_SRC / NXM_OF_IP_DST
110 * - NXM_OF_TCP_SRC / NXM_OF_TCP_DST
112 NX_HASH_FIELDS_SYMMETRIC_L4
115 /* This command enables or disables an Open vSwitch extension that allows a
116 * controller to specify the OpenFlow table to which a flow should be added,
117 * instead of having the switch decide which table is most appropriate as
118 * required by OpenFlow 1.0. Because NXM was designed as an extension to
119 * OpenFlow 1.0, the extension applies equally to ofp10_flow_mod and
120 * nx_flow_mod. By default, the extension is disabled.
122 * When this feature is enabled, Open vSwitch treats struct ofp10_flow_mod's
123 * and struct nx_flow_mod's 16-bit 'command' member as two separate fields.
124 * The upper 8 bits are used as the table ID, the lower 8 bits specify the
125 * command as usual. A table ID of 0xff is treated like a wildcarded table ID.
127 * The specific treatment of the table ID depends on the type of flow mod:
129 * - OFPFC_ADD: Given a specific table ID, the flow is always placed in that
130 * table. If an identical flow already exists in that table only, then it
131 * is replaced. If the flow cannot be placed in the specified table,
132 * either because the table is full or because the table cannot support
133 * flows of the given type, the switch replies with an OFPFMFC_TABLE_FULL
134 * error. (A controller can distinguish these cases by comparing the
135 * current and maximum number of entries reported in ofp_table_stats.)
137 * If the table ID is wildcarded, the switch picks an appropriate table
138 * itself. If an identical flow already exist in the selected flow table,
139 * then it is replaced. The choice of table might depend on the flows
140 * that are already in the switch; for example, if one table fills up then
141 * the switch might fall back to another one.
143 * - OFPFC_MODIFY, OFPFC_DELETE: Given a specific table ID, only flows
144 * within that table are matched and modified or deleted. If the table ID
145 * is wildcarded, flows within any table may be matched and modified or
148 * - OFPFC_MODIFY_STRICT, OFPFC_DELETE_STRICT: Given a specific table ID,
149 * only a flow within that table may be matched and modified or deleted.
150 * If the table ID is wildcarded and exactly one flow within any table
151 * matches, then it is modified or deleted; if flows in more than one
152 * table match, then none is modified or deleted.
154 struct nx_flow_mod_table_id {
155 uint8_t set; /* Nonzero to enable, zero to disable. */
158 OFP_ASSERT(sizeof(struct nx_flow_mod_table_id) == 8);
160 enum nx_packet_in_format {
161 NXPIF_OPENFLOW10 = 0, /* Standard OpenFlow 1.0 compatible. */
162 NXPIF_NXM = 1 /* Nicira Extended. */
165 /* NXT_SET_PACKET_IN_FORMAT request. */
166 struct nx_set_packet_in_format {
167 ovs_be32 format; /* One of NXPIF_*. */
169 OFP_ASSERT(sizeof(struct nx_set_packet_in_format) == 4);
171 /* NXT_PACKET_IN (analogous to OFPT_PACKET_IN).
173 * NXT_PACKET_IN is similar to the OpenFlow 1.2 OFPT_PACKET_IN. The
176 * - NXT_PACKET_IN includes the cookie of the rule that triggered the
177 * message. (OpenFlow 1.3 OFPT_PACKET_IN also includes the cookie.)
179 * - The metadata fields use NXM (instead of OXM) field numbers.
181 * Open vSwitch 1.9.0 and later omits metadata fields that are zero (as allowed
182 * by OpenFlow 1.2). Earlier versions included all implemented metadata
185 * Open vSwitch does not include non-metadata in the nx_match, because by
186 * definition that information can be found in the packet itself. The format
187 * and the standards allow this, however, so controllers should be prepared to
188 * tolerate future changes.
190 * The NXM format is convenient for reporting metadata values, but it is
191 * important not to interpret the format as matching against a flow, because it
192 * does not. Nothing is being matched; arbitrary metadata masks would not be
195 * Whereas in most cases a controller can expect to only get back NXM fields
196 * that it set up itself (e.g. flow dumps will ordinarily report only NXM
197 * fields from flows that the controller added), NXT_PACKET_IN messages might
198 * contain fields that the controller does not understand, because the switch
199 * might support fields (new registers, new protocols, etc.) that the
200 * controller does not. The controller must prepared to tolerate these.
