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.
290 * OpenFlow 1.2+ introduced OpenFlow Extensible Match (OXM), adapting
291 * the design of NXM. The format of NXM and OXM are compatible.
297 * An nx_match is a sequence of zero or more "nxm_entry"s, which are
298 * type-length-value (TLV) entries, each 5 to 259 (inclusive) bytes long.
299 * "nxm_entry"s are not aligned on or padded to any multibyte boundary. The
300 * first 4 bytes of an nxm_entry are its "header", followed by the entry's
303 * An nxm_entry's header is interpreted as a 32-bit word in network byte order:
305 * |<-------------------- nxm_type ------------------>|
308 * +----------------------------------+---------------+--+------------------+
309 * | nxm_vendor | nxm_field |hm| nxm_length |
310 * +----------------------------------+---------------+--+------------------+
312 * The most-significant 23 bits of the header are collectively "nxm_type".
313 * Bits 16...31 are "nxm_vendor", one of OFPXMC12_* values. In case of
314 * NXM, it's either OFPXMC12_NXM_0 or OFPXMC12_NXM_1.
315 * Bits 9...15 are "nxm_field", which is a vendor-specific value. nxm_type
316 * normally designates a protocol header, such as the Ethernet type, but it
317 * can also refer to packet metadata, such as the switch port on which a packet
320 * Bit 8 is "nxm_hasmask" (labeled "hm" above for space reasons). The meaning
321 * of this bit is explained later.
323 * The least-significant 8 bits are "nxm_length", a positive integer. The
324 * length of the nxm_entry, including the header, is exactly 4 + nxm_length
327 * For a given nxm_vendor, nxm_field, and nxm_hasmask value, nxm_length is a
328 * constant. It is included only to allow software to minimally parse
329 * "nxm_entry"s of unknown types. (Similarly, for a given nxm_vendor,
330 * nxm_field, and nxm_length, nxm_hasmask is a constant.)
336 * A zero-length nx_match (one with no "nxm_entry"s) matches every packet.
338 * An nxm_entry places a constraint on the packets matched by the nx_match:
340 * - If nxm_hasmask is 0, the nxm_entry's body contains a value for the
341 * field, called "nxm_value". The nx_match matches only packets in which
342 * the field equals nxm_value.
344 * - If nxm_hasmask is 1, then the nxm_entry's body contains a value for the
345 * field (nxm_value), followed by a bitmask of the same length as the
346 * value, called "nxm_mask". For each 1-bit in position J in nxm_mask, the
347 * nx_match matches only packets for which bit J in the given field's value
348 * matches bit J in nxm_value. A 0-bit in nxm_mask causes the
349 * corresponding bit in nxm_value is ignored (it should be 0; Open vSwitch
350 * may enforce this someday), as is the corresponding bit in the field's
351 * value. (The sense of the nxm_mask bits is the opposite of that used by
352 * the "wildcards" member of struct ofp10_match.)
354 * When nxm_hasmask is 1, nxm_length is always even.
356 * An all-zero-bits nxm_mask is equivalent to omitting the nxm_entry
357 * entirely. An all-one-bits nxm_mask is equivalent to specifying 0 for
360 * When there are multiple "nxm_entry"s, all of the constraints must be met.
366 * Masks may be restricted:
368 * - Some nxm_types may not support masked wildcards, that is, nxm_hasmask
369 * must always be 0 when these fields are specified. For example, the
370 * field that identifies the port on which a packet was received may not be
373 * - Some nxm_types that do support masked wildcards may only support certain
374 * nxm_mask patterns. For example, fields that have IPv4 address values
375 * may be restricted to CIDR masks.
377 * These restrictions should be noted in specifications for individual fields.
378 * A switch may accept an nxm_hasmask or nxm_mask value that the specification
379 * disallows, if the switch correctly implements support for that nxm_hasmask
380 * or nxm_mask value. A switch must reject an attempt to set up a flow that
381 * contains a nxm_hasmask or nxm_mask value that it does not support.
384 * Prerequisite Restrictions
385 * =========================
387 * The presence of an nxm_entry with a given nxm_type may be restricted based
388 * on the presence of or values of other "nxm_entry"s. For example:
390 * - An nxm_entry for nxm_type=NXM_OF_IP_TOS is allowed only if it is
391 * preceded by another entry with nxm_type=NXM_OF_ETH_TYPE, nxm_hasmask=0,
392 * and nxm_value=0x0800. That is, matching on the IP source address is
393 * allowed only if the Ethernet type is explicitly set to IP.
