]> git.proxmox.com Git - mirror_ovs.git/blob - include/openflow/nicira-ext.h
Add table_id to NXM flow_removed messages.
[mirror_ovs.git] / include / openflow / nicira-ext.h
1 /*
2 * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
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.
15 */
16
17 #ifndef OPENFLOW_NICIRA_EXT_H
18 #define OPENFLOW_NICIRA_EXT_H 1
19
20 #include "openflow/openflow.h"
21 #include "openvswitch/types.h"
22
23 /* The following vendor extensions, proposed by Nicira, are not yet
24 * standardized, so they are not included in openflow.h. Some of them may be
25 * suitable for standardization; others we never expect to standardize. */
26
27 #define NX_VENDOR_ID 0x00002320
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
39 * and NXVC_VENDOR_ERROR as code, followed by struct nx_vendor_error with
40 * vendor-specific details, followed by at least 64 bytes of the failed
41 * request.
42 *
43 * It would be better to have a type-specific vendor extension, e.g. so that
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. */
58 enum 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. */
63 struct 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. */
71
72 /* Header for Nicira vendor requests and replies. */
73 struct nicira_header {
74 struct ofp_header header;
75 ovs_be32 vendor; /* NX_VENDOR_ID. */
76 ovs_be32 subtype; /* See the NXT numbers in ofp-msgs.h. */
77 };
78 OFP_ASSERT(sizeof(struct nicira_header) == 16);
79
80 /* Header for Nicira vendor stats request and reply messages in OpenFlow
81 * 1.0. */
82 struct nicira10_stats_msg {
83 struct ofp10_vendor_stats_msg vsm; /* Vendor NX_VENDOR_ID. */
84 ovs_be32 subtype; /* One of NXST_* below. */
85 uint8_t pad[4]; /* Align to 64-bits. */
86 };
87 OFP_ASSERT(sizeof(struct nicira10_stats_msg) == 24);
88
89 /* Header for Nicira vendor stats request and reply messages in OpenFlow
90 * 1.1. */
91 struct nicira11_stats_msg {
92 struct ofp11_vendor_stats_msg vsm; /* Vendor NX_VENDOR_ID. */
93 ovs_be32 subtype; /* One of NXST_* below. */
94 };
95 OFP_ASSERT(sizeof(struct nicira11_stats_msg) == 24);
96
97 /* Fields to use when hashing flows. */
98 enum 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
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 *
121 * When this feature is enabled, Open vSwitch treats struct ofp10_flow_mod's
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
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.)
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 */
153 struct nx_flow_mod_table_id {
154 uint8_t set; /* Nonzero to enable, zero to disable. */
155 uint8_t pad[7];
156 };
157 OFP_ASSERT(sizeof(struct nx_flow_mod_table_id) == 8);
158
159 enum 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. */
165 struct nx_set_packet_in_format {
166 ovs_be32 format; /* One of NXPIF_*. */
167 };
168 OFP_ASSERT(sizeof(struct nx_set_packet_in_format) == 4);
169
170 /* NXT_PACKET_IN (analogous to OFPT_PACKET_IN).
171 *
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.
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
199 * controller does not. The controller must prepared to tolerate these.
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. */
203 struct nx_packet_in {
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
221 /* uint8_t nxm_fields[...]; */ /* NXM headers. */
222 /* uint8_t pad[2]; */ /* Align to 64 bit + 16 bit. */
223 /* uint8_t data[0]; */ /* Ethernet frame. */
224 };
225 OFP_ASSERT(sizeof(struct nx_packet_in) == 24);
226
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 *
243 * Slave controllers do not receive OFPT_PACKET_IN or OFPT_FLOW_REMOVED
244 * messages, but they do receive OFPT_PORT_STATUS messages.
245 */
246 struct nx_role_request {
247 ovs_be32 role; /* One of NX_ROLE_*. */
248 };
249 OFP_ASSERT(sizeof(struct nx_role_request) == 4);
250
251 enum 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 };
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.
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).
273 */
274 struct nx_async_config {
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 };
279 OFP_ASSERT(sizeof(struct nx_async_config) == 24);
280 \f
281 /* Nicira vendor flow actions. */
282
283 enum nx_action_subtype {
284 NXAST_SNAT__OBSOLETE, /* No longer used. */
285 NXAST_RESUBMIT, /* struct nx_action_resubmit */
286 NXAST_SET_TUNNEL, /* struct nx_action_set_tunnel */
287 NXAST_DROP_SPOOFED_ARP__OBSOLETE,
288 NXAST_SET_QUEUE, /* struct nx_action_set_queue */
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 */
292 NXAST_NOTE, /* struct nx_action_note */
293 NXAST_SET_TUNNEL64, /* struct nx_action_set_tunnel64 */
294 NXAST_MULTIPATH, /* struct nx_action_multipath */
295 NXAST_AUTOPATH__OBSOLETE, /* No longer used. */
296 NXAST_BUNDLE, /* struct nx_action_bundle */
297 NXAST_BUNDLE_LOAD, /* struct nx_action_bundle */
298 NXAST_RESUBMIT_TABLE, /* struct nx_action_resubmit */
299 NXAST_OUTPUT_REG, /* struct nx_action_output_reg */
300 NXAST_LEARN, /* struct nx_action_learn */
301 NXAST_EXIT, /* struct nx_action_header */
302 NXAST_DEC_TTL, /* struct nx_action_header */
303 NXAST_FIN_TIMEOUT, /* struct nx_action_fin_timeout */
304 NXAST_CONTROLLER, /* struct nx_action_controller */
305 NXAST_DEC_TTL_CNT_IDS, /* struct nx_action_cnt_ids */
306 NXAST_WRITE_METADATA, /* struct nx_action_write_metadata */
307 NXAST_PUSH_MPLS, /* struct nx_action_push_mpls */
308 NXAST_POP_MPLS, /* struct nx_action_pop_mpls */
309 NXAST_SET_MPLS_TTL, /* struct nx_action_ttl */
310 NXAST_DEC_MPLS_TTL, /* struct nx_action_header */
311 NXAST_STACK_PUSH, /* struct nx_action_stack */
312 NXAST_STACK_POP, /* struct nx_action_stack */
313 };
314
315 /* Header for Nicira-defined actions. */
316 struct nx_action_header {
317 ovs_be16 type; /* OFPAT_VENDOR. */
318 ovs_be16 len; /* Length is 16. */
319 ovs_be32 vendor; /* NX_VENDOR_ID. */
320 ovs_be16 subtype; /* NXAST_*. */
321 uint8_t pad[6];
322 };
323 OFP_ASSERT(sizeof(struct nx_action_header) == 16);
324
325 /* Action structures for NXAST_RESUBMIT and NXAST_RESUBMIT_TABLE.
326 *
327 * These actions search one of the switch's flow tables:
328 *
329 * - For NXAST_RESUBMIT_TABLE only, if the 'table' member is not 255, then
330 * it specifies the table to search.
331 *
332 * - Otherwise (for NXAST_RESUBMIT_TABLE with a 'table' of 255, or for
333 * NXAST_RESUBMIT regardless of 'table'), it searches the current flow
334 * table, that is, the OpenFlow flow table that contains the flow from
335 * which this action was obtained. If this action did not come from a
336 * flow table (e.g. it came from an OFPT_PACKET_OUT message), then table 0
337 * is the current table.
338 *
339 * The flow table lookup uses a flow that may be slightly modified from the
340 * original lookup:
341 *
342 * - For NXAST_RESUBMIT, the 'in_port' member of struct nx_action_resubmit
343 * is used as the flow's in_port.
344 *
345 * - For NXAST_RESUBMIT_TABLE, if the 'in_port' member is not OFPP_IN_PORT,
346 * then its value is used as the flow's in_port. Otherwise, the original
347 * in_port is used.
348 *
349 * - If actions that modify the flow (e.g. OFPAT_SET_VLAN_VID) precede the
350 * resubmit action, then the flow is updated with the new values.
351 *
352 * Following the lookup, the original in_port is restored.
353 *
354 * If the modified flow matched in the flow table, then the corresponding
355 * actions are executed. Afterward, actions following the resubmit in the
356 * original set of actions, if any, are executed; any changes made to the
357 * packet (e.g. changes to VLAN) by secondary actions persist when those
358 * actions are executed, although the original in_port is restored.
359 *
360 * Resubmit actions may be used any number of times within a set of actions.
361 *
362 * Resubmit actions may nest to an implementation-defined depth. Beyond this
363 * implementation-defined depth, further resubmit actions are simply ignored.
364 *
365 * NXAST_RESUBMIT ignores 'table' and 'pad'. NXAST_RESUBMIT_TABLE requires
366 * 'pad' to be all-bits-zero.
367 *
368 * Open vSwitch 1.0.1 and earlier did not support recursion. Open vSwitch
369 * before 1.2.90 did not support NXAST_RESUBMIT_TABLE.
370 */
371 struct nx_action_resubmit {
372 ovs_be16 type; /* OFPAT_VENDOR. */
373 ovs_be16 len; /* Length is 16. */
374 ovs_be32 vendor; /* NX_VENDOR_ID. */
375 ovs_be16 subtype; /* NXAST_RESUBMIT. */
376 ovs_be16 in_port; /* New in_port for checking flow table. */
377 uint8_t table; /* NXAST_RESUBMIT_TABLE: table to use. */
378 uint8_t pad[3];
379 };
380 OFP_ASSERT(sizeof(struct nx_action_resubmit) == 16);
381
382 /* Action structure for NXAST_SET_TUNNEL.
383 *
384 * Sets the encapsulating tunnel ID to a 32-bit value. The most-significant 32
385 * bits of the tunnel ID are set to 0. */
386 struct nx_action_set_tunnel {
387 ovs_be16 type; /* OFPAT_VENDOR. */
388 ovs_be16 len; /* Length is 16. */
389 ovs_be32 vendor; /* NX_VENDOR_ID. */
390 ovs_be16 subtype; /* NXAST_SET_TUNNEL. */
391 uint8_t pad[2];
392 ovs_be32 tun_id; /* Tunnel ID. */
393 };
394 OFP_ASSERT(sizeof(struct nx_action_set_tunnel) == 16);
395
396 /* Action structure for NXAST_SET_TUNNEL64.
397 *
398 * Sets the encapsulating tunnel ID to a 64-bit value. */
399 struct nx_action_set_tunnel64 {
400 ovs_be16 type; /* OFPAT_VENDOR. */
401 ovs_be16 len; /* Length is 24. */
402 ovs_be32 vendor; /* NX_VENDOR_ID. */
403 ovs_be16 subtype; /* NXAST_SET_TUNNEL64. */
404 uint8_t pad[6];
405 ovs_be64 tun_id; /* Tunnel ID. */
406 };
407 OFP_ASSERT(sizeof(struct nx_action_set_tunnel64) == 24);
408
409 /* Action structure for NXAST_SET_QUEUE.
410 *
411 * Set the queue that should be used when packets are output. This is similar
412 * to the OpenFlow OFPAT_ENQUEUE action, but does not take the output port as
413 * an argument. This allows the queue to be defined before the port is
414 * known. */
415 struct nx_action_set_queue {
416 ovs_be16 type; /* OFPAT_VENDOR. */
417 ovs_be16 len; /* Length is 16. */
418 ovs_be32 vendor; /* NX_VENDOR_ID. */
419 ovs_be16 subtype; /* NXAST_SET_QUEUE. */
420 uint8_t pad[2];
421 ovs_be32 queue_id; /* Where to enqueue packets. */
422 };
423 OFP_ASSERT(sizeof(struct nx_action_set_queue) == 16);
424
425 /* Action structure for NXAST_POP_QUEUE.
426 *
427 * Restores the queue to the value it was before any NXAST_SET_QUEUE actions
428 * were used. Only the original queue can be restored this way; no stack is
429 * maintained. */
430 struct nx_action_pop_queue {
431 ovs_be16 type; /* OFPAT_VENDOR. */
432 ovs_be16 len; /* Length is 16. */
433 ovs_be32 vendor; /* NX_VENDOR_ID. */
434 ovs_be16 subtype; /* NXAST_POP_QUEUE. */
435 uint8_t pad[6];
436 };
437 OFP_ASSERT(sizeof(struct nx_action_pop_queue) == 16);
438
439 /* Action structure for NXAST_REG_MOVE.
440 *
441 * Copies src[src_ofs:src_ofs+n_bits] to dst[dst_ofs:dst_ofs+n_bits], where
442 * a[b:c] denotes the bits within 'a' numbered 'b' through 'c' (not including
443 * bit 'c'). Bit numbering starts at 0 for the least-significant bit, 1 for
444 * the next most significant bit, and so on.
445 *
446 * 'src' and 'dst' are nxm_header values with nxm_hasmask=0. (It doesn't make
447 * sense to use nxm_hasmask=1 because the action does not do any kind of
448 * matching; it uses the actual value of a field.)