202 * The 'cookie' field has no meaning when 'reason' is OFPR_NO_MATCH. In this
203 * case it should be UINT64_MAX. */
204 struct nx_packet_in {
205 ovs_be32 buffer_id; /* ID assigned by datapath. */
206 ovs_be16 total_len; /* Full length of frame. */
207 uint8_t reason; /* Reason packet is sent (one of OFPR_*). */
208 uint8_t table_id; /* ID of the table that was looked up. */
209 ovs_be64 cookie; /* Cookie of the rule that was looked up. */
210 ovs_be16 match_len; /* Size of nx_match. */
211 uint8_t pad[6]; /* Align to 64-bits. */
213 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
214 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
215 * all-zero bytes, then
216 * - Exactly 2 all-zero padding bytes, then
217 * - An Ethernet frame whose length is inferred from nxh.header.length.
219 * The padding bytes preceding the Ethernet frame ensure that the IP
220 * header (if any) following the Ethernet header is 32-bit aligned. */
222 /* uint8_t nxm_fields[...]; */ /* NXM headers. */
223 /* uint8_t pad[2]; */ /* Align to 64 bit + 16 bit. */
224 /* uint8_t data[0]; */ /* Ethernet frame. */
226 OFP_ASSERT(sizeof(struct nx_packet_in) == 24);
228 /* Configures the "role" of the sending controller. The default role is:
230 * - Other (NX_ROLE_OTHER), which allows the controller access to all
233 * The other possible roles are a related pair:
235 * - Master (NX_ROLE_MASTER) is equivalent to Other, except that there may
236 * be at most one Master controller at a time: when a controller
237 * configures itself as Master, any existing Master is demoted to the
240 * - Slave (NX_ROLE_SLAVE) allows the controller read-only access to
241 * OpenFlow features. In particular attempts to modify the flow table
242 * will be rejected with an OFPBRC_EPERM error.
244 * Slave controllers do not receive OFPT_PACKET_IN or OFPT_FLOW_REMOVED
245 * messages, but they do receive OFPT_PORT_STATUS messages.
247 struct nx_role_request {
248 ovs_be32 role; /* One of NX_ROLE_*. */
250 OFP_ASSERT(sizeof(struct nx_role_request) == 4);
253 NX_ROLE_OTHER, /* Default role, full access. */
254 NX_ROLE_MASTER, /* Full access, at most one. */
255 NX_ROLE_SLAVE /* Read-only access. */
258 /* NXT_SET_ASYNC_CONFIG.
260 * Sent by a controller, this message configures the asynchronous messages that
261 * the controller wants to receive. Element 0 in each array specifies messages
262 * of interest when the controller has an "other" or "master" role; element 1,
263 * when the controller has a "slave" role.
265 * Each array element is a bitmask in which a 0-bit disables receiving a
266 * particular message and a 1-bit enables receiving it. Each bit controls the
267 * message whose 'reason' corresponds to the bit index. For example, the bit
268 * with value 1<<2 == 4 in port_status_mask[1] determines whether the
269 * controller will receive OFPT_PORT_STATUS messages with reason OFPPR_MODIFY
270 * (value 2) when the controller has a "slave" role.
272 * As a side effect, for service controllers, this message changes the
273 * miss_send_len from default of zero to OFP_DEFAULT_MISS_SEND_LEN (128).
275 struct nx_async_config {
276 ovs_be32 packet_in_mask[2]; /* Bitmasks of OFPR_* values. */
277 ovs_be32 port_status_mask[2]; /* Bitmasks of OFPRR_* values. */
278 ovs_be32 flow_removed_mask[2]; /* Bitmasks of OFPPR_* values. */
280 OFP_ASSERT(sizeof(struct nx_async_config) == 24);
282 /* Flexible flow specifications (aka NXM = Nicira Extended Match).
284 * OpenFlow 1.0 has "struct ofp10_match" for specifying flow matches. This
285 * structure is fixed-length and hence difficult to extend. This section
286 * describes a more flexible, variable-length flow match, called "nx_match" for
287 * short, that is also supported by Open vSwitch. This section also defines a
288 * replacement for each OpenFlow message that includes struct ofp10_match.
294 * An nx_match is a sequence of zero or more "nxm_entry"s, which are
295 * type-length-value (TLV) entries, each 5 to 259 (inclusive) bytes long.
296 * "nxm_entry"s are not aligned on or padded to any multibyte boundary. The
297 * first 4 bytes of an nxm_entry are its "header", followed by the entry's
300 * An nxm_entry's header is interpreted as a 32-bit word in network byte order:
302 * |<-------------------- nxm_type ------------------>|
305 * +----------------------------------+---------------+--+------------------+
306 * | nxm_vendor | nxm_field |hm| nxm_length |
307 * +----------------------------------+---------------+--+------------------+
309 * The most-significant 23 bits of the header are collectively "nxm_type".