395 * - An nxm_entry for nxm_type=NXM_OF_TCP_SRC is allowed only if it is
396 * preceded by an entry with nxm_type=NXM_OF_ETH_TYPE, nxm_hasmask=0, and
397 * nxm_value either 0x0800 or 0x86dd, and another with
398 * nxm_type=NXM_OF_IP_PROTO, nxm_hasmask=0, nxm_value=6, in that order.
399 * That is, matching on the TCP source port is allowed only if the Ethernet
400 * type is IP or IPv6 and the IP protocol is TCP.
402 * These restrictions should be noted in specifications for individual fields.
403 * A switch may implement relaxed versions of these restrictions. A switch
404 * must reject an attempt to set up a flow that violates its restrictions.
407 * Ordering Restrictions
408 * =====================
410 * An nxm_entry that has prerequisite restrictions must appear after the
411 * "nxm_entry"s for its prerequisites. Ordering of "nxm_entry"s within an
412 * nx_match is not otherwise constrained.
414 * Any given nxm_type may appear in an nx_match at most once.
420 * These examples show the format of a single nxm_entry with particular
421 * nxm_hasmask and nxm_length values. The diagrams are labeled with field
422 * numbers and byte indexes.
425 * 8-bit nxm_value, nxm_hasmask=1, nxm_length=2:
428 * +------------+---+---+
430 * +------------+---+---+
433 * 16-bit nxm_value, nxm_hasmask=0, nxm_length=2:
436 * +------------+------+
438 * +------------+------+
441 * 32-bit nxm_value, nxm_hasmask=0, nxm_length=4:
444 * +------------+-------------+
445 * | header | nxm_value |
446 * +------------+-------------+
449 * 48-bit nxm_value, nxm_hasmask=0, nxm_length=6:
452 * +------------+------------------+
453 * | header | nxm_value |
454 * +------------+------------------+
457 * 48-bit nxm_value, nxm_hasmask=1, nxm_length=12:
460 * +------------+------------------+------------------+
461 * | header | nxm_value | nxm_mask |
462 * +------------+------------------+------------------+
468 * A switch should report an error in an nx_match using error type
469 * OFPET_BAD_REQUEST and one of the NXBRC_NXM_* codes. Ideally the switch
470 * should report a specific error code, if one is assigned for the particular
471 * problem, but NXBRC_NXM_INVALID is also available to report a generic
475 /* Number of registers allocated NXM field IDs. */
476 #define NXM_NX_MAX_REGS 16
478 /* Bits in the value of NXM_NX_IP_FRAG. */
479 #define NX_IP_FRAG_ANY (1 << 0) /* Is this a fragment? */
480 #define NX_IP_FRAG_LATER (1 << 1) /* Is this a fragment with nonzero offset? */
482 /* ## --------------------- ## */
483 /* ## Requests and replies. ## */
484 /* ## --------------------- ## */
486 enum nx_flow_format {
487 NXFF_OPENFLOW10 = 0, /* Standard OpenFlow 1.0 compatible. */
488 NXFF_NXM = 2 /* Nicira extended match. */
491 /* NXT_SET_FLOW_FORMAT request. */
492 struct nx_set_flow_format {
493 ovs_be32 format; /* One of NXFF_*. */
495 OFP_ASSERT(sizeof(struct nx_set_flow_format) == 4);
497 /* NXT_FLOW_MOD (analogous to OFPT_FLOW_MOD).
499 * It is possible to limit flow deletions and modifications to certain
500 * cookies by using the NXM_NX_COOKIE(_W) matches. The "cookie" field
501 * is used only to add or modify flow cookies.
504 ovs_be64 cookie; /* Opaque controller-issued identifier. */
505 ovs_be16 command; /* OFPFC_* + possibly a table ID (see comment
506 * on struct nx_flow_mod_table_id). */
507 ovs_be16 idle_timeout; /* Idle time before discarding (seconds). */
508 ovs_be16 hard_timeout; /* Max time before discarding (seconds). */
509 ovs_be16 priority; /* Priority level of flow entry. */
510 ovs_be32 buffer_id; /* Buffered packet to apply to (or -1).
