2 * Copyright (c) 2008, 2009, 2010, 2011, 2012 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
17 #ifndef OPENFLOW_NICIRA_EXT_H
18 #define OPENFLOW_NICIRA_EXT_H 1
20 #include "openflow/openflow.h"
21 #include "openvswitch/types.h"
23 /* The following vendor extensions, proposed by Nicira, are not yet
24 * standardized, so they are not included in openflow.h. Some of them may be
25 * suitable for standardization; others we never expect to standardize. */
27 #define NX_VENDOR_ID 0x00002320
29 /* Nicira vendor-specific error messages extension.
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.
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
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.
49 * Currently this extension is only implemented by Open vSwitch, but it seems
50 * like a reasonable candidate for future standardization.
53 /* This is a random number to avoid accidental collision with any other
54 * vendor's extension. */
55 #define NXET_VENDOR 0xb0c2
57 /* ofp_error msg 'code' values for NXET_VENDOR. */
59 NXVC_VENDOR_ERROR /* 'data' contains struct nx_vendor_error. */
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. */
70 /* Nicira vendor requests and replies. */
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; /* One of NXT_* below. */
78 OFP_ASSERT(sizeof(struct nicira_header) == 16);
80 /* Values for the 'subtype' member of struct nicira_header. */
83 NXT_STATUS_REQUEST__OBSOLETE = 0,
84 NXT_STATUS_REPLY__OBSOLETE = 1,
85 NXT_ACT_SET_CONFIG__OBSOLETE = 2,
86 NXT_ACT_GET_CONFIG__OBSOLETE = 3,
87 NXT_COMMAND_REQUEST__OBSOLETE = 4,
88 NXT_COMMAND_REPLY__OBSOLETE = 5,
89 NXT_FLOW_END_CONFIG__OBSOLETE = 6,
90 NXT_FLOW_END__OBSOLETE = 7,
91 NXT_MGMT__OBSOLETE = 8,
92 NXT_TUN_ID_FROM_COOKIE__OBSOLETE = 9,
94 /* Controller role support. The request body is struct nx_role_request.
95 * The reply echos the request. */
96 NXT_ROLE_REQUEST = 10,
99 /* Flexible flow specification (aka NXM = Nicira Extended Match). */
100 NXT_SET_FLOW_FORMAT = 12, /* Set flow format. */
101 NXT_FLOW_MOD = 13, /* Analogous to OFPT_FLOW_MOD. */
102 NXT_FLOW_REMOVED = 14, /* Analogous to OFPT_FLOW_REMOVED. */
104 /* Use the upper 8 bits of the 'command' member in struct ofp_flow_mod to
105 * designate the table to which a flow is to be added? See the big comment
106 * on struct nx_flow_mod_table_id for more information. */
107 NXT_FLOW_MOD_TABLE_ID = 15,
109 /* Alternative PACKET_IN message formats. */
110 NXT_SET_PACKET_IN_FORMAT = 16, /* Set Packet In format. */
111 NXT_PACKET_IN = 17, /* Nicira Packet In. */
113 /* Are the idle_age and hard_age members in struct nx_flow_stats supported?
114 * If so, the switch does not reply to this message (which consists only of
115 * a "struct nicira_header"). If not, the switch sends an error reply. */
118 NXT_SET_ASYNC_CONFIG = 19, /* struct nx_async_config. */
119 NXT_SET_CONTROLLER_ID = 20, /* struct nx_controller_id. */
122 /* Header for Nicira vendor stats request and reply messages. */
123 struct nicira_stats_msg {
124 struct ofp_vendor_stats_msg vsm; /* Vendor NX_VENDOR_ID. */
125 ovs_be32 subtype; /* One of NXST_* below. */
126 uint8_t pad[4]; /* Align to 64-bits. */
128 OFP_ASSERT(sizeof(struct nicira_stats_msg) == 24);
130 /* Values for the 'subtype' member of struct nicira_stats_msg. */
131 enum nicira_stats_type {
132 /* Flexible flow specification (aka NXM = Nicira Extended Match). */
133 NXST_FLOW, /* Analogous to OFPST_FLOW. */
134 NXST_AGGREGATE /* Analogous to OFPST_AGGREGATE. */
137 /* Fields to use when hashing flows. */
138 enum nx_hash_fields {
139 /* Ethernet source address (NXM_OF_ETH_SRC) only. */
140 NX_HASH_FIELDS_ETH_SRC,
142 /* L2 through L4, symmetric across src/dst. Specifically, each of the
143 * following fields, if present, is hashed (slashes separate symmetric
146 * - NXM_OF_ETH_DST / NXM_OF_ETH_SRC
148 * - The VID bits from NXM_OF_VLAN_TCI, ignoring PCP and CFI.
150 * - NXM_OF_IP_SRC / NXM_OF_IP_DST
151 * - NXM_OF_TCP_SRC / NXM_OF_TCP_DST
153 NX_HASH_FIELDS_SYMMETRIC_L4
156 /* This command enables or disables an Open vSwitch extension that allows a
157 * controller to specify the OpenFlow table to which a flow should be added,
158 * instead of having the switch decide which table is most appropriate as
159 * required by OpenFlow 1.0. By default, the extension is disabled.