449 *
450 * The following nxm_header values are potentially acceptable as 'src':
451 *
452 * - NXM_OF_IN_PORT
453 * - NXM_OF_ETH_DST
454 * - NXM_OF_ETH_SRC
455 * - NXM_OF_ETH_TYPE
456 * - NXM_OF_VLAN_TCI
457 * - NXM_OF_IP_TOS
458 * - NXM_OF_IP_PROTO
459 * - NXM_OF_IP_SRC
460 * - NXM_OF_IP_DST
461 * - NXM_OF_TCP_SRC
462 * - NXM_OF_TCP_DST
463 * - NXM_OF_UDP_SRC
464 * - NXM_OF_UDP_DST
465 * - NXM_OF_ICMP_TYPE
466 * - NXM_OF_ICMP_CODE
467 * - NXM_OF_ARP_OP
468 * - NXM_OF_ARP_SPA
469 * - NXM_OF_ARP_TPA
470 * - NXM_NX_TUN_ID
471 * - NXM_NX_ARP_SHA
472 * - NXM_NX_ARP_THA
473 * - NXM_NX_ICMPV6_TYPE
474 * - NXM_NX_ICMPV6_CODE
475 * - NXM_NX_ND_SLL
476 * - NXM_NX_ND_TLL
477 * - NXM_NX_REG(idx) for idx in the switch's accepted range.
478 *
479 * The following nxm_header values are potentially acceptable as 'dst':
480 *
481 * - NXM_OF_ETH_DST
482 * - NXM_OF_ETH_SRC
483 * - NXM_OF_IP_TOS
484 * - NXM_OF_IP_SRC
485 * - NXM_OF_IP_DST
486 * - NXM_OF_TCP_SRC
487 * - NXM_OF_TCP_DST
488 * - NXM_OF_UDP_SRC
489 * - NXM_OF_UDP_DST
490 * Modifying any of the above fields changes the corresponding packet
491 * header.
492 *
493 * - NXM_NX_REG(idx) for idx in the switch's accepted range.
494 *
495 * - NXM_OF_VLAN_TCI. Modifying this field's value has side effects on the
496 * packet's 802.1Q header. Setting a value with CFI=0 removes the 802.1Q
497 * header (if any), ignoring the other bits. Setting a value with CFI=1
498 * adds or modifies the 802.1Q header appropriately, setting the TCI field
499 * to the field's new value (with the CFI bit masked out).
500 *
501 * - NXM_NX_TUN_ID. Modifying this value modifies the tunnel ID used for the
502 * packet's next tunnel encapsulation.
503 *
504 * A given nxm_header value may be used as 'src' or 'dst' only on a flow whose
505 * nx_match satisfies its prerequisites. For example, NXM_OF_IP_TOS may be
506 * used only if the flow's nx_match includes an nxm_entry that specifies
507 * nxm_type=NXM_OF_ETH_TYPE, nxm_hasmask=0, and nxm_value=0x0800.
508 *
509 * The switch will reject actions for which src_ofs+n_bits is greater than the
510 * width of 'src' or dst_ofs+n_bits is greater than the width of 'dst' with
511 * error type OFPET_BAD_ACTION, code OFPBAC_BAD_ARGUMENT.
512 *
513 * This action behaves properly when 'src' overlaps with 'dst', that is, it
514 * behaves as if 'src' were copied out to a temporary buffer, then the
515 * temporary buffer copied to 'dst'.
516 */
517 struct nx_action_reg_move {
518 ovs_be16 type; /* OFPAT_VENDOR. */
519 ovs_be16 len; /* Length is 24. */
520 ovs_be32 vendor; /* NX_VENDOR_ID. */
521 ovs_be16 subtype; /* NXAST_REG_MOVE. */
522 ovs_be16 n_bits; /* Number of bits. */
523 ovs_be16 src_ofs; /* Starting bit offset in source. */
524 ovs_be16 dst_ofs; /* Starting bit offset in destination. */
525 ovs_be32 src; /* Source register. */
526 ovs_be32 dst; /* Destination register. */
527 };
528 OFP_ASSERT(sizeof(struct nx_action_reg_move) == 24);
529
530 /* Action structure for NXAST_REG_LOAD.
531 *
532 * Copies value[0:n_bits] to dst[ofs:ofs+n_bits], where a[b:c] denotes the bits
533 * within 'a' numbered 'b' through 'c' (not including bit 'c'). Bit numbering
534 * starts at 0 for the least-significant bit, 1 for the next most significant
535 * bit, and so on.
536 *
537 * 'dst' is an nxm_header with nxm_hasmask=0. See the documentation for
538 * NXAST_REG_MOVE, above, for the permitted fields and for the side effects of
539 * loading them.
540 *
541 * The 'ofs' and 'n_bits' fields are combined into a single 'ofs_nbits' field
542 * to avoid enlarging the structure by another 8 bytes. To allow 'n_bits' to
543 * take a value between 1 and 64 (inclusive) while taking up only 6 bits, it is
544 * also stored as one less than its true value:
545 *
546 * 15 6 5 0
547 * +------------------------------+------------------+
548 * | ofs | n_bits - 1 |
549 * +------------------------------+------------------+
550 *
551 * The switch will reject actions for which ofs+n_bits is greater than the
552 * width of 'dst', or in which any bits in 'value' with value 2**n_bits or
553 * greater are set to 1, with error type OFPET_BAD_ACTION, code
554 * OFPBAC_BAD_ARGUMENT.
555 */
556 struct nx_action_reg_load {
557 ovs_be16 type; /* OFPAT_VENDOR. */
558 ovs_be16 len; /* Length is 24. */
559 ovs_be32 vendor; /* NX_VENDOR_ID. */
560 ovs_be16 subtype; /* NXAST_REG_LOAD. */
561 ovs_be16 ofs_nbits; /* (ofs << 6) | (n_bits - 1). */
562 ovs_be32 dst; /* Destination register. */
563 ovs_be64 value; /* Immediate value. */
564 };
565 OFP_ASSERT(sizeof(struct nx_action_reg_load) == 24);
566
567 /* Action structure for NXAST_STACK_PUSH and NXAST_STACK_POP.
568 *
569 * Pushes (or pops) field[offset: offset + n_bits] to (or from)
570 * top of the stack.
571 */
572 struct nx_action_stack {
573 ovs_be16 type; /* OFPAT_VENDOR. */
574 ovs_be16 len; /* Length is 16. */
575 ovs_be32 vendor; /* NX_VENDOR_ID. */
576 ovs_be16 subtype; /* NXAST_REG_PUSH or NXAST_REG_POP. */
577 ovs_be16 offset; /* Bit offset into the field. */
578 ovs_be32 field; /* The field used for push or pop. */
579 ovs_be16 n_bits; /* (n_bits + 1) bits of the field. */
580 uint8_t zero[6]; /* Reserved, must be zero. */
581 };
582 OFP_ASSERT(sizeof(struct nx_action_stack) == 24);
583
584 /* Action structure for NXAST_NOTE.
585 *
586 * This action has no effect. It is variable length. The switch does not
587 * attempt to interpret the user-defined 'note' data in any way. A controller
588 * can use this action to attach arbitrary metadata to a flow.
589 *
590 * This action might go away in the future.
591 */
592 struct nx_action_note {
593 ovs_be16 type; /* OFPAT_VENDOR. */
594 ovs_be16 len; /* A multiple of 8, but at least 16. */
595 ovs_be32 vendor; /* NX_VENDOR_ID. */
596 ovs_be16 subtype; /* NXAST_NOTE. */
597 uint8_t note[6]; /* Start of user-defined data. */
598 /* Possibly followed by additional user-defined data. */
599 };
600 OFP_ASSERT(sizeof(struct nx_action_note) == 16);
601
602 /* Action structure for NXAST_MULTIPATH.
603 *
604 * This action performs the following steps in sequence:
605 *
606 * 1. Hashes the fields designated by 'fields', one of NX_HASH_FIELDS_*.
607 * Refer to the definition of "enum nx_mp_fields" for details.
608 *
609 * The 'basis' value is used as a universal hash parameter, that is,
610 * different values of 'basis' yield different hash functions. The
611 * particular universal hash function used is implementation-defined.
612 *
613 * The hashed fields' values are drawn from the current state of the
614 * flow, including all modifications that have been made by actions up to
615 * this point.
616 *
617 * 2. Applies the multipath link choice algorithm specified by 'algorithm',
618 * one of NX_MP_ALG_*. Refer to the definition of "enum nx_mp_algorithm"
619 * for details.
620 *
621 * The output of the algorithm is 'link', an unsigned integer less than
622 * or equal to 'max_link'.
623 *
624 * Some algorithms use 'arg' as an additional argument.
625 *
626 * 3. Stores 'link' in dst[ofs:ofs+n_bits]. The format and semantics of
627 * 'dst' and 'ofs_nbits' are similar to those for the NXAST_REG_LOAD
628 * action.
629 *
630 * The switch will reject actions that have an unknown 'fields', or an unknown
631 * 'algorithm', or in which ofs+n_bits is greater than the width of 'dst', or
632 * in which 'max_link' is greater than or equal to 2**n_bits, with error type
633 * OFPET_BAD_ACTION, code OFPBAC_BAD_ARGUMENT.
634 */
635 struct nx_action_multipath {
636 ovs_be16 type; /* OFPAT_VENDOR. */
637 ovs_be16 len; /* Length is 32. */
638 ovs_be32 vendor; /* NX_VENDOR_ID. */
639 ovs_be16 subtype; /* NXAST_MULTIPATH. */
640
641 /* What fields to hash and how. */
642 ovs_be16 fields; /* One of NX_HASH_FIELDS_*. */
643 ovs_be16 basis; /* Universal hash parameter. */
644 ovs_be16 pad0;
645
646 /* Multipath link choice algorithm to apply to hash value. */
647 ovs_be16 algorithm; /* One of NX_MP_ALG_*. */
648 ovs_be16 max_link; /* Number of output links, minus 1. */
649 ovs_be32 arg; /* Algorithm-specific argument. */
650 ovs_be16 pad1;
651
652 /* Where to store the result. */
653 ovs_be16 ofs_nbits; /* (ofs << 6) | (n_bits - 1). */
654 ovs_be32 dst; /* Destination. */
655 };
656 OFP_ASSERT(sizeof(struct nx_action_multipath) == 32);
657
658 /* NXAST_MULTIPATH: Multipath link choice algorithm to apply.
659 *
660 * In the descriptions below, 'n_links' is max_link + 1. */
661 enum nx_mp_algorithm {
662 /* link = hash(flow) % n_links.
663 *
664 * Redistributes all traffic when n_links changes. O(1) performance. See
665 * RFC 2992.
666 *
667 * Use UINT16_MAX for max_link to get a raw hash value. */
668 NX_MP_ALG_MODULO_N,
669
670 /* link = hash(flow) / (MAX_HASH / n_links).
671 *
672 * Redistributes between one-quarter and one-half of traffic when n_links
673 * changes. O(1) performance. See RFC 2992.
674 */
675 NX_MP_ALG_HASH_THRESHOLD,
676
677 /* for i in [0,n_links):
678 * weights[i] = hash(flow, i)
679 * link = { i such that weights[i] >= weights[j] for all j != i }
680 *
681 * Redistributes 1/n_links of traffic when n_links changes. O(n_links)
682 * performance. If n_links is greater than a threshold (currently 64, but
683 * subject to change), Open vSwitch will substitute another algorithm
684 * automatically. See RFC 2992. */
685 NX_MP_ALG_HRW, /* Highest Random Weight. */
686
687 /* i = 0
688 * repeat:
689 * i = i + 1
690 * link = hash(flow, i) % arg
691 * while link > max_link
692 *
693 * Redistributes 1/n_links of traffic when n_links changes. O(1)
694 * performance when arg/max_link is bounded by a constant.
695 *
696 * Redistributes all traffic when arg changes.
697 *
698 * arg must be greater than max_link and for best performance should be no
699 * more than approximately max_link * 2. If arg is outside the acceptable
700 * range, Open vSwitch will automatically substitute the least power of 2
701 * greater than max_link.
702 *
703 * This algorithm is specific to Open vSwitch.
704 */
705 NX_MP_ALG_ITER_HASH /* Iterative Hash. */
706 };
707 \f
708 /* Action structure for NXAST_LEARN.
709 *
710 * This action adds or modifies a flow in an OpenFlow table, similar to
711 * OFPT_FLOW_MOD with OFPFC_MODIFY_STRICT as 'command'. The new flow has the
712 * specified idle timeout, hard timeout, priority, cookie, and flags. The new
713 * flow's match criteria and actions are built by applying each of the series
714 * of flow_mod_spec elements included as part of the action.
715 *
716 * A flow_mod_spec starts with a 16-bit header. A header that is all-bits-0 is
717 * a no-op used for padding the action as a whole to a multiple of 8 bytes in
718 * length. Otherwise, the flow_mod_spec can be thought of as copying 'n_bits'
719 * bits from a source to a destination. In this case, the header contains
720 * multiple fields:
721 *
722 * 15 14 13 12 11 10 0
723 * +------+---+------+---------------------------------+
724 * | 0 |src| dst | n_bits |
725 * +------+---+------+---------------------------------+
726 *
727 * The meaning and format of a flow_mod_spec depends on 'src' and 'dst'. The
728 * following table summarizes the meaning of each possible combination.