310 * Bits 16...31 are "nxm_vendor", one of the NXM_VENDOR_* values below. Bits
311 * 9...15 are "nxm_field", which is a vendor-specific value. nxm_type normally
312 * designates a protocol header, such as the Ethernet type, but it can also
313 * refer to packet metadata, such as the switch port on which a packet arrived.
315 * Bit 8 is "nxm_hasmask" (labeled "hm" above for space reasons). The meaning
316 * of this bit is explained later.
318 * The least-significant 8 bits are "nxm_length", a positive integer. The
319 * length of the nxm_entry, including the header, is exactly 4 + nxm_length
322 * For a given nxm_vendor, nxm_field, and nxm_hasmask value, nxm_length is a
323 * constant. It is included only to allow software to minimally parse
324 * "nxm_entry"s of unknown types. (Similarly, for a given nxm_vendor,
325 * nxm_field, and nxm_length, nxm_hasmask is a constant.)
331 * A zero-length nx_match (one with no "nxm_entry"s) matches every packet.
333 * An nxm_entry places a constraint on the packets matched by the nx_match:
335 * - If nxm_hasmask is 0, the nxm_entry's body contains a value for the
336 * field, called "nxm_value". The nx_match matches only packets in which
337 * the field equals nxm_value.
339 * - If nxm_hasmask is 1, then the nxm_entry's body contains a value for the
340 * field (nxm_value), followed by a bitmask of the same length as the
341 * value, called "nxm_mask". For each 1-bit in position J in nxm_mask, the
342 * nx_match matches only packets for which bit J in the given field's value
343 * matches bit J in nxm_value. A 0-bit in nxm_mask causes the
344 * corresponding bit in nxm_value is ignored (it should be 0; Open vSwitch
345 * may enforce this someday), as is the corresponding bit in the field's
346 * value. (The sense of the nxm_mask bits is the opposite of that used by
347 * the "wildcards" member of struct ofp10_match.)
349 * When nxm_hasmask is 1, nxm_length is always even.
351 * An all-zero-bits nxm_mask is equivalent to omitting the nxm_entry
352 * entirely. An all-one-bits nxm_mask is equivalent to specifying 0 for
355 * When there are multiple "nxm_entry"s, all of the constraints must be met.
361 * Masks may be restricted:
363 * - Some nxm_types may not support masked wildcards, that is, nxm_hasmask
364 * must always be 0 when these fields are specified. For example, the
365 * field that identifies the port on which a packet was received may not be
368 * - Some nxm_types that do support masked wildcards may only support certain
369 * nxm_mask patterns. For example, fields that have IPv4 address values
370 * may be restricted to CIDR masks.
372 * These restrictions should be noted in specifications for individual fields.
373 * A switch may accept an nxm_hasmask or nxm_mask value that the specification
374 * disallows, if the switch correctly implements support for that nxm_hasmask
375 * or nxm_mask value. A switch must reject an attempt to set up a flow that
376 * contains a nxm_hasmask or nxm_mask value that it does not support.
379 * Prerequisite Restrictions
380 * =========================
382 * The presence of an nxm_entry with a given nxm_type may be restricted based
383 * on the presence of or values of other "nxm_entry"s. For example:
385 * - An nxm_entry for nxm_type=NXM_OF_IP_TOS is allowed only if it is
386 * preceded by another entry with nxm_type=NXM_OF_ETH_TYPE, nxm_hasmask=0,
387 * and nxm_value=0x0800. That is, matching on the IP source address is
388 * allowed only if the Ethernet type is explicitly set to IP.
390 * - An nxm_entry for nxm_type=NXM_OF_TCP_SRC is allowed only if it is
391 * preceded by an entry with nxm_type=NXM_OF_ETH_TYPE, nxm_hasmask=0, and
392 * nxm_value either 0x0800 or 0x86dd, and another with
393 * nxm_type=NXM_OF_IP_PROTO, nxm_hasmask=0, nxm_value=6, in that order.
394 * That is, matching on the TCP source port is allowed only if the Ethernet
395 * type is IP or IPv6 and the IP protocol is TCP.
397 * These restrictions should be noted in specifications for individual fields.
398 * A switch may implement relaxed versions of these restrictions. A switch
399 * must reject an attempt to set up a flow that violates its restrictions.
402 * Ordering Restrictions
403 * =====================
405 * An nxm_entry that has prerequisite restrictions must appear after the
406 * "nxm_entry"s for its prerequisites. Ordering of "nxm_entry"s within an
407 * nx_match is not otherwise constrained.