161 * When this feature is enabled, Open vSwitch treats struct ofp_flow_mod's
162 * 16-bit 'command' member as two separate fields. The upper 8 bits are used
163 * as the table ID, the lower 8 bits specify the command as usual. A table ID
164 * of 0xff is treated like a wildcarded table ID.
166 * The specific treatment of the table ID depends on the type of flow mod:
168 * - OFPFC_ADD: Given a specific table ID, the flow is always placed in that
169 * table. If an identical flow already exists in that table only, then it
170 * is replaced. If the flow cannot be placed in the specified table,
171 * either because the table is full or because the table cannot support
172 * flows of the given type, the switch replies with an
173 * OFPFMFC_ALL_TABLES_FULL error. (A controller can distinguish these
174 * cases by comparing the current and maximum number of entries reported
175 * in ofp_table_stats.)
177 * If the table ID is wildcarded, the switch picks an appropriate table
178 * itself. If an identical flow already exist in the selected flow table,
179 * then it is replaced. The choice of table might depend on the flows
180 * that are already in the switch; for example, if one table fills up then
181 * the switch might fall back to another one.
183 * - OFPFC_MODIFY, OFPFC_DELETE: Given a specific table ID, only flows
184 * within that table are matched and modified or deleted. If the table ID
185 * is wildcarded, flows within any table may be matched and modified or
188 * - OFPFC_MODIFY_STRICT, OFPFC_DELETE_STRICT: Given a specific table ID,
189 * only a flow within that table may be matched and modified or deleted.
190 * If the table ID is wildcarded and exactly one flow within any table
191 * matches, then it is modified or deleted; if flows in more than one
192 * table match, then none is modified or deleted.
194 struct nx_flow_mod_table_id {
195 struct ofp_header header;
196 ovs_be32 vendor; /* NX_VENDOR_ID. */
197 ovs_be32 subtype; /* NXT_FLOW_MOD_TABLE_ID. */
198 uint8_t set; /* Nonzero to enable, zero to disable. */
201 OFP_ASSERT(sizeof(struct nx_flow_mod_table_id) == 24);
203 enum nx_packet_in_format {
204 NXPIF_OPENFLOW10 = 0, /* Standard OpenFlow 1.0 compatible. */
205 NXPIF_NXM = 1 /* Nicira Extended. */
208 /* NXT_SET_PACKET_IN_FORMAT request. */
209 struct nx_set_packet_in_format {
210 struct nicira_header nxh;
211 ovs_be32 format; /* One of NXPIF_*. */
213 OFP_ASSERT(sizeof(struct nx_set_packet_in_format) == 20);
215 /* NXT_PACKET_IN (analogous to OFPT_PACKET_IN).
217 * The NXT_PACKET_IN format is intended to model the OpenFlow-1.2 PACKET_IN
218 * with some minor tweaks. Most notably NXT_PACKET_IN includes the cookie of
219 * the rule which triggered the NXT_PACKET_IN message, and the match fields are
222 * The match fields in the NXT_PACKET_IN are intended to contain flow
223 * processing metadata collected at the time the NXT_PACKET_IN message was
224 * triggered. It is minimally required to contain the NXM_OF_IN_PORT of the
225 * packet, but may include other NXM headers such as flow registers. The match
226 * fields are allowed to contain non-metadata (e.g. NXM_OF_ETH_SRC etc).
227 * However, this information can typically be found in the packet directly, so
228 * it may be redundant.
230 * Whereas in most cases a controller can expect to only get back NXM fields
231 * that it set up itself (e.g. flow dumps will ordinarily report only NXM
232 * fields from flows that the controller added), NXT_PACKET_IN messages might
233 * contain fields that the controller does not understand, because the switch
234 * might support fields (new registers, new protocols, etc.) that the
235 * controller does not. Â The controller must prepared to tolerate these.
237 * The 'cookie' and 'table_id' fields have no meaning when 'reason' is
238 * OFPR_NO_MATCH. In this case they should be set to 0. */
239 struct nx_packet_in {
240 struct nicira_header nxh;
241 ovs_be32 buffer_id; /* ID assigned by datapath. */
242 ovs_be16 total_len; /* Full length of frame. */
243 uint8_t reason; /* Reason packet is sent (one of OFPR_*). */
244 uint8_t table_id; /* ID of the table that was looked up. */
245 ovs_be64 cookie; /* Cookie of the rule that was looked up. */
246 ovs_be16 match_len; /* Size of nx_match. */
247 uint8_t pad[6]; /* Align to 64-bits. */
249 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
250 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
251 * all-zero bytes, then
252 * - Exactly 2 all-zero padding bytes, then
253 * - An Ethernet frame whose length is inferred from nxh.header.length.