511 Not meaningful for OFPFC_DELETE*. */
512 ovs_be16 out_port; /* For OFPFC_DELETE* commands, require
513 matching entries to include this as an
514 output port. A value of OFPP_NONE
515 indicates no restriction. */
516 ovs_be16 flags; /* One of OFPFF_*. */
517 ovs_be16 match_len; /* Size of nx_match. */
518 uint8_t pad[6]; /* Align to 64-bits. */
520 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
521 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
522 * all-zero bytes, then
523 * - Actions to fill out the remainder of the message length (always a
527 OFP_ASSERT(sizeof(struct nx_flow_mod) == 32);
529 /* NXT_FLOW_REMOVED (analogous to OFPT_FLOW_REMOVED).
531 * 'table_id' is present only in Open vSwitch 1.11 and later. In earlier
532 * versions of Open vSwitch, this is a padding byte that is always zeroed.
533 * Therefore, a 'table_id' value of 0 indicates that the table ID is not known,
534 * and other values may be interpreted as one more than the flow's former table
536 struct nx_flow_removed {
537 ovs_be64 cookie; /* Opaque controller-issued identifier. */
538 ovs_be16 priority; /* Priority level of flow entry. */
539 uint8_t reason; /* One of OFPRR_*. */
540 uint8_t table_id; /* Flow's former table ID, plus one. */
541 ovs_be32 duration_sec; /* Time flow was alive in seconds. */
542 ovs_be32 duration_nsec; /* Time flow was alive in nanoseconds beyond
544 ovs_be16 idle_timeout; /* Idle timeout from original flow mod. */
545 ovs_be16 match_len; /* Size of nx_match. */
546 ovs_be64 packet_count;
549 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
550 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
553 OFP_ASSERT(sizeof(struct nx_flow_removed) == 40);
555 /* Nicira vendor stats request of type NXST_FLOW (analogous to OFPST_FLOW
558 * It is possible to limit matches to certain cookies by using the
559 * NXM_NX_COOKIE and NXM_NX_COOKIE_W matches.
561 struct nx_flow_stats_request {
562 ovs_be16 out_port; /* Require matching entries to include this
563 as an output port. A value of OFPP_NONE
564 indicates no restriction. */
565 ovs_be16 match_len; /* Length of nx_match. */
566 uint8_t table_id; /* ID of table to read (from ofp_table_stats)
567 or 0xff for all tables. */
568 uint8_t pad[3]; /* Align to 64 bits. */
570 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
571 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
572 * all-zero bytes, which must also exactly fill out the length of the
576 OFP_ASSERT(sizeof(struct nx_flow_stats_request) == 8);
578 /* Body for Nicira vendor stats reply of type NXST_FLOW (analogous to
581 * The values of 'idle_age' and 'hard_age' are only meaningful when talking to
582 * a switch that implements the NXT_FLOW_AGE extension. Zero means that the
583 * true value is unknown, perhaps because hardware does not track the value.
584 * (Zero is also the value that one should ordinarily expect to see talking to
585 * a switch that does not implement NXT_FLOW_AGE, since those switches zero the
586 * padding bytes that these fields replaced.) A nonzero value X represents X-1
587 * seconds. A value of 65535 represents 65534 or more seconds.
589 * 'idle_age' is the number of seconds that the flow has been idle, that is,
590 * the number of seconds since a packet passed through the flow. 'hard_age' is
591 * the number of seconds since the flow was last modified (e.g. OFPFC_MODIFY or
592 * OFPFC_MODIFY_STRICT). (The 'duration_*' fields are the elapsed time since
593 * the flow was added, regardless of subsequent modifications.)
595 * For a flow with an idle or hard timeout, 'idle_age' or 'hard_age',
596 * respectively, will ordinarily be smaller than the timeout, but flow
597 * expiration times are only approximate and so one must be prepared to
598 * tolerate expirations that occur somewhat early or late.