409 * Any given nxm_type may appear in an nx_match at most once.
415 * These examples show the format of a single nxm_entry with particular
416 * nxm_hasmask and nxm_length values. The diagrams are labeled with field
417 * numbers and byte indexes.
420 * 8-bit nxm_value, nxm_hasmask=1, nxm_length=2:
423 * +------------+---+---+
425 * +------------+---+---+
428 * 16-bit nxm_value, nxm_hasmask=0, nxm_length=2:
431 * +------------+------+
433 * +------------+------+
436 * 32-bit nxm_value, nxm_hasmask=0, nxm_length=4:
439 * +------------+-------------+
440 * | header | nxm_value |
441 * +------------+-------------+
444 * 48-bit nxm_value, nxm_hasmask=0, nxm_length=6:
447 * +------------+------------------+
448 * | header | nxm_value |
449 * +------------+------------------+
452 * 48-bit nxm_value, nxm_hasmask=1, nxm_length=12:
455 * +------------+------------------+------------------+
456 * | header | nxm_value | nxm_mask |
457 * +------------+------------------+------------------+
463 * A switch should report an error in an nx_match using error type
464 * OFPET_BAD_REQUEST and one of the NXBRC_NXM_* codes. Ideally the switch
465 * should report a specific error code, if one is assigned for the particular
466 * problem, but NXBRC_NXM_INVALID is also available to report a generic
470 #define NXM_HEADER__(VENDOR, FIELD, HASMASK, LENGTH) \
471 (((VENDOR) << 16) | ((FIELD) << 9) | ((HASMASK) << 8) | (LENGTH))
472 #define NXM_HEADER(VENDOR, FIELD, LENGTH) \
473 NXM_HEADER__(VENDOR, FIELD, 0, LENGTH)
474 #define NXM_HEADER_W(VENDOR, FIELD, LENGTH) \
475 NXM_HEADER__(VENDOR, FIELD, 1, (LENGTH) * 2)
476 #define NXM_VENDOR(HEADER) ((HEADER) >> 16)
477 #define NXM_FIELD(HEADER) (((HEADER) >> 9) & 0x7f)
478 #define NXM_TYPE(HEADER) (((HEADER) >> 9) & 0x7fffff)
479 #define NXM_HASMASK(HEADER) (((HEADER) >> 8) & 1)
480 #define NXM_LENGTH(HEADER) ((HEADER) & 0xff)
482 #define NXM_MAKE_WILD_HEADER(HEADER) \
483 NXM_HEADER_W(NXM_VENDOR(HEADER), NXM_FIELD(HEADER), NXM_LENGTH(HEADER))
485 /* Number of registers allocated NXM field IDs. */
486 #define NXM_NX_MAX_REGS 16
488 /* Bits in the value of NXM_NX_IP_FRAG. */
489 #define NX_IP_FRAG_ANY (1 << 0) /* Is this a fragment? */
490 #define NX_IP_FRAG_LATER (1 << 1) /* Is this a fragment with nonzero offset? */
494 * This may be used to gain the OpenFlow 1.1-like ability to restrict
495 * certain NXM-based Flow Mod and Flow Stats Request messages to flows
496 * with specific cookies. See the "nx_flow_mod" and "nx_flow_stats_request"
497 * structure definitions for more details. This match is otherwise not
502 * Format: 64-bit integer in network byte order.
504 * Masking: Arbitrary masks. */
505 #define NXM_NX_COOKIE NXM_HEADER (0x0001, 30, 8)
506 #define NXM_NX_COOKIE_W NXM_HEADER_W(0x0001, 30, 8)
509 /* ## --------------------- ## */
510 /* ## Requests and replies. ## */
511 /* ## --------------------- ## */
513 enum nx_flow_format {
514 NXFF_OPENFLOW10 = 0, /* Standard OpenFlow 1.0 compatible. */
515 NXFF_NXM = 2 /* Nicira extended match. */
518 /* NXT_SET_FLOW_FORMAT request. */
519 struct nx_set_flow_format {
520 ovs_be32 format; /* One of NXFF_*. */
522 OFP_ASSERT(sizeof(struct nx_set_flow_format) == 4);
524 /* NXT_FLOW_MOD (analogous to OFPT_FLOW_MOD).
526 * It is possible to limit flow deletions and modifications to certain
527 * cookies by using the NXM_NX_COOKIE(_W) matches. The "cookie" field
528 * is used only to add or modify flow cookies.