255 * The padding bytes preceding the Ethernet frame ensure that the IP
256 * header (if any) following the Ethernet header is 32-bit aligned. */
258 /* uint8_t nxm_fields[...]; */ /* Match. */
259 /* uint8_t pad[2]; */ /* Align to 64 bit + 16 bit. */
260 /* uint8_t data[0]; */ /* Ethernet frame. */
262 OFP_ASSERT(sizeof(struct nx_packet_in) == 40);
264 /* Configures the "role" of the sending controller. The default role is:
266 * - Other (NX_ROLE_OTHER), which allows the controller access to all
269 * The other possible roles are a related pair:
271 * - Master (NX_ROLE_MASTER) is equivalent to Other, except that there may
272 * be at most one Master controller at a time: when a controller
273 * configures itself as Master, any existing Master is demoted to the
276 * - Slave (NX_ROLE_SLAVE) allows the controller read-only access to
277 * OpenFlow features. In particular attempts to modify the flow table
278 * will be rejected with an OFPBRC_EPERM error.
280 * Slave controllers do not receive OFPT_PACKET_IN or OFPT_FLOW_REMOVED
281 * messages, but they do receive OFPT_PORT_STATUS messages.
283 struct nx_role_request {
284 struct nicira_header nxh;
285 ovs_be32 role; /* One of NX_ROLE_*. */
287 OFP_ASSERT(sizeof(struct nx_role_request) == 20);
290 NX_ROLE_OTHER, /* Default role, full access. */
291 NX_ROLE_MASTER, /* Full access, at most one. */
292 NX_ROLE_SLAVE /* Read-only access. */
295 /* NXT_SET_ASYNC_CONFIG.
297 * Sent by a controller, this message configures the asynchronous messages that
298 * the controller wants to receive. Element 0 in each array specifies messages
299 * of interest when the controller has an "other" or "master" role; element 1,
300 * when the controller has a "slave" role.
302 * Each array element is a bitmask in which a 0-bit disables receiving a
303 * particular message and a 1-bit enables receiving it. Each bit controls the
304 * message whose 'reason' corresponds to the bit index. For example, the bit
305 * with value 1<<2 == 4 in port_status_mask[1] determines whether the
306 * controller will receive OFPT_PORT_STATUS messages with reason OFPPR_MODIFY
307 * (value 2) when the controller has a "slave" role.
309 * As a side effect, for service controllers, this message changes the
310 * miss_send_len from default of zero to OFP_DEFAULT_MISS_SEND_LEN (128).
312 struct nx_async_config {
313 struct nicira_header nxh;
314 ovs_be32 packet_in_mask[2]; /* Bitmasks of OFPR_* values. */
315 ovs_be32 port_status_mask[2]; /* Bitmasks of OFPRR_* values. */
316 ovs_be32 flow_removed_mask[2]; /* Bitmasks of OFPPR_* values. */
318 OFP_ASSERT(sizeof(struct nx_async_config) == 40);
320 /* Nicira vendor flow actions. */
322 enum nx_action_subtype {
323 NXAST_SNAT__OBSOLETE, /* No longer used. */
324 NXAST_RESUBMIT, /* struct nx_action_resubmit */
325 NXAST_SET_TUNNEL, /* struct nx_action_set_tunnel */
326 NXAST_DROP_SPOOFED_ARP__OBSOLETE,
327 NXAST_SET_QUEUE, /* struct nx_action_set_queue */
328 NXAST_POP_QUEUE, /* struct nx_action_pop_queue */
329 NXAST_REG_MOVE, /* struct nx_action_reg_move */
330 NXAST_REG_LOAD, /* struct nx_action_reg_load */
331 NXAST_NOTE, /* struct nx_action_note */
332 NXAST_SET_TUNNEL64, /* struct nx_action_set_tunnel64 */
333 NXAST_MULTIPATH, /* struct nx_action_multipath */
334 NXAST_AUTOPATH, /* struct nx_action_autopath */
335 NXAST_BUNDLE, /* struct nx_action_bundle */
336 NXAST_BUNDLE_LOAD, /* struct nx_action_bundle */
337 NXAST_RESUBMIT_TABLE, /* struct nx_action_resubmit */
338 NXAST_OUTPUT_REG, /* struct nx_action_output_reg */
339 NXAST_LEARN, /* struct nx_action_learn */
340 NXAST_EXIT, /* struct nx_action_header */
341 NXAST_DEC_TTL, /* struct nx_action_header */
342 NXAST_FIN_TIMEOUT, /* struct nx_action_fin_timeout */
343 NXAST_CONTROLLER, /* struct nx_action_controller */
346 /* Header for Nicira-defined actions. */
347 struct nx_action_header {
348 ovs_be16 type; /* OFPAT_VENDOR. */
349 ovs_be16 len; /* Length is 16. */
350 ovs_be32 vendor; /* NX_VENDOR_ID. */
351 ovs_be16 subtype; /* NXAST_*. */
354 OFP_ASSERT(sizeof(struct nx_action_header) == 16);
356 /* Action structures for NXAST_RESUBMIT and NXAST_RESUBMIT_TABLE.
358 * These actions search one of the switch's flow tables:
360 * - For NXAST_RESUBMIT_TABLE only, if the 'table' member is not 255, then
361 * it specifies the table to search.
363 * - Otherwise (for NXAST_RESUBMIT_TABLE with a 'table' of 255, or for
364 * NXAST_RESUBMIT regardless of 'table'), it searches the current flow
365 * table, that is, the OpenFlow flow table that contains the flow from
366 * which this action was obtained. If this action did not come from a
367 * flow table (e.g. it came from an OFPT_PACKET_OUT message), then table 0
368 * is the current table.