600 struct nx_flow_stats {
601 ovs_be16 length; /* Length of this entry. */
602 uint8_t table_id; /* ID of table flow came from. */
604 ovs_be32 duration_sec; /* Time flow has been alive in seconds. */
605 ovs_be32 duration_nsec; /* Time flow has been alive in nanoseconds
606 beyond duration_sec. */
607 ovs_be16 priority; /* Priority of the entry. */
608 ovs_be16 idle_timeout; /* Number of seconds idle before expiration. */
609 ovs_be16 hard_timeout; /* Number of seconds before expiration. */
610 ovs_be16 match_len; /* Length of nx_match. */
611 ovs_be16 idle_age; /* Seconds since last packet, plus one. */
612 ovs_be16 hard_age; /* Seconds since last modification, plus one. */
613 ovs_be64 cookie; /* Opaque controller-issued identifier. */
614 ovs_be64 packet_count; /* Number of packets, UINT64_MAX if unknown. */
615 ovs_be64 byte_count; /* Number of bytes, UINT64_MAX if unknown. */
617 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
618 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
619 * all-zero bytes, then
620 * - Actions to fill out the remainder 'length' bytes (always a multiple
624 OFP_ASSERT(sizeof(struct nx_flow_stats) == 48);
626 /* Nicira vendor stats request of type NXST_AGGREGATE (analogous to
627 * OFPST_AGGREGATE request).
629 * The reply format is identical to the reply format for OFPST_AGGREGATE,
630 * except for the header. */
631 struct nx_aggregate_stats_request {
632 ovs_be16 out_port; /* Require matching entries to include this
633 as an output port. A value of OFPP_NONE
634 indicates no restriction. */
635 ovs_be16 match_len; /* Length of nx_match. */
636 uint8_t table_id; /* ID of table to read (from ofp_table_stats)
637 or 0xff for all tables. */
638 uint8_t pad[3]; /* Align to 64 bits. */
640 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
641 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
642 * all-zero bytes, which must also exactly fill out the length of the
646 OFP_ASSERT(sizeof(struct nx_aggregate_stats_request) == 8);
648 /* NXT_SET_CONTROLLER_ID.
650 * Each OpenFlow controller connection has a 16-bit identifier that is
651 * initially 0. This message changes the connection's ID to 'id'.
653 * Controller connection IDs need not be unique.
655 * The NXAST_CONTROLLER action is the only current user of controller
657 struct nx_controller_id {
658 uint8_t zero[6]; /* Must be zero. */
659 ovs_be16 controller_id; /* New controller connection ID. */
661 OFP_ASSERT(sizeof(struct nx_controller_id) == 8);
663 /* Flow Table Monitoring
664 * =====================
666 * NXST_FLOW_MONITOR allows a controller to keep track of changes to OpenFlow
667 * flow table(s) or subsets of them, with the following workflow:
669 * 1. The controller sends an NXST_FLOW_MONITOR request to begin monitoring
670 * flows. The 'id' in the request must be unique among all monitors that
671 * the controller has started and not yet canceled on this OpenFlow
674 * 2. The switch responds with an NXST_FLOW_MONITOR reply. If the request's
675 * 'flags' included NXFMF_INITIAL, the reply includes all the flows that
676 * matched the request at the time of the request (with event NXFME_ADDED).
677 * If 'flags' did not include NXFMF_INITIAL, the reply is empty.
679 * The reply uses the xid of the request (as do all replies to OpenFlow
682 * 3. Whenever a change to a flow table entry matches some outstanding monitor
683 * request's criteria and flags, the switch sends a notification to the
684 * controller as an additional NXST_FLOW_MONITOR reply with xid 0.
686 * When multiple outstanding monitors match a single change, only a single
687 * notification is sent. This merged notification includes the information
688 * requested in any of the individual monitors. That is, if any of the
689 * matching monitors requests actions (NXFMF_ACTIONS), the notification
690 * includes actions, and if any of the monitors request full changes for the
691 * controller's own changes (NXFMF_OWN), the controller's own changes will
692 * be included in full.
694 * 4. The controller may cancel a monitor with NXT_FLOW_MONITOR_CANCEL. No
695 * further notifications will be sent on the basis of the canceled monitor
702 * OpenFlow messages for flow monitor notifications can overflow the buffer
703 * space available to the switch, either temporarily (e.g. due to network
704 * conditions slowing OpenFlow traffic) or more permanently (e.g. the sustained
705 * rate of flow table change exceeds the network bandwidth between switch and
708 * When Open vSwitch's notification buffer space reaches a limiting threshold,
709 * OVS reacts as follows:
711 * 1. OVS sends an NXT_FLOW_MONITOR_PAUSED message to the controller, following
712 * all the already queued notifications. After it receives this message,
713 * the controller knows that its view of the flow table, as represented by
714 * flow monitor notifications, is incomplete.
716 * 2. As long as the notification buffer is not empty:
718 * - NXMFE_ADD and NXFME_MODIFIED notifications will not be sent.