531 ovs_be64 cookie; /* Opaque controller-issued identifier. */
532 ovs_be16 command; /* OFPFC_* + possibly a table ID (see comment
533 * on struct nx_flow_mod_table_id). */
534 ovs_be16 idle_timeout; /* Idle time before discarding (seconds). */
535 ovs_be16 hard_timeout; /* Max time before discarding (seconds). */
536 ovs_be16 priority; /* Priority level of flow entry. */
537 ovs_be32 buffer_id; /* Buffered packet to apply to (or -1).
538 Not meaningful for OFPFC_DELETE*. */
539 ovs_be16 out_port; /* For OFPFC_DELETE* commands, require
540 matching entries to include this as an
541 output port. A value of OFPP_NONE
542 indicates no restriction. */
543 ovs_be16 flags; /* One of OFPFF_*. */
544 ovs_be16 match_len; /* Size of nx_match. */
545 uint8_t pad[6]; /* Align to 64-bits. */
547 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
548 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
549 * all-zero bytes, then
550 * - Actions to fill out the remainder of the message length (always a
554 OFP_ASSERT(sizeof(struct nx_flow_mod) == 32);
556 /* NXT_FLOW_REMOVED (analogous to OFPT_FLOW_REMOVED).
558 * 'table_id' is present only in Open vSwitch 1.11 and later. In earlier
559 * versions of Open vSwitch, this is a padding byte that is always zeroed.
560 * Therefore, a 'table_id' value of 0 indicates that the table ID is not known,
561 * and other values may be interpreted as one more than the flow's former table
563 struct nx_flow_removed {
564 ovs_be64 cookie; /* Opaque controller-issued identifier. */
565 ovs_be16 priority; /* Priority level of flow entry. */
566 uint8_t reason; /* One of OFPRR_*. */
567 uint8_t table_id; /* Flow's former table ID, plus one. */
568 ovs_be32 duration_sec; /* Time flow was alive in seconds. */
569 ovs_be32 duration_nsec; /* Time flow was alive in nanoseconds beyond
571 ovs_be16 idle_timeout; /* Idle timeout from original flow mod. */
572 ovs_be16 match_len; /* Size of nx_match. */
573 ovs_be64 packet_count;
576 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
577 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
580 OFP_ASSERT(sizeof(struct nx_flow_removed) == 40);
582 /* Nicira vendor stats request of type NXST_FLOW (analogous to OFPST_FLOW
585 * It is possible to limit matches to certain cookies by using the
586 * NXM_NX_COOKIE and NXM_NX_COOKIE_W matches.
588 struct nx_flow_stats_request {
589 ovs_be16 out_port; /* Require matching entries to include this
590 as an output port. A value of OFPP_NONE
591 indicates no restriction. */
592 ovs_be16 match_len; /* Length of nx_match. */
593 uint8_t table_id; /* ID of table to read (from ofp_table_stats)
594 or 0xff for all tables. */
595 uint8_t pad[3]; /* Align to 64 bits. */
597 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
598 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
599 * all-zero bytes, which must also exactly fill out the length of the
603 OFP_ASSERT(sizeof(struct nx_flow_stats_request) == 8);
605 /* Body for Nicira vendor stats reply of type NXST_FLOW (analogous to
608 * The values of 'idle_age' and 'hard_age' are only meaningful when talking to
609 * a switch that implements the NXT_FLOW_AGE extension. Zero means that the
610 * true value is unknown, perhaps because hardware does not track the value.
611 * (Zero is also the value that one should ordinarily expect to see talking to
612 * a switch that does not implement NXT_FLOW_AGE, since those switches zero the
613 * padding bytes that these fields replaced.) A nonzero value X represents X-1
614 * seconds. A value of 65535 represents 65534 or more seconds.
616 * 'idle_age' is the number of seconds that the flow has been idle, that is,
617 * the number of seconds since a packet passed through the flow. 'hard_age' is
618 * the number of seconds since the flow was last modified (e.g. OFPFC_MODIFY or
619 * OFPFC_MODIFY_STRICT). (The 'duration_*' fields are the elapsed time since
620 * the flow was added, regardless of subsequent modifications.)
622 * For a flow with an idle or hard timeout, 'idle_age' or 'hard_age',
623 * respectively, will ordinarily be smaller than the timeout, but flow
624 * expiration times are only approximate and so one must be prepared to
625 * tolerate expirations that occur somewhat early or late.