370 * The flow table lookup uses a flow that may be slightly modified from the
373 * - For NXAST_RESUBMIT, the 'in_port' member of struct nx_action_resubmit
374 * is used as the flow's in_port.
376 * - For NXAST_RESUBMIT_TABLE, if the 'in_port' member is not OFPP_IN_PORT,
377 * then its value is used as the flow's in_port. Otherwise, the original
380 * - If actions that modify the flow (e.g. OFPAT_SET_VLAN_VID) precede the
381 * resubmit action, then the flow is updated with the new values.
383 * Following the lookup, the original in_port is restored.
385 * If the modified flow matched in the flow table, then the corresponding
386 * actions are executed. Afterward, actions following the resubmit in the
387 * original set of actions, if any, are executed; any changes made to the
388 * packet (e.g. changes to VLAN) by secondary actions persist when those
389 * actions are executed, although the original in_port is restored.
391 * Resubmit actions may be used any number of times within a set of actions.
393 * Resubmit actions may nest to an implementation-defined depth. Beyond this
394 * implementation-defined depth, further resubmit actions are simply ignored.
396 * NXAST_RESUBMIT ignores 'table' and 'pad'. NXAST_RESUBMIT_TABLE requires
397 * 'pad' to be all-bits-zero.
399 * Open vSwitch 1.0.1 and earlier did not support recursion. Open vSwitch
400 * before 1.2.90 did not support NXAST_RESUBMIT_TABLE.
402 struct nx_action_resubmit {
403 ovs_be16 type; /* OFPAT_VENDOR. */
404 ovs_be16 len; /* Length is 16. */
405 ovs_be32 vendor; /* NX_VENDOR_ID. */
406 ovs_be16 subtype; /* NXAST_RESUBMIT. */
407 ovs_be16 in_port; /* New in_port for checking flow table. */
408 uint8_t table; /* NXAST_RESUBMIT_TABLE: table to use. */
411 OFP_ASSERT(sizeof(struct nx_action_resubmit) == 16);
413 /* Action structure for NXAST_SET_TUNNEL.
415 * Sets the encapsulating tunnel ID to a 32-bit value. The most-significant 32
416 * bits of the tunnel ID are set to 0. */
417 struct nx_action_set_tunnel {
418 ovs_be16 type; /* OFPAT_VENDOR. */
419 ovs_be16 len; /* Length is 16. */
420 ovs_be32 vendor; /* NX_VENDOR_ID. */
421 ovs_be16 subtype; /* NXAST_SET_TUNNEL. */
423 ovs_be32 tun_id; /* Tunnel ID. */
425 OFP_ASSERT(sizeof(struct nx_action_set_tunnel) == 16);
427 /* Action structure for NXAST_SET_TUNNEL64.
429 * Sets the encapsulating tunnel ID to a 64-bit value. */
430 struct nx_action_set_tunnel64 {
431 ovs_be16 type; /* OFPAT_VENDOR. */
432 ovs_be16 len; /* Length is 24. */
433 ovs_be32 vendor; /* NX_VENDOR_ID. */
434 ovs_be16 subtype; /* NXAST_SET_TUNNEL64. */
436 ovs_be64 tun_id; /* Tunnel ID. */
438 OFP_ASSERT(sizeof(struct nx_action_set_tunnel64) == 24);
440 /* Action structure for NXAST_SET_QUEUE.
442 * Set the queue that should be used when packets are output. This is similar
443 * to the OpenFlow OFPAT_ENQUEUE action, but does not take the output port as
444 * an argument. This allows the queue to be defined before the port is
446 struct nx_action_set_queue {
447 ovs_be16 type; /* OFPAT_VENDOR. */
448 ovs_be16 len; /* Length is 16. */
449 ovs_be32 vendor; /* NX_VENDOR_ID. */
450 ovs_be16 subtype; /* NXAST_SET_QUEUE. */
452 ovs_be32 queue_id; /* Where to enqueue packets. */
454 OFP_ASSERT(sizeof(struct nx_action_set_queue) == 16);
456 /* Action structure for NXAST_POP_QUEUE.
458 * Restores the queue to the value it was before any NXAST_SET_QUEUE actions
459 * were used. Only the original queue can be restored this way; no stack is
461 struct nx_action_pop_queue {
462 ovs_be16 type; /* OFPAT_VENDOR. */
463 ovs_be16 len; /* Length is 16. */
464 ovs_be32 vendor; /* NX_VENDOR_ID. */
465 ovs_be16 subtype; /* NXAST_POP_QUEUE. */
468 OFP_ASSERT(sizeof(struct nx_action_pop_queue) == 16);
470 /* Action structure for NXAST_REG_MOVE.
472 * Copies src[src_ofs:src_ofs+n_bits] to dst[dst_ofs:dst_ofs+n_bits], where
473 * a[b:c] denotes the bits within 'a' numbered 'b' through 'c' (not including
474 * bit 'c'). Bit numbering starts at 0 for the least-significant bit, 1 for
475 * the next most significant bit, and so on.
477 * 'src' and 'dst' are nxm_header values with nxm_hasmask=0. (It doesn't make
478 * sense to use nxm_hasmask=1 because the action does not do any kind of
479 * matching; it uses the actual value of a field.)