720 * - NXFME_DELETED notifications will still be sent, but only for flows
721 * that existed before OVS sent NXT_FLOW_MONITOR_PAUSED.
723 * - NXFME_ABBREV notifications will not be sent. They are treated as
724 * the expanded version (and therefore only the NXFME_DELETED
725 * components, if any, are sent).
727 * 3. When the notification buffer empties, OVS sends NXFME_ADD notifications
728 * for flows added since the buffer reached its limit and NXFME_MODIFIED
729 * notifications for flows that existed before the limit was reached and
730 * changed after the limit was reached.
732 * 4. OVS sends an NXT_FLOW_MONITOR_RESUMED message to the controller. After
733 * it receives this message, the controller knows that its view of the flow
734 * table, as represented by flow monitor notifications, is again complete.
736 * This allows the maximum buffer space requirement for notifications to be
737 * bounded by the limit plus the maximum number of supported flows.
740 * "Flow Removed" messages
741 * =======================
743 * The flow monitor mechanism is independent of OFPT_FLOW_REMOVED and
744 * NXT_FLOW_REMOVED. Flow monitor updates for deletion are sent if
745 * NXFMF_DELETE is set on a monitor, regardless of whether the
746 * OFPFF_SEND_FLOW_REM flag was set when the flow was added. */
748 /* NXST_FLOW_MONITOR request.
750 * The NXST_FLOW_MONITOR request's body consists of an array of zero or more
751 * instances of this structure. The request arranges to monitor the flows
752 * that match the specified criteria, which are interpreted in the same way as
755 * 'id' identifies a particular monitor for the purpose of allowing it to be
756 * canceled later with NXT_FLOW_MONITOR_CANCEL. 'id' must be unique among
757 * existing monitors that have not already been canceled.
759 * The reply includes the initial flow matches for monitors that have the
760 * NXFMF_INITIAL flag set. No single flow will be included in the reply more
761 * than once, even if more than one requested monitor matches that flow. The
762 * reply will be empty if none of the monitors has NXFMF_INITIAL set or if none
763 * of the monitors initially matches any flows.
765 * For NXFMF_ADD, an event will be reported if 'out_port' matches against the
766 * actions of the flow being added or, for a flow that is replacing an existing
767 * flow, if 'out_port' matches against the actions of the flow being replaced.
768 * For NXFMF_DELETE, 'out_port' matches against the actions of a flow being
769 * deleted. For NXFMF_MODIFY, an event will be reported if 'out_port' matches
770 * either the old or the new actions. */
771 struct nx_flow_monitor_request {
772 ovs_be32 id; /* Controller-assigned ID for this monitor. */
773 ovs_be16 flags; /* NXFMF_*. */
774 ovs_be16 out_port; /* Required output port, if not OFPP_NONE. */
775 ovs_be16 match_len; /* Length of nx_match. */
776 uint8_t table_id; /* One table's ID or 0xff for all tables. */
777 uint8_t zeros[5]; /* Align to 64 bits (must be zero). */
779 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
780 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
783 OFP_ASSERT(sizeof(struct nx_flow_monitor_request) == 16);
785 /* 'flags' bits in struct nx_flow_monitor_request. */
786 enum nx_flow_monitor_flags {
787 /* When to send updates. */
788 NXFMF_INITIAL = 1 << 0, /* Initially matching flows. */
789 NXFMF_ADD = 1 << 1, /* New matching flows as they are added. */
790 NXFMF_DELETE = 1 << 2, /* Old matching flows as they are removed. */
791 NXFMF_MODIFY = 1 << 3, /* Matching flows as they are changed. */
793 /* What to include in updates. */
794 NXFMF_ACTIONS = 1 << 4, /* If set, actions are included. */
795 NXFMF_OWN = 1 << 5, /* If set, include own changes in full. */
798 /* NXST_FLOW_MONITOR reply header.
800 * The body of an NXST_FLOW_MONITOR reply is an array of variable-length
801 * structures, each of which begins with this header. The 'length' member may
802 * be used to traverse the array, and the 'event' member may be used to
803 * determine the particular structure.
805 * Every instance is a multiple of 8 bytes long. */
806 struct nx_flow_update_header {
807 ovs_be16 length; /* Length of this entry. */
808 ovs_be16 event; /* One of NXFME_*. */
809 /* ...other data depending on 'event'... */
811 OFP_ASSERT(sizeof(struct nx_flow_update_header) == 4);
813 /* 'event' values in struct nx_flow_update_header. */
814 enum nx_flow_update_event {
815 /* struct nx_flow_update_full. */
816 NXFME_ADDED = 0, /* Flow was added. */
817 NXFME_DELETED = 1, /* Flow was deleted. */
818 NXFME_MODIFIED = 2, /* Flow (generally its actions) was changed. */
820 /* struct nx_flow_update_abbrev. */
821 NXFME_ABBREV = 3, /* Abbreviated reply. */
824 /* NXST_FLOW_MONITOR reply for NXFME_ADDED, NXFME_DELETED, and
826 struct nx_flow_update_full {
827 ovs_be16 length; /* Length is 24. */
828 ovs_be16 event; /* One of NXFME_*. */
829 ovs_be16 reason; /* OFPRR_* for NXFME_DELETED, else zero. */
830 ovs_be16 priority; /* Priority of the entry. */
831 ovs_be16 idle_timeout; /* Number of seconds idle before expiration. */
832 ovs_be16 hard_timeout; /* Number of seconds before expiration. */
833 ovs_be16 match_len; /* Length of nx_match. */
834 uint8_t table_id; /* ID of flow's table. */
835 uint8_t pad; /* Reserved, currently zeroed. */
836 ovs_be64 cookie; /* Opaque controller-issued identifier. */
838 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
839 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
840 * all-zero bytes, then
841 * - Actions to fill out the remainder 'length' bytes (always a multiple
842 * of 8). If NXFMF_ACTIONS was not specified, or 'event' is
843 * NXFME_DELETED, no actions are included.
846 OFP_ASSERT(sizeof(struct nx_flow_update_full) == 24);
848 /* NXST_FLOW_MONITOR reply for NXFME_ABBREV.
850 * When the controller does not specify NXFMF_OWN in a monitor request, any
851 * flow tables changes due to the controller's own requests (on the same
852 * OpenFlow channel) will be abbreviated, when possible, to this form, which
853 * simply specifies the 'xid' of the OpenFlow request (e.g. an OFPT_FLOW_MOD or
854 * NXT_FLOW_MOD) that caused the change.
856 * Some changes cannot be abbreviated and will be sent in full:
858 * - Changes that only partially succeed. This can happen if, for example,
859 * a flow_mod with type OFPFC_MODIFY affects multiple flows, but only some
860 * of those modifications succeed (e.g. due to hardware limitations).
862 * This cannot occur with the Open vSwitch software datapath. This also
863 * cannot occur in Open vSwitch 2.4 and later, because these versions only
864 * execute any flow modifications if all of them will succeed.
866 * - Changes that race with conflicting changes made by other controllers or
867 * other flow_mods (not separated by barriers) by the same controller.
869 * This cannot occur with the current Open vSwitch implementation
870 * (regardless of datapath) because Open vSwitch internally serializes
871 * potentially conflicting changes.
873 * - Changes that occur when flow notification is paused (see "Buffer
874 * Management" above).
876 * A flow_mod that does not change the flow table will not trigger any
877 * notification, even an abbreviated one. For example, a "modify" or "delete"
878 * flow_mod that does not match any flows will not trigger a notification.
879 * Whether an "add" or "modify" that specifies all the same parameters that a
880 * flow already has triggers a notification is unspecified and subject to
881 * change in future versions of Open vSwitch.
883 * OVS will always send the notifications for a given flow table change before
884 * the reply to a OFPT_BARRIER_REQUEST request that follows the flow table
885 * change. Thus, if the controller does not receive an abbreviated (or
886 * unabbreviated) notification for a flow_mod before the next
887 * OFPT_BARRIER_REPLY, it will never receive one. */
888 struct nx_flow_update_abbrev {
889 ovs_be16 length; /* Length is 8. */
890 ovs_be16 event; /* NXFME_ABBREV. */
891 ovs_be32 xid; /* Controller-specified xid from flow_mod. */
893 OFP_ASSERT(sizeof(struct nx_flow_update_abbrev) == 8);
895 /* NXT_FLOW_MONITOR_CANCEL.
897 * Used by a controller to cancel an outstanding monitor. */
898 struct nx_flow_monitor_cancel {
899 ovs_be32 id; /* 'id' from nx_flow_monitor_request. */
901 OFP_ASSERT(sizeof(struct nx_flow_monitor_cancel) == 4);
903 #endif /* openflow/nicira-ext.h */