627 struct nx_flow_stats {
628 ovs_be16 length; /* Length of this entry. */
629 uint8_t table_id; /* ID of table flow came from. */
631 ovs_be32 duration_sec; /* Time flow has been alive in seconds. */
632 ovs_be32 duration_nsec; /* Time flow has been alive in nanoseconds
633 beyond duration_sec. */
634 ovs_be16 priority; /* Priority of the entry. */
635 ovs_be16 idle_timeout; /* Number of seconds idle before expiration. */
636 ovs_be16 hard_timeout; /* Number of seconds before expiration. */
637 ovs_be16 match_len; /* Length of nx_match. */
638 ovs_be16 idle_age; /* Seconds since last packet, plus one. */
639 ovs_be16 hard_age; /* Seconds since last modification, plus one. */
640 ovs_be64 cookie; /* Opaque controller-issued identifier. */
641 ovs_be64 packet_count; /* Number of packets, UINT64_MAX if unknown. */
642 ovs_be64 byte_count; /* Number of bytes, UINT64_MAX if unknown. */
644 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
645 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
646 * all-zero bytes, then
647 * - Actions to fill out the remainder 'length' bytes (always a multiple
651 OFP_ASSERT(sizeof(struct nx_flow_stats) == 48);
653 /* Nicira vendor stats request of type NXST_AGGREGATE (analogous to
654 * OFPST_AGGREGATE request).
656 * The reply format is identical to the reply format for OFPST_AGGREGATE,
657 * except for the header. */
658 struct nx_aggregate_stats_request {
659 ovs_be16 out_port; /* Require matching entries to include this
660 as an output port. A value of OFPP_NONE
661 indicates no restriction. */
662 ovs_be16 match_len; /* Length of nx_match. */
663 uint8_t table_id; /* ID of table to read (from ofp_table_stats)
664 or 0xff for all tables. */
665 uint8_t pad[3]; /* Align to 64 bits. */
667 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
668 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
669 * all-zero bytes, which must also exactly fill out the length of the
673 OFP_ASSERT(sizeof(struct nx_aggregate_stats_request) == 8);
675 /* NXT_SET_CONTROLLER_ID.
677 * Each OpenFlow controller connection has a 16-bit identifier that is
678 * initially 0. This message changes the connection's ID to 'id'.
680 * Controller connection IDs need not be unique.
682 * The NXAST_CONTROLLER action is the only current user of controller
684 struct nx_controller_id {
685 uint8_t zero[6]; /* Must be zero. */
686 ovs_be16 controller_id; /* New controller connection ID. */
688 OFP_ASSERT(sizeof(struct nx_controller_id) == 8);
690 /* Flow Table Monitoring
691 * =====================
693 * NXST_FLOW_MONITOR allows a controller to keep track of changes to OpenFlow
694 * flow table(s) or subsets of them, with the following workflow:
696 * 1. The controller sends an NXST_FLOW_MONITOR request to begin monitoring
697 * flows. The 'id' in the request must be unique among all monitors that
698 * the controller has started and not yet canceled on this OpenFlow
701 * 2. The switch responds with an NXST_FLOW_MONITOR reply. If the request's
702 * 'flags' included NXFMF_INITIAL, the reply includes all the flows that
703 * matched the request at the time of the request (with event NXFME_ADDED).
704 * If 'flags' did not include NXFMF_INITIAL, the reply is empty.
706 * The reply uses the xid of the request (as do all replies to OpenFlow
709 * 3. Whenever a change to a flow table entry matches some outstanding monitor
710 * request's criteria and flags, the switch sends a notification to the
711 * controller as an additional NXST_FLOW_MONITOR reply with xid 0.
713 * When multiple outstanding monitors match a single change, only a single
714 * notification is sent. This merged notification includes the information
715 * requested in any of the individual monitors. That is, if any of the
716 * matching monitors requests actions (NXFMF_ACTIONS), the notification
717 * includes actions, and if any of the monitors request full changes for the
718 * controller's own changes (NXFMF_OWN), the controller's own changes will
719 * be included in full.
721 * 4. The controller may cancel a monitor with NXT_FLOW_MONITOR_CANCEL. No
722 * further notifications will be sent on the basis of the canceled monitor
729 * OpenFlow messages for flow monitor notifications can overflow the buffer
730 * space available to the switch, either temporarily (e.g. due to network
731 * conditions slowing OpenFlow traffic) or more permanently (e.g. the sustained
732 * rate of flow table change exceeds the network bandwidth between switch and
735 * When Open vSwitch's notification buffer space reaches a limiting threshold,
736 * OVS reacts as follows:
738 * 1. OVS sends an NXT_FLOW_MONITOR_PAUSED message to the controller, following
739 * all the already queued notifications. After it receives this message,
740 * the controller knows that its view of the flow table, as represented by
741 * flow monitor notifications, is incomplete.
743 * 2. As long as the notification buffer is not empty:
745 * - NXMFE_ADD and NXFME_MODIFIED notifications will not be sent.
747 * - NXFME_DELETED notifications will still be sent, but only for flows
748 * that existed before OVS sent NXT_FLOW_MONITOR_PAUSED.
750 * - NXFME_ABBREV notifications will not be sent. They are treated as
751 * the expanded version (and therefore only the NXFME_DELETED
752 * components, if any, are sent).
754 * 3. When the notification buffer empties, OVS sends NXFME_ADD notifications
755 * for flows added since the buffer reached its limit and NXFME_MODIFIED
756 * notifications for flows that existed before the limit was reached and
757 * changed after the limit was reached.
759 * 4. OVS sends an NXT_FLOW_MONITOR_RESUMED message to the controller. After
760 * it receives this message, the controller knows that its view of the flow
761 * table, as represented by flow monitor notifications, is again complete.
763 * This allows the maximum buffer space requirement for notifications to be
764 * bounded by the limit plus the maximum number of supported flows.
767 * "Flow Removed" messages
768 * =======================
770 * The flow monitor mechanism is independent of OFPT_FLOW_REMOVED and
771 * NXT_FLOW_REMOVED. Flow monitor updates for deletion are sent if
772 * NXFMF_DELETE is set on a monitor, regardless of whether the
773 * OFPFF_SEND_FLOW_REM flag was set when the flow was added. */
775 /* NXST_FLOW_MONITOR request.
777 * The NXST_FLOW_MONITOR request's body consists of an array of zero or more
778 * instances of this structure. The request arranges to monitor the flows
779 * that match the specified criteria, which are interpreted in the same way as
782 * 'id' identifies a particular monitor for the purpose of allowing it to be
783 * canceled later with NXT_FLOW_MONITOR_CANCEL. 'id' must be unique among
784 * existing monitors that have not already been canceled.
786 * The reply includes the initial flow matches for monitors that have the
787 * NXFMF_INITIAL flag set. No single flow will be included in the reply more
788 * than once, even if more than one requested monitor matches that flow. The
789 * reply will be empty if none of the monitors has NXFMF_INITIAL set or if none
790 * of the monitors initially matches any flows.
792 * For NXFMF_ADD, an event will be reported if 'out_port' matches against the
793 * actions of the flow being added or, for a flow that is replacing an existing
794 * flow, if 'out_port' matches against the actions of the flow being replaced.
795 * For NXFMF_DELETE, 'out_port' matches against the actions of a flow being
796 * deleted. For NXFMF_MODIFY, an event will be reported if 'out_port' matches
797 * either the old or the new actions. */
798 struct nx_flow_monitor_request {
799 ovs_be32 id; /* Controller-assigned ID for this monitor. */
800 ovs_be16 flags; /* NXFMF_*. */
801 ovs_be16 out_port; /* Required output port, if not OFPP_NONE. */
802 ovs_be16 match_len; /* Length of nx_match. */
803 uint8_t table_id; /* One table's ID or 0xff for all tables. */
804 uint8_t zeros[5]; /* Align to 64 bits (must be zero). */
806 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
807 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
810 OFP_ASSERT(sizeof(struct nx_flow_monitor_request) == 16);
812 /* 'flags' bits in struct nx_flow_monitor_request. */
813 enum nx_flow_monitor_flags {
814 /* When to send updates. */
815 NXFMF_INITIAL = 1 << 0, /* Initially matching flows. */
816 NXFMF_ADD = 1 << 1, /* New matching flows as they are added. */
817 NXFMF_DELETE = 1 << 2, /* Old matching flows as they are removed. */
818 NXFMF_MODIFY = 1 << 3, /* Matching flows as they are changed. */
820 /* What to include in updates. */
821 NXFMF_ACTIONS = 1 << 4, /* If set, actions are included. */
822 NXFMF_OWN = 1 << 5, /* If set, include own changes in full. */
825 /* NXST_FLOW_MONITOR reply header.
827 * The body of an NXST_FLOW_MONITOR reply is an array of variable-length
828 * structures, each of which begins with this header. The 'length' member may
829 * be used to traverse the array, and the 'event' member may be used to
830 * determine the particular structure.
832 * Every instance is a multiple of 8 bytes long. */
833 struct nx_flow_update_header {
834 ovs_be16 length; /* Length of this entry. */
835 ovs_be16 event; /* One of NXFME_*. */
836 /* ...other data depending on 'event'... */
838 OFP_ASSERT(sizeof(struct nx_flow_update_header) == 4);
840 /* 'event' values in struct nx_flow_update_header. */
841 enum nx_flow_update_event {
842 /* struct nx_flow_update_full. */
843 NXFME_ADDED = 0, /* Flow was added. */
844 NXFME_DELETED = 1, /* Flow was deleted. */
845 NXFME_MODIFIED = 2, /* Flow (generally its actions) was changed. */
847 /* struct nx_flow_update_abbrev. */
848 NXFME_ABBREV = 3, /* Abbreviated reply. */
851 /* NXST_FLOW_MONITOR reply for NXFME_ADDED, NXFME_DELETED, and
853 struct nx_flow_update_full {
854 ovs_be16 length; /* Length is 24. */
855 ovs_be16 event; /* One of NXFME_*. */
856 ovs_be16 reason; /* OFPRR_* for NXFME_DELETED, else zero. */
857 ovs_be16 priority; /* Priority of the entry. */
858 ovs_be16 idle_timeout; /* Number of seconds idle before expiration. */
859 ovs_be16 hard_timeout; /* Number of seconds before expiration. */
860 ovs_be16 match_len; /* Length of nx_match. */
861 uint8_t table_id; /* ID of flow's table. */
862 uint8_t pad; /* Reserved, currently zeroed. */
863 ovs_be64 cookie; /* Opaque controller-issued identifier. */
865 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
866 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
867 * all-zero bytes, then
868 * - Actions to fill out the remainder 'length' bytes (always a multiple
869 * of 8). If NXFMF_ACTIONS was not specified, or 'event' is
870 * NXFME_DELETED, no actions are included.
873 OFP_ASSERT(sizeof(struct nx_flow_update_full) == 24);
875 /* NXST_FLOW_MONITOR reply for NXFME_ABBREV.
877 * When the controller does not specify NXFMF_OWN in a monitor request, any
878 * flow tables changes due to the controller's own requests (on the same
879 * OpenFlow channel) will be abbreviated, when possible, to this form, which
880 * simply specifies the 'xid' of the OpenFlow request (e.g. an OFPT_FLOW_MOD or
881 * NXT_FLOW_MOD) that caused the change.
883 * Some changes cannot be abbreviated and will be sent in full:
885 * - Changes that only partially succeed. This can happen if, for example,
886 * a flow_mod with type OFPFC_MODIFY affects multiple flows, but only some
887 * of those modifications succeed (e.g. due to hardware limitations).
889 * This cannot occur with the Open vSwitch software datapath. This also
890 * cannot occur in Open vSwitch 2.4 and later, because these versions only
891 * execute any flow modifications if all of them will succeed.
893 * - Changes that race with conflicting changes made by other controllers or
894 * other flow_mods (not separated by barriers) by the same controller.
896 * This cannot occur with the current Open vSwitch implementation
897 * (regardless of datapath) because Open vSwitch internally serializes
898 * potentially conflicting changes.
900 * - Changes that occur when flow notification is paused (see "Buffer
901 * Management" above).
903 * A flow_mod that does not change the flow table will not trigger any
904 * notification, even an abbreviated one. For example, a "modify" or "delete"
905 * flow_mod that does not match any flows will not trigger a notification.
906 * Whether an "add" or "modify" that specifies all the same parameters that a
907 * flow already has triggers a notification is unspecified and subject to
908 * change in future versions of Open vSwitch.
910 * OVS will always send the notifications for a given flow table change before
911 * the reply to a OFPT_BARRIER_REQUEST request that follows the flow table
912 * change. Thus, if the controller does not receive an abbreviated (or
913 * unabbreviated) notification for a flow_mod before the next
914 * OFPT_BARRIER_REPLY, it will never receive one. */
915 struct nx_flow_update_abbrev {
916 ovs_be16 length; /* Length is 8. */
917 ovs_be16 event; /* NXFME_ABBREV. */
918 ovs_be32 xid; /* Controller-specified xid from flow_mod. */
920 OFP_ASSERT(sizeof(struct nx_flow_update_abbrev) == 8);
922 /* NXT_FLOW_MONITOR_CANCEL.
924 * Used by a controller to cancel an outstanding monitor. */
925 struct nx_flow_monitor_cancel {
926 ovs_be32 id; /* 'id' from nx_flow_monitor_request. */
928 OFP_ASSERT(sizeof(struct nx_flow_monitor_cancel) == 4);
930 #endif /* openflow/nicira-ext.h */