481 * The following nxm_header values are potentially acceptable as 'src':
504 * - NXM_NX_ICMPV6_TYPE
505 * - NXM_NX_ICMPV6_CODE
508 * - NXM_NX_REG(idx) for idx in the switch's accepted range.
510 * The following nxm_header values are potentially acceptable as 'dst':
521 * Modifying any of the above fields changes the corresponding packet
524 * - NXM_NX_REG(idx) for idx in the switch's accepted range.
526 * - NXM_OF_VLAN_TCI. Modifying this field's value has side effects on the
527 * packet's 802.1Q header. Setting a value with CFI=0 removes the 802.1Q
528 * header (if any), ignoring the other bits. Setting a value with CFI=1
529 * adds or modifies the 802.1Q header appropriately, setting the TCI field
530 * to the field's new value (with the CFI bit masked out).
532 * - NXM_NX_TUN_ID. Modifying this value modifies the tunnel ID used for the
533 * packet's next tunnel encapsulation.
535 * A given nxm_header value may be used as 'src' or 'dst' only on a flow whose
536 * nx_match satisfies its prerequisites. For example, NXM_OF_IP_TOS may be
537 * used only if the flow's nx_match includes an nxm_entry that specifies
538 * nxm_type=NXM_OF_ETH_TYPE, nxm_hasmask=0, and nxm_value=0x0800.
540 * The switch will reject actions for which src_ofs+n_bits is greater than the
541 * width of 'src' or dst_ofs+n_bits is greater than the width of 'dst' with
542 * error type OFPET_BAD_ACTION, code OFPBAC_BAD_ARGUMENT.
544 struct nx_action_reg_move {
545 ovs_be16 type; /* OFPAT_VENDOR. */
546 ovs_be16 len; /* Length is 24. */
547 ovs_be32 vendor; /* NX_VENDOR_ID. */
548 ovs_be16 subtype; /* NXAST_REG_MOVE. */
549 ovs_be16 n_bits; /* Number of bits. */
550 ovs_be16 src_ofs; /* Starting bit offset in source. */
551 ovs_be16 dst_ofs; /* Starting bit offset in destination. */
552 ovs_be32 src; /* Source register. */
553 ovs_be32 dst; /* Destination register. */
555 OFP_ASSERT(sizeof(struct nx_action_reg_move) == 24);
557 /* Action structure for NXAST_REG_LOAD.
559 * Copies value[0:n_bits] to dst[ofs:ofs+n_bits], where a[b:c] denotes the bits
560 * within 'a' numbered 'b' through 'c' (not including bit 'c'). Bit numbering
561 * starts at 0 for the least-significant bit, 1 for the next most significant
564 * 'dst' is an nxm_header with nxm_hasmask=0. See the documentation for
565 * NXAST_REG_MOVE, above, for the permitted fields and for the side effects of
568 * The 'ofs' and 'n_bits' fields are combined into a single 'ofs_nbits' field
569 * to avoid enlarging the structure by another 8 bytes. To allow 'n_bits' to
570 * take a value between 1 and 64 (inclusive) while taking up only 6 bits, it is
571 * also stored as one less than its true value:
574 * +------------------------------+------------------+
575 * | ofs | n_bits - 1 |
576 * +------------------------------+------------------+
578 * The switch will reject actions for which ofs+n_bits is greater than the
579 * width of 'dst', or in which any bits in 'value' with value 2**n_bits or
580 * greater are set to 1, with error type OFPET_BAD_ACTION, code
581 * OFPBAC_BAD_ARGUMENT.
583 struct nx_action_reg_load {
584 ovs_be16 type; /* OFPAT_VENDOR. */
585 ovs_be16 len; /* Length is 24. */
586 ovs_be32 vendor; /* NX_VENDOR_ID. */
587 ovs_be16 subtype; /* NXAST_REG_LOAD. */
588 ovs_be16 ofs_nbits; /* (ofs << 6) | (n_bits - 1). */
589 ovs_be32 dst; /* Destination register. */
590 ovs_be64 value; /* Immediate value. */
592 OFP_ASSERT(sizeof(struct nx_action_reg_load) == 24);
594 /* Action structure for NXAST_NOTE.
596 * This action has no effect. It is variable length. The switch does not
597 * attempt to interpret the user-defined 'note' data in any way. A controller
598 * can use this action to attach arbitrary metadata to a flow.
600 * This action might go away in the future.
602 struct nx_action_note {
603 ovs_be16 type; /* OFPAT_VENDOR. */
604 ovs_be16 len; /* A multiple of 8, but at least 16. */
605 ovs_be32 vendor; /* NX_VENDOR_ID. */
606 ovs_be16 subtype; /* NXAST_NOTE. */
607 uint8_t note[6]; /* Start of user-defined data. */
608 /* Possibly followed by additional user-defined data. */
610 OFP_ASSERT(sizeof(struct nx_action_note) == 16);
612 /* Action structure for NXAST_MULTIPATH.
614 * This action performs the following steps in sequence:
616 * 1. Hashes the fields designated by 'fields', one of NX_HASH_FIELDS_*.
617 * Refer to the definition of "enum nx_mp_fields" for details.