729 * Details follow the table:
730 *
731 * src dst meaning
732 * --- --- ----------------------------------------------------------
733 * 0 0 Add match criteria based on value in a field.
734 * 1 0 Add match criteria based on an immediate value.
735 * 0 1 Add NXAST_REG_LOAD action to copy field into a different field.
736 * 1 1 Add NXAST_REG_LOAD action to load immediate value into a field.
737 * 0 2 Add OFPAT_OUTPUT action to output to port from specified field.
738 * All other combinations are undefined and not allowed.
739 *
740 * The flow_mod_spec header is followed by a source specification and a
741 * destination specification. The format and meaning of the source
742 * specification depends on 'src':
743 *
744 * - If 'src' is 0, the source bits are taken from a field in the flow to
745 * which this action is attached. (This should be a wildcarded field. If
746 * its value is fully specified then the source bits being copied have
747 * constant values.)
748 *
749 * The source specification is an ovs_be32 'field' and an ovs_be16 'ofs'.
750 * 'field' is an nxm_header with nxm_hasmask=0, and 'ofs' the starting bit
751 * offset within that field. The source bits are field[ofs:ofs+n_bits-1].
752 * 'field' and 'ofs' are subject to the same restrictions as the source
753 * field in NXAST_REG_MOVE.
754 *
755 * - If 'src' is 1, the source bits are a constant value. The source
756 * specification is (n_bits+15)/16*2 bytes long. Taking those bytes as a
757 * number in network order, the source bits are the 'n_bits'
758 * least-significant bits. The switch will report an error if other bits
759 * in the constant are nonzero.
760 *
761 * The flow_mod_spec destination specification, for 'dst' of 0 or 1, is an
762 * ovs_be32 'field' and an ovs_be16 'ofs'. 'field' is an nxm_header with
763 * nxm_hasmask=0 and 'ofs' is a starting bit offset within that field. The
764 * meaning of the flow_mod_spec depends on 'dst':
765 *
766 * - If 'dst' is 0, the flow_mod_spec specifies match criteria for the new
767 * flow. The new flow matches only if bits field[ofs:ofs+n_bits-1] in a
768 * packet equal the source bits. 'field' may be any nxm_header with
769 * nxm_hasmask=0 that is allowed in NXT_FLOW_MOD.
770 *
771 * Order is significant. Earlier flow_mod_specs must satisfy any
772 * prerequisites for matching fields specified later, by copying constant
773 * values into prerequisite fields.
774 *
775 * The switch will reject flow_mod_specs that do not satisfy NXM masking
776 * restrictions.
777 *
778 * - If 'dst' is 1, the flow_mod_spec specifies an NXAST_REG_LOAD action for
779 * the new flow. The new flow copies the source bits into
780 * field[ofs:ofs+n_bits-1]. Actions are executed in the same order as the
781 * flow_mod_specs.
782 *
783 * A single NXAST_REG_LOAD action writes no more than 64 bits, so n_bits
784 * greater than 64 yields multiple NXAST_REG_LOAD actions.
785 *
786 * The flow_mod_spec destination spec for 'dst' of 2 (when 'src' is 0) is
787 * empty. It has the following meaning:
788 *
789 * - The flow_mod_spec specifies an OFPAT_OUTPUT action for the new flow.
790 * The new flow outputs to the OpenFlow port specified by the source field.
791 * Of the special output ports with value OFPP_MAX or larger, OFPP_IN_PORT,
792 * OFPP_FLOOD, OFPP_LOCAL, and OFPP_ALL are supported. Other special ports
793 * may not be used.
794 *
795 * Resource Management
796 * -------------------
797 *
798 * A switch has a finite amount of flow table space available for learning.
799 * When this space is exhausted, no new learning table entries will be learned
800 * until some existing flow table entries expire. The controller should be
801 * prepared to handle this by flooding (which can be implemented as a
802 * low-priority flow).
803 *
804 * If a learned flow matches a single TCP stream with a relatively long
805 * timeout, one may make the best of resource constraints by setting
806 * 'fin_idle_timeout' or 'fin_hard_timeout' (both measured in seconds), or
807 * both, to shorter timeouts. When either of these is specified as a nonzero
808 * value, OVS adds a NXAST_FIN_TIMEOUT action, with the specified timeouts, to
809 * the learned flow.
810 *
811 * Examples
812 * --------
813 *
814 * The following examples give a prose description of the flow_mod_specs along
815 * with informal notation for how those would be represented and a hex dump of
816 * the bytes that would be required.
817 *
818 * These examples could work with various nx_action_learn parameters. Typical
819 * values would be idle_timeout=OFP_FLOW_PERMANENT, hard_timeout=60,
820 * priority=OFP_DEFAULT_PRIORITY, flags=0, table_id=10.
821 *
822 * 1. Learn input port based on the source MAC, with lookup into
823 * NXM_NX_REG1[16:31] by resubmit to in_port=99:
824 *
825 * Match on in_port=99:
826 * ovs_be16(src=1, dst=0, n_bits=16), 20 10
827 * ovs_be16(99), 00 63
828 * ovs_be32(NXM_OF_IN_PORT), ovs_be16(0) 00 00 00 02 00 00
829 *
830 * Match Ethernet destination on Ethernet source from packet:
831 * ovs_be16(src=0, dst=0, n_bits=48), 00 30
832 * ovs_be32(NXM_OF_ETH_SRC), ovs_be16(0) 00 00 04 06 00 00
833 * ovs_be32(NXM_OF_ETH_DST), ovs_be16(0) 00 00 02 06 00 00
834 *
835 * Set NXM_NX_REG1[16:31] to the packet's input port:
836 * ovs_be16(src=0, dst=1, n_bits=16), 08 10
837 * ovs_be32(NXM_OF_IN_PORT), ovs_be16(0) 00 00 00 02 00 00
838 * ovs_be32(NXM_NX_REG1), ovs_be16(16) 00 01 02 04 00 10
839 *
840 * Given a packet that arrived on port A with Ethernet source address B,
841 * this would set up the flow "in_port=99, dl_dst=B,
842 * actions=load:A->NXM_NX_REG1[16..31]".
843 *
844 * In syntax accepted by ovs-ofctl, this action is: learn(in_port=99,
845 * NXM_OF_ETH_DST[]=NXM_OF_ETH_SRC[],
846 * load:NXM_OF_IN_PORT[]->NXM_NX_REG1[16..31])
847 *
848 * 2. Output to input port based on the source MAC and VLAN VID, with lookup
849 * into NXM_NX_REG1[16:31]:
850 *
851 * Match on same VLAN ID as packet:
852 * ovs_be16(src=0, dst=0, n_bits=12), 00 0c
853 * ovs_be32(NXM_OF_VLAN_TCI), ovs_be16(0) 00 00 08 02 00 00
854 * ovs_be32(NXM_OF_VLAN_TCI), ovs_be16(0) 00 00 08 02 00 00
855 *
856 * Match Ethernet destination on Ethernet source from packet:
857 * ovs_be16(src=0, dst=0, n_bits=48), 00 30
858 * ovs_be32(NXM_OF_ETH_SRC), ovs_be16(0) 00 00 04 06 00 00
859 * ovs_be32(NXM_OF_ETH_DST), ovs_be16(0) 00 00 02 06 00 00
860 *
861 * Output to the packet's input port:
862 * ovs_be16(src=0, dst=2, n_bits=16), 10 10
863 * ovs_be32(NXM_OF_IN_PORT), ovs_be16(0) 00 00 00 02 00 00
864 *
865 * Given a packet that arrived on port A with Ethernet source address B in
866 * VLAN C, this would set up the flow "dl_dst=B, vlan_vid=C,
867 * actions=output:A".
868 *
869 * In syntax accepted by ovs-ofctl, this action is:
870 * learn(NXM_OF_VLAN_TCI[0..11], NXM_OF_ETH_DST[]=NXM_OF_ETH_SRC[],
871 * output:NXM_OF_IN_PORT[])
872 *
873 * 3. Here's a recipe for a very simple-minded MAC learning switch. It uses a
874 * 10-second MAC expiration time to make it easier to see what's going on
875 *
876 * ovs-vsctl del-controller br0
877 * ovs-ofctl del-flows br0
878 * ovs-ofctl add-flow br0 "table=0 actions=learn(table=1, \
879 hard_timeout=10, NXM_OF_VLAN_TCI[0..11], \
880 NXM_OF_ETH_DST[]=NXM_OF_ETH_SRC[], \
881 output:NXM_OF_IN_PORT[]), resubmit(,1)"
882 * ovs-ofctl add-flow br0 "table=1 priority=0 actions=flood"
883 *
884 * You can then dump the MAC learning table with:
885 *
886 * ovs-ofctl dump-flows br0 table=1
887 *
888 * Usage Advice
889 * ------------
890 *
891 * For best performance, segregate learned flows into a table that is not used
892 * for any other flows except possibly for a lowest-priority "catch-all" flow
893 * (a flow with no match criteria). If different learning actions specify
894 * different match criteria, use different tables for the learned flows.
895 *
896 * The meaning of 'hard_timeout' and 'idle_timeout' can be counterintuitive.
897 * These timeouts apply to the flow that is added, which means that a flow with
898 * an idle timeout will expire when no traffic has been sent *to* the learned
899 * address. This is not usually the intent in MAC learning; instead, we want
900 * the MAC learn entry to expire when no traffic has been sent *from* the
901 * learned address. Use a hard timeout for that.
902 */
903 struct nx_action_learn {
904 ovs_be16 type; /* OFPAT_VENDOR. */
905 ovs_be16 len; /* At least 24. */
906 ovs_be32 vendor; /* NX_VENDOR_ID. */
907 ovs_be16 subtype; /* NXAST_LEARN. */
908 ovs_be16 idle_timeout; /* Idle time before discarding (seconds). */
909 ovs_be16 hard_timeout; /* Max time before discarding (seconds). */
910 ovs_be16 priority; /* Priority level of flow entry. */
911 ovs_be64 cookie; /* Cookie for new flow. */
912 ovs_be16 flags; /* Either 0 or OFPFF_SEND_FLOW_REM. */
913 uint8_t table_id; /* Table to insert flow entry. */
914 uint8_t pad; /* Must be zero. */
915 ovs_be16 fin_idle_timeout; /* Idle timeout after FIN, if nonzero. */
916 ovs_be16 fin_hard_timeout; /* Hard timeout after FIN, if nonzero. */
917 /* Followed by a sequence of flow_mod_spec elements, as described above,
918 * until the end of the action is reached. */
919 };
920 OFP_ASSERT(sizeof(struct nx_action_learn) == 32);
921
922 #define NX_LEARN_N_BITS_MASK 0x3ff
923
924 #define NX_LEARN_SRC_FIELD (0 << 13) /* Copy from field. */
925 #define NX_LEARN_SRC_IMMEDIATE (1 << 13) /* Copy from immediate value. */
926 #define NX_LEARN_SRC_MASK (1 << 13)
927
928 #define NX_LEARN_DST_MATCH (0 << 11) /* Add match criterion. */
929 #define NX_LEARN_DST_LOAD (1 << 11) /* Add NXAST_REG_LOAD action. */
930 #define NX_LEARN_DST_OUTPUT (2 << 11) /* Add OFPAT_OUTPUT action. */
931 #define NX_LEARN_DST_RESERVED (3 << 11) /* Not yet defined. */
932 #define NX_LEARN_DST_MASK (3 << 11)
933
934 /* Action structure for NXAST_FIN_TIMEOUT.
935 *
936 * This action changes the idle timeout or hard timeout, or both, of this
937 * OpenFlow rule when the rule matches a TCP packet with the FIN or RST flag.
938 * When such a packet is observed, the action reduces the rule's idle timeout
939 * to 'fin_idle_timeout' and its hard timeout to 'fin_hard_timeout'. This
940 * action has no effect on an existing timeout that is already shorter than the
941 * one that the action specifies. A 'fin_idle_timeout' or 'fin_hard_timeout'
942 * of zero has no effect on the respective timeout.
943 *
944 * 'fin_idle_timeout' and 'fin_hard_timeout' are measured in seconds.
945 * 'fin_hard_timeout' specifies time since the flow's creation, not since the
946 * receipt of the FIN or RST.
947 *
948 * This is useful for quickly discarding learned TCP flows that otherwise will
949 * take a long time to expire.
950 *
951 * This action is intended for use with an OpenFlow rule that matches only a
952 * single TCP flow. If the rule matches multiple TCP flows (e.g. it wildcards
953 * all TCP traffic, or all TCP traffic to a particular port), then any FIN or
954 * RST in any of those flows will cause the entire OpenFlow rule to expire
955 * early, which is not normally desirable.
956 */
957 struct nx_action_fin_timeout {
958 ovs_be16 type; /* OFPAT_VENDOR. */
959 ovs_be16 len; /* 16. */
960 ovs_be32 vendor; /* NX_VENDOR_ID. */
961 ovs_be16 subtype; /* NXAST_FIN_TIMEOUT. */
962 ovs_be16 fin_idle_timeout; /* New idle timeout, if nonzero. */
963 ovs_be16 fin_hard_timeout; /* New hard timeout, if nonzero. */
964 ovs_be16 pad; /* Must be zero. */
965 };
966 OFP_ASSERT(sizeof(struct nx_action_fin_timeout) == 16);
967 \f
968 /* Action structure for NXAST_BUNDLE and NXAST_BUNDLE_LOAD.
969 *
970 * The bundle actions choose a slave from a supplied list of options.
971 * NXAST_BUNDLE outputs to its selection. NXAST_BUNDLE_LOAD writes its
972 * selection to a register.
973 *
974 * The list of possible slaves follows the nx_action_bundle structure. The size
975 * of each slave is governed by its type as indicated by the 'slave_type'
976 * parameter. The list of slaves should be padded at its end with zeros to make
977 * the total length of the action a multiple of 8.
978 *
979 * Switches infer from the 'slave_type' parameter the size of each slave. All
980 * implementations must support the NXM_OF_IN_PORT 'slave_type' which indicates
981 * that the slaves are OpenFlow port numbers with NXM_LENGTH(NXM_OF_IN_PORT) ==
982 * 2 byte width. Switches should reject actions which indicate unknown or
983 * unsupported slave types.
984 *
985 * Switches use a strategy dictated by the 'algorithm' parameter to choose a
986 * slave. If the switch does not support the specified 'algorithm' parameter,
987 * it should reject the action.
988 *
989 * Several algorithms take into account liveness when selecting slaves. The
990 * liveness of a slave is implementation defined (with one exception), but will
991 * generally take into account things like its carrier status and the results
992 * of any link monitoring protocols which happen to be running on it. In order
993 * to give controllers a place-holder value, the OFPP_NONE port is always
994 * considered live.
995 *
996 * Some slave selection strategies require the use of a hash function, in which
997 * case the 'fields' and 'basis' parameters should be populated. The 'fields'
998 * parameter (one of NX_HASH_FIELDS_*) designates which parts of the flow to
999 * hash. Refer to the definition of "enum nx_hash_fields" for details. The
1000 * 'basis' parameter is used as a universal hash parameter. Different values
1001 * of 'basis' yield different hash results.
1002 *
1003 * The 'zero' parameter at the end of the action structure is reserved for
1004 * future use. Switches are required to reject actions which have nonzero
1005 * bytes in the 'zero' field.
1006 *
1007 * NXAST_BUNDLE actions should have 'ofs_nbits' and 'dst' zeroed. Switches
1008 * should reject actions which have nonzero bytes in either of these fields.
1009 *
1010 * NXAST_BUNDLE_LOAD stores the OpenFlow port number of the selected slave in
1011 * dst[ofs:ofs+n_bits]. The format and semantics of 'dst' and 'ofs_nbits' are
1012 * similar to those for the NXAST_REG_LOAD action. */
1013 struct nx_action_bundle {
1014 ovs_be16 type; /* OFPAT_VENDOR. */
1015 ovs_be16 len; /* Length including slaves. */
1016 ovs_be32 vendor; /* NX_VENDOR_ID. */
1017 ovs_be16 subtype; /* NXAST_BUNDLE or NXAST_BUNDLE_LOAD. */
1018
1019 /* Slave choice algorithm to apply to hash value. */
1020 ovs_be16 algorithm; /* One of NX_BD_ALG_*. */
1021
1022 /* What fields to hash and how. */
1023 ovs_be16 fields; /* One of NX_HASH_FIELDS_*. */
1024 ovs_be16 basis; /* Universal hash parameter. */
1025
1026 ovs_be32 slave_type; /* NXM_OF_IN_PORT. */
1027 ovs_be16 n_slaves; /* Number of slaves. */
1028
1029 ovs_be16 ofs_nbits; /* (ofs << 6) | (n_bits - 1). */
1030 ovs_be32 dst; /* Destination. */
1031
1032 uint8_t zero[4]; /* Reserved. Must be zero. */
1033 };
1034 OFP_ASSERT(sizeof(struct nx_action_bundle) == 32);
1035
1036 /* NXAST_BUNDLE: Bundle slave choice algorithm to apply.
1037 *
1038 * In the descriptions below, 'slaves' is the list of possible slaves in the
1039 * order they appear in the OpenFlow action. */
1040 enum nx_bd_algorithm {
1041 /* Chooses the first live slave listed in the bundle.
1042 *
1043 * O(n_slaves) performance. */
1044 NX_BD_ALG_ACTIVE_BACKUP,
1045
1046 /* for i in [0,n_slaves):
1047 * weights[i] = hash(flow, i)
1048 * slave = { slaves[i] such that weights[i] >= weights[j] for all j != i }
1049 *
1050 * Redistributes 1/n_slaves of traffic when a slave's liveness changes.
1051 * O(n_slaves) performance.
1052 *
1053 * Uses the 'fields' and 'basis' parameters. */
1054 NX_BD_ALG_HRW /* Highest Random Weight. */
1055 };
1056
1057 \f
1058 /* Action structure for NXAST_DEC_TTL_CNT_IDS.
1059 *
1060 * If the packet is not IPv4 or IPv6, does nothing. For IPv4 or IPv6, if the
1061 * TTL or hop limit is at least 2, decrements it by 1. Otherwise, if TTL or
1062 * hop limit is 0 or 1, sends a packet-in to the controllers with each of the
1063 * 'n_controllers' controller IDs specified in 'cnt_ids'.
1064 *
1065 * (This differs from NXAST_DEC_TTL in that for NXAST_DEC_TTL the packet-in is
1066 * sent only to controllers with id 0.)
1067 */
1068 struct nx_action_cnt_ids {
1069 ovs_be16 type; /* OFPAT_VENDOR. */
1070 ovs_be16 len; /* Length including slaves. */
1071 ovs_be32 vendor; /* NX_VENDOR_ID. */
1072 ovs_be16 subtype; /* NXAST_DEC_TTL_CNT_IDS. */
1073
1074 ovs_be16 n_controllers; /* Number of controllers. */
1075 uint8_t zeros[4]; /* Must be zero. */
1076
1077 /* Followed by 1 or more controller ids.
1078 *
1079 * uint16_t cnt_ids[]; // Controller ids.
1080 * uint8_t pad[]; // Must be 0 to 8-byte align cnt_ids[].
1081 */
1082 };
1083 OFP_ASSERT(sizeof(struct nx_action_cnt_ids) == 16);
1084
1085 \f
1086 /* Action structure for NXAST_OUTPUT_REG.
1087 *
1088 * Outputs to the OpenFlow port number written to src[ofs:ofs+nbits].
1089 *
1090 * The format and semantics of 'src' and 'ofs_nbits' are similar to those for
1091 * the NXAST_REG_LOAD action.
1092 *
1093 * The acceptable nxm_header values for 'src' are the same as the acceptable
1094 * nxm_header values for the 'src' field of NXAST_REG_MOVE.
1095 *
1096 * The 'max_len' field indicates the number of bytes to send when the chosen
1097 * port is OFPP_CONTROLLER. Its semantics are equivalent to the 'max_len'
1098 * field of OFPAT_OUTPUT.
1099 *
1100 * The 'zero' field is required to be zeroed for forward compatibility. */
1101 struct nx_action_output_reg {
1102 ovs_be16 type; /* OFPAT_VENDOR. */
1103 ovs_be16 len; /* 24. */
1104 ovs_be32 vendor; /* NX_VENDOR_ID. */
1105 ovs_be16 subtype; /* NXAST_OUTPUT_REG. */
1106
1107 ovs_be16 ofs_nbits; /* (ofs << 6) | (n_bits - 1). */
1108 ovs_be32 src; /* Source. */
1109
1110 ovs_be16 max_len; /* Max length to send to controller. */
1111
1112 uint8_t zero[6]; /* Reserved, must be zero. */
1113 };
1114 OFP_ASSERT(sizeof(struct nx_action_output_reg) == 24);
1115 \f
1116 /* NXAST_EXIT
1117 *
1118 * Discontinues action processing.
1119 *
1120 * The NXAST_EXIT action causes the switch to immediately halt processing
1121 * actions for the flow. Any actions which have already been processed are
1122 * executed by the switch. However, any further actions, including those which
1123 * may be in different tables, or different levels of the NXAST_RESUBMIT
1124 * hierarchy, will be ignored.
1125 *
1126 * Uses the nx_action_header structure. */
1127 \f
1128 /* Flexible flow specifications (aka NXM = Nicira Extended Match).
1129 *
1130 * OpenFlow 1.0 has "struct ofp10_match" for specifying flow matches. This
1131 * structure is fixed-length and hence difficult to extend. This section
1132 * describes a more flexible, variable-length flow match, called "nx_match" for
1133 * short, that is also supported by Open vSwitch. This section also defines a
1134 * replacement for each OpenFlow message that includes struct ofp10_match.
1135 *
1136 *
1137 * Format
1138 * ======
1139 *
1140 * An nx_match is a sequence of zero or more "nxm_entry"s, which are
1141 * type-length-value (TLV) entries, each 5 to 259 (inclusive) bytes long.
1142 * "nxm_entry"s are not aligned on or padded to any multibyte boundary. The
1143 * first 4 bytes of an nxm_entry are its "header", followed by the entry's
1144 * "body".
1145 *
1146 * An nxm_entry's header is interpreted as a 32-bit word in network byte order:
1147 *
1148 * |<-------------------- nxm_type ------------------>|
1149 * | |
1150 * |31 16 15 9| 8 7 0
1151 * +----------------------------------+---------------+--+------------------+
1152 * | nxm_vendor | nxm_field |hm| nxm_length |
1153 * +----------------------------------+---------------+--+------------------+
1154 *
1155 * The most-significant 23 bits of the header are collectively "nxm_type".
1156 * Bits 16...31 are "nxm_vendor", one of the NXM_VENDOR_* values below. Bits
1157 * 9...15 are "nxm_field", which is a vendor-specific value. nxm_type normally
1158 * designates a protocol header, such as the Ethernet type, but it can also
1159 * refer to packet metadata, such as the switch port on which a packet arrived.
1160 *
1161 * Bit 8 is "nxm_hasmask" (labeled "hm" above for space reasons). The meaning
1162 * of this bit is explained later.
1163 *
1164 * The least-significant 8 bits are "nxm_length", a positive integer. The
1165 * length of the nxm_entry, including the header, is exactly 4 + nxm_length
1166 * bytes.
1167 *
1168 * For a given nxm_vendor, nxm_field, and nxm_hasmask value, nxm_length is a
1169 * constant. It is included only to allow software to minimally parse
1170 * "nxm_entry"s of unknown types. (Similarly, for a given nxm_vendor,
1171 * nxm_field, and nxm_length, nxm_hasmask is a constant.)
1172 *
1173 *
1174 * Semantics
1175 * =========
1176 *
1177 * A zero-length nx_match (one with no "nxm_entry"s) matches every packet.
1178 *
1179 * An nxm_entry places a constraint on the packets matched by the nx_match:
1180 *
1181 * - If nxm_hasmask is 0, the nxm_entry's body contains a value for the
1182 * field, called "nxm_value". The nx_match matches only packets in which
1183 * the field equals nxm_value.
1184 *
1185 * - If nxm_hasmask is 1, then the nxm_entry's body contains a value for the
1186 * field (nxm_value), followed by a bitmask of the same length as the
1187 * value, called "nxm_mask". For each 1-bit in position J in nxm_mask, the
1188 * nx_match matches only packets for which bit J in the given field's value
1189 * matches bit J in nxm_value. A 0-bit in nxm_mask causes the
1190 * corresponding bit in nxm_value is ignored (it should be 0; Open vSwitch
1191 * may enforce this someday), as is the corresponding bit in the field's
1192 * value. (The sense of the nxm_mask bits is the opposite of that used by
1193 * the "wildcards" member of struct ofp10_match.)
1194 *
1195 * When nxm_hasmask is 1, nxm_length is always even.
1196 *
1197 * An all-zero-bits nxm_mask is equivalent to omitting the nxm_entry
1198 * entirely. An all-one-bits nxm_mask is equivalent to specifying 0 for
1199 * nxm_hasmask.
1200 *
1201 * When there are multiple "nxm_entry"s, all of the constraints must be met.
1202 *
1203 *
1204 * Mask Restrictions
1205 * =================
1206 *
1207 * Masks may be restricted:
1208 *
1209 * - Some nxm_types may not support masked wildcards, that is, nxm_hasmask
1210 * must always be 0 when these fields are specified. For example, the
1211 * field that identifies the port on which a packet was received may not be
1212 * masked.
1213 *
1214 * - Some nxm_types that do support masked wildcards may only support certain
1215 * nxm_mask patterns. For example, fields that have IPv4 address values
1216 * may be restricted to CIDR masks.
1217 *
1218 * These restrictions should be noted in specifications for individual fields.
1219 * A switch may accept an nxm_hasmask or nxm_mask value that the specification
1220 * disallows, if the switch correctly implements support for that nxm_hasmask
1221 * or nxm_mask value. A switch must reject an attempt to set up a flow that
1222 * contains a nxm_hasmask or nxm_mask value that it does not support.
1223 *
1224 *
1225 * Prerequisite Restrictions
1226 * =========================
1227 *
1228 * The presence of an nxm_entry with a given nxm_type may be restricted based
1229 * on the presence of or values of other "nxm_entry"s. For example:
1230 *
1231 * - An nxm_entry for nxm_type=NXM_OF_IP_TOS is allowed only if it is
1232 * preceded by another entry with nxm_type=NXM_OF_ETH_TYPE, nxm_hasmask=0,
1233 * and nxm_value=0x0800. That is, matching on the IP source address is
1234 * allowed only if the Ethernet type is explicitly set to IP.
1235 *
1236 * - An nxm_entry for nxm_type=NXM_OF_TCP_SRC is allowed only if it is
1237 * preceded by an entry with nxm_type=NXM_OF_ETH_TYPE, nxm_hasmask=0, and
1238 * nxm_value either 0x0800 or 0x86dd, and another with
1239 * nxm_type=NXM_OF_IP_PROTO, nxm_hasmask=0, nxm_value=6, in that order.
1240 * That is, matching on the TCP source port is allowed only if the Ethernet
1241 * type is IP or IPv6 and the IP protocol is TCP.
1242 *
1243 * These restrictions should be noted in specifications for individual fields.
1244 * A switch may implement relaxed versions of these restrictions. A switch
1245 * must reject an attempt to set up a flow that violates its restrictions.
1246 *
1247 *
1248 * Ordering Restrictions
1249 * =====================
1250 *
1251 * An nxm_entry that has prerequisite restrictions must appear after the
1252 * "nxm_entry"s for its prerequisites. Ordering of "nxm_entry"s within an
1253 * nx_match is not otherwise constrained.
1254 *
1255 * Any given nxm_type may appear in an nx_match at most once.
1256 *
1257 *
1258 * nxm_entry Examples
1259 * ==================
1260 *
1261 * These examples show the format of a single nxm_entry with particular
1262 * nxm_hasmask and nxm_length values. The diagrams are labeled with field
1263 * numbers and byte indexes.
1264 *
1265 *
1266 * 8-bit nxm_value, nxm_hasmask=1, nxm_length=2:
1267 *
1268 * 0 3 4 5
1269 * +------------+---+---+
1270 * | header | v | m |
1271 * +------------+---+---+
1272 *
1273 *
1274 * 16-bit nxm_value, nxm_hasmask=0, nxm_length=2:
1275 *
1276 * 0 3 4 5
1277 * +------------+------+
1278 * | header | value|
1279 * +------------+------+
1280 *
1281 *
1282 * 32-bit nxm_value, nxm_hasmask=0, nxm_length=4:
1283 *
1284 * 0 3 4 7
1285 * +------------+-------------+
1286 * | header | nxm_value |
1287 * +------------+-------------+
1288 *
1289 *
1290 * 48-bit nxm_value, nxm_hasmask=0, nxm_length=6:
1291 *
1292 * 0 3 4 9
1293 * +------------+------------------+
1294 * | header | nxm_value |
1295 * +------------+------------------+
1296 *
1297 *
1298 * 48-bit nxm_value, nxm_hasmask=1, nxm_length=12:
1299 *
1300 * 0 3 4 9 10 15
1301 * +------------+------------------+------------------+
1302 * | header | nxm_value | nxm_mask |
1303 * +------------+------------------+------------------+
1304 *
1305 *
1306 * Error Reporting
1307 * ===============
1308 *
1309 * A switch should report an error in an nx_match using error type
1310 * OFPET_BAD_REQUEST and one of the NXBRC_NXM_* codes. Ideally the switch
1311 * should report a specific error code, if one is assigned for the particular
1312 * problem, but NXBRC_NXM_INVALID is also available to report a generic
1313 * nx_match error.
1314 */
1315
1316 #define NXM_HEADER__(VENDOR, FIELD, HASMASK, LENGTH) \
1317 (((VENDOR) << 16) | ((FIELD) << 9) | ((HASMASK) << 8) | (LENGTH))
1318 #define NXM_HEADER(VENDOR, FIELD, LENGTH) \
1319 NXM_HEADER__(VENDOR, FIELD, 0, LENGTH)
1320 #define NXM_HEADER_W(VENDOR, FIELD, LENGTH) \
1321 NXM_HEADER__(VENDOR, FIELD, 1, (LENGTH) * 2)
1322 #define NXM_VENDOR(HEADER) ((HEADER) >> 16)
1323 #define NXM_FIELD(HEADER) (((HEADER) >> 9) & 0x7f)
1324 #define NXM_TYPE(HEADER) (((HEADER) >> 9) & 0x7fffff)
1325 #define NXM_HASMASK(HEADER) (((HEADER) >> 8) & 1)
1326 #define NXM_LENGTH(HEADER) ((HEADER) & 0xff)
1327
1328 #define NXM_MAKE_WILD_HEADER(HEADER) \
1329 NXM_HEADER_W(NXM_VENDOR(HEADER), NXM_FIELD(HEADER), NXM_LENGTH(HEADER))
1330
1331 /* ## ------------------------------- ## */
1332 /* ## OpenFlow 1.0-compatible fields. ## */
1333 /* ## ------------------------------- ## */
1334
1335 /* Physical or virtual port on which the packet was received.
1336 *
1337 * Prereqs: None.
1338 *
1339 * Format: 16-bit integer in network byte order.
1340 *
1341 * Masking: Not maskable. */
1342 #define NXM_OF_IN_PORT NXM_HEADER (0x0000, 0, 2)
1343
1344 /* Source or destination address in Ethernet header.
1345 *
1346 * Prereqs: None.
1347 *
1348 * Format: 48-bit Ethernet MAC address.
1349 *
1350 * Masking: Fully maskable, in versions 1.8 and later. Earlier versions only
1351 * supported the following masks for NXM_OF_ETH_DST_W: 00:00:00:00:00:00,
1352 * fe:ff:ff:ff:ff:ff, 01:00:00:00:00:00, ff:ff:ff:ff:ff:ff. */
1353 #define NXM_OF_ETH_DST NXM_HEADER (0x0000, 1, 6)
1354 #define NXM_OF_ETH_DST_W NXM_HEADER_W(0x0000, 1, 6)
1355 #define NXM_OF_ETH_SRC NXM_HEADER (0x0000, 2, 6)
1356 #define NXM_OF_ETH_SRC_W NXM_HEADER_W(0x0000, 2, 6)
1357
1358 /* Packet's Ethernet type.
1359 *
1360 * For an Ethernet II packet this is taken from the Ethernet header. For an
1361 * 802.2 LLC+SNAP header with OUI 00-00-00 this is taken from the SNAP header.
1362 * A packet that has neither format has value 0x05ff
1363 * (OFP_DL_TYPE_NOT_ETH_TYPE).
1364 *
1365 * For a packet with an 802.1Q header, this is the type of the encapsulated
1366 * frame.
1367 *
1368 * Prereqs: None.
1369 *
1370 * Format: 16-bit integer in network byte order.
1371 *
1372 * Masking: Not maskable. */
1373 #define NXM_OF_ETH_TYPE NXM_HEADER (0x0000, 3, 2)
1374
1375 /* 802.1Q TCI.
1376 *
1377 * For a packet with an 802.1Q header, this is the Tag Control Information
1378 * (TCI) field, with the CFI bit forced to 1. For a packet with no 802.1Q
1379 * header, this has value 0.
1380 *
1381 * Prereqs: None.
1382 *
1383 * Format: 16-bit integer in network byte order.
1384 *
1385 * Masking: Arbitrary masks.
1386 *
1387 * This field can be used in various ways:
1388 *
1389 * - If it is not constrained at all, the nx_match matches packets without
1390 * an 802.1Q header or with an 802.1Q header that has any TCI value.
1391 *
1392 * - Testing for an exact match with 0 matches only packets without an
1393 * 802.1Q header.
1394 *
1395 * - Testing for an exact match with a TCI value with CFI=1 matches packets
1396 * that have an 802.1Q header with a specified VID and PCP.
1397 *
1398 * - Testing for an exact match with a nonzero TCI value with CFI=0 does
1399 * not make sense. The switch may reject this combination.
1400 *
1401 * - Testing with a specific VID and CFI=1, with nxm_mask=0x1fff, matches
1402 * packets that have an 802.1Q header with that VID (and any PCP).
1403 *
1404 * - Testing with a specific PCP and CFI=1, with nxm_mask=0xf000, matches
1405 * packets that have an 802.1Q header with that PCP (and any VID).
1406 *
1407 * - Testing with nxm_value=0, nxm_mask=0x0fff matches packets with no 802.1Q
1408 * header or with an 802.1Q header with a VID of 0.
1409 *
1410 * - Testing with nxm_value=0, nxm_mask=0xe000 matches packets with no 802.1Q
1411 * header or with an 802.1Q header with a PCP of 0.
1412 *
1413 * - Testing with nxm_value=0, nxm_mask=0xefff matches packets with no 802.1Q
1414 * header or with an 802.1Q header with both VID and PCP of 0.
1415 */
1416 #define NXM_OF_VLAN_TCI NXM_HEADER (0x0000, 4, 2)
1417 #define NXM_OF_VLAN_TCI_W NXM_HEADER_W(0x0000, 4, 2)
1418
1419 /* The "type of service" byte of the IP header, with the ECN bits forced to 0.
1420 *
1421 * Prereqs: NXM_OF_ETH_TYPE must be either 0x0800 or 0x86dd.
1422 *
1423 * Format: 8-bit integer with 2 least-significant bits forced to 0.
1424 *
1425 * Masking: Not maskable. */
1426 #define NXM_OF_IP_TOS NXM_HEADER (0x0000, 5, 1)
1427
1428 /* The "protocol" byte in the IP header.
1429 *
1430 * Prereqs: NXM_OF_ETH_TYPE must be either 0x0800 or 0x86dd.
1431 *
1432 * Format: 8-bit integer.
1433 *
1434 * Masking: Not maskable. */
1435 #define NXM_OF_IP_PROTO NXM_HEADER (0x0000, 6, 1)
1436
1437 /* The source or destination address in the IP header.
1438 *
1439 * Prereqs: NXM_OF_ETH_TYPE must match 0x0800 exactly.
1440 *
1441 * Format: 32-bit integer in network byte order.
1442 *
1443 * Masking: Fully maskable, in Open vSwitch 1.8 and later. In earlier
1444 * versions, only CIDR masks are allowed, that is, masks that consist of N
1445 * high-order bits set to 1 and the other 32-N bits set to 0. */
1446 #define NXM_OF_IP_SRC NXM_HEADER (0x0000, 7, 4)
1447 #define NXM_OF_IP_SRC_W NXM_HEADER_W(0x0000, 7, 4)
1448 #define NXM_OF_IP_DST NXM_HEADER (0x0000, 8, 4)
1449 #define NXM_OF_IP_DST_W NXM_HEADER_W(0x0000, 8, 4)
1450
1451 /* The source or destination port in the TCP header.
1452 *
1453 * Prereqs:
1454 * NXM_OF_ETH_TYPE must be either 0x0800 or 0x86dd.
1455 * NXM_OF_IP_PROTO must match 6 exactly.
1456 *
1457 * Format: 16-bit integer in network byte order.
1458 *
1459 * Masking: Fully maskable, in Open vSwitch 1.6 and later. Not maskable, in
1460 * earlier versions. */
1461 #define NXM_OF_TCP_SRC NXM_HEADER (0x0000, 9, 2)
1462 #define NXM_OF_TCP_SRC_W NXM_HEADER_W(0x0000, 9, 2)
1463 #define NXM_OF_TCP_DST NXM_HEADER (0x0000, 10, 2)
1464 #define NXM_OF_TCP_DST_W NXM_HEADER_W(0x0000, 10, 2)
1465
1466 /* The source or destination port in the UDP header.
1467 *
1468 * Prereqs:
1469 * NXM_OF_ETH_TYPE must match either 0x0800 or 0x86dd.
1470 * NXM_OF_IP_PROTO must match 17 exactly.
1471 *
1472 * Format: 16-bit integer in network byte order.
1473 *
1474 * Masking: Fully maskable, in Open vSwitch 1.6 and later. Not maskable, in
1475 * earlier versions. */
1476 #define NXM_OF_UDP_SRC NXM_HEADER (0x0000, 11, 2)
1477 #define NXM_OF_UDP_SRC_W NXM_HEADER_W(0x0000, 11, 2)
1478 #define NXM_OF_UDP_DST NXM_HEADER (0x0000, 12, 2)
1479 #define NXM_OF_UDP_DST_W NXM_HEADER_W(0x0000, 12, 2)
1480
1481 /* The type or code in the ICMP header.
1482 *
1483 * Prereqs:
1484 * NXM_OF_ETH_TYPE must match 0x0800 exactly.
1485 * NXM_OF_IP_PROTO must match 1 exactly.
1486 *
1487 * Format: 8-bit integer.
1488 *
1489 * Masking: Not maskable. */
1490 #define NXM_OF_ICMP_TYPE NXM_HEADER (0x0000, 13, 1)
1491 #define NXM_OF_ICMP_CODE NXM_HEADER (0x0000, 14, 1)
1492
1493 /* ARP opcode.
1494 *
1495 * For an Ethernet+IP ARP packet, the opcode in the ARP header. Always 0
1496 * otherwise. Only ARP opcodes between 1 and 255 should be specified for
1497 * matching.
1498 *
1499 * Prereqs: NXM_OF_ETH_TYPE must match either 0x0806 or 0x8035.
1500 *
1501 * Format: 16-bit integer in network byte order.
1502 *
1503 * Masking: Not maskable. */
1504 #define NXM_OF_ARP_OP NXM_HEADER (0x0000, 15, 2)
1505
1506 /* For an Ethernet+IP ARP packet, the source or target protocol address
1507 * in the ARP header. Always 0 otherwise.
1508 *
1509 * Prereqs: NXM_OF_ETH_TYPE must match either 0x0806 or 0x8035.
1510 *
1511 * Format: 32-bit integer in network byte order.
1512 *
1513 * Masking: Fully maskable, in Open vSwitch 1.8 and later. In earlier
1514 * versions, only CIDR masks are allowed, that is, masks that consist of N
1515 * high-order bits set to 1 and the other 32-N bits set to 0. */
1516 #define NXM_OF_ARP_SPA NXM_HEADER (0x0000, 16, 4)
1517 #define NXM_OF_ARP_SPA_W NXM_HEADER_W(0x0000, 16, 4)
1518 #define NXM_OF_ARP_TPA NXM_HEADER (0x0000, 17, 4)
1519 #define NXM_OF_ARP_TPA_W NXM_HEADER_W(0x0000, 17, 4)
1520
1521 /* ## ------------------------ ## */
1522 /* ## Nicira match extensions. ## */
1523 /* ## ------------------------ ## */
1524
1525 /* Metadata registers.
1526 *
1527 * Registers initially have value 0. Actions allow register values to be
1528 * manipulated.
1529 *
1530 * Prereqs: None.
1531 *
1532 * Format: Array of 32-bit integer registers. Space is reserved for up to
1533 * NXM_NX_MAX_REGS registers, but switches may implement fewer.
1534 *
1535 * Masking: Arbitrary masks. */
1536 #define NXM_NX_MAX_REGS 16
1537 #define NXM_NX_REG(IDX) NXM_HEADER (0x0001, IDX, 4)
1538 #define NXM_NX_REG_W(IDX) NXM_HEADER_W(0x0001, IDX, 4)
1539 #define NXM_NX_REG_IDX(HEADER) NXM_FIELD(HEADER)
1540 #define NXM_IS_NX_REG(HEADER) (!((((HEADER) ^ NXM_NX_REG0)) & 0xffffe1ff))
1541 #define NXM_IS_NX_REG_W(HEADER) (!((((HEADER) ^ NXM_NX_REG0_W)) & 0xffffe1ff))
1542 #define NXM_NX_REG0 NXM_HEADER (0x0001, 0, 4)
1543 #define NXM_NX_REG0_W NXM_HEADER_W(0x0001, 0, 4)
1544 #define NXM_NX_REG1 NXM_HEADER (0x0001, 1, 4)
1545 #define NXM_NX_REG1_W NXM_HEADER_W(0x0001, 1, 4)
1546 #define NXM_NX_REG2 NXM_HEADER (0x0001, 2, 4)
1547 #define NXM_NX_REG2_W NXM_HEADER_W(0x0001, 2, 4)
1548 #define NXM_NX_REG3 NXM_HEADER (0x0001, 3, 4)
1549 #define NXM_NX_REG3_W NXM_HEADER_W(0x0001, 3, 4)
1550 #define NXM_NX_REG4 NXM_HEADER (0x0001, 4, 4)
1551 #define NXM_NX_REG4_W NXM_HEADER_W(0x0001, 4, 4)
1552 #define NXM_NX_REG5 NXM_HEADER (0x0001, 5, 4)
1553 #define NXM_NX_REG5_W NXM_HEADER_W(0x0001, 5, 4)
1554 #define NXM_NX_REG6 NXM_HEADER (0x0001, 6, 4)
1555 #define NXM_NX_REG6_W NXM_HEADER_W(0x0001, 6, 4)
1556 #define NXM_NX_REG7 NXM_HEADER (0x0001, 7, 4)
1557 #define NXM_NX_REG7_W NXM_HEADER_W(0x0001, 7, 4)
1558
1559 /* Tunnel ID.
1560 *
1561 * For a packet received via a GRE, VXLAN or LISP tunnel including a (32-bit)
1562 * key, the key is stored in the low 32-bits and the high bits are zeroed. For
1563 * other packets, the value is 0.
1564 *
1565 * All zero bits, for packets not received via a keyed tunnel.
1566 *
1567 * Prereqs: None.
1568 *
1569 * Format: 64-bit integer in network byte order.
1570 *
1571 * Masking: Arbitrary masks. */
1572 #define NXM_NX_TUN_ID NXM_HEADER (0x0001, 16, 8)
1573 #define NXM_NX_TUN_ID_W NXM_HEADER_W(0x0001, 16, 8)
1574
1575 /* For an Ethernet+IP ARP packet, the source or target hardware address
1576 * in the ARP header. Always 0 otherwise.
1577 *
1578 * Prereqs: NXM_OF_ETH_TYPE must match either 0x0806 or 0x8035.
1579 *
1580 * Format: 48-bit Ethernet MAC address.
1581 *
1582 * Masking: Not maskable. */
1583 #define NXM_NX_ARP_SHA NXM_HEADER (0x0001, 17, 6)
1584 #define NXM_NX_ARP_THA NXM_HEADER (0x0001, 18, 6)
1585
1586 /* The source or destination address in the IPv6 header.
1587 *
1588 * Prereqs: NXM_OF_ETH_TYPE must match 0x86dd exactly.
1589 *
1590 * Format: 128-bit IPv6 address.
1591 *
1592 * Masking: Fully maskable, in Open vSwitch 1.8 and later. In previous
1593 * versions, only CIDR masks are allowed, that is, masks that consist of N
1594 * high-order bits set to 1 and the other 128-N bits set to 0. */
1595 #define NXM_NX_IPV6_SRC NXM_HEADER (0x0001, 19, 16)
1596 #define NXM_NX_IPV6_SRC_W NXM_HEADER_W(0x0001, 19, 16)
1597 #define NXM_NX_IPV6_DST NXM_HEADER (0x0001, 20, 16)
1598 #define NXM_NX_IPV6_DST_W NXM_HEADER_W(0x0001, 20, 16)
1599
1600 /* The type or code in the ICMPv6 header.
1601 *
1602 * Prereqs:
1603 * NXM_OF_ETH_TYPE must match 0x86dd exactly.
1604 * NXM_OF_IP_PROTO must match 58 exactly.
1605 *
1606 * Format: 8-bit integer.
1607 *
1608 * Masking: Not maskable. */
1609 #define NXM_NX_ICMPV6_TYPE NXM_HEADER (0x0001, 21, 1)
1610 #define NXM_NX_ICMPV6_CODE NXM_HEADER (0x0001, 22, 1)
1611
1612 /* The target address in an IPv6 Neighbor Discovery message.
1613 *
1614 * Prereqs:
1615 * NXM_OF_ETH_TYPE must match 0x86dd exactly.
1616 * NXM_OF_IP_PROTO must match 58 exactly.
1617 * NXM_OF_ICMPV6_TYPE must be either 135 or 136.
1618 *
1619 * Format: 128-bit IPv6 address.
1620 *
1621 * Masking: Fully maskable, in Open vSwitch 1.8 and later. In previous
1622 * versions, only CIDR masks are allowed, that is, masks that consist of N
1623 * high-order bits set to 1 and the other 128-N bits set to 0. */
1624 #define NXM_NX_ND_TARGET NXM_HEADER (0x0001, 23, 16)
1625 #define NXM_NX_ND_TARGET_W NXM_HEADER_W (0x0001, 23, 16)
1626
1627 /* The source link-layer address option in an IPv6 Neighbor Discovery
1628 * message.
1629 *
1630 * Prereqs:
1631 * NXM_OF_ETH_TYPE must match 0x86dd exactly.
1632 * NXM_OF_IP_PROTO must match 58 exactly.
1633 * NXM_OF_ICMPV6_TYPE must be exactly 135.
1634 *
1635 * Format: 48-bit Ethernet MAC address.
1636 *
1637 * Masking: Not maskable. */
1638 #define NXM_NX_ND_SLL NXM_HEADER (0x0001, 24, 6)
1639
1640 /* The target link-layer address option in an IPv6 Neighbor Discovery
1641 * message.
1642 *
1643 * Prereqs:
1644 * NXM_OF_ETH_TYPE must match 0x86dd exactly.
1645 * NXM_OF_IP_PROTO must match 58 exactly.
1646 * NXM_OF_ICMPV6_TYPE must be exactly 136.
1647 *
1648 * Format: 48-bit Ethernet MAC address.
1649 *
1650 * Masking: Not maskable. */
1651 #define NXM_NX_ND_TLL NXM_HEADER (0x0001, 25, 6)
1652
1653 /* IP fragment information.
1654 *
1655 * Prereqs:
1656 * NXM_OF_ETH_TYPE must be either 0x0800 or 0x86dd.
1657 *
1658 * Format: 8-bit value with one of the values 0, 1, or 3, as described below.
1659 *
1660 * Masking: Fully maskable.
1661 *
1662 * This field has three possible values:
1663 *
1664 * - A packet that is not an IP fragment has value 0.
1665 *
1666 * - A packet that is an IP fragment with offset 0 (the first fragment) has
1667 * bit 0 set and thus value 1.
1668 *
1669 * - A packet that is an IP fragment with nonzero offset has bits 0 and 1 set
1670 * and thus value 3.
1671 *
1672 * NX_IP_FRAG_ANY and NX_IP_FRAG_LATER are declared to symbolically represent
1673 * the meanings of bits 0 and 1.
1674 *
1675 * The switch may reject matches against values that can never appear.
1676 *
1677 * It is important to understand how this field interacts with the OpenFlow IP
1678 * fragment handling mode:
1679 *
1680 * - In OFPC_FRAG_DROP mode, the OpenFlow switch drops all IP fragments
1681 * before they reach the flow table, so every packet that is available for
1682 * matching will have value 0 in this field.
1683 *
1684 * - Open vSwitch does not implement OFPC_FRAG_REASM mode, but if it did then
1685 * IP fragments would be reassembled before they reached the flow table and
1686 * again every packet available for matching would always have value 0.
1687 *
1688 * - In OFPC_FRAG_NORMAL mode, all three values are possible, but OpenFlow
1689 * 1.0 says that fragments' transport ports are always 0, even for the
1690 * first fragment, so this does not provide much extra information.
1691 *
1692 * - In OFPC_FRAG_NX_MATCH mode, all three values are possible. For
1693 * fragments with offset 0, Open vSwitch makes L4 header information
1694 * available.
1695 */
1696 #define NXM_NX_IP_FRAG NXM_HEADER (0x0001, 26, 1)
1697 #define NXM_NX_IP_FRAG_W NXM_HEADER_W(0x0001, 26, 1)
1698
1699 /* Bits in the value of NXM_NX_IP_FRAG. */
1700 #define NX_IP_FRAG_ANY (1 << 0) /* Is this a fragment? */
1701 #define NX_IP_FRAG_LATER (1 << 1) /* Is this a fragment with nonzero offset? */
1702
1703 /* The flow label in the IPv6 header.
1704 *
1705 * Prereqs: NXM_OF_ETH_TYPE must match 0x86dd exactly.
1706 *
1707 * Format: 20-bit IPv6 flow label in least-significant bits.
1708 *
1709 * Masking: Not maskable. */
1710 #define NXM_NX_IPV6_LABEL NXM_HEADER (0x0001, 27, 4)
1711
1712 /* The ECN of the IP header.
1713 *
1714 * Prereqs: NXM_OF_ETH_TYPE must be either 0x0800 or 0x86dd.
1715 *
1716 * Format: ECN in the low-order 2 bits.
1717 *
1718 * Masking: Not maskable. */
1719 #define NXM_NX_IP_ECN NXM_HEADER (0x0001, 28, 1)
1720
1721 /* The time-to-live/hop limit of the IP header.
1722 *
1723 * Prereqs: NXM_OF_ETH_TYPE must be either 0x0800 or 0x86dd.
1724 *
1725 * Format: 8-bit integer.
1726 *
1727 * Masking: Not maskable. */
1728 #define NXM_NX_IP_TTL NXM_HEADER (0x0001, 29, 1)
1729
1730 /* Flow cookie.
1731 *
1732 * This may be used to gain the OpenFlow 1.1-like ability to restrict
1733 * certain NXM-based Flow Mod and Flow Stats Request messages to flows
1734 * with specific cookies. See the "nx_flow_mod" and "nx_flow_stats_request"
1735 * structure definitions for more details. This match is otherwise not
1736 * allowed.
1737 *
1738 * Prereqs: None.
1739 *
1740 * Format: 64-bit integer in network byte order.
1741 *
1742 * Masking: Arbitrary masks. */
1743 #define NXM_NX_COOKIE NXM_HEADER (0x0001, 30, 8)
1744 #define NXM_NX_COOKIE_W NXM_HEADER_W(0x0001, 30, 8)
1745
1746 /* ## --------------------- ## */
1747 /* ## Requests and replies. ## */
1748 /* ## --------------------- ## */
1749
1750 enum nx_flow_format {
1751 NXFF_OPENFLOW10 = 0, /* Standard OpenFlow 1.0 compatible. */
1752 NXFF_NXM = 2 /* Nicira extended match. */
1753 };
1754
1755 /* NXT_SET_FLOW_FORMAT request. */
1756 struct nx_set_flow_format {
1757 ovs_be32 format; /* One of NXFF_*. */
1758 };
1759 OFP_ASSERT(sizeof(struct nx_set_flow_format) == 4);
1760
1761 /* NXT_FLOW_MOD (analogous to OFPT_FLOW_MOD).
1762 *
1763 * It is possible to limit flow deletions and modifications to certain
1764 * cookies by using the NXM_NX_COOKIE(_W) matches. The "cookie" field
1765 * is used only to add or modify flow cookies.
1766 */
1767 struct nx_flow_mod {
1768 ovs_be64 cookie; /* Opaque controller-issued identifier. */
1769 ovs_be16 command; /* One of OFPFC_*. */
1770 ovs_be16 idle_timeout; /* Idle time before discarding (seconds). */
1771 ovs_be16 hard_timeout; /* Max time before discarding (seconds). */
1772 ovs_be16 priority; /* Priority level of flow entry. */
1773 ovs_be32 buffer_id; /* Buffered packet to apply to (or -1).
1774 Not meaningful for OFPFC_DELETE*. */
1775 ovs_be16 out_port; /* For OFPFC_DELETE* commands, require
1776 matching entries to include this as an
1777 output port. A value of OFPP_NONE
1778 indicates no restriction. */
1779 ovs_be16 flags; /* One of OFPFF_*. */
1780 ovs_be16 match_len; /* Size of nx_match. */
1781 uint8_t pad[6]; /* Align to 64-bits. */
1782 /* Followed by:
1783 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
1784 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
1785 * all-zero bytes, then
1786 * - Actions to fill out the remainder of the message length (always a
1787 * multiple of 8).
1788 */
1789 };
1790 OFP_ASSERT(sizeof(struct nx_flow_mod) == 32);
1791
1792 /* NXT_FLOW_REMOVED (analogous to OFPT_FLOW_REMOVED).
1793 *
1794 * 'table_id' is present only in Open vSwitch 1.11 and later. In earlier
1795 * versions of Open vSwitch, this is a padding byte that is always zeroed.
1796 * Therefore, a 'table_id' value of 0 indicates that the table ID is not known,
1797 * and other values may be interpreted as one more than the flow's former table
1798 * ID. */
1799 struct nx_flow_removed {
1800 ovs_be64 cookie; /* Opaque controller-issued identifier. */
1801 ovs_be16 priority; /* Priority level of flow entry. */
1802 uint8_t reason; /* One of OFPRR_*. */
1803 uint8_t table_id; /* Flow's former table ID, plus one. */
1804 ovs_be32 duration_sec; /* Time flow was alive in seconds. */
1805 ovs_be32 duration_nsec; /* Time flow was alive in nanoseconds beyond
1806 duration_sec. */
1807 ovs_be16 idle_timeout; /* Idle timeout from original flow mod. */
1808 ovs_be16 match_len; /* Size of nx_match. */
1809 ovs_be64 packet_count;
1810 ovs_be64 byte_count;
1811 /* Followed by:
1812 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
1813 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
1814 * all-zero bytes. */
1815 };
1816 OFP_ASSERT(sizeof(struct nx_flow_removed) == 40);
1817
1818 /* Nicira vendor stats request of type NXST_FLOW (analogous to OFPST_FLOW
1819 * request).
1820 *
1821 * It is possible to limit matches to certain cookies by using the
1822 * NXM_NX_COOKIE and NXM_NX_COOKIE_W matches.
1823 */
1824 struct nx_flow_stats_request {
1825 ovs_be16 out_port; /* Require matching entries to include this
1826 as an output port. A value of OFPP_NONE
1827 indicates no restriction. */
1828 ovs_be16 match_len; /* Length of nx_match. */
1829 uint8_t table_id; /* ID of table to read (from ofp_table_stats)
1830 or 0xff for all tables. */
1831 uint8_t pad[3]; /* Align to 64 bits. */
1832 /* Followed by:
1833 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
1834 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
1835 * all-zero bytes, which must also exactly fill out the length of the
1836 * message.
1837 */
1838 };
1839 OFP_ASSERT(sizeof(struct nx_flow_stats_request) == 8);
1840
1841 /* Body for Nicira vendor stats reply of type NXST_FLOW (analogous to
1842 * OFPST_FLOW reply).
1843 *
1844 * The values of 'idle_age' and 'hard_age' are only meaningful when talking to
1845 * a switch that implements the NXT_FLOW_AGE extension. Zero means that the
1846 * true value is unknown, perhaps because hardware does not track the value.
1847 * (Zero is also the value that one should ordinarily expect to see talking to
1848 * a switch that does not implement NXT_FLOW_AGE, since those switches zero the
1849 * padding bytes that these fields replaced.) A nonzero value X represents X-1
1850 * seconds. A value of 65535 represents 65534 or more seconds.
1851 *
1852 * 'idle_age' is the number of seconds that the flow has been idle, that is,
1853 * the number of seconds since a packet passed through the flow. 'hard_age' is
1854 * the number of seconds since the flow was last modified (e.g. OFPFC_MODIFY or
1855 * OFPFC_MODIFY_STRICT). (The 'duration_*' fields are the elapsed time since
1856 * the flow was added, regardless of subsequent modifications.)
1857 *
1858 * For a flow with an idle or hard timeout, 'idle_age' or 'hard_age',
1859 * respectively, will ordinarily be smaller than the timeout, but flow
1860 * expiration times are only approximate and so one must be prepared to
1861 * tolerate expirations that occur somewhat early or late.
1862 */
1863 struct nx_flow_stats {
1864 ovs_be16 length; /* Length of this entry. */
1865 uint8_t table_id; /* ID of table flow came from. */
1866 uint8_t pad;
1867 ovs_be32 duration_sec; /* Time flow has been alive in seconds. */
1868 ovs_be32 duration_nsec; /* Time flow has been alive in nanoseconds
1869 beyond duration_sec. */
1870 ovs_be16 priority; /* Priority of the entry. */
1871 ovs_be16 idle_timeout; /* Number of seconds idle before expiration. */
1872 ovs_be16 hard_timeout; /* Number of seconds before expiration. */
1873 ovs_be16 match_len; /* Length of nx_match. */
1874 ovs_be16 idle_age; /* Seconds since last packet, plus one. */
1875 ovs_be16 hard_age; /* Seconds since last modification, plus one. */
1876 ovs_be64 cookie; /* Opaque controller-issued identifier. */
1877 ovs_be64 packet_count; /* Number of packets, UINT64_MAX if unknown. */
1878 ovs_be64 byte_count; /* Number of bytes, UINT64_MAX if unknown. */
1879 /* Followed by:
1880 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
1881 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
1882 * all-zero bytes, then
1883 * - Actions to fill out the remainder 'length' bytes (always a multiple
1884 * of 8).
1885 */
1886 };
1887 OFP_ASSERT(sizeof(struct nx_flow_stats) == 48);
1888
1889 /* Nicira vendor stats request of type NXST_AGGREGATE (analogous to
1890 * OFPST_AGGREGATE request).
1891 *
1892 * The reply format is identical to the reply format for OFPST_AGGREGATE,
1893 * except for the header. */
1894 struct nx_aggregate_stats_request {
1895 ovs_be16 out_port; /* Require matching entries to include this
1896 as an output port. A value of OFPP_NONE
1897 indicates no restriction. */
1898 ovs_be16 match_len; /* Length of nx_match. */
1899 uint8_t table_id; /* ID of table to read (from ofp_table_stats)
1900 or 0xff for all tables. */
1901 uint8_t pad[3]; /* Align to 64 bits. */
1902 /* Followed by:
1903 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
1904 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
1905 * all-zero bytes, which must also exactly fill out the length of the
1906 * message.
1907 */
1908 };
1909 OFP_ASSERT(sizeof(struct nx_aggregate_stats_request) == 8);
1910 \f
1911 /* NXT_SET_CONTROLLER_ID.
1912 *
1913 * Each OpenFlow controller connection has a 16-bit identifier that is
1914 * initially 0. This message changes the connection's ID to 'id'.
1915 *
1916 * Controller connection IDs need not be unique.
1917 *
1918 * The NXAST_CONTROLLER action is the only current user of controller
1919 * connection IDs. */
1920 struct nx_controller_id {
1921 uint8_t zero[6]; /* Must be zero. */
1922 ovs_be16 controller_id; /* New controller connection ID. */
1923 };
1924 OFP_ASSERT(sizeof(struct nx_controller_id) == 8);
1925
1926 /* Action structure for NXAST_CONTROLLER.
1927 *
1928 * This generalizes using OFPAT_OUTPUT to send a packet to OFPP_CONTROLLER. In
1929 * addition to the 'max_len' that OFPAT_OUTPUT supports, it also allows
1930 * specifying:
1931 *
1932 * - 'reason': The reason code to use in the ofp_packet_in or nx_packet_in.
1933 *
1934 * - 'controller_id': The ID of the controller connection to which the
1935 * ofp_packet_in should be sent. The ofp_packet_in or nx_packet_in is
1936 * sent only to controllers that have the specified controller connection
1937 * ID. See "struct nx_controller_id" for more information. */
1938 struct nx_action_controller {
1939 ovs_be16 type; /* OFPAT_VENDOR. */
1940 ovs_be16 len; /* Length is 16. */
1941 ovs_be32 vendor; /* NX_VENDOR_ID. */
1942 ovs_be16 subtype; /* NXAST_CONTROLLER. */
1943 ovs_be16 max_len; /* Maximum length to send to controller. */
1944 ovs_be16 controller_id; /* Controller ID to send packet-in. */
1945 uint8_t reason; /* enum ofp_packet_in_reason (OFPR_*). */
1946 uint8_t zero; /* Must be zero. */
1947 };
1948 OFP_ASSERT(sizeof(struct nx_action_controller) == 16);
1949 \f
1950 /* Flow Table Monitoring
1951 * =====================
1952 *
1953 * NXST_FLOW_MONITOR allows a controller to keep track of changes to OpenFlow
1954 * flow table(s) or subsets of them, with the following workflow:
1955 *
1956 * 1. The controller sends an NXST_FLOW_MONITOR request to begin monitoring
1957 * flows. The 'id' in the request must be unique among all monitors that
1958 * the controller has started and not yet canceled on this OpenFlow
1959 * connection.
1960 *
1961 * 2. The switch responds with an NXST_FLOW_MONITOR reply. If the request's
1962 * 'flags' included NXFMF_INITIAL, the reply includes all the flows that
1963 * matched the request at the time of the request (with event NXFME_ADDED).
1964 * If 'flags' did not include NXFMF_INITIAL, the reply is empty.
1965 *
1966 * The reply uses the xid of the request (as do all replies to OpenFlow
1967 * requests).
1968 *
1969 * 3. Whenever a change to a flow table entry matches some outstanding monitor
1970 * request's criteria and flags, the switch sends a notification to the
1971 * controller as an additional NXST_FLOW_MONITOR reply with xid 0.
1972 *
1973 * When multiple outstanding monitors match a single change, only a single
1974 * notification is sent. This merged notification includes the information
1975 * requested in any of the individual monitors. That is, if any of the
1976 * matching monitors requests actions (NXFMF_ACTIONS), the notification
1977 * includes actions, and if any of the monitors request full changes for the
1978 * controller's own changes (NXFMF_OWN), the controller's own changes will
1979 * be included in full.
1980 *
1981 * 4. The controller may cancel a monitor with NXT_FLOW_MONITOR_CANCEL. No
1982 * further notifications will be sent on the basis of the canceled monitor
1983 * afterward.
1984 *
1985 *
1986 * Buffer Management
1987 * =================
1988 *
1989 * OpenFlow messages for flow monitor notifications can overflow the buffer
1990 * space available to the switch, either temporarily (e.g. due to network
1991 * conditions slowing OpenFlow traffic) or more permanently (e.g. the sustained
1992 * rate of flow table change exceeds the network bandwidth between switch and
1993 * controller).
1994 *
1995 * When Open vSwitch's notification buffer space reaches a limiting threshold,
1996 * OVS reacts as follows:
1997 *
1998 * 1. OVS sends an NXT_FLOW_MONITOR_PAUSED message to the controller, following
1999 * all the already queued notifications. After it receives this message,
2000 * the controller knows that its view of the flow table, as represented by
2001 * flow monitor notifications, is incomplete.
2002 *
2003 * 2. As long as the notification buffer is not empty:
2004 *
2005 * - NXMFE_ADD and NXFME_MODIFIED notifications will not be sent.
2006 *
2007 * - NXFME_DELETED notifications will still be sent, but only for flows
2008 * that existed before OVS sent NXT_FLOW_MONITOR_PAUSED.
2009 *
2010 * - NXFME_ABBREV notifications will not be sent. They are treated as
2011 * the expanded version (and therefore only the NXFME_DELETED
2012 * components, if any, are sent).
2013 *
2014 * 3. When the notification buffer empties, OVS sends NXFME_ADD notifications
2015 * for flows added since the buffer reached its limit and NXFME_MODIFIED
2016 * notifications for flows that existed before the limit was reached and
2017 * changed after the limit was reached.
2018 *
2019 * 4. OVS sends an NXT_FLOW_MONITOR_RESUMED message to the controller. After
2020 * it receives this message, the controller knows that its view of the flow
2021 * table, as represented by flow monitor notifications, is again complete.
2022 *
2023 * This allows the maximum buffer space requirement for notifications to be
2024 * bounded by the limit plus the maximum number of supported flows.
2025 *
2026 *
2027 * "Flow Removed" messages
2028 * =======================
2029 *
2030 * The flow monitor mechanism is independent of OFPT_FLOW_REMOVED and
2031 * NXT_FLOW_REMOVED. Flow monitor updates for deletion are sent if
2032 * NXFMF_DELETE is set on a monitor, regardless of whether the
2033 * OFPFF_SEND_FLOW_REM flag was set when the flow was added. */
2034
2035 /* NXST_FLOW_MONITOR request.
2036 *
2037 * The NXST_FLOW_MONITOR request's body consists of an array of zero or more
2038 * instances of this structure. The request arranges to monitor the flows
2039 * that match the specified criteria, which are interpreted in the same way as
2040 * for NXST_FLOW.
2041 *
2042 * 'id' identifies a particular monitor for the purpose of allowing it to be
2043 * canceled later with NXT_FLOW_MONITOR_CANCEL. 'id' must be unique among
2044 * existing monitors that have not already been canceled.
2045 *
2046 * The reply includes the initial flow matches for monitors that have the
2047 * NXFMF_INITIAL flag set. No single flow will be included in the reply more
2048 * than once, even if more than one requested monitor matches that flow. The
2049 * reply will be empty if none of the monitors has NXFMF_INITIAL set or if none
2050 * of the monitors initially matches any flows.
2051 *
2052 * For NXFMF_ADD, an event will be reported if 'out_port' matches against the
2053 * actions of the flow being added or, for a flow that is replacing an existing
2054 * flow, if 'out_port' matches against the actions of the flow being replaced.
2055 * For NXFMF_DELETE, 'out_port' matches against the actions of a flow being
2056 * deleted. For NXFMF_MODIFY, an event will be reported if 'out_port' matches
2057 * either the old or the new actions. */
2058 struct nx_flow_monitor_request {
2059 ovs_be32 id; /* Controller-assigned ID for this monitor. */
2060 ovs_be16 flags; /* NXFMF_*. */
2061 ovs_be16 out_port; /* Required output port, if not OFPP_NONE. */
2062 ovs_be16 match_len; /* Length of nx_match. */
2063 uint8_t table_id; /* One table's ID or 0xff for all tables. */
2064 uint8_t zeros[5]; /* Align to 64 bits (must be zero). */
2065 /* Followed by:
2066 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
2067 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
2068 * all-zero bytes. */
2069 };
2070 OFP_ASSERT(sizeof(struct nx_flow_monitor_request) == 16);
2071
2072 /* 'flags' bits in struct nx_flow_monitor_request. */
2073 enum nx_flow_monitor_flags {
2074 /* When to send updates. */
2075 NXFMF_INITIAL = 1 << 0, /* Initially matching flows. */
2076 NXFMF_ADD = 1 << 1, /* New matching flows as they are added. */
2077 NXFMF_DELETE = 1 << 2, /* Old matching flows as they are removed. */
2078 NXFMF_MODIFY = 1 << 3, /* Matching flows as they are changed. */
2079
2080 /* What to include in updates. */
2081 NXFMF_ACTIONS = 1 << 4, /* If set, actions are included. */
2082 NXFMF_OWN = 1 << 5, /* If set, include own changes in full. */
2083 };
2084
2085 /* NXST_FLOW_MONITOR reply header.
2086 *
2087 * The body of an NXST_FLOW_MONITOR reply is an array of variable-length
2088 * structures, each of which begins with this header. The 'length' member may
2089 * be used to traverse the array, and the 'event' member may be used to
2090 * determine the particular structure.
2091 *
2092 * Every instance is a multiple of 8 bytes long. */
2093 struct nx_flow_update_header {
2094 ovs_be16 length; /* Length of this entry. */
2095 ovs_be16 event; /* One of NXFME_*. */
2096 /* ...other data depending on 'event'... */
2097 };
2098 OFP_ASSERT(sizeof(struct nx_flow_update_header) == 4);
2099
2100 /* 'event' values in struct nx_flow_update_header. */
2101 enum nx_flow_update_event {
2102 /* struct nx_flow_update_full. */
2103 NXFME_ADDED = 0, /* Flow was added. */
2104 NXFME_DELETED = 1, /* Flow was deleted. */
2105 NXFME_MODIFIED = 2, /* Flow (generally its actions) was changed. */
2106
2107 /* struct nx_flow_update_abbrev. */
2108 NXFME_ABBREV = 3, /* Abbreviated reply. */
2109 };
2110
2111 /* NXST_FLOW_MONITOR reply for NXFME_ADDED, NXFME_DELETED, and
2112 * NXFME_MODIFIED. */
2113 struct nx_flow_update_full {
2114 ovs_be16 length; /* Length is 24. */
2115 ovs_be16 event; /* One of NXFME_*. */
2116 ovs_be16 reason; /* OFPRR_* for NXFME_DELETED, else zero. */
2117 ovs_be16 priority; /* Priority of the entry. */
2118 ovs_be16 idle_timeout; /* Number of seconds idle before expiration. */
2119 ovs_be16 hard_timeout; /* Number of seconds before expiration. */
2120 ovs_be16 match_len; /* Length of nx_match. */
2121 uint8_t table_id; /* ID of flow's table. */
2122 uint8_t pad; /* Reserved, currently zeroed. */
2123 ovs_be64 cookie; /* Opaque controller-issued identifier. */
2124 /* Followed by:
2125 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
2126 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
2127 * all-zero bytes, then
2128 * - Actions to fill out the remainder 'length' bytes (always a multiple
2129 * of 8). If NXFMF_ACTIONS was not specified, or 'event' is
2130 * NXFME_DELETED, no actions are included.
2131 */
2132 };
2133 OFP_ASSERT(sizeof(struct nx_flow_update_full) == 24);
2134
2135 /* NXST_FLOW_MONITOR reply for NXFME_ABBREV.
2136 *
2137 * When the controller does not specify NXFMF_OWN in a monitor request, any
2138 * flow tables changes due to the controller's own requests (on the same
2139 * OpenFlow channel) will be abbreviated, when possible, to this form, which
2140 * simply specifies the 'xid' of the OpenFlow request (e.g. an OFPT_FLOW_MOD or
2141 * NXT_FLOW_MOD) that caused the change.
2142 *
2143 * Some changes cannot be abbreviated and will be sent in full:
2144 *
2145 * - Changes that only partially succeed. This can happen if, for example,
2146 * a flow_mod with type OFPFC_MODIFY affects multiple flows, but only some
2147 * of those modifications succeed (e.g. due to hardware limitations).
2148 *
2149 * This cannot occur with the current implementation of the Open vSwitch
2150 * software datapath. It could happen with other datapath implementations.
2151 *
2152 * - Changes that race with conflicting changes made by other controllers or
2153 * other flow_mods (not separated by barriers) by the same controller.
2154 *
2155 * This cannot occur with the current Open vSwitch implementation
2156 * (regardless of datapath) because Open vSwitch internally serializes
2157 * potentially conflicting changes.
2158 *
2159 * A flow_mod that does not change the flow table will not trigger any
2160 * notification, even an abbreviated one. For example, a "modify" or "delete"
2161 * flow_mod that does not match any flows will not trigger a notification.
2162 * Whether an "add" or "modify" that specifies all the same parameters that a
2163 * flow already has triggers a notification is unspecified and subject to
2164 * change in future versions of Open vSwitch.
2165 *
2166 * OVS will always send the notifications for a given flow table change before
2167 * the reply to a OFPT_BARRIER_REQUEST request that follows the flow table
2168 * change. Thus, if the controller does not receive an abbreviated (or
2169 * unabbreviated) notification for a flow_mod before the next
2170 * OFPT_BARRIER_REPLY, it will never receive one. */
2171 struct nx_flow_update_abbrev {
2172 ovs_be16 length; /* Length is 8. */
2173 ovs_be16 event; /* NXFME_ABBREV. */
2174 ovs_be32 xid; /* Controller-specified xid from flow_mod. */
2175 };
2176 OFP_ASSERT(sizeof(struct nx_flow_update_abbrev) == 8);
2177
2178 /* NXT_FLOW_MONITOR_CANCEL.
2179 *
2180 * Used by a controller to cancel an outstanding monitor. */
2181 struct nx_flow_monitor_cancel {
2182 ovs_be32 id; /* 'id' from nx_flow_monitor_request. */
2183 };
2184 OFP_ASSERT(sizeof(struct nx_flow_monitor_cancel) == 4);
2185
2186 /* Action structure for NXAST_WRITE_METADATA.
2187 *
2188 * Modifies the 'mask' bits of the metadata value. */
2189 struct nx_action_write_metadata {
2190 ovs_be16 type; /* OFPAT_VENDOR. */
2191 ovs_be16 len; /* Length is 32. */
2192 ovs_be32 vendor; /* NX_VENDOR_ID. */
2193 ovs_be16 subtype; /* NXAST_WRITE_METADATA. */
2194 uint8_t zeros[6]; /* Must be zero. */
2195 ovs_be64 metadata; /* Metadata register. */
2196 ovs_be64 mask; /* Metadata mask. */
2197 };
2198 OFP_ASSERT(sizeof(struct nx_action_write_metadata) == 32);
2199
2200 /* Action structure for NXAST_PUSH_MPLS. */
2201 struct nx_action_push_mpls {
2202 ovs_be16 type; /* OFPAT_VENDOR. */
2203 ovs_be16 len; /* Length is 8. */
2204 ovs_be32 vendor; /* NX_VENDOR_ID. */
2205 ovs_be16 subtype; /* NXAST_PUSH_MPLS. */
2206 ovs_be16 ethertype; /* Ethertype */
2207 uint8_t pad[4];
2208 };
2209 OFP_ASSERT(sizeof(struct nx_action_push_mpls) == 16);
2210
2211 /* Action structure for NXAST_POP_MPLS. */
2212 struct nx_action_pop_mpls {
2213 ovs_be16 type; /* OFPAT_VENDOR. */
2214 ovs_be16 len; /* Length is 8. */
2215 ovs_be32 vendor; /* NX_VENDOR_ID. */
2216 ovs_be16 subtype; /* NXAST_POP_MPLS. */
2217 ovs_be16 ethertype; /* Ethertype */
2218 uint8_t pad[4];
2219 };
2220 OFP_ASSERT(sizeof(struct nx_action_pop_mpls) == 16);
2221
2222 /* Action structure for NXAST_SET_MPLS_TTL. */
2223 struct nx_action_mpls_ttl {
2224 ovs_be16 type; /* OFPAT_VENDOR. */
2225 ovs_be16 len; /* Length is 8. */
2226 ovs_be32 vendor; /* NX_VENDOR_ID. */
2227 ovs_be16 subtype; /* NXAST_SET_MPLS_TTL. */
2228 uint8_t ttl; /* TTL */
2229 uint8_t pad[5];
2230 };
2231 OFP_ASSERT(sizeof(struct nx_action_mpls_ttl) == 16);
2232
2233 #endif /* openflow/nicira-ext.h */