]> git.proxmox.com Git - mirror_ovs.git/blame - include/openflow/nicira-ext.h
rconn: Maintain original allowed_versions instead of substituting for zero.
[mirror_ovs.git] / include / openflow / nicira-ext.h
CommitLineData
064af421 1/*
e0edde6f 2 * Copyright (c) 2008, 2009, 2010, 2011, 2012 Nicira, Inc.
a14bc59f
BP
3 *
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:
7 *
8 * http://www.apache.org/licenses/LICENSE-2.0
9 *
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.
064af421
BP
15 */
16
17#ifndef OPENFLOW_NICIRA_EXT_H
18#define OPENFLOW_NICIRA_EXT_H 1
19
20#include "openflow/openflow.h"
26c112c2 21#include "openvswitch/types.h"
064af421 22
e0edde6f 23/* The following vendor extensions, proposed by Nicira, are not yet
26c112c2
BP
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. */
064af421
BP
26
27#define NX_VENDOR_ID 0x00002320
26c112c2
BP
28\f
29/* Nicira vendor-specific error messages extension.
30 *
31 * OpenFlow 1.0 has a set of predefined error types (OFPET_*) and codes (which
32 * are specific to each type). It does not have any provision for
33 * vendor-specific error codes, and it does not even provide "generic" error
34 * codes that can apply to problems not anticipated by the OpenFlow
35 * specification authors.
36 *
37 * This extension attempts to address the problem by adding a generic "error
38 * vendor extension". The extension works as follows: use NXET_VENDOR as type
6eb59a8f 39 * and NXVC_VENDOR_ERROR as code, followed by struct nx_vendor_error with
26c112c2
BP
40 * vendor-specific details, followed by at least 64 bytes of the failed
41 * request.
42 *
a23aab1f 43 * It would be better to have a type-specific vendor extension, e.g. so that
26c112c2
BP
44 * OFPET_BAD_ACTION could be used with vendor-specific code values. But
45 * OFPET_BAD_ACTION and most other standardized types already specify that
46 * their 'data' values are (the start of) the OpenFlow message being replied
47 * to, so there is no room to insert a vendor ID.
48 *
49 * Currently this extension is only implemented by Open vSwitch, but it seems
50 * like a reasonable candidate for future standardization.
51 */
52
53/* This is a random number to avoid accidental collision with any other
54 * vendor's extension. */
55#define NXET_VENDOR 0xb0c2
56
57/* ofp_error msg 'code' values for NXET_VENDOR. */
58enum nx_vendor_code {
59 NXVC_VENDOR_ERROR /* 'data' contains struct nx_vendor_error. */
60};
61
62/* 'data' for 'type' == NXET_VENDOR, 'code' == NXVC_VENDOR_ERROR. */
63struct nx_vendor_error {
64 ovs_be32 vendor; /* Vendor ID as in struct ofp_vendor_header. */
65 ovs_be16 type; /* Vendor-defined type. */
66 ovs_be16 code; /* Vendor-defined subtype. */
67 /* Followed by at least the first 64 bytes of the failed request. */
68};
69\f
70/* Nicira vendor requests and replies. */
064af421 71
09246b99
BP
72/* Header for Nicira vendor requests and replies. */
73struct nicira_header {
74 struct ofp_header header;
75 ovs_be32 vendor; /* NX_VENDOR_ID. */
982697a4 76 ovs_be32 subtype; /* See the NXT numbers in ofp-msgs.h. */
09246b99
BP
77};
78OFP_ASSERT(sizeof(struct nicira_header) == 16);
79
982697a4
BP
80/* Header for Nicira vendor stats request and reply messages in OpenFlow
81 * 1.0. */
82struct nicira10_stats_msg {
83 struct ofp10_vendor_stats_msg vsm; /* Vendor NX_VENDOR_ID. */
09246b99
BP
84 ovs_be32 subtype; /* One of NXST_* below. */
85 uint8_t pad[4]; /* Align to 64-bits. */
064af421 86};
982697a4 87OFP_ASSERT(sizeof(struct nicira10_stats_msg) == 24);
064af421 88
982697a4
BP
89/* Header for Nicira vendor stats request and reply messages in OpenFlow
90 * 1.1. */
91struct nicira11_stats_msg {
92 struct ofp11_vendor_stats_msg vsm; /* Vendor NX_VENDOR_ID. */
93 ovs_be32 subtype; /* One of NXST_* below. */
f9bfea14 94};
982697a4 95OFP_ASSERT(sizeof(struct nicira11_stats_msg) == 24);
f9bfea14 96
520e9a2a
EJ
97/* Fields to use when hashing flows. */
98enum nx_hash_fields {
99 /* Ethernet source address (NXM_OF_ETH_SRC) only. */
100 NX_HASH_FIELDS_ETH_SRC,
101
102 /* L2 through L4, symmetric across src/dst. Specifically, each of the
103 * following fields, if present, is hashed (slashes separate symmetric
104 * pairs):
105 *
106 * - NXM_OF_ETH_DST / NXM_OF_ETH_SRC
107 * - NXM_OF_ETH_TYPE
108 * - The VID bits from NXM_OF_VLAN_TCI, ignoring PCP and CFI.
109 * - NXM_OF_IP_PROTO
110 * - NXM_OF_IP_SRC / NXM_OF_IP_DST
111 * - NXM_OF_TCP_SRC / NXM_OF_TCP_DST
112 */
113 NX_HASH_FIELDS_SYMMETRIC_L4
114};
115
6c1491fb
BP
116/* This command enables or disables an Open vSwitch extension that allows a
117 * controller to specify the OpenFlow table to which a flow should be added,
118 * instead of having the switch decide which table is most appropriate as
119 * required by OpenFlow 1.0. By default, the extension is disabled.
120 *
35805806 121 * When this feature is enabled, Open vSwitch treats struct ofp10_flow_mod's
6c1491fb
BP
122 * 16-bit 'command' member as two separate fields. The upper 8 bits are used
123 * as the table ID, the lower 8 bits specify the command as usual. A table ID
124 * of 0xff is treated like a wildcarded table ID.
125 *
126 * The specific treatment of the table ID depends on the type of flow mod:
127 *
128 * - OFPFC_ADD: Given a specific table ID, the flow is always placed in that
129 * table. If an identical flow already exists in that table only, then it
130 * is replaced. If the flow cannot be placed in the specified table,
131 * either because the table is full or because the table cannot support
be2b69d1
BP
132 * flows of the given type, the switch replies with an OFPFMFC_TABLE_FULL
133 * error. (A controller can distinguish these cases by comparing the
134 * current and maximum number of entries reported in ofp_table_stats.)
6c1491fb
BP
135 *
136 * If the table ID is wildcarded, the switch picks an appropriate table
137 * itself. If an identical flow already exist in the selected flow table,
138 * then it is replaced. The choice of table might depend on the flows
139 * that are already in the switch; for example, if one table fills up then
140 * the switch might fall back to another one.
141 *
142 * - OFPFC_MODIFY, OFPFC_DELETE: Given a specific table ID, only flows
143 * within that table are matched and modified or deleted. If the table ID
144 * is wildcarded, flows within any table may be matched and modified or
145 * deleted.
146 *
147 * - OFPFC_MODIFY_STRICT, OFPFC_DELETE_STRICT: Given a specific table ID,
148 * only a flow within that table may be matched and modified or deleted.
149 * If the table ID is wildcarded and exactly one flow within any table
150 * matches, then it is modified or deleted; if flows in more than one
151 * table match, then none is modified or deleted.
152 */
73dbf4ab 153struct nx_flow_mod_table_id {
6c1491fb
BP
154 uint8_t set; /* Nonzero to enable, zero to disable. */
155 uint8_t pad[7];
156};
982697a4 157OFP_ASSERT(sizeof(struct nx_flow_mod_table_id) == 8);
6c1491fb 158
54834960
EJ
159enum nx_packet_in_format {
160 NXPIF_OPENFLOW10 = 0, /* Standard OpenFlow 1.0 compatible. */
161 NXPIF_NXM = 1 /* Nicira Extended. */
162};
163
164/* NXT_SET_PACKET_IN_FORMAT request. */
73dbf4ab 165struct nx_set_packet_in_format {
54834960
EJ
166 ovs_be32 format; /* One of NXPIF_*. */
167};
982697a4 168OFP_ASSERT(sizeof(struct nx_set_packet_in_format) == 4);
54834960
EJ
169
170/* NXT_PACKET_IN (analogous to OFPT_PACKET_IN).
171 *
42edbe39
BP
172 * NXT_PACKET_IN is similar to the OpenFlow 1.2 OFPT_PACKET_IN. The
173 * differences are:
174 *
175 * - NXT_PACKET_IN includes the cookie of the rule that triggered the
176 * message. (OpenFlow 1.3 OFPT_PACKET_IN also includes the cookie.)
177 *
178 * - The metadata fields use NXM (instead of OXM) field numbers.
179 *
180 * Open vSwitch 1.9.0 and later omits metadata fields that are zero (as allowed
181 * by OpenFlow 1.2). Earlier versions included all implemented metadata
182 * fields.
183 *
184 * Open vSwitch does not include non-metadata in the nx_match, because by
185 * definition that information can be found in the packet itself. The format
186 * and the standards allow this, however, so controllers should be prepared to
187 * tolerate future changes.
188 *
189 * The NXM format is convenient for reporting metadata values, but it is
190 * important not to interpret the format as matching against a flow, because it
191 * does not. Nothing is being matched; arbitrary metadata masks would not be
192 * meaningful.
54834960
EJ
193 *
194 * Whereas in most cases a controller can expect to only get back NXM fields
195 * that it set up itself (e.g. flow dumps will ordinarily report only NXM
196 * fields from flows that the controller added), NXT_PACKET_IN messages might
197 * contain fields that the controller does not understand, because the switch
198 * might support fields (new registers, new protocols, etc.) that the
42edbe39 199 * controller does not. The controller must prepared to tolerate these.
54834960
EJ
200 *
201 * The 'cookie' and 'table_id' fields have no meaning when 'reason' is
202 * OFPR_NO_MATCH. In this case they should be set to 0. */
73dbf4ab 203struct nx_packet_in {
54834960
EJ
204 ovs_be32 buffer_id; /* ID assigned by datapath. */
205 ovs_be16 total_len; /* Full length of frame. */
206 uint8_t reason; /* Reason packet is sent (one of OFPR_*). */
207 uint8_t table_id; /* ID of the table that was looked up. */
208 ovs_be64 cookie; /* Cookie of the rule that was looked up. */
209 ovs_be16 match_len; /* Size of nx_match. */
210 uint8_t pad[6]; /* Align to 64-bits. */
211 /* Followed by:
212 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
213 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
214 * all-zero bytes, then
215 * - Exactly 2 all-zero padding bytes, then
216 * - An Ethernet frame whose length is inferred from nxh.header.length.
217 *
218 * The padding bytes preceding the Ethernet frame ensure that the IP
219 * header (if any) following the Ethernet header is 32-bit aligned. */
220
42edbe39 221 /* uint8_t nxm_fields[...]; */ /* NXM headers. */
54834960
EJ
222 /* uint8_t pad[2]; */ /* Align to 64 bit + 16 bit. */
223 /* uint8_t data[0]; */ /* Ethernet frame. */
224};
982697a4 225OFP_ASSERT(sizeof(struct nx_packet_in) == 24);
54834960 226
9deba63b
BP
227/* Configures the "role" of the sending controller. The default role is:
228 *
229 * - Other (NX_ROLE_OTHER), which allows the controller access to all
230 * OpenFlow features.
231 *
232 * The other possible roles are a related pair:
233 *
234 * - Master (NX_ROLE_MASTER) is equivalent to Other, except that there may
235 * be at most one Master controller at a time: when a controller
236 * configures itself as Master, any existing Master is demoted to the
237 * Slave role.
238 *
239 * - Slave (NX_ROLE_SLAVE) allows the controller read-only access to
240 * OpenFlow features. In particular attempts to modify the flow table
241 * will be rejected with an OFPBRC_EPERM error.
242 *
197a992f
BP
243 * Slave controllers do not receive OFPT_PACKET_IN or OFPT_FLOW_REMOVED
244 * messages, but they do receive OFPT_PORT_STATUS messages.
9deba63b
BP
245 */
246struct nx_role_request {
459749fe 247 ovs_be32 role; /* One of NX_ROLE_*. */
9deba63b 248};
982697a4 249OFP_ASSERT(sizeof(struct nx_role_request) == 4);
9deba63b
BP
250
251enum nx_role {
252 NX_ROLE_OTHER, /* Default role, full access. */
253 NX_ROLE_MASTER, /* Full access, at most one. */
254 NX_ROLE_SLAVE /* Read-only access. */
255};
80d5aefd
BP
256
257/* NXT_SET_ASYNC_CONFIG.
258 *
259 * Sent by a controller, this message configures the asynchronous messages that
260 * the controller wants to receive. Element 0 in each array specifies messages
261 * of interest when the controller has an "other" or "master" role; element 1,
262 * when the controller has a "slave" role.
263 *
264 * Each array element is a bitmask in which a 0-bit disables receiving a
265 * particular message and a 1-bit enables receiving it. Each bit controls the
266 * message whose 'reason' corresponds to the bit index. For example, the bit
267 * with value 1<<2 == 4 in port_status_mask[1] determines whether the
268 * controller will receive OFPT_PORT_STATUS messages with reason OFPPR_MODIFY
269 * (value 2) when the controller has a "slave" role.
4550b647
MM
270 *
271 * As a side effect, for service controllers, this message changes the
272 * miss_send_len from default of zero to OFP_DEFAULT_MISS_SEND_LEN (128).
80d5aefd
BP
273 */
274struct nx_async_config {
80d5aefd
BP
275 ovs_be32 packet_in_mask[2]; /* Bitmasks of OFPR_* values. */
276 ovs_be32 port_status_mask[2]; /* Bitmasks of OFPRR_* values. */
277 ovs_be32 flow_removed_mask[2]; /* Bitmasks of OFPPR_* values. */
278};
982697a4 279OFP_ASSERT(sizeof(struct nx_async_config) == 24);
26c112c2
BP
280\f
281/* Nicira vendor flow actions. */
9deba63b 282
064af421 283enum nx_action_subtype {
39b72738
BP
284 NXAST_SNAT__OBSOLETE, /* No longer used. */
285 NXAST_RESUBMIT, /* struct nx_action_resubmit */
286 NXAST_SET_TUNNEL, /* struct nx_action_set_tunnel */
6c222e55 287 NXAST_DROP_SPOOFED_ARP__OBSOLETE,
39b72738 288 NXAST_SET_QUEUE, /* struct nx_action_set_queue */
b6c9e612
BP
289 NXAST_POP_QUEUE, /* struct nx_action_pop_queue */
290 NXAST_REG_MOVE, /* struct nx_action_reg_move */
291 NXAST_REG_LOAD, /* struct nx_action_reg_load */
b9298d3f
BP
292 NXAST_NOTE, /* struct nx_action_note */
293 NXAST_SET_TUNNEL64, /* struct nx_action_set_tunnel64 */
3b6a2571 294 NXAST_MULTIPATH, /* struct nx_action_multipath */
c51c638a 295 NXAST_AUTOPATH__DEPRECATED, /* struct nx_action_autopath */
a368bb53 296 NXAST_BUNDLE, /* struct nx_action_bundle */
29901626 297 NXAST_BUNDLE_LOAD, /* struct nx_action_bundle */
f694937d 298 NXAST_RESUBMIT_TABLE, /* struct nx_action_resubmit */
75a75043 299 NXAST_OUTPUT_REG, /* struct nx_action_output_reg */
848e8809 300 NXAST_LEARN, /* struct nx_action_learn */
f0fd1a17
PS
301 NXAST_EXIT, /* struct nx_action_header */
302 NXAST_DEC_TTL, /* struct nx_action_header */
0e553d9c 303 NXAST_FIN_TIMEOUT, /* struct nx_action_fin_timeout */
a7349929 304 NXAST_CONTROLLER, /* struct nx_action_controller */
c2d967a5 305 NXAST_DEC_TTL_CNT_IDS, /* struct nx_action_cnt_ids */
4cceacb9 306 NXAST_WRITE_METADATA, /* struct nx_action_write_metadata */
39b72738 307};
eedc0097 308
39b72738
BP
309/* Header for Nicira-defined actions. */
310struct nx_action_header {
459749fe
BP
311 ovs_be16 type; /* OFPAT_VENDOR. */
312 ovs_be16 len; /* Length is 16. */
313 ovs_be32 vendor; /* NX_VENDOR_ID. */
314 ovs_be16 subtype; /* NXAST_*. */
39b72738 315 uint8_t pad[6];
064af421 316};
39b72738 317OFP_ASSERT(sizeof(struct nx_action_header) == 16);
064af421 318
29901626 319/* Action structures for NXAST_RESUBMIT and NXAST_RESUBMIT_TABLE.
39b72738 320 *
29901626 321 * These actions search one of the switch's flow tables:
39b72738 322 *
29901626
BP
323 * - For NXAST_RESUBMIT_TABLE only, if the 'table' member is not 255, then
324 * it specifies the table to search.
39b72738 325 *
29901626
BP
326 * - Otherwise (for NXAST_RESUBMIT_TABLE with a 'table' of 255, or for
327 * NXAST_RESUBMIT regardless of 'table'), it searches the current flow
328 * table, that is, the OpenFlow flow table that contains the flow from
329 * which this action was obtained. If this action did not come from a
330 * flow table (e.g. it came from an OFPT_PACKET_OUT message), then table 0
331 * is the current table.
332 *
333 * The flow table lookup uses a flow that may be slightly modified from the
334 * original lookup:
335 *
336 * - For NXAST_RESUBMIT, the 'in_port' member of struct nx_action_resubmit
337 * is used as the flow's in_port.
338 *
339 * - For NXAST_RESUBMIT_TABLE, if the 'in_port' member is not OFPP_IN_PORT,
340 * then its value is used as the flow's in_port. Otherwise, the original
341 * in_port is used.
342 *
343 * - If actions that modify the flow (e.g. OFPAT_SET_VLAN_VID) precede the
344 * resubmit action, then the flow is updated with the new values.
39b72738
BP
345 *
346 * Following the lookup, the original in_port is restored.
347 *
348 * If the modified flow matched in the flow table, then the corresponding
29901626 349 * actions are executed. Afterward, actions following the resubmit in the
39b72738
BP
350 * original set of actions, if any, are executed; any changes made to the
351 * packet (e.g. changes to VLAN) by secondary actions persist when those
352 * actions are executed, although the original in_port is restored.
353 *
29901626
BP
354 * Resubmit actions may be used any number of times within a set of actions.
355 *
356 * Resubmit actions may nest to an implementation-defined depth. Beyond this
357 * implementation-defined depth, further resubmit actions are simply ignored.
358 *
359 * NXAST_RESUBMIT ignores 'table' and 'pad'. NXAST_RESUBMIT_TABLE requires
360 * 'pad' to be all-bits-zero.
39b72738 361 *
29901626
BP
362 * Open vSwitch 1.0.1 and earlier did not support recursion. Open vSwitch
363 * before 1.2.90 did not support NXAST_RESUBMIT_TABLE.
39b72738 364 */
064af421 365struct nx_action_resubmit {
459749fe
BP
366 ovs_be16 type; /* OFPAT_VENDOR. */
367 ovs_be16 len; /* Length is 16. */
368 ovs_be32 vendor; /* NX_VENDOR_ID. */
369 ovs_be16 subtype; /* NXAST_RESUBMIT. */
370 ovs_be16 in_port; /* New in_port for checking flow table. */
29901626
BP
371 uint8_t table; /* NXAST_RESUBMIT_TABLE: table to use. */
372 uint8_t pad[3];
064af421
BP
373};
374OFP_ASSERT(sizeof(struct nx_action_resubmit) == 16);
375
39b72738
BP
376/* Action structure for NXAST_SET_TUNNEL.
377 *
b9298d3f
BP
378 * Sets the encapsulating tunnel ID to a 32-bit value. The most-significant 32
379 * bits of the tunnel ID are set to 0. */
659586ef 380struct nx_action_set_tunnel {
459749fe
BP
381 ovs_be16 type; /* OFPAT_VENDOR. */
382 ovs_be16 len; /* Length is 16. */
383 ovs_be32 vendor; /* NX_VENDOR_ID. */
384 ovs_be16 subtype; /* NXAST_SET_TUNNEL. */
659586ef 385 uint8_t pad[2];
459749fe 386 ovs_be32 tun_id; /* Tunnel ID. */
659586ef
JG
387};
388OFP_ASSERT(sizeof(struct nx_action_set_tunnel) == 16);
389
b9298d3f
BP
390/* Action structure for NXAST_SET_TUNNEL64.
391 *
392 * Sets the encapsulating tunnel ID to a 64-bit value. */
393struct nx_action_set_tunnel64 {
394 ovs_be16 type; /* OFPAT_VENDOR. */
9c59112b 395 ovs_be16 len; /* Length is 24. */
b9298d3f
BP
396 ovs_be32 vendor; /* NX_VENDOR_ID. */
397 ovs_be16 subtype; /* NXAST_SET_TUNNEL64. */
398 uint8_t pad[6];
399 ovs_be64 tun_id; /* Tunnel ID. */
400};
401OFP_ASSERT(sizeof(struct nx_action_set_tunnel64) == 24);
402
39b72738
BP
403/* Action structure for NXAST_SET_QUEUE.
404 *
405 * Set the queue that should be used when packets are output. This is similar
406 * to the OpenFlow OFPAT_ENQUEUE action, but does not take the output port as
407 * an argument. This allows the queue to be defined before the port is
408 * known. */
eedc0097 409struct nx_action_set_queue {
459749fe
BP
410 ovs_be16 type; /* OFPAT_VENDOR. */
411 ovs_be16 len; /* Length is 16. */
412 ovs_be32 vendor; /* NX_VENDOR_ID. */
413 ovs_be16 subtype; /* NXAST_SET_QUEUE. */
eedc0097 414 uint8_t pad[2];
459749fe 415 ovs_be32 queue_id; /* Where to enqueue packets. */
eedc0097
JP
416};
417OFP_ASSERT(sizeof(struct nx_action_set_queue) == 16);
418
39b72738
BP
419/* Action structure for NXAST_POP_QUEUE.
420 *
421 * Restores the queue to the value it was before any NXAST_SET_QUEUE actions
422 * were used. Only the original queue can be restored this way; no stack is
423 * maintained. */
424struct nx_action_pop_queue {
459749fe
BP
425 ovs_be16 type; /* OFPAT_VENDOR. */
426 ovs_be16 len; /* Length is 16. */
427 ovs_be32 vendor; /* NX_VENDOR_ID. */
428 ovs_be16 subtype; /* NXAST_POP_QUEUE. */
064af421
BP
429 uint8_t pad[6];
430};
39b72738 431OFP_ASSERT(sizeof(struct nx_action_pop_queue) == 16);
064af421 432
b6c9e612
BP
433/* Action structure for NXAST_REG_MOVE.
434 *
435 * Copies src[src_ofs:src_ofs+n_bits] to dst[dst_ofs:dst_ofs+n_bits], where
436 * a[b:c] denotes the bits within 'a' numbered 'b' through 'c' (not including
437 * bit 'c'). Bit numbering starts at 0 for the least-significant bit, 1 for
438 * the next most significant bit, and so on.
439 *
60d5e0d8
BP
440 * 'src' and 'dst' are nxm_header values with nxm_hasmask=0. (It doesn't make
441 * sense to use nxm_hasmask=1 because the action does not do any kind of
442 * matching; it uses the actual value of a field.)
443 *
444 * The following nxm_header values are potentially acceptable as 'src':
b6c9e612
BP
445 *
446 * - NXM_OF_IN_PORT
447 * - NXM_OF_ETH_DST
448 * - NXM_OF_ETH_SRC
449 * - NXM_OF_ETH_TYPE
450 * - NXM_OF_VLAN_TCI
451 * - NXM_OF_IP_TOS
452 * - NXM_OF_IP_PROTO
453 * - NXM_OF_IP_SRC
454 * - NXM_OF_IP_DST
455 * - NXM_OF_TCP_SRC
456 * - NXM_OF_TCP_DST
457 * - NXM_OF_UDP_SRC
458 * - NXM_OF_UDP_DST
459 * - NXM_OF_ICMP_TYPE
460 * - NXM_OF_ICMP_CODE
461 * - NXM_OF_ARP_OP
462 * - NXM_OF_ARP_SPA
463 * - NXM_OF_ARP_TPA
464 * - NXM_NX_TUN_ID
bad68a99
JP
465 * - NXM_NX_ARP_SHA
466 * - NXM_NX_ARP_THA
d31f1109
JP
467 * - NXM_NX_ICMPV6_TYPE
468 * - NXM_NX_ICMPV6_CODE
685a51a5
JP
469 * - NXM_NX_ND_SLL
470 * - NXM_NX_ND_TLL
b6c9e612
BP
471 * - NXM_NX_REG(idx) for idx in the switch's accepted range.
472 *
473 * The following nxm_header values are potentially acceptable as 'dst':
474 *
54d54726
BP
475 * - NXM_OF_ETH_DST
476 * - NXM_OF_ETH_SRC
477 * - NXM_OF_IP_TOS
478 * - NXM_OF_IP_SRC
479 * - NXM_OF_IP_DST
480 * - NXM_OF_TCP_SRC
481 * - NXM_OF_TCP_DST
482 * - NXM_OF_UDP_SRC
483 * - NXM_OF_UDP_DST
484 * Modifying any of the above fields changes the corresponding packet
485 * header.
486 *
b6c9e612
BP
487 * - NXM_NX_REG(idx) for idx in the switch's accepted range.
488 *
489 * - NXM_OF_VLAN_TCI. Modifying this field's value has side effects on the
490 * packet's 802.1Q header. Setting a value with CFI=0 removes the 802.1Q
491 * header (if any), ignoring the other bits. Setting a value with CFI=1
492 * adds or modifies the 802.1Q header appropriately, setting the TCI field
493 * to the field's new value (with the CFI bit masked out).
494 *
495 * - NXM_NX_TUN_ID. Modifying this value modifies the tunnel ID used for the
496 * packet's next tunnel encapsulation.
497 *
498 * A given nxm_header value may be used as 'src' or 'dst' only on a flow whose
499 * nx_match satisfies its prerequisites. For example, NXM_OF_IP_TOS may be
500 * used only if the flow's nx_match includes an nxm_entry that specifies
501 * nxm_type=NXM_OF_ETH_TYPE, nxm_hasmask=0, and nxm_value=0x0800.
502 *
503 * The switch will reject actions for which src_ofs+n_bits is greater than the
504 * width of 'src' or dst_ofs+n_bits is greater than the width of 'dst' with
505 * error type OFPET_BAD_ACTION, code OFPBAC_BAD_ARGUMENT.
a478fee3
BP
506 *
507 * This action behaves properly when 'src' overlaps with 'dst', that is, it
508 * behaves as if 'src' were copied out to a temporary buffer, then the
509 * temporary buffer copied to 'dst'.
b6c9e612
BP
510 */
511struct nx_action_reg_move {
512 ovs_be16 type; /* OFPAT_VENDOR. */
9c59112b 513 ovs_be16 len; /* Length is 24. */
b6c9e612
BP
514 ovs_be32 vendor; /* NX_VENDOR_ID. */
515 ovs_be16 subtype; /* NXAST_REG_MOVE. */
516 ovs_be16 n_bits; /* Number of bits. */
517 ovs_be16 src_ofs; /* Starting bit offset in source. */
518 ovs_be16 dst_ofs; /* Starting bit offset in destination. */
519 ovs_be32 src; /* Source register. */
520 ovs_be32 dst; /* Destination register. */
521};
522OFP_ASSERT(sizeof(struct nx_action_reg_move) == 24);
523
524/* Action structure for NXAST_REG_LOAD.
525 *
526 * Copies value[0:n_bits] to dst[ofs:ofs+n_bits], where a[b:c] denotes the bits
527 * within 'a' numbered 'b' through 'c' (not including bit 'c'). Bit numbering
528 * starts at 0 for the least-significant bit, 1 for the next most significant
529 * bit, and so on.
530 *
7b064a79
BP
531 * 'dst' is an nxm_header with nxm_hasmask=0. See the documentation for
532 * NXAST_REG_MOVE, above, for the permitted fields and for the side effects of
533 * loading them.
b6c9e612
BP
534 *
535 * The 'ofs' and 'n_bits' fields are combined into a single 'ofs_nbits' field
536 * to avoid enlarging the structure by another 8 bytes. To allow 'n_bits' to
537 * take a value between 1 and 64 (inclusive) while taking up only 6 bits, it is
538 * also stored as one less than its true value:
539 *
540 * 15 6 5 0
541 * +------------------------------+------------------+
542 * | ofs | n_bits - 1 |
543 * +------------------------------+------------------+
544 *
545 * The switch will reject actions for which ofs+n_bits is greater than the
546 * width of 'dst', or in which any bits in 'value' with value 2**n_bits or
547 * greater are set to 1, with error type OFPET_BAD_ACTION, code
548 * OFPBAC_BAD_ARGUMENT.
549 */
550struct nx_action_reg_load {
551 ovs_be16 type; /* OFPAT_VENDOR. */
9c59112b 552 ovs_be16 len; /* Length is 24. */
b6c9e612
BP
553 ovs_be32 vendor; /* NX_VENDOR_ID. */
554 ovs_be16 subtype; /* NXAST_REG_LOAD. */
555 ovs_be16 ofs_nbits; /* (ofs << 6) | (n_bits - 1). */
556 ovs_be32 dst; /* Destination register. */
557 ovs_be64 value; /* Immediate value. */
558};
559OFP_ASSERT(sizeof(struct nx_action_reg_load) == 24);
560
96fc46e8
BP
561/* Action structure for NXAST_NOTE.
562 *
563 * This action has no effect. It is variable length. The switch does not
564 * attempt to interpret the user-defined 'note' data in any way. A controller
565 * can use this action to attach arbitrary metadata to a flow.
566 *
567 * This action might go away in the future.
568 */
569struct nx_action_note {
459749fe
BP
570 ovs_be16 type; /* OFPAT_VENDOR. */
571 ovs_be16 len; /* A multiple of 8, but at least 16. */
572 ovs_be32 vendor; /* NX_VENDOR_ID. */
573 ovs_be16 subtype; /* NXAST_NOTE. */
96fc46e8
BP
574 uint8_t note[6]; /* Start of user-defined data. */
575 /* Possibly followed by additional user-defined data. */
576};
577OFP_ASSERT(sizeof(struct nx_action_note) == 16);
578
53ddd40a
BP
579/* Action structure for NXAST_MULTIPATH.
580 *
581 * This action performs the following steps in sequence:
582 *
520e9a2a 583 * 1. Hashes the fields designated by 'fields', one of NX_HASH_FIELDS_*.
53ddd40a
BP
584 * Refer to the definition of "enum nx_mp_fields" for details.
585 *
586 * The 'basis' value is used as a universal hash parameter, that is,
587 * different values of 'basis' yield different hash functions. The
588 * particular universal hash function used is implementation-defined.
589 *
590 * The hashed fields' values are drawn from the current state of the
591 * flow, including all modifications that have been made by actions up to
592 * this point.
593 *
594 * 2. Applies the multipath link choice algorithm specified by 'algorithm',
595 * one of NX_MP_ALG_*. Refer to the definition of "enum nx_mp_algorithm"
596 * for details.
597 *
598 * The output of the algorithm is 'link', an unsigned integer less than
599 * or equal to 'max_link'.
600 *
601 * Some algorithms use 'arg' as an additional argument.
602 *
603 * 3. Stores 'link' in dst[ofs:ofs+n_bits]. The format and semantics of
338bd6a0 604 * 'dst' and 'ofs_nbits' are similar to those for the NXAST_REG_LOAD
43edca57 605 * action.
53ddd40a
BP
606 *
607 * The switch will reject actions that have an unknown 'fields', or an unknown
608 * 'algorithm', or in which ofs+n_bits is greater than the width of 'dst', or
609 * in which 'max_link' is greater than or equal to 2**n_bits, with error type
610 * OFPET_BAD_ACTION, code OFPBAC_BAD_ARGUMENT.
611 */
612struct nx_action_multipath {
613 ovs_be16 type; /* OFPAT_VENDOR. */
614 ovs_be16 len; /* Length is 32. */
615 ovs_be32 vendor; /* NX_VENDOR_ID. */
616 ovs_be16 subtype; /* NXAST_MULTIPATH. */
617
618 /* What fields to hash and how. */
520e9a2a 619 ovs_be16 fields; /* One of NX_HASH_FIELDS_*. */
53ddd40a
BP
620 ovs_be16 basis; /* Universal hash parameter. */
621 ovs_be16 pad0;
622
623 /* Multipath link choice algorithm to apply to hash value. */
624 ovs_be16 algorithm; /* One of NX_MP_ALG_*. */
625 ovs_be16 max_link; /* Number of output links, minus 1. */
626 ovs_be32 arg; /* Algorithm-specific argument. */
627 ovs_be16 pad1;
628
629 /* Where to store the result. */
630 ovs_be16 ofs_nbits; /* (ofs << 6) | (n_bits - 1). */
43edca57 631 ovs_be32 dst; /* Destination. */
53ddd40a
BP
632};
633OFP_ASSERT(sizeof(struct nx_action_multipath) == 32);
634
53ddd40a
BP
635/* NXAST_MULTIPATH: Multipath link choice algorithm to apply.
636 *
637 * In the descriptions below, 'n_links' is max_link + 1. */
638enum nx_mp_algorithm {
639 /* link = hash(flow) % n_links.
640 *
641 * Redistributes all traffic when n_links changes. O(1) performance. See
642 * RFC 2992.
643 *
644 * Use UINT16_MAX for max_link to get a raw hash value. */
645 NX_MP_ALG_MODULO_N,
646
647 /* link = hash(flow) / (MAX_HASH / n_links).
648 *
649 * Redistributes between one-quarter and one-half of traffic when n_links
650 * changes. O(1) performance. See RFC 2992.
651 */
652 NX_MP_ALG_HASH_THRESHOLD,
653
654 /* for i in [0,n_links):
655 * weights[i] = hash(flow, i)
656 * link = { i such that weights[i] >= weights[j] for all j != i }
657 *
658 * Redistributes 1/n_links of traffic when n_links changes. O(n_links)
659 * performance. If n_links is greater than a threshold (currently 64, but
660 * subject to change), Open vSwitch will substitute another algorithm
661 * automatically. See RFC 2992. */
662 NX_MP_ALG_HRW, /* Highest Random Weight. */
663
664 /* i = 0
665 * repeat:
666 * i = i + 1
667 * link = hash(flow, i) % arg
668 * while link > max_link
669 *
670 * Redistributes 1/n_links of traffic when n_links changes. O(1)
671 * performance when arg/max_link is bounded by a constant.
672 *
673 * Redistributes all traffic when arg changes.
674 *
675 * arg must be greater than max_link and for best performance should be no
676 * more than approximately max_link * 2. If arg is outside the acceptable
677 * range, Open vSwitch will automatically substitute the least power of 2
678 * greater than max_link.
679 *
680 * This algorithm is specific to Open vSwitch.
681 */
682 NX_MP_ALG_ITER_HASH /* Iterative Hash. */
683};
09246b99 684\f
75a75043
BP
685/* Action structure for NXAST_LEARN.
686 *
687 * This action adds or modifies a flow in an OpenFlow table, similar to
688 * OFPT_FLOW_MOD with OFPFC_MODIFY_STRICT as 'command'. The new flow has the
689 * specified idle timeout, hard timeout, priority, cookie, and flags. The new
690 * flow's match criteria and actions are built by applying each of the series
691 * of flow_mod_spec elements included as part of the action.
692 *
693 * A flow_mod_spec starts with a 16-bit header. A header that is all-bits-0 is
694 * a no-op used for padding the action as a whole to a multiple of 8 bytes in
695 * length. Otherwise, the flow_mod_spec can be thought of as copying 'n_bits'
696 * bits from a source to a destination. In this case, the header contains
697 * multiple fields:
698 *
699 * 15 14 13 12 11 10 0
700 * +------+---+------+---------------------------------+
701 * | 0 |src| dst | n_bits |
702 * +------+---+------+---------------------------------+
703 *
704 * The meaning and format of a flow_mod_spec depends on 'src' and 'dst'. The
705 * following table summarizes the meaning of each possible combination.
706 * Details follow the table:
707 *
708 * src dst meaning
709 * --- --- ----------------------------------------------------------
710 * 0 0 Add match criteria based on value in a field.
711 * 1 0 Add match criteria based on an immediate value.
712 * 0 1 Add NXAST_REG_LOAD action to copy field into a different field.
713 * 1 1 Add NXAST_REG_LOAD action to load immediate value into a field.
714 * 0 2 Add OFPAT_OUTPUT action to output to port from specified field.
715 * All other combinations are undefined and not allowed.
716 *
717 * The flow_mod_spec header is followed by a source specification and a
718 * destination specification. The format and meaning of the source
719 * specification depends on 'src':
720 *
721 * - If 'src' is 0, the source bits are taken from a field in the flow to
722 * which this action is attached. (This should be a wildcarded field. If
723 * its value is fully specified then the source bits being copied have
724 * constant values.)
725 *
726 * The source specification is an ovs_be32 'field' and an ovs_be16 'ofs'.
727 * 'field' is an nxm_header with nxm_hasmask=0, and 'ofs' the starting bit
728 * offset within that field. The source bits are field[ofs:ofs+n_bits-1].
729 * 'field' and 'ofs' are subject to the same restrictions as the source
730 * field in NXAST_REG_MOVE.
731 *
732 * - If 'src' is 1, the source bits are a constant value. The source
733 * specification is (n_bits+15)/16*2 bytes long. Taking those bytes as a
734 * number in network order, the source bits are the 'n_bits'
735 * least-significant bits. The switch will report an error if other bits
736 * in the constant are nonzero.
737 *
738 * The flow_mod_spec destination specification, for 'dst' of 0 or 1, is an
739 * ovs_be32 'field' and an ovs_be16 'ofs'. 'field' is an nxm_header with
740 * nxm_hasmask=0 and 'ofs' is a starting bit offset within that field. The
741 * meaning of the flow_mod_spec depends on 'dst':
742 *
743 * - If 'dst' is 0, the flow_mod_spec specifies match criteria for the new
744 * flow. The new flow matches only if bits field[ofs:ofs+n_bits-1] in a
745 * packet equal the source bits. 'field' may be any nxm_header with
746 * nxm_hasmask=0 that is allowed in NXT_FLOW_MOD.
747 *
748 * Order is significant. Earlier flow_mod_specs must satisfy any
749 * prerequisites for matching fields specified later, by copying constant
750 * values into prerequisite fields.
751 *
752 * The switch will reject flow_mod_specs that do not satisfy NXM masking
753 * restrictions.
754 *
755 * - If 'dst' is 1, the flow_mod_spec specifies an NXAST_REG_LOAD action for
756 * the new flow. The new flow copies the source bits into
757 * field[ofs:ofs+n_bits-1]. Actions are executed in the same order as the
758 * flow_mod_specs.
759 *
337b9cec
BP
760 * A single NXAST_REG_LOAD action writes no more than 64 bits, so n_bits
761 * greater than 64 yields multiple NXAST_REG_LOAD actions.
762 *
75a75043
BP
763 * The flow_mod_spec destination spec for 'dst' of 2 (when 'src' is 0) is
764 * empty. It has the following meaning:
765 *
766 * - The flow_mod_spec specifies an OFPAT_OUTPUT action for the new flow.
767 * The new flow outputs to the OpenFlow port specified by the source field.
768 * Of the special output ports with value OFPP_MAX or larger, OFPP_IN_PORT,
769 * OFPP_FLOOD, OFPP_LOCAL, and OFPP_ALL are supported. Other special ports
770 * may not be used.
771 *
772 * Resource Management
773 * -------------------
774 *
775 * A switch has a finite amount of flow table space available for learning.
776 * When this space is exhausted, no new learning table entries will be learned
777 * until some existing flow table entries expire. The controller should be
778 * prepared to handle this by flooding (which can be implemented as a
779 * low-priority flow).
780 *
0e553d9c
BP
781 * If a learned flow matches a single TCP stream with a relatively long
782 * timeout, one may make the best of resource constraints by setting
783 * 'fin_idle_timeout' or 'fin_hard_timeout' (both measured in seconds), or
784 * both, to shorter timeouts. When either of these is specified as a nonzero
785 * value, OVS adds a NXAST_FIN_TIMEOUT action, with the specified timeouts, to
786 * the learned flow.
787 *
75a75043
BP
788 * Examples
789 * --------
790 *
791 * The following examples give a prose description of the flow_mod_specs along
792 * with informal notation for how those would be represented and a hex dump of
793 * the bytes that would be required.
794 *
795 * These examples could work with various nx_action_learn parameters. Typical
796 * values would be idle_timeout=OFP_FLOW_PERMANENT, hard_timeout=60,
797 * priority=OFP_DEFAULT_PRIORITY, flags=0, table_id=10.
798 *
799 * 1. Learn input port based on the source MAC, with lookup into
800 * NXM_NX_REG1[16:31] by resubmit to in_port=99:
801 *
802 * Match on in_port=99:
803 * ovs_be16(src=1, dst=0, n_bits=16), 20 10
804 * ovs_be16(99), 00 63
805 * ovs_be32(NXM_OF_IN_PORT), ovs_be16(0) 00 00 00 02 00 00
806 *
807 * Match Ethernet destination on Ethernet source from packet:
808 * ovs_be16(src=0, dst=0, n_bits=48), 00 30
809 * ovs_be32(NXM_OF_ETH_SRC), ovs_be16(0) 00 00 04 06 00 00
810 * ovs_be32(NXM_OF_ETH_DST), ovs_be16(0) 00 00 02 06 00 00
811 *
812 * Set NXM_NX_REG1[16:31] to the packet's input port:
813 * ovs_be16(src=0, dst=1, n_bits=16), 08 10
814 * ovs_be32(NXM_OF_IN_PORT), ovs_be16(0) 00 00 00 02 00 00
815 * ovs_be32(NXM_NX_REG1), ovs_be16(16) 00 01 02 04 00 10
816 *
817 * Given a packet that arrived on port A with Ethernet source address B,
818 * this would set up the flow "in_port=99, dl_dst=B,
819 * actions=load:A->NXM_NX_REG1[16..31]".
820 *
821 * In syntax accepted by ovs-ofctl, this action is: learn(in_port=99,
eddfcba7
BP
822 * NXM_OF_ETH_DST[]=NXM_OF_ETH_SRC[],
823 * load:NXM_OF_IN_PORT[]->NXM_NX_REG1[16..31])
75a75043
BP
824 *
825 * 2. Output to input port based on the source MAC and VLAN VID, with lookup
826 * into NXM_NX_REG1[16:31]:
827 *
828 * Match on same VLAN ID as packet:
829 * ovs_be16(src=0, dst=0, n_bits=12), 00 0c
830 * ovs_be32(NXM_OF_VLAN_TCI), ovs_be16(0) 00 00 08 02 00 00
831 * ovs_be32(NXM_OF_VLAN_TCI), ovs_be16(0) 00 00 08 02 00 00
832 *
833 * Match Ethernet destination on Ethernet source from packet:
834 * ovs_be16(src=0, dst=0, n_bits=48), 00 30
835 * ovs_be32(NXM_OF_ETH_SRC), ovs_be16(0) 00 00 04 06 00 00
836 * ovs_be32(NXM_OF_ETH_DST), ovs_be16(0) 00 00 02 06 00 00
837 *
838 * Output to the packet's input port:
839 * ovs_be16(src=0, dst=2, n_bits=16), 10 10
840 * ovs_be32(NXM_OF_IN_PORT), ovs_be16(0) 00 00 00 02 00 00
841 *
842 * Given a packet that arrived on port A with Ethernet source address B in
843 * VLAN C, this would set up the flow "dl_dst=B, vlan_vid=C,
844 * actions=output:A".
845 *
846 * In syntax accepted by ovs-ofctl, this action is:
847 * learn(NXM_OF_VLAN_TCI[0..11], NXM_OF_ETH_DST[]=NXM_OF_ETH_SRC[],
848 * output:NXM_OF_IN_PORT[])
849 *
850 * 3. Here's a recipe for a very simple-minded MAC learning switch. It uses a
851 * 10-second MAC expiration time to make it easier to see what's going on
852 *
853 * ovs-vsctl del-controller br0
854 * ovs-ofctl del-flows br0
855 * ovs-ofctl add-flow br0 "table=0 actions=learn(table=1, \
856 hard_timeout=10, NXM_OF_VLAN_TCI[0..11], \
857 NXM_OF_ETH_DST[]=NXM_OF_ETH_SRC[], \
858 output:NXM_OF_IN_PORT[]), resubmit(,1)"
859 * ovs-ofctl add-flow br0 "table=1 priority=0 actions=flood"
860 *
861 * You can then dump the MAC learning table with:
862 *
863 * ovs-ofctl dump-flows br0 table=1
864 *
865 * Usage Advice
866 * ------------
867 *
868 * For best performance, segregate learned flows into a table that is not used
869 * for any other flows except possibly for a lowest-priority "catch-all" flow
870 * (a flow with no match criteria). If different learning actions specify
871 * different match criteria, use different tables for the learned flows.
872 *
873 * The meaning of 'hard_timeout' and 'idle_timeout' can be counterintuitive.
874 * These timeouts apply to the flow that is added, which means that a flow with
875 * an idle timeout will expire when no traffic has been sent *to* the learned
876 * address. This is not usually the intent in MAC learning; instead, we want
877 * the MAC learn entry to expire when no traffic has been sent *from* the
878 * learned address. Use a hard timeout for that.
879 */
880struct nx_action_learn {
881 ovs_be16 type; /* OFPAT_VENDOR. */
882 ovs_be16 len; /* At least 24. */
883 ovs_be32 vendor; /* NX_VENDOR_ID. */
884 ovs_be16 subtype; /* NXAST_LEARN. */
885 ovs_be16 idle_timeout; /* Idle time before discarding (seconds). */
886 ovs_be16 hard_timeout; /* Max time before discarding (seconds). */
887 ovs_be16 priority; /* Priority level of flow entry. */
888 ovs_be64 cookie; /* Cookie for new flow. */
889 ovs_be16 flags; /* Either 0 or OFPFF_SEND_FLOW_REM. */
890 uint8_t table_id; /* Table to insert flow entry. */
0e553d9c
BP
891 uint8_t pad; /* Must be zero. */
892 ovs_be16 fin_idle_timeout; /* Idle timeout after FIN, if nonzero. */
893 ovs_be16 fin_hard_timeout; /* Hard timeout after FIN, if nonzero. */
75a75043
BP
894 /* Followed by a sequence of flow_mod_spec elements, as described above,
895 * until the end of the action is reached. */
896};
897OFP_ASSERT(sizeof(struct nx_action_learn) == 32);
898
899#define NX_LEARN_N_BITS_MASK 0x3ff
900
901#define NX_LEARN_SRC_FIELD (0 << 13) /* Copy from field. */
902#define NX_LEARN_SRC_IMMEDIATE (1 << 13) /* Copy from immediate value. */
903#define NX_LEARN_SRC_MASK (1 << 13)
904
905#define NX_LEARN_DST_MATCH (0 << 11) /* Add match criterion. */
906#define NX_LEARN_DST_LOAD (1 << 11) /* Add NXAST_REG_LOAD action. */
907#define NX_LEARN_DST_OUTPUT (2 << 11) /* Add OFPAT_OUTPUT action. */
908#define NX_LEARN_DST_RESERVED (3 << 11) /* Not yet defined. */
909#define NX_LEARN_DST_MASK (3 << 11)
0e553d9c
BP
910
911/* Action structure for NXAST_FIN_TIMEOUT.
912 *
913 * This action changes the idle timeout or hard timeout, or both, of this
914 * OpenFlow rule when the rule matches a TCP packet with the FIN or RST flag.
915 * When such a packet is observed, the action reduces the rule's idle timeout
916 * to 'fin_idle_timeout' and its hard timeout to 'fin_hard_timeout'. This
917 * action has no effect on an existing timeout that is already shorter than the
918 * one that the action specifies. A 'fin_idle_timeout' or 'fin_hard_timeout'
919 * of zero has no effect on the respective timeout.
920 *
921 * 'fin_idle_timeout' and 'fin_hard_timeout' are measured in seconds.
922 * 'fin_hard_timeout' specifies time since the flow's creation, not since the
923 * receipt of the FIN or RST.
924 *
925 * This is useful for quickly discarding learned TCP flows that otherwise will
926 * take a long time to expire.
927 *
928 * This action is intended for use with an OpenFlow rule that matches only a
929 * single TCP flow. If the rule matches multiple TCP flows (e.g. it wildcards
930 * all TCP traffic, or all TCP traffic to a particular port), then any FIN or
931 * RST in any of those flows will cause the entire OpenFlow rule to expire
932 * early, which is not normally desirable.
933 */
934struct nx_action_fin_timeout {
935 ovs_be16 type; /* OFPAT_VENDOR. */
936 ovs_be16 len; /* 16. */
937 ovs_be32 vendor; /* NX_VENDOR_ID. */
938 ovs_be16 subtype; /* NXAST_FIN_TIMEOUT. */
939 ovs_be16 fin_idle_timeout; /* New idle timeout, if nonzero. */
940 ovs_be16 fin_hard_timeout; /* New hard timeout, if nonzero. */
941 ovs_be16 pad; /* Must be zero. */
942};
943OFP_ASSERT(sizeof(struct nx_action_fin_timeout) == 16);
75a75043 944\f
3b6a2571
EJ
945/* Action structure for NXAST_AUTOPATH.
946 *
947 * This action performs the following steps in sequence:
948 *
949 * 1. Hashes the flow using an implementation-defined hash function.
950 *
951 * The hashed fields' values are drawn from the current state of the
952 * flow, including all modifications that have been made by actions up to
953 * this point.
954 *
955 * 2. Selects an OpenFlow 'port'.
956 *
957 * 'port' is selected in an implementation-defined manner, taking into
958 * account 'id' and the hash value calculated in step 1.
959 *
960 * Generally a switch will have been configured with a set of ports that
961 * may be chosen given 'id'. The switch may take into account any number
962 * of factors when choosing 'port' from its configured set. Factors may
963 * include carrier, load, and the results of configuration protocols such
964 * as LACP.
965 *
966 * 3. Stores 'port' in dst[ofs:ofs+n_bits].
967 *
968 * The format and semantics of 'dst' and 'ofs_nbits' are similar to those
43edca57 969 * for the NXAST_REG_LOAD action.
3b6a2571
EJ
970 *
971 * The switch will reject actions in which ofs+n_bits is greater than the width
972 * of 'dst', with error type OFPET_BAD_ACTION, code OFPBAC_BAD_ARGUMENT.
973 */
974struct nx_action_autopath {
975 ovs_be16 type; /* OFPAT_VENDOR. */
2716c204 976 ovs_be16 len; /* Length is 24. */
3b6a2571 977 ovs_be32 vendor; /* NX_VENDOR_ID. */
eac7014f 978 ovs_be16 subtype; /* NXAST_AUTOPATH. */
3b6a2571
EJ
979
980 /* Where to store the result. */
981 ovs_be16 ofs_nbits; /* (ofs << 6) | (n_bits - 1). */
43edca57 982 ovs_be32 dst; /* Destination. */
3b6a2571
EJ
983
984 ovs_be32 id; /* Autopath ID. */
985 ovs_be32 pad;
986};
987OFP_ASSERT(sizeof(struct nx_action_autopath) == 24);
988\f
a368bb53 989/* Action structure for NXAST_BUNDLE and NXAST_BUNDLE_LOAD.
daff3353 990 *
a368bb53
EJ
991 * The bundle actions choose a slave from a supplied list of options.
992 * NXAST_BUNDLE outputs to its selection. NXAST_BUNDLE_LOAD writes its
993 * selection to a register.
daff3353
EJ
994 *
995 * The list of possible slaves follows the nx_action_bundle structure. The size
996 * of each slave is governed by its type as indicated by the 'slave_type'
997 * parameter. The list of slaves should be padded at its end with zeros to make
998 * the total length of the action a multiple of 8.
999 *
1000 * Switches infer from the 'slave_type' parameter the size of each slave. All
1001 * implementations must support the NXM_OF_IN_PORT 'slave_type' which indicates
1002 * that the slaves are OpenFlow port numbers with NXM_LENGTH(NXM_OF_IN_PORT) ==
1003 * 2 byte width. Switches should reject actions which indicate unknown or
1004 * unsupported slave types.
1005 *
1006 * Switches use a strategy dictated by the 'algorithm' parameter to choose a
1007 * slave. If the switch does not support the specified 'algorithm' parameter,
1008 * it should reject the action.
1009 *
439e4d8c
EJ
1010 * Several algorithms take into account liveness when selecting slaves. The
1011 * liveness of a slave is implementation defined (with one exception), but will
1012 * generally take into account things like its carrier status and the results
1013 * of any link monitoring protocols which happen to be running on it. In order
1014 * to give controllers a place-holder value, the OFPP_NONE port is always
1015 * considered live.
1016 *
daff3353
EJ
1017 * Some slave selection strategies require the use of a hash function, in which
1018 * case the 'fields' and 'basis' parameters should be populated. The 'fields'
1019 * parameter (one of NX_HASH_FIELDS_*) designates which parts of the flow to
1020 * hash. Refer to the definition of "enum nx_hash_fields" for details. The
1021 * 'basis' parameter is used as a universal hash parameter. Different values
1022 * of 'basis' yield different hash results.
1023 *
1024 * The 'zero' parameter at the end of the action structure is reserved for
1025 * future use. Switches are required to reject actions which have nonzero
a368bb53
EJ
1026 * bytes in the 'zero' field.
1027 *
1028 * NXAST_BUNDLE actions should have 'ofs_nbits' and 'dst' zeroed. Switches
1029 * should reject actions which have nonzero bytes in either of these fields.
1030 *
1031 * NXAST_BUNDLE_LOAD stores the OpenFlow port number of the selected slave in
1032 * dst[ofs:ofs+n_bits]. The format and semantics of 'dst' and 'ofs_nbits' are
1033 * similar to those for the NXAST_REG_LOAD action. */
daff3353
EJ
1034struct nx_action_bundle {
1035 ovs_be16 type; /* OFPAT_VENDOR. */
1036 ovs_be16 len; /* Length including slaves. */
1037 ovs_be32 vendor; /* NX_VENDOR_ID. */
799d2bfa 1038 ovs_be16 subtype; /* NXAST_BUNDLE or NXAST_BUNDLE_LOAD. */
daff3353
EJ
1039
1040 /* Slave choice algorithm to apply to hash value. */
1041 ovs_be16 algorithm; /* One of NX_BD_ALG_*. */
1042
1043 /* What fields to hash and how. */
735bc64a 1044 ovs_be16 fields; /* One of NX_HASH_FIELDS_*. */
daff3353
EJ
1045 ovs_be16 basis; /* Universal hash parameter. */
1046
1047 ovs_be32 slave_type; /* NXM_OF_IN_PORT. */
1048 ovs_be16 n_slaves; /* Number of slaves. */
1049
a368bb53
EJ
1050 ovs_be16 ofs_nbits; /* (ofs << 6) | (n_bits - 1). */
1051 ovs_be32 dst; /* Destination. */
1052
1053 uint8_t zero[4]; /* Reserved. Must be zero. */
daff3353
EJ
1054};
1055OFP_ASSERT(sizeof(struct nx_action_bundle) == 32);
1056
1057/* NXAST_BUNDLE: Bundle slave choice algorithm to apply.
1058 *
1059 * In the descriptions below, 'slaves' is the list of possible slaves in the
1060 * order they appear in the OpenFlow action. */
1061enum nx_bd_algorithm {
1062 /* Chooses the first live slave listed in the bundle.
1063 *
1064 * O(n_slaves) performance. */
1065 NX_BD_ALG_ACTIVE_BACKUP,
1066
1067 /* for i in [0,n_slaves):
1068 * weights[i] = hash(flow, i)
1069 * slave = { slaves[i] such that weights[i] >= weights[j] for all j != i }
1070 *
1071 * Redistributes 1/n_slaves of traffic when a slave's liveness changes.
1072 * O(n_slaves) performance.
1073 *
1074 * Uses the 'fields' and 'basis' parameters. */
1075 NX_BD_ALG_HRW /* Highest Random Weight. */
1076};
c2d967a5
MM
1077
1078\f
1079/* Action structure for NXAST_DEC_TTL_CNT_IDS.
1080 *
1081 * If the packet is not IPv4 or IPv6, does nothing. For IPv4 or IPv6, if the
1082 * TTL or hop limit is at least 2, decrements it by 1. Otherwise, if TTL or
1083 * hop limit is 0 or 1, sends a packet-in to the controllers with each of the
1084 * 'n_controllers' controller IDs specified in 'cnt_ids'.
1085 *
1086 * (This differs from NXAST_DEC_TTL in that for NXAST_DEC_TTL the packet-in is
1087 * sent only to controllers with id 0.)
1088 */
1089struct nx_action_cnt_ids {
1090 ovs_be16 type; /* OFPAT_VENDOR. */
1091 ovs_be16 len; /* Length including slaves. */
1092 ovs_be32 vendor; /* NX_VENDOR_ID. */
1093 ovs_be16 subtype; /* NXAST_DEC_TTL_CNT_IDS. */
1094
1095 ovs_be16 n_controllers; /* Number of controllers. */
1096 uint8_t zeros[4]; /* Must be zero. */
1097
1098 /* Followed by 1 or more controller ids.
1099 *
1100 * uint16_t cnt_ids[]; // Controller ids.
1101 * uint8_t pad[]; // Must be 0 to 8-byte align cnt_ids[].
1102 */
1103};
1104OFP_ASSERT(sizeof(struct nx_action_cnt_ids) == 16);
1105
daff3353 1106\f
f694937d
EJ
1107/* Action structure for NXAST_OUTPUT_REG.
1108 *
1109 * Outputs to the OpenFlow port number written to src[ofs:ofs+nbits].
1110 *
1111 * The format and semantics of 'src' and 'ofs_nbits' are similar to those for
1112 * the NXAST_REG_LOAD action.
1113 *
1114 * The acceptable nxm_header values for 'src' are the same as the acceptable
1115 * nxm_header values for the 'src' field of NXAST_REG_MOVE.
1116 *
1117 * The 'max_len' field indicates the number of bytes to send when the chosen
1118 * port is OFPP_CONTROLLER. Its semantics are equivalent to the 'max_len'
1119 * field of OFPAT_OUTPUT.
1120 *
1121 * The 'zero' field is required to be zeroed for forward compatibility. */
1122struct nx_action_output_reg {
1123 ovs_be16 type; /* OFPAT_VENDOR. */
1124 ovs_be16 len; /* 24. */
1125 ovs_be32 vendor; /* NX_VENDOR_ID. */
1126 ovs_be16 subtype; /* NXAST_OUTPUT_REG. */
1127
1128 ovs_be16 ofs_nbits; /* (ofs << 6) | (n_bits - 1). */
1129 ovs_be32 src; /* Source. */
1130
1131 ovs_be16 max_len; /* Max length to send to controller. */
1132
1133 uint8_t zero[6]; /* Reserved, must be zero. */
1134};
1135OFP_ASSERT(sizeof(struct nx_action_output_reg) == 24);
1136\f
848e8809
EJ
1137/* NXAST_EXIT
1138 *
1139 * Discontinues action processing.
1140 *
1141 * The NXAST_EXIT action causes the switch to immediately halt processing
1142 * actions for the flow. Any actions which have already been processed are
1143 * executed by the switch. However, any further actions, including those which
1144 * may be in different tables, or different levels of the NXAST_RESUBMIT
1145 * hierarchy, will be ignored.
1146 *
1147 * Uses the nx_action_header structure. */
1148\f
09246b99
BP
1149/* Flexible flow specifications (aka NXM = Nicira Extended Match).
1150 *
eec25dc1 1151 * OpenFlow 1.0 has "struct ofp10_match" for specifying flow matches. This
09246b99
BP
1152 * structure is fixed-length and hence difficult to extend. This section
1153 * describes a more flexible, variable-length flow match, called "nx_match" for
1154 * short, that is also supported by Open vSwitch. This section also defines a
eec25dc1 1155 * replacement for each OpenFlow message that includes struct ofp10_match.
09246b99
BP
1156 *
1157 *
1158 * Format
1159 * ======
1160 *
1161 * An nx_match is a sequence of zero or more "nxm_entry"s, which are
1162 * type-length-value (TLV) entries, each 5 to 259 (inclusive) bytes long.
1163 * "nxm_entry"s are not aligned on or padded to any multibyte boundary. The
1164 * first 4 bytes of an nxm_entry are its "header", followed by the entry's
1165 * "body".
1166 *
1167 * An nxm_entry's header is interpreted as a 32-bit word in network byte order:
1168 *
1169 * |<-------------------- nxm_type ------------------>|
1170 * | |
1171 * |31 16 15 9| 8 7 0
1172 * +----------------------------------+---------------+--+------------------+
1173 * | nxm_vendor | nxm_field |hm| nxm_length |
1174 * +----------------------------------+---------------+--+------------------+
1175 *
1176 * The most-significant 23 bits of the header are collectively "nxm_type".
1177 * Bits 16...31 are "nxm_vendor", one of the NXM_VENDOR_* values below. Bits
1178 * 9...15 are "nxm_field", which is a vendor-specific value. nxm_type normally
1179 * designates a protocol header, such as the Ethernet type, but it can also
1180 * refer to packet metadata, such as the switch port on which a packet arrived.
1181 *
1182 * Bit 8 is "nxm_hasmask" (labeled "hm" above for space reasons). The meaning
1183 * of this bit is explained later.
1184 *
1185 * The least-significant 8 bits are "nxm_length", a positive integer. The
1186 * length of the nxm_entry, including the header, is exactly 4 + nxm_length
1187 * bytes.
1188 *
1189 * For a given nxm_vendor, nxm_field, and nxm_hasmask value, nxm_length is a
1190 * constant. It is included only to allow software to minimally parse
1191 * "nxm_entry"s of unknown types. (Similarly, for a given nxm_vendor,
1192 * nxm_field, and nxm_length, nxm_hasmask is a constant.)
1193 *
1194 *
1195 * Semantics
1196 * =========
1197 *
1198 * A zero-length nx_match (one with no "nxm_entry"s) matches every packet.
1199 *
1200 * An nxm_entry places a constraint on the packets matched by the nx_match:
1201 *
1202 * - If nxm_hasmask is 0, the nxm_entry's body contains a value for the
1203 * field, called "nxm_value". The nx_match matches only packets in which
1204 * the field equals nxm_value.
1205 *
1206 * - If nxm_hasmask is 1, then the nxm_entry's body contains a value for the
1207 * field (nxm_value), followed by a bitmask of the same length as the
1208 * value, called "nxm_mask". For each 1-bit in position J in nxm_mask, the
1209 * nx_match matches only packets for which bit J in the given field's value
1210 * matches bit J in nxm_value. A 0-bit in nxm_mask causes the
1211 * corresponding bits in nxm_value and the field's value to be ignored.
1212 * (The sense of the nxm_mask bits is the opposite of that used by the
eec25dc1 1213 * "wildcards" member of struct ofp10_match.)
09246b99
BP
1214 *
1215 * When nxm_hasmask is 1, nxm_length is always even.
1216 *
1217 * An all-zero-bits nxm_mask is equivalent to omitting the nxm_entry
1218 * entirely. An all-one-bits nxm_mask is equivalent to specifying 0 for
1219 * nxm_hasmask.
1220 *
1221 * When there are multiple "nxm_entry"s, all of the constraints must be met.
1222 *
1223 *
1224 * Mask Restrictions
1225 * =================
1226 *
1227 * Masks may be restricted:
1228 *
1229 * - Some nxm_types may not support masked wildcards, that is, nxm_hasmask
1230 * must always be 0 when these fields are specified. For example, the
1231 * field that identifies the port on which a packet was received may not be
1232 * masked.
1233 *
1234 * - Some nxm_types that do support masked wildcards may only support certain
1235 * nxm_mask patterns. For example, fields that have IPv4 address values
1236 * may be restricted to CIDR masks.
1237 *
1238 * These restrictions should be noted in specifications for individual fields.
1239 * A switch may accept an nxm_hasmask or nxm_mask value that the specification
1240 * disallows, if the switch correctly implements support for that nxm_hasmask
1241 * or nxm_mask value. A switch must reject an attempt to set up a flow that
1242 * contains a nxm_hasmask or nxm_mask value that it does not support.
1243 *
1244 *
1245 * Prerequisite Restrictions
1246 * =========================
1247 *
1248 * The presence of an nxm_entry with a given nxm_type may be restricted based
1249 * on the presence of or values of other "nxm_entry"s. For example:
1250 *
1251 * - An nxm_entry for nxm_type=NXM_OF_IP_TOS is allowed only if it is
1252 * preceded by another entry with nxm_type=NXM_OF_ETH_TYPE, nxm_hasmask=0,
1253 * and nxm_value=0x0800. That is, matching on the IP source address is
1254 * allowed only if the Ethernet type is explicitly set to IP.
1255 *
e51df1a0
BP
1256 * - An nxm_entry for nxm_type=NXM_OF_TCP_SRC is allowed only if it is
1257 * preceded by an entry with nxm_type=NXM_OF_ETH_TYPE, nxm_hasmask=0, and
1258 * nxm_value either 0x0800 or 0x86dd, and another with
1259 * nxm_type=NXM_OF_IP_PROTO, nxm_hasmask=0, nxm_value=6, in that order.
1260 * That is, matching on the TCP source port is allowed only if the Ethernet
1261 * type is IP or IPv6 and the IP protocol is TCP.
09246b99
BP
1262 *
1263 * These restrictions should be noted in specifications for individual fields.
1264 * A switch may implement relaxed versions of these restrictions. A switch
1265 * must reject an attempt to set up a flow that violates its restrictions.
1266 *
1267 *
1268 * Ordering Restrictions
1269 * =====================
1270 *
1271 * An nxm_entry that has prerequisite restrictions must appear after the
1272 * "nxm_entry"s for its prerequisites. Ordering of "nxm_entry"s within an
1273 * nx_match is not otherwise constrained.
1274 *
1275 * Any given nxm_type may appear in an nx_match at most once.
1276 *
1277 *
1278 * nxm_entry Examples
1279 * ==================
1280 *
1281 * These examples show the format of a single nxm_entry with particular
1282 * nxm_hasmask and nxm_length values. The diagrams are labeled with field
1283 * numbers and byte indexes.
1284 *
1285 *
be86ea7a 1286 * 8-bit nxm_value, nxm_hasmask=1, nxm_length=2:
09246b99
BP
1287 *
1288 * 0 3 4 5
1289 * +------------+---+---+
1290 * | header | v | m |
1291 * +------------+---+---+
1292 *
1293 *
1294 * 16-bit nxm_value, nxm_hasmask=0, nxm_length=2:
1295 *
1296 * 0 3 4 5
1297 * +------------+------+
1298 * | header | value|
1299 * +------------+------+
1300 *
1301 *
1302 * 32-bit nxm_value, nxm_hasmask=0, nxm_length=4:
1303 *
1304 * 0 3 4 7
1305 * +------------+-------------+
1306 * | header | nxm_value |
1307 * +------------+-------------+
1308 *
1309 *
1310 * 48-bit nxm_value, nxm_hasmask=0, nxm_length=6:
1311 *
1312 * 0 3 4 9
1313 * +------------+------------------+
1314 * | header | nxm_value |
1315 * +------------+------------------+
1316 *
1317 *
1318 * 48-bit nxm_value, nxm_hasmask=1, nxm_length=12:
1319 *
1320 * 0 3 4 9 10 15
1321 * +------------+------------------+------------------+
1322 * | header | nxm_value | nxm_mask |
1323 * +------------+------------------+------------------+
1324 *
1325 *
1326 * Error Reporting
1327 * ===============
1328 *
1329 * A switch should report an error in an nx_match using error type
1330 * OFPET_BAD_REQUEST and one of the NXBRC_NXM_* codes. Ideally the switch
1331 * should report a specific error code, if one is assigned for the particular
1332 * problem, but NXBRC_NXM_INVALID is also available to report a generic
1333 * nx_match error.
1334 */
1335
1336#define NXM_HEADER__(VENDOR, FIELD, HASMASK, LENGTH) \
1337 (((VENDOR) << 16) | ((FIELD) << 9) | ((HASMASK) << 8) | (LENGTH))
1338#define NXM_HEADER(VENDOR, FIELD, LENGTH) \
1339 NXM_HEADER__(VENDOR, FIELD, 0, LENGTH)
1340#define NXM_HEADER_W(VENDOR, FIELD, LENGTH) \
1341 NXM_HEADER__(VENDOR, FIELD, 1, (LENGTH) * 2)
1342#define NXM_VENDOR(HEADER) ((HEADER) >> 16)
1343#define NXM_FIELD(HEADER) (((HEADER) >> 9) & 0x7f)
1344#define NXM_TYPE(HEADER) (((HEADER) >> 9) & 0x7fffff)
1345#define NXM_HASMASK(HEADER) (((HEADER) >> 8) & 1)
1346#define NXM_LENGTH(HEADER) ((HEADER) & 0xff)
1347
1348#define NXM_MAKE_WILD_HEADER(HEADER) \
1349 NXM_HEADER_W(NXM_VENDOR(HEADER), NXM_FIELD(HEADER), NXM_LENGTH(HEADER))
1350
1351/* ## ------------------------------- ## */
1352/* ## OpenFlow 1.0-compatible fields. ## */
1353/* ## ------------------------------- ## */
1354
1355/* Physical or virtual port on which the packet was received.
1356 *
1357 * Prereqs: None.
1358 *
1359 * Format: 16-bit integer in network byte order.
1360 *
1361 * Masking: Not maskable. */
1362#define NXM_OF_IN_PORT NXM_HEADER (0x0000, 0, 2)
1363
1364/* Source or destination address in Ethernet header.
1365 *
1366 * Prereqs: None.
1367 *
1368 * Format: 48-bit Ethernet MAC address.
1369 *
73c0ce34
JS
1370 * Masking: Fully maskable, in versions 1.8 and later. Earlier versions only
1371 * supported the following masks for NXM_OF_ETH_DST_W: 00:00:00:00:00:00,
1372 * fe:ff:ff:ff:ff:ff, 01:00:00:00:00:00, ff:ff:ff:ff:ff:ff. */
09246b99 1373#define NXM_OF_ETH_DST NXM_HEADER (0x0000, 1, 6)
1e37a2d7 1374#define NXM_OF_ETH_DST_W NXM_HEADER_W(0x0000, 1, 6)
09246b99 1375#define NXM_OF_ETH_SRC NXM_HEADER (0x0000, 2, 6)
73c0ce34 1376#define NXM_OF_ETH_SRC_W NXM_HEADER_W(0x0000, 2, 6)
09246b99
BP
1377
1378/* Packet's Ethernet type.
1379 *
1380 * For an Ethernet II packet this is taken from the Ethernet header. For an
1381 * 802.2 LLC+SNAP header with OUI 00-00-00 this is taken from the SNAP header.
1382 * A packet that has neither format has value 0x05ff
1383 * (OFP_DL_TYPE_NOT_ETH_TYPE).
1384 *
1385 * For a packet with an 802.1Q header, this is the type of the encapsulated
1386 * frame.
1387 *
1388 * Prereqs: None.
1389 *
1390 * Format: 16-bit integer in network byte order.
1391 *
1392 * Masking: Not maskable. */
1393#define NXM_OF_ETH_TYPE NXM_HEADER (0x0000, 3, 2)
1394
1395/* 802.1Q TCI.
1396 *
1397 * For a packet with an 802.1Q header, this is the Tag Control Information
1398 * (TCI) field, with the CFI bit forced to 1. For a packet with no 802.1Q
1399 * header, this has value 0.
1400 *
1401 * Prereqs: None.
1402 *
1403 * Format: 16-bit integer in network byte order.
1404 *
1405 * Masking: Arbitrary masks.
1406 *
1407 * This field can be used in various ways:
1408 *
1409 * - If it is not constrained at all, the nx_match matches packets without
1410 * an 802.1Q header or with an 802.1Q header that has any TCI value.
1411 *
1412 * - Testing for an exact match with 0 matches only packets without an
1413 * 802.1Q header.
1414 *
1415 * - Testing for an exact match with a TCI value with CFI=1 matches packets
1416 * that have an 802.1Q header with a specified VID and PCP.
1417 *
1418 * - Testing for an exact match with a nonzero TCI value with CFI=0 does
1419 * not make sense. The switch may reject this combination.
1420 *
1421 * - Testing with a specific VID and CFI=1, with nxm_mask=0x1fff, matches
1422 * packets that have an 802.1Q header with that VID (and any PCP).
1423 *
1424 * - Testing with a specific PCP and CFI=1, with nxm_mask=0xf000, matches
1425 * packets that have an 802.1Q header with that PCP (and any VID).
1426 *
8ca67760 1427 * - Testing with nxm_value=0, nxm_mask=0x0fff matches packets with no 802.1Q
09246b99 1428 * header or with an 802.1Q header with a VID of 0.
8ca67760
BP
1429 *
1430 * - Testing with nxm_value=0, nxm_mask=0xe000 matches packets with no 802.1Q
1431 * header or with an 802.1Q header with a PCP of 0.
1432 *
1433 * - Testing with nxm_value=0, nxm_mask=0xefff matches packets with no 802.1Q
1434 * header or with an 802.1Q header with both VID and PCP of 0.
09246b99
BP
1435 */
1436#define NXM_OF_VLAN_TCI NXM_HEADER (0x0000, 4, 2)
1437#define NXM_OF_VLAN_TCI_W NXM_HEADER_W(0x0000, 4, 2)
1438
1439/* The "type of service" byte of the IP header, with the ECN bits forced to 0.
1440 *
d31f1109 1441 * Prereqs: NXM_OF_ETH_TYPE must be either 0x0800 or 0x86dd.
09246b99
BP
1442 *
1443 * Format: 8-bit integer with 2 least-significant bits forced to 0.
1444 *
1445 * Masking: Not maskable. */
1446#define NXM_OF_IP_TOS NXM_HEADER (0x0000, 5, 1)
1447
1448/* The "protocol" byte in the IP header.
1449 *
d31f1109 1450 * Prereqs: NXM_OF_ETH_TYPE must be either 0x0800 or 0x86dd.
09246b99
BP
1451 *
1452 * Format: 8-bit integer.
1453 *
1454 * Masking: Not maskable. */
1455#define NXM_OF_IP_PROTO NXM_HEADER (0x0000, 6, 1)
1456
1457/* The source or destination address in the IP header.
1458 *
1459 * Prereqs: NXM_OF_ETH_TYPE must match 0x0800 exactly.
1460 *
1461 * Format: 32-bit integer in network byte order.
1462 *
c08201d6
BP
1463 * Masking: Fully maskable, in Open vSwitch 1.8 and later. In earlier
1464 * versions, only CIDR masks are allowed, that is, masks that consist of N
09246b99
BP
1465 * high-order bits set to 1 and the other 32-N bits set to 0. */
1466#define NXM_OF_IP_SRC NXM_HEADER (0x0000, 7, 4)
1467#define NXM_OF_IP_SRC_W NXM_HEADER_W(0x0000, 7, 4)
1468#define NXM_OF_IP_DST NXM_HEADER (0x0000, 8, 4)
1469#define NXM_OF_IP_DST_W NXM_HEADER_W(0x0000, 8, 4)
1470
1471/* The source or destination port in the TCP header.
1472 *
1473 * Prereqs:
d31f1109 1474 * NXM_OF_ETH_TYPE must be either 0x0800 or 0x86dd.
09246b99
BP
1475 * NXM_OF_IP_PROTO must match 6 exactly.
1476 *
1477 * Format: 16-bit integer in network byte order.
1478 *
73f33563
BP
1479 * Masking: Fully maskable, in Open vSwitch 1.6 and later. Not maskable, in
1480 * earlier versions. */
09246b99 1481#define NXM_OF_TCP_SRC NXM_HEADER (0x0000, 9, 2)
73f33563 1482#define NXM_OF_TCP_SRC_W NXM_HEADER_W(0x0000, 9, 2)
09246b99 1483#define NXM_OF_TCP_DST NXM_HEADER (0x0000, 10, 2)
73f33563 1484#define NXM_OF_TCP_DST_W NXM_HEADER_W(0x0000, 10, 2)
09246b99
BP
1485
1486/* The source or destination port in the UDP header.
1487 *
1488 * Prereqs:
d31f1109 1489 * NXM_OF_ETH_TYPE must match either 0x0800 or 0x86dd.
09246b99
BP
1490 * NXM_OF_IP_PROTO must match 17 exactly.
1491 *
1492 * Format: 16-bit integer in network byte order.
1493 *
73f33563
BP
1494 * Masking: Fully maskable, in Open vSwitch 1.6 and later. Not maskable, in
1495 * earlier versions. */
09246b99 1496#define NXM_OF_UDP_SRC NXM_HEADER (0x0000, 11, 2)
73f33563 1497#define NXM_OF_UDP_SRC_W NXM_HEADER_W(0x0000, 11, 2)
09246b99 1498#define NXM_OF_UDP_DST NXM_HEADER (0x0000, 12, 2)
73f33563 1499#define NXM_OF_UDP_DST_W NXM_HEADER_W(0x0000, 12, 2)
09246b99
BP
1500
1501/* The type or code in the ICMP header.
1502 *
1503 * Prereqs:
1504 * NXM_OF_ETH_TYPE must match 0x0800 exactly.
1505 * NXM_OF_IP_PROTO must match 1 exactly.
1506 *
1507 * Format: 8-bit integer.
1508 *
1509 * Masking: Not maskable. */
1510#define NXM_OF_ICMP_TYPE NXM_HEADER (0x0000, 13, 1)
1511#define NXM_OF_ICMP_CODE NXM_HEADER (0x0000, 14, 1)
1512
1513/* ARP opcode.
1514 *
1515 * For an Ethernet+IP ARP packet, the opcode in the ARP header. Always 0
1516 * otherwise. Only ARP opcodes between 1 and 255 should be specified for
1517 * matching.
1518 *
8087f5ff 1519 * Prereqs: NXM_OF_ETH_TYPE must match either 0x0806 or 0x8035.
09246b99
BP
1520 *
1521 * Format: 16-bit integer in network byte order.
1522 *
1523 * Masking: Not maskable. */
1524#define NXM_OF_ARP_OP NXM_HEADER (0x0000, 15, 2)
1525
1526/* For an Ethernet+IP ARP packet, the source or target protocol address
1527 * in the ARP header. Always 0 otherwise.
1528 *
8087f5ff 1529 * Prereqs: NXM_OF_ETH_TYPE must match either 0x0806 or 0x8035.
09246b99
BP
1530 *
1531 * Format: 32-bit integer in network byte order.
1532 *
c08201d6
BP
1533 * Masking: Fully maskable, in Open vSwitch 1.8 and later. In earlier
1534 * versions, only CIDR masks are allowed, that is, masks that consist of N
09246b99
BP
1535 * high-order bits set to 1 and the other 32-N bits set to 0. */
1536#define NXM_OF_ARP_SPA NXM_HEADER (0x0000, 16, 4)
1537#define NXM_OF_ARP_SPA_W NXM_HEADER_W(0x0000, 16, 4)
1538#define NXM_OF_ARP_TPA NXM_HEADER (0x0000, 17, 4)
1539#define NXM_OF_ARP_TPA_W NXM_HEADER_W(0x0000, 17, 4)
1540
1541/* ## ------------------------ ## */
1542/* ## Nicira match extensions. ## */
1543/* ## ------------------------ ## */
1544
b6c9e612
BP
1545/* Metadata registers.
1546 *
1547 * Registers initially have value 0. Actions allow register values to be
1548 * manipulated.
1549 *
1550 * Prereqs: None.
1551 *
1552 * Format: Array of 32-bit integer registers. Space is reserved for up to
1553 * NXM_NX_MAX_REGS registers, but switches may implement fewer.
1554 *
1555 * Masking: Arbitrary masks. */
1556#define NXM_NX_MAX_REGS 16
1557#define NXM_NX_REG(IDX) NXM_HEADER (0x0001, IDX, 4)
1558#define NXM_NX_REG_W(IDX) NXM_HEADER_W(0x0001, IDX, 4)
1559#define NXM_NX_REG_IDX(HEADER) NXM_FIELD(HEADER)
19e7439c
BP
1560#define NXM_IS_NX_REG(HEADER) (!((((HEADER) ^ NXM_NX_REG0)) & 0xffffe1ff))
1561#define NXM_IS_NX_REG_W(HEADER) (!((((HEADER) ^ NXM_NX_REG0_W)) & 0xffffe1ff))
b6c9e612
BP
1562#define NXM_NX_REG0 NXM_HEADER (0x0001, 0, 4)
1563#define NXM_NX_REG0_W NXM_HEADER_W(0x0001, 0, 4)
1564#define NXM_NX_REG1 NXM_HEADER (0x0001, 1, 4)
1565#define NXM_NX_REG1_W NXM_HEADER_W(0x0001, 1, 4)
1566#define NXM_NX_REG2 NXM_HEADER (0x0001, 2, 4)
1567#define NXM_NX_REG2_W NXM_HEADER_W(0x0001, 2, 4)
1568#define NXM_NX_REG3 NXM_HEADER (0x0001, 3, 4)
1569#define NXM_NX_REG3_W NXM_HEADER_W(0x0001, 3, 4)
d2c0fed9
JP
1570#define NXM_NX_REG4 NXM_HEADER (0x0001, 4, 4)
1571#define NXM_NX_REG4_W NXM_HEADER_W(0x0001, 4, 4)
e9358af6
EJ
1572#define NXM_NX_REG5 NXM_HEADER (0x0001, 5, 4)
1573#define NXM_NX_REG5_W NXM_HEADER_W(0x0001, 5, 4)
1574#define NXM_NX_REG6 NXM_HEADER (0x0001, 6, 4)
1575#define NXM_NX_REG6_W NXM_HEADER_W(0x0001, 6, 4)
1576#define NXM_NX_REG7 NXM_HEADER (0x0001, 7, 4)
1577#define NXM_NX_REG7_W NXM_HEADER_W(0x0001, 7, 4)
b6c9e612 1578
09246b99
BP
1579/* Tunnel ID.
1580 *
1581 * For a packet received via GRE tunnel including a (32-bit) key, the key is
1582 * stored in the low 32-bits and the high bits are zeroed. For other packets,
1583 * the value is 0.
1584 *
1585 * Prereqs: None.
1586 *
1587 * Format: 64-bit integer in network byte order.
1588 *
1589 * Masking: Arbitrary masks. */
1590#define NXM_NX_TUN_ID NXM_HEADER (0x0001, 16, 8)
1591#define NXM_NX_TUN_ID_W NXM_HEADER_W(0x0001, 16, 8)
1592
bad68a99
JP
1593/* For an Ethernet+IP ARP packet, the source or target hardware address
1594 * in the ARP header. Always 0 otherwise.
1595 *
8087f5ff 1596 * Prereqs: NXM_OF_ETH_TYPE must match either 0x0806 or 0x8035.
bad68a99
JP
1597 *
1598 * Format: 48-bit Ethernet MAC address.
1599 *
1600 * Masking: Not maskable. */
1601#define NXM_NX_ARP_SHA NXM_HEADER (0x0001, 17, 6)
1602#define NXM_NX_ARP_THA NXM_HEADER (0x0001, 18, 6)
1603
d31f1109
JP
1604/* The source or destination address in the IPv6 header.
1605 *
1606 * Prereqs: NXM_OF_ETH_TYPE must match 0x86dd exactly.
1607 *
1608 * Format: 128-bit IPv6 address.
1609 *
ff0b06ee
BP
1610 * Masking: Fully maskable, in Open vSwitch 1.8 and later. In previous
1611 * versions, only CIDR masks are allowed, that is, masks that consist of N
d31f1109
JP
1612 * high-order bits set to 1 and the other 128-N bits set to 0. */
1613#define NXM_NX_IPV6_SRC NXM_HEADER (0x0001, 19, 16)
1614#define NXM_NX_IPV6_SRC_W NXM_HEADER_W(0x0001, 19, 16)
1615#define NXM_NX_IPV6_DST NXM_HEADER (0x0001, 20, 16)
1616#define NXM_NX_IPV6_DST_W NXM_HEADER_W(0x0001, 20, 16)
1617
1618/* The type or code in the ICMPv6 header.
1619 *
1620 * Prereqs:
1621 * NXM_OF_ETH_TYPE must match 0x86dd exactly.
1622 * NXM_OF_IP_PROTO must match 58 exactly.
1623 *
1624 * Format: 8-bit integer.
1625 *
1626 * Masking: Not maskable. */
1627#define NXM_NX_ICMPV6_TYPE NXM_HEADER (0x0001, 21, 1)
1628#define NXM_NX_ICMPV6_CODE NXM_HEADER (0x0001, 22, 1)
1629
685a51a5
JP
1630/* The target address in an IPv6 Neighbor Discovery message.
1631 *
1632 * Prereqs:
1633 * NXM_OF_ETH_TYPE must match 0x86dd exactly.
1634 * NXM_OF_IP_PROTO must match 58 exactly.
1635 * NXM_OF_ICMPV6_TYPE must be either 135 or 136.
1636 *
1637 * Format: 128-bit IPv6 address.
1638 *
ff0b06ee
BP
1639 * Masking: Fully maskable, in Open vSwitch 1.8 and later. In previous
1640 * versions, only CIDR masks are allowed, that is, masks that consist of N
47284b1f
AA
1641 * high-order bits set to 1 and the other 128-N bits set to 0. */
1642#define NXM_NX_ND_TARGET NXM_HEADER (0x0001, 23, 16)
1643#define NXM_NX_ND_TARGET_W NXM_HEADER_W (0x0001, 23, 16)
685a51a5
JP
1644
1645/* The source link-layer address option in an IPv6 Neighbor Discovery
1646 * message.
1647 *
1648 * Prereqs:
1649 * NXM_OF_ETH_TYPE must match 0x86dd exactly.
1650 * NXM_OF_IP_PROTO must match 58 exactly.
1651 * NXM_OF_ICMPV6_TYPE must be exactly 135.
1652 *
1653 * Format: 48-bit Ethernet MAC address.
1654 *
1655 * Masking: Not maskable. */
1656#define NXM_NX_ND_SLL NXM_HEADER (0x0001, 24, 6)
1657
1658/* The target link-layer address option in an IPv6 Neighbor Discovery
1659 * message.
1660 *
1661 * Prereqs:
1662 * NXM_OF_ETH_TYPE must match 0x86dd exactly.
1663 * NXM_OF_IP_PROTO must match 58 exactly.
1664 * NXM_OF_ICMPV6_TYPE must be exactly 136.
1665 *
1666 * Format: 48-bit Ethernet MAC address.
1667 *
1668 * Masking: Not maskable. */
1669#define NXM_NX_ND_TLL NXM_HEADER (0x0001, 25, 6)
1670
7257b535
BP
1671/* IP fragment information.
1672 *
1673 * Prereqs:
1674 * NXM_OF_ETH_TYPE must be either 0x0800 or 0x86dd.
1675 *
1676 * Format: 8-bit value with one of the values 0, 1, or 3, as described below.
1677 *
1678 * Masking: Fully maskable.
1679 *
1680 * This field has three possible values:
1681 *
1682 * - A packet that is not an IP fragment has value 0.
1683 *
1684 * - A packet that is an IP fragment with offset 0 (the first fragment) has
1685 * bit 0 set and thus value 1.
1686 *
1687 * - A packet that is an IP fragment with nonzero offset has bits 0 and 1 set
1688 * and thus value 3.
1689 *
1690 * NX_IP_FRAG_ANY and NX_IP_FRAG_LATER are declared to symbolically represent
1691 * the meanings of bits 0 and 1.
1692 *
1693 * The switch may reject matches against values that can never appear.
1694 *
1695 * It is important to understand how this field interacts with the OpenFlow IP
1696 * fragment handling mode:
1697 *
1698 * - In OFPC_FRAG_DROP mode, the OpenFlow switch drops all IP fragments
1699 * before they reach the flow table, so every packet that is available for
1700 * matching will have value 0 in this field.
1701 *
1702 * - Open vSwitch does not implement OFPC_FRAG_REASM mode, but if it did then
1703 * IP fragments would be reassembled before they reached the flow table and
1704 * again every packet available for matching would always have value 0.
1705 *
1706 * - In OFPC_FRAG_NORMAL mode, all three values are possible, but OpenFlow
1707 * 1.0 says that fragments' transport ports are always 0, even for the
1708 * first fragment, so this does not provide much extra information.
1709 *
1710 * - In OFPC_FRAG_NX_MATCH mode, all three values are possible. For
1711 * fragments with offset 0, Open vSwitch makes L4 header information
1712 * available.
1713 */
1714#define NXM_NX_IP_FRAG NXM_HEADER (0x0001, 26, 1)
1715#define NXM_NX_IP_FRAG_W NXM_HEADER_W(0x0001, 26, 1)
1716
1717/* Bits in the value of NXM_NX_IP_FRAG. */
1718#define NX_IP_FRAG_ANY (1 << 0) /* Is this a fragment? */
1719#define NX_IP_FRAG_LATER (1 << 1) /* Is this a fragment with nonzero offset? */
d31f1109 1720
fa8223b7
JP
1721/* The flow label in the IPv6 header.
1722 *
1723 * Prereqs: NXM_OF_ETH_TYPE must match 0x86dd exactly.
1724 *
1725 * Format: 20-bit IPv6 flow label in least-significant bits.
1726 *
1727 * Masking: Not maskable. */
1728#define NXM_NX_IPV6_LABEL NXM_HEADER (0x0001, 27, 4)
1729
530180fd
JP
1730/* The ECN of the IP header.
1731 *
1732 * Prereqs: NXM_OF_ETH_TYPE must be either 0x0800 or 0x86dd.
1733 *
1734 * Format: ECN in the low-order 2 bits.
1735 *
1736 * Masking: Not maskable. */
1737#define NXM_NX_IP_ECN NXM_HEADER (0x0001, 28, 1)
1738
a61680c6
JP
1739/* The time-to-live/hop limit of the IP header.
1740 *
1741 * Prereqs: NXM_OF_ETH_TYPE must be either 0x0800 or 0x86dd.
1742 *
1743 * Format: 8-bit integer.
1744 *
1745 * Masking: Not maskable. */
1746#define NXM_NX_IP_TTL NXM_HEADER (0x0001, 29, 1)
1747
e729e793
JP
1748/* Flow cookie.
1749 *
1750 * This may be used to gain the OpenFlow 1.1-like ability to restrict
1751 * certain NXM-based Flow Mod and Flow Stats Request messages to flows
1752 * with specific cookies. See the "nx_flow_mod" and "nx_flow_stats_request"
1753 * structure definitions for more details. This match is otherwise not
1754 * allowed.
1755 *
1756 * Prereqs: None.
1757 *
1758 * Format: 64-bit integer in network byte order.
1759 *
1760 * Masking: Arbitrary masks. */
1761#define NXM_NX_COOKIE NXM_HEADER (0x0001, 30, 8)
1762#define NXM_NX_COOKIE_W NXM_HEADER_W(0x0001, 30, 8)
1763
09246b99
BP
1764/* ## --------------------- ## */
1765/* ## Requests and replies. ## */
1766/* ## --------------------- ## */
1767
492f7572 1768enum nx_flow_format {
09246b99 1769 NXFF_OPENFLOW10 = 0, /* Standard OpenFlow 1.0 compatible. */
310f3699 1770 NXFF_NXM = 2 /* Nicira extended match. */
09246b99
BP
1771};
1772
1773/* NXT_SET_FLOW_FORMAT request. */
73dbf4ab 1774struct nx_set_flow_format {
09246b99
BP
1775 ovs_be32 format; /* One of NXFF_*. */
1776};
982697a4 1777OFP_ASSERT(sizeof(struct nx_set_flow_format) == 4);
09246b99 1778
e729e793
JP
1779/* NXT_FLOW_MOD (analogous to OFPT_FLOW_MOD).
1780 *
1781 * It is possible to limit flow deletions and modifications to certain
623e1caf
JP
1782 * cookies by using the NXM_NX_COOKIE(_W) matches. The "cookie" field
1783 * is used only to add or modify flow cookies.
e729e793 1784 */
09246b99 1785struct nx_flow_mod {
09246b99
BP
1786 ovs_be64 cookie; /* Opaque controller-issued identifier. */
1787 ovs_be16 command; /* One of OFPFC_*. */
1788 ovs_be16 idle_timeout; /* Idle time before discarding (seconds). */
1789 ovs_be16 hard_timeout; /* Max time before discarding (seconds). */
1790 ovs_be16 priority; /* Priority level of flow entry. */
1791 ovs_be32 buffer_id; /* Buffered packet to apply to (or -1).
1792 Not meaningful for OFPFC_DELETE*. */
1793 ovs_be16 out_port; /* For OFPFC_DELETE* commands, require
1794 matching entries to include this as an
1795 output port. A value of OFPP_NONE
1796 indicates no restriction. */
1797 ovs_be16 flags; /* One of OFPFF_*. */
1798 ovs_be16 match_len; /* Size of nx_match. */
1799 uint8_t pad[6]; /* Align to 64-bits. */
1800 /* Followed by:
1801 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
1802 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
1803 * all-zero bytes, then
1804 * - Actions to fill out the remainder of the message length (always a
1805 * multiple of 8).
1806 */
1807};
982697a4 1808OFP_ASSERT(sizeof(struct nx_flow_mod) == 32);
09246b99
BP
1809
1810/* NXT_FLOW_REMOVED (analogous to OFPT_FLOW_REMOVED). */
1811struct nx_flow_removed {
09246b99
BP
1812 ovs_be64 cookie; /* Opaque controller-issued identifier. */
1813 ovs_be16 priority; /* Priority level of flow entry. */
1814 uint8_t reason; /* One of OFPRR_*. */
1815 uint8_t pad[1]; /* Align to 32-bits. */
1816 ovs_be32 duration_sec; /* Time flow was alive in seconds. */
1817 ovs_be32 duration_nsec; /* Time flow was alive in nanoseconds beyond
1818 duration_sec. */
1819 ovs_be16 idle_timeout; /* Idle timeout from original flow mod. */
1820 ovs_be16 match_len; /* Size of nx_match. */
1821 ovs_be64 packet_count;
1822 ovs_be64 byte_count;
1823 /* Followed by:
1824 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
1825 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
1826 * all-zero bytes. */
1827};
982697a4 1828OFP_ASSERT(sizeof(struct nx_flow_removed) == 40);
09246b99
BP
1829
1830/* Nicira vendor stats request of type NXST_FLOW (analogous to OFPST_FLOW
e729e793
JP
1831 * request).
1832 *
1833 * It is possible to limit matches to certain cookies by using the
1834 * NXM_NX_COOKIE and NXM_NX_COOKIE_W matches.
1835 */
09246b99 1836struct nx_flow_stats_request {
09246b99
BP
1837 ovs_be16 out_port; /* Require matching entries to include this
1838 as an output port. A value of OFPP_NONE
1839 indicates no restriction. */
1840 ovs_be16 match_len; /* Length of nx_match. */
1841 uint8_t table_id; /* ID of table to read (from ofp_table_stats)
1842 or 0xff for all tables. */
1843 uint8_t pad[3]; /* Align to 64 bits. */
1844 /* Followed by:
1845 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
1846 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
1847 * all-zero bytes, which must also exactly fill out the length of the
1848 * message.
1849 */
1850};
982697a4 1851OFP_ASSERT(sizeof(struct nx_flow_stats_request) == 8);
09246b99
BP
1852
1853/* Body for Nicira vendor stats reply of type NXST_FLOW (analogous to
f27f2134
BP
1854 * OFPST_FLOW reply).
1855 *
1856 * The values of 'idle_age' and 'hard_age' are only meaningful when talking to
1857 * a switch that implements the NXT_FLOW_AGE extension. Zero means that the
1858 * true value is unknown, perhaps because hardware does not track the value.
1859 * (Zero is also the value that one should ordinarily expect to see talking to
1860 * a switch that does not implement NXT_FLOW_AGE, since those switches zero the
1861 * padding bytes that these fields replaced.) A nonzero value X represents X-1
1862 * seconds. A value of 65535 represents 65534 or more seconds.
1863 *
1864 * 'idle_age' is the number of seconds that the flow has been idle, that is,
1865 * the number of seconds since a packet passed through the flow. 'hard_age' is
1866 * the number of seconds since the flow was last modified (e.g. OFPFC_MODIFY or
1867 * OFPFC_MODIFY_STRICT). (The 'duration_*' fields are the elapsed time since
1868 * the flow was added, regardless of subsequent modifications.)
1869 *
1870 * For a flow with an idle or hard timeout, 'idle_age' or 'hard_age',
1871 * respectively, will ordinarily be smaller than the timeout, but flow
1872 * expiration times are only approximate and so one must be prepared to
1873 * tolerate expirations that occur somewhat early or late.
1874 */
09246b99
BP
1875struct nx_flow_stats {
1876 ovs_be16 length; /* Length of this entry. */
1877 uint8_t table_id; /* ID of table flow came from. */
1878 uint8_t pad;
1879 ovs_be32 duration_sec; /* Time flow has been alive in seconds. */
1880 ovs_be32 duration_nsec; /* Time flow has been alive in nanoseconds
1881 beyond duration_sec. */
cc75d2c7 1882 ovs_be16 priority; /* Priority of the entry. */
09246b99
BP
1883 ovs_be16 idle_timeout; /* Number of seconds idle before expiration. */
1884 ovs_be16 hard_timeout; /* Number of seconds before expiration. */
1885 ovs_be16 match_len; /* Length of nx_match. */
f27f2134
BP
1886 ovs_be16 idle_age; /* Seconds since last packet, plus one. */
1887 ovs_be16 hard_age; /* Seconds since last modification, plus one. */
09246b99 1888 ovs_be64 cookie; /* Opaque controller-issued identifier. */
5e9d0469
BP
1889 ovs_be64 packet_count; /* Number of packets, UINT64_MAX if unknown. */
1890 ovs_be64 byte_count; /* Number of bytes, UINT64_MAX if unknown. */
09246b99
BP
1891 /* Followed by:
1892 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
1893 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
1894 * all-zero bytes, then
1895 * - Actions to fill out the remainder 'length' bytes (always a multiple
1896 * of 8).
1897 */
1898};
1899OFP_ASSERT(sizeof(struct nx_flow_stats) == 48);
1900
1901/* Nicira vendor stats request of type NXST_AGGREGATE (analogous to
a814ba0f
BP
1902 * OFPST_AGGREGATE request).
1903 *
1904 * The reply format is identical to the reply format for OFPST_AGGREGATE,
1905 * except for the header. */
09246b99 1906struct nx_aggregate_stats_request {
09246b99
BP
1907 ovs_be16 out_port; /* Require matching entries to include this
1908 as an output port. A value of OFPP_NONE
1909 indicates no restriction. */
1910 ovs_be16 match_len; /* Length of nx_match. */
1911 uint8_t table_id; /* ID of table to read (from ofp_table_stats)
1912 or 0xff for all tables. */
1913 uint8_t pad[3]; /* Align to 64 bits. */
1914 /* Followed by:
1915 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
1916 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
1917 * all-zero bytes, which must also exactly fill out the length of the
1918 * message.
1919 */
1920};
982697a4 1921OFP_ASSERT(sizeof(struct nx_aggregate_stats_request) == 8);
a7349929
BP
1922\f
1923/* NXT_SET_CONTROLLER_ID.
1924 *
1925 * Each OpenFlow controller connection has a 16-bit identifier that is
1926 * initially 0. This message changes the connection's ID to 'id'.
1927 *
1928 * Controller connection IDs need not be unique.
1929 *
1930 * The NXAST_CONTROLLER action is the only current user of controller
1931 * connection IDs. */
1932struct nx_controller_id {
a7349929
BP
1933 uint8_t zero[6]; /* Must be zero. */
1934 ovs_be16 controller_id; /* New controller connection ID. */
1935};
982697a4 1936OFP_ASSERT(sizeof(struct nx_controller_id) == 8);
a7349929
BP
1937
1938/* Action structure for NXAST_CONTROLLER.
1939 *
1940 * This generalizes using OFPAT_OUTPUT to send a packet to OFPP_CONTROLLER. In
1941 * addition to the 'max_len' that OFPAT_OUTPUT supports, it also allows
1942 * specifying:
1943 *
1944 * - 'reason': The reason code to use in the ofp_packet_in or nx_packet_in.
1945 *
1946 * - 'controller_id': The ID of the controller connection to which the
1947 * ofp_packet_in should be sent. The ofp_packet_in or nx_packet_in is
1948 * sent only to controllers that have the specified controller connection
1949 * ID. See "struct nx_controller_id" for more information. */
1950struct nx_action_controller {
1951 ovs_be16 type; /* OFPAT_VENDOR. */
1952 ovs_be16 len; /* Length is 16. */
1953 ovs_be32 vendor; /* NX_VENDOR_ID. */
1954 ovs_be16 subtype; /* NXAST_CONTROLLER. */
1955 ovs_be16 max_len; /* Maximum length to send to controller. */
1956 ovs_be16 controller_id; /* Controller ID to send packet-in. */
1957 uint8_t reason; /* enum ofp_packet_in_reason (OFPR_*). */
1958 uint8_t zero; /* Must be zero. */
1959};
1960OFP_ASSERT(sizeof(struct nx_action_controller) == 16);
2b07c8b1
BP
1961\f
1962/* Flow Table Monitoring
1963 * =====================
1964 *
1965 * NXST_FLOW_MONITOR allows a controller to keep track of changes to OpenFlow
1966 * flow table(s) or subsets of them, with the following workflow:
1967 *
1968 * 1. The controller sends an NXST_FLOW_MONITOR request to begin monitoring
1969 * flows. The 'id' in the request must be unique among all monitors that
1970 * the controller has started and not yet canceled on this OpenFlow
1971 * connection.
1972 *
1973 * 2. The switch responds with an NXST_FLOW_MONITOR reply. If the request's
1974 * 'flags' included NXFMF_INITIAL, the reply includes all the flows that
1975 * matched the request at the time of the request (with event NXFME_ADDED).
1976 * If 'flags' did not include NXFMF_INITIAL, the reply is empty.
1977 *
1978 * The reply uses the xid of the request (as do all replies to OpenFlow
1979 * requests).
1980 *
1981 * 3. Whenever a change to a flow table entry matches some outstanding monitor
1982 * request's criteria and flags, the switch sends a notification to the
1983 * controller as an additional NXST_FLOW_MONITOR reply with xid 0.
1984 *
1985 * When multiple outstanding monitors match a single change, only a single
1986 * notification is sent. This merged notification includes the information
1987 * requested in any of the individual monitors. That is, if any of the
1988 * matching monitors requests actions (NXFMF_ACTIONS), the notification
1989 * includes actions, and if any of the monitors request full changes for the
1990 * controller's own changes (NXFMF_OWN), the controller's own changes will
1991 * be included in full.
1992 *
1993 * 4. The controller may cancel a monitor with NXT_FLOW_MONITOR_CANCEL. No
1994 * further notifications will be sent on the basis of the canceled monitor
1995 * afterward.
1996 *
1997 *
1998 * Buffer Management
1999 * =================
2000 *
2001 * OpenFlow messages for flow monitor notifications can overflow the buffer
2002 * space available to the switch, either temporarily (e.g. due to network
2003 * conditions slowing OpenFlow traffic) or more permanently (e.g. the sustained
2004 * rate of flow table change exceeds the network bandwidth between switch and
2005 * controller).
2006 *
2007 * When Open vSwitch's notification buffer space reaches a limiting threshold,
2008 * OVS reacts as follows:
2009 *
2010 * 1. OVS sends an NXT_FLOW_MONITOR_PAUSED message to the controller, following
2011 * all the already queued notifications. After it receives this message,
2012 * the controller knows that its view of the flow table, as represented by
2013 * flow monitor notifications, is incomplete.
2014 *
2015 * 2. As long as the notification buffer is not empty:
2016 *
2017 * - NXMFE_ADD and NXFME_MODIFIED notifications will not be sent.
2018 *
2019 * - NXFME_DELETED notifications will still be sent, but only for flows
2020 * that existed before OVS sent NXT_FLOW_MONITOR_PAUSED.
2021 *
2022 * - NXFME_ABBREV notifications will not be sent. They are treated as
2023 * the expanded version (and therefore only the NXFME_DELETED
2024 * components, if any, are sent).
2025 *
2026 * 3. When the notification buffer empties, OVS sends NXFME_ADD notifications
2027 * for flows added since the buffer reached its limit and NXFME_MODIFIED
2028 * notifications for flows that existed before the limit was reached and
2029 * changed after the limit was reached.
2030 *
2031 * 4. OVS sends an NXT_FLOW_MONITOR_RESUMED message to the controller. After
2032 * it receives this message, the controller knows that its view of the flow
2033 * table, as represented by flow monitor notifications, is again complete.
2034 *
2035 * This allows the maximum buffer space requirement for notifications to be
2036 * bounded by the limit plus the maximum number of supported flows.
2037 *
2038 *
2039 * "Flow Removed" messages
2040 * =======================
2041 *
2042 * The flow monitor mechanism is independent of OFPT_FLOW_REMOVED and
2043 * NXT_FLOW_REMOVED. Flow monitor updates for deletion are sent if
2044 * NXFMF_DELETE is set on a monitor, regardless of whether the
2045 * OFPFF_SEND_FLOW_REM flag was set when the flow was added. */
2046
2047/* NXST_FLOW_MONITOR request.
2048 *
2049 * The NXST_FLOW_MONITOR request's body consists of an array of zero or more
2050 * instances of this structure. The request arranges to monitor the flows
2051 * that match the specified criteria, which are interpreted in the same way as
2052 * for NXST_FLOW.
2053 *
2054 * 'id' identifies a particular monitor for the purpose of allowing it to be
2055 * canceled later with NXT_FLOW_MONITOR_CANCEL. 'id' must be unique among
2056 * existing monitors that have not already been canceled.
2057 *
2058 * The reply includes the initial flow matches for monitors that have the
2059 * NXFMF_INITIAL flag set. No single flow will be included in the reply more
2060 * than once, even if more than one requested monitor matches that flow. The
2061 * reply will be empty if none of the monitors has NXFMF_INITIAL set or if none
2062 * of the monitors initially matches any flows.
2063 *
2064 * For NXFMF_ADD, an event will be reported if 'out_port' matches against the
2065 * actions of the flow being added or, for a flow that is replacing an existing
2066 * flow, if 'out_port' matches against the actions of the flow being replaced.
2067 * For NXFMF_DELETE, 'out_port' matches against the actions of a flow being
2068 * deleted. For NXFMF_MODIFY, an event will be reported if 'out_port' matches
2069 * either the old or the new actions. */
2070struct nx_flow_monitor_request {
2071 ovs_be32 id; /* Controller-assigned ID for this monitor. */
2072 ovs_be16 flags; /* NXFMF_*. */
2073 ovs_be16 out_port; /* Required output port, if not OFPP_NONE. */
2074 ovs_be16 match_len; /* Length of nx_match. */
2075 uint8_t table_id; /* One table's ID or 0xff for all tables. */
2076 uint8_t zeros[5]; /* Align to 64 bits (must be zero). */
2077 /* Followed by:
2078 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
2079 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
2080 * all-zero bytes. */
2081};
2082OFP_ASSERT(sizeof(struct nx_flow_monitor_request) == 16);
2083
2084/* 'flags' bits in struct nx_flow_monitor_request. */
2085enum nx_flow_monitor_flags {
2086 /* When to send updates. */
2087 NXFMF_INITIAL = 1 << 0, /* Initially matching flows. */
2088 NXFMF_ADD = 1 << 1, /* New matching flows as they are added. */
2089 NXFMF_DELETE = 1 << 2, /* Old matching flows as they are removed. */
2090 NXFMF_MODIFY = 1 << 3, /* Matching flows as they are changed. */
2091
2092 /* What to include in updates. */
2093 NXFMF_ACTIONS = 1 << 4, /* If set, actions are included. */
2094 NXFMF_OWN = 1 << 5, /* If set, include own changes in full. */
2095};
2096
2097/* NXST_FLOW_MONITOR reply header.
2098 *
2099 * The body of an NXST_FLOW_MONITOR reply is an array of variable-length
2100 * structures, each of which begins with this header. The 'length' member may
2101 * be used to traverse the array, and the 'event' member may be used to
2102 * determine the particular structure.
2103 *
2104 * Every instance is a multiple of 8 bytes long. */
2105struct nx_flow_update_header {
2106 ovs_be16 length; /* Length of this entry. */
2107 ovs_be16 event; /* One of NXFME_*. */
2108 /* ...other data depending on 'event'... */
2109};
2110OFP_ASSERT(sizeof(struct nx_flow_update_header) == 4);
2111
2112/* 'event' values in struct nx_flow_update_header. */
2113enum nx_flow_update_event {
2114 /* struct nx_flow_update_full. */
2115 NXFME_ADDED = 0, /* Flow was added. */
2116 NXFME_DELETED = 1, /* Flow was deleted. */
2117 NXFME_MODIFIED = 2, /* Flow (generally its actions) was changed. */
2118
2119 /* struct nx_flow_update_abbrev. */
2120 NXFME_ABBREV = 3, /* Abbreviated reply. */
2121};
2122
2123/* NXST_FLOW_MONITOR reply for NXFME_ADDED, NXFME_DELETED, and
2124 * NXFME_MODIFIED. */
2125struct nx_flow_update_full {
2126 ovs_be16 length; /* Length is 24. */
2127 ovs_be16 event; /* One of NXFME_*. */
2128 ovs_be16 reason; /* OFPRR_* for NXFME_DELETED, else zero. */
2129 ovs_be16 priority; /* Priority of the entry. */
2130 ovs_be16 idle_timeout; /* Number of seconds idle before expiration. */
2131 ovs_be16 hard_timeout; /* Number of seconds before expiration. */
2132 ovs_be16 match_len; /* Length of nx_match. */
2133 uint8_t table_id; /* ID of flow's table. */
2134 uint8_t pad; /* Reserved, currently zeroed. */
2135 ovs_be64 cookie; /* Opaque controller-issued identifier. */
2136 /* Followed by:
2137 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
2138 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
2139 * all-zero bytes, then
2140 * - Actions to fill out the remainder 'length' bytes (always a multiple
2141 * of 8). If NXFMF_ACTIONS was not specified, or 'event' is
2142 * NXFME_DELETED, no actions are included.
2143 */
2144};
2145OFP_ASSERT(sizeof(struct nx_flow_update_full) == 24);
2146
2147/* NXST_FLOW_MONITOR reply for NXFME_ABBREV.
2148 *
2149 * When the controller does not specify NXFMF_OWN in a monitor request, any
2150 * flow tables changes due to the controller's own requests (on the same
2151 * OpenFlow channel) will be abbreviated, when possible, to this form, which
2152 * simply specifies the 'xid' of the OpenFlow request (e.g. an OFPT_FLOW_MOD or
2153 * NXT_FLOW_MOD) that caused the change.
2154 *
2155 * Some changes cannot be abbreviated and will be sent in full:
2156 *
2157 * - Changes that only partially succeed. This can happen if, for example,
2158 * a flow_mod with type OFPFC_MODIFY affects multiple flows, but only some
2159 * of those modifications succeed (e.g. due to hardware limitations).
2160 *
2161 * This cannot occur with the current implementation of the Open vSwitch
2162 * software datapath. It could happen with other datapath implementations.
2163 *
2164 * - Changes that race with conflicting changes made by other controllers or
2165 * other flow_mods (not separated by barriers) by the same controller.
2166 *
2167 * This cannot occur with the current Open vSwitch implementation
2168 * (regardless of datapath) because Open vSwitch internally serializes
2169 * potentially conflicting changes.
2170 *
2171 * A flow_mod that does not change the flow table will not trigger any
2172 * notification, even an abbreviated one. For example, a "modify" or "delete"
2173 * flow_mod that does not match any flows will not trigger a notification.
2174 * Whether an "add" or "modify" that specifies all the same parameters that a
2175 * flow already has triggers a notification is unspecified and subject to
2176 * change in future versions of Open vSwitch.
2177 *
2178 * OVS will always send the notifications for a given flow table change before
b10a4760
BP
2179 * the reply to a OFPT_BARRIER_REQUEST request that follows the flow table
2180 * change. Thus, if the controller does not receive an abbreviated (or
2181 * unabbreviated) notification for a flow_mod before the next
2182 * OFPT_BARRIER_REPLY, it will never receive one. */
2b07c8b1
BP
2183struct nx_flow_update_abbrev {
2184 ovs_be16 length; /* Length is 8. */
2185 ovs_be16 event; /* NXFME_ABBREV. */
2186 ovs_be32 xid; /* Controller-specified xid from flow_mod. */
2187};
2188OFP_ASSERT(sizeof(struct nx_flow_update_abbrev) == 8);
2189
982697a4
BP
2190/* NXT_FLOW_MONITOR_CANCEL.
2191 *
2192 * Used by a controller to cancel an outstanding monitor. */
2b07c8b1 2193struct nx_flow_monitor_cancel {
2b07c8b1
BP
2194 ovs_be32 id; /* 'id' from nx_flow_monitor_request. */
2195};
982697a4 2196OFP_ASSERT(sizeof(struct nx_flow_monitor_cancel) == 4);
659586ef 2197
4cceacb9
JS
2198/* Action structure for NXAST_WRITE_METADATA.
2199 *
2200 * Modifies the 'mask' bits of the metadata value. */
2201struct nx_action_write_metadata {
2202 ovs_be16 type; /* OFPAT_VENDOR. */
2203 ovs_be16 len; /* Length is 32. */
2204 ovs_be32 vendor; /* NX_VENDOR_ID. */
2205 ovs_be16 subtype; /* NXAST_WRITE_METADATA. */
2206 uint8_t zeros[6]; /* Must be zero. */
2207 ovs_be64 metadata; /* Metadata register. */
2208 ovs_be64 mask; /* Metadata mask. */
2209};
2210OFP_ASSERT(sizeof(struct nx_action_write_metadata) == 32);
2211
064af421 2212#endif /* openflow/nicira-ext.h */