619 * The 'basis' value is used as a universal hash parameter, that is,
620 * different values of 'basis' yield different hash functions. The
621 * particular universal hash function used is implementation-defined.
623 * The hashed fields' values are drawn from the current state of the
624 * flow, including all modifications that have been made by actions up to
627 * 2. Applies the multipath link choice algorithm specified by 'algorithm',
628 * one of NX_MP_ALG_*. Refer to the definition of "enum nx_mp_algorithm"
631 * The output of the algorithm is 'link', an unsigned integer less than
632 * or equal to 'max_link'.
634 * Some algorithms use 'arg' as an additional argument.
636 * 3. Stores 'link' in dst[ofs:ofs+n_bits]. The format and semantics of
637 * 'dst' and 'ofs_nbits' are similar to those for the NXAST_REG_LOAD
640 * The switch will reject actions that have an unknown 'fields', or an unknown
641 * 'algorithm', or in which ofs+n_bits is greater than the width of 'dst', or
642 * in which 'max_link' is greater than or equal to 2**n_bits, with error type
643 * OFPET_BAD_ACTION, code OFPBAC_BAD_ARGUMENT.
645 struct nx_action_multipath {
646 ovs_be16 type; /* OFPAT_VENDOR. */
647 ovs_be16 len; /* Length is 32. */
648 ovs_be32 vendor; /* NX_VENDOR_ID. */
649 ovs_be16 subtype; /* NXAST_MULTIPATH. */
651 /* What fields to hash and how. */
652 ovs_be16 fields; /* One of NX_HASH_FIELDS_*. */
653 ovs_be16 basis; /* Universal hash parameter. */
656 /* Multipath link choice algorithm to apply to hash value. */
657 ovs_be16 algorithm; /* One of NX_MP_ALG_*. */
658 ovs_be16 max_link; /* Number of output links, minus 1. */
659 ovs_be32 arg; /* Algorithm-specific argument. */
662 /* Where to store the result. */
663 ovs_be16 ofs_nbits; /* (ofs << 6) | (n_bits - 1). */
664 ovs_be32 dst; /* Destination. */
666 OFP_ASSERT(sizeof(struct nx_action_multipath) == 32);
668 /* NXAST_MULTIPATH: Multipath link choice algorithm to apply.
670 * In the descriptions below, 'n_links' is max_link + 1. */
671 enum nx_mp_algorithm {
672 /* link = hash(flow) % n_links.
674 * Redistributes all traffic when n_links changes. O(1) performance. See
677 * Use UINT16_MAX for max_link to get a raw hash value. */
680 /* link = hash(flow) / (MAX_HASH / n_links).
682 * Redistributes between one-quarter and one-half of traffic when n_links
683 * changes. O(1) performance. See RFC 2992.
685 NX_MP_ALG_HASH_THRESHOLD,
687 /* for i in [0,n_links):
688 * weights[i] = hash(flow, i)
689 * link = { i such that weights[i] >= weights[j] for all j != i }
691 * Redistributes 1/n_links of traffic when n_links changes. O(n_links)
692 * performance. If n_links is greater than a threshold (currently 64, but
693 * subject to change), Open vSwitch will substitute another algorithm
694 * automatically. See RFC 2992. */
695 NX_MP_ALG_HRW, /* Highest Random Weight. */
700 * link = hash(flow, i) % arg
701 * while link > max_link
703 * Redistributes 1/n_links of traffic when n_links changes. O(1)
704 * performance when arg/max_link is bounded by a constant.
706 * Redistributes all traffic when arg changes.
708 * arg must be greater than max_link and for best performance should be no
709 * more than approximately max_link * 2. If arg is outside the acceptable
710 * range, Open vSwitch will automatically substitute the least power of 2
711 * greater than max_link.
713 * This algorithm is specific to Open vSwitch.
715 NX_MP_ALG_ITER_HASH /* Iterative Hash. */
718 /* Action structure for NXAST_LEARN.
720 * This action adds or modifies a flow in an OpenFlow table, similar to
721 * OFPT_FLOW_MOD with OFPFC_MODIFY_STRICT as 'command'. The new flow has the
722 * specified idle timeout, hard timeout, priority, cookie, and flags. The new
723 * flow's match criteria and actions are built by applying each of the series
724 * of flow_mod_spec elements included as part of the action.
726 * A flow_mod_spec starts with a 16-bit header. A header that is all-bits-0 is
727 * a no-op used for padding the action as a whole to a multiple of 8 bytes in
728 * length. Otherwise, the flow_mod_spec can be thought of as copying 'n_bits'
729 * bits from a source to a destination. In this case, the header contains
732 * 15 14 13 12 11 10 0
733 * +------+---+------+---------------------------------+
734 * | 0 |src| dst | n_bits |
735 * +------+---+------+---------------------------------+
737 * The meaning and format of a flow_mod_spec depends on 'src' and 'dst'. The
738 * following table summarizes the meaning of each possible combination.
739 * Details follow the table:
742 * --- --- ----------------------------------------------------------
743 * 0 0 Add match criteria based on value in a field.
744 * 1 0 Add match criteria based on an immediate value.
745 * 0 1 Add NXAST_REG_LOAD action to copy field into a different field.
746 * 1 1 Add NXAST_REG_LOAD action to load immediate value into a field.
747 * 0 2 Add OFPAT_OUTPUT action to output to port from specified field.
748 * All other combinations are undefined and not allowed.
750 * The flow_mod_spec header is followed by a source specification and a
751 * destination specification. The format and meaning of the source
752 * specification depends on 'src':
754 * - If 'src' is 0, the source bits are taken from a field in the flow to
755 * which this action is attached. (This should be a wildcarded field. If
756 * its value is fully specified then the source bits being copied have
759 * The source specification is an ovs_be32 'field' and an ovs_be16 'ofs'.
760 * 'field' is an nxm_header with nxm_hasmask=0, and 'ofs' the starting bit
761 * offset within that field. The source bits are field[ofs:ofs+n_bits-1].
762 * 'field' and 'ofs' are subject to the same restrictions as the source
763 * field in NXAST_REG_MOVE.
765 * - If 'src' is 1, the source bits are a constant value. The source
766 * specification is (n_bits+15)/16*2 bytes long. Taking those bytes as a
767 * number in network order, the source bits are the 'n_bits'
768 * least-significant bits. The switch will report an error if other bits
769 * in the constant are nonzero.
771 * The flow_mod_spec destination specification, for 'dst' of 0 or 1, is an
772 * ovs_be32 'field' and an ovs_be16 'ofs'. 'field' is an nxm_header with
773 * nxm_hasmask=0 and 'ofs' is a starting bit offset within that field. The
774 * meaning of the flow_mod_spec depends on 'dst':
776 * - If 'dst' is 0, the flow_mod_spec specifies match criteria for the new
777 * flow. The new flow matches only if bits field[ofs:ofs+n_bits-1] in a
778 * packet equal the source bits. 'field' may be any nxm_header with
779 * nxm_hasmask=0 that is allowed in NXT_FLOW_MOD.
781 * Order is significant. Earlier flow_mod_specs must satisfy any
782 * prerequisites for matching fields specified later, by copying constant
783 * values into prerequisite fields.
785 * The switch will reject flow_mod_specs that do not satisfy NXM masking
788 * - If 'dst' is 1, the flow_mod_spec specifies an NXAST_REG_LOAD action for
789 * the new flow. The new flow copies the source bits into
790 * field[ofs:ofs+n_bits-1]. Actions are executed in the same order as the
793 * A single NXAST_REG_LOAD action writes no more than 64 bits, so n_bits
794 * greater than 64 yields multiple NXAST_REG_LOAD actions.
796 * The flow_mod_spec destination spec for 'dst' of 2 (when 'src' is 0) is
797 * empty. It has the following meaning:
799 * - The flow_mod_spec specifies an OFPAT_OUTPUT action for the new flow.
800 * The new flow outputs to the OpenFlow port specified by the source field.
801 * Of the special output ports with value OFPP_MAX or larger, OFPP_IN_PORT,
802 * OFPP_FLOOD, OFPP_LOCAL, and OFPP_ALL are supported. Other special ports
805 * Resource Management
806 * -------------------
808 * A switch has a finite amount of flow table space available for learning.
809 * When this space is exhausted, no new learning table entries will be learned
810 * until some existing flow table entries expire. The controller should be
811 * prepared to handle this by flooding (which can be implemented as a
812 * low-priority flow).
814 * If a learned flow matches a single TCP stream with a relatively long
815 * timeout, one may make the best of resource constraints by setting
816 * 'fin_idle_timeout' or 'fin_hard_timeout' (both measured in seconds), or
817 * both, to shorter timeouts. When either of these is specified as a nonzero
818 * value, OVS adds a NXAST_FIN_TIMEOUT action, with the specified timeouts, to
824 * The following examples give a prose description of the flow_mod_specs along
825 * with informal notation for how those would be represented and a hex dump of
826 * the bytes that would be required.
828 * These examples could work with various nx_action_learn parameters. Typical
829 * values would be idle_timeout=OFP_FLOW_PERMANENT, hard_timeout=60,
830 * priority=OFP_DEFAULT_PRIORITY, flags=0, table_id=10.
832 * 1. Learn input port based on the source MAC, with lookup into
833 * NXM_NX_REG1[16:31] by resubmit to in_port=99:
835 * Match on in_port=99:
836 * ovs_be16(src=1, dst=0, n_bits=16), 20 10
837 * ovs_be16(99), 00 63
838 * ovs_be32(NXM_OF_IN_PORT), ovs_be16(0) 00 00 00 02 00 00
840 * Match Ethernet destination on Ethernet source from packet:
841 * ovs_be16(src=0, dst=0, n_bits=48), 00 30
842 * ovs_be32(NXM_OF_ETH_SRC), ovs_be16(0) 00 00 04 06 00 00
843 * ovs_be32(NXM_OF_ETH_DST), ovs_be16(0) 00 00 02 06 00 00
845 * Set NXM_NX_REG1[16:31] to the packet's input port:
846 * ovs_be16(src=0, dst=1, n_bits=16), 08 10
847 * ovs_be32(NXM_OF_IN_PORT), ovs_be16(0) 00 00 00 02 00 00
848 * ovs_be32(NXM_NX_REG1), ovs_be16(16) 00 01 02 04 00 10
850 * Given a packet that arrived on port A with Ethernet source address B,
851 * this would set up the flow "in_port=99, dl_dst=B,
852 * actions=load:A->NXM_NX_REG1[16..31]".
854 * In syntax accepted by ovs-ofctl, this action is: learn(in_port=99,
855 * NXM_OF_ETH_DST[]=NXM_OF_ETH_SRC[],
856 * load:NXM_OF_IN_PORT[]->NXM_NX_REG1[16..31])
858 * 2. Output to input port based on the source MAC and VLAN VID, with lookup
859 * into NXM_NX_REG1[16:31]:
861 * Match on same VLAN ID as packet:
862 * ovs_be16(src=0, dst=0, n_bits=12), 00 0c
863 * ovs_be32(NXM_OF_VLAN_TCI), ovs_be16(0) 00 00 08 02 00 00
864 * ovs_be32(NXM_OF_VLAN_TCI), ovs_be16(0) 00 00 08 02 00 00
866 * Match Ethernet destination on Ethernet source from packet:
867 * ovs_be16(src=0, dst=0, n_bits=48), 00 30
868 * ovs_be32(NXM_OF_ETH_SRC), ovs_be16(0) 00 00 04 06 00 00
869 * ovs_be32(NXM_OF_ETH_DST), ovs_be16(0) 00 00 02 06 00 00
871 * Output to the packet's input port:
872 * ovs_be16(src=0, dst=2, n_bits=16), 10 10
873 * ovs_be32(NXM_OF_IN_PORT), ovs_be16(0) 00 00 00 02 00 00
875 * Given a packet that arrived on port A with Ethernet source address B in
876 * VLAN C, this would set up the flow "dl_dst=B, vlan_vid=C,
879 * In syntax accepted by ovs-ofctl, this action is:
880 * learn(NXM_OF_VLAN_TCI[0..11], NXM_OF_ETH_DST[]=NXM_OF_ETH_SRC[],
881 * output:NXM_OF_IN_PORT[])
883 * 3. Here's a recipe for a very simple-minded MAC learning switch. It uses a
884 * 10-second MAC expiration time to make it easier to see what's going on
886 * ovs-vsctl del-controller br0
887 * ovs-ofctl del-flows br0
888 * ovs-ofctl add-flow br0 "table=0 actions=learn(table=1, \
889 hard_timeout=10, NXM_OF_VLAN_TCI[0..11], \
890 NXM_OF_ETH_DST[]=NXM_OF_ETH_SRC[], \
891 output:NXM_OF_IN_PORT[]), resubmit(,1)"
892 * ovs-ofctl add-flow br0 "table=1 priority=0 actions=flood"
894 * You can then dump the MAC learning table with:
896 * ovs-ofctl dump-flows br0 table=1
901 * For best performance, segregate learned flows into a table that is not used
902 * for any other flows except possibly for a lowest-priority "catch-all" flow
903 * (a flow with no match criteria). If different learning actions specify
904 * different match criteria, use different tables for the learned flows.
906 * The meaning of 'hard_timeout' and 'idle_timeout' can be counterintuitive.
907 * These timeouts apply to the flow that is added, which means that a flow with
908 * an idle timeout will expire when no traffic has been sent *to* the learned
909 * address. This is not usually the intent in MAC learning; instead, we want
910 * the MAC learn entry to expire when no traffic has been sent *from* the
911 * learned address. Use a hard timeout for that.
913 struct nx_action_learn {
914 ovs_be16 type; /* OFPAT_VENDOR. */
915 ovs_be16 len; /* At least 24. */
916 ovs_be32 vendor; /* NX_VENDOR_ID. */
917 ovs_be16 subtype; /* NXAST_LEARN. */
918 ovs_be16 idle_timeout; /* Idle time before discarding (seconds). */
919 ovs_be16 hard_timeout; /* Max time before discarding (seconds). */
920 ovs_be16 priority; /* Priority level of flow entry. */
921 ovs_be64 cookie; /* Cookie for new flow. */
922 ovs_be16 flags; /* Either 0 or OFPFF_SEND_FLOW_REM. */
923 uint8_t table_id; /* Table to insert flow entry. */
924 uint8_t pad; /* Must be zero. */
925 ovs_be16 fin_idle_timeout; /* Idle timeout after FIN, if nonzero. */
926 ovs_be16 fin_hard_timeout; /* Hard timeout after FIN, if nonzero. */
927 /* Followed by a sequence of flow_mod_spec elements, as described above,
928 * until the end of the action is reached. */
930 OFP_ASSERT(sizeof(struct nx_action_learn) == 32);
932 #define NX_LEARN_N_BITS_MASK 0x3ff
934 #define NX_LEARN_SRC_FIELD (0 << 13) /* Copy from field. */
935 #define NX_LEARN_SRC_IMMEDIATE (1 << 13) /* Copy from immediate value. */
936 #define NX_LEARN_SRC_MASK (1 << 13)
938 #define NX_LEARN_DST_MATCH (0 << 11) /* Add match criterion. */
939 #define NX_LEARN_DST_LOAD (1 << 11) /* Add NXAST_REG_LOAD action. */
940 #define NX_LEARN_DST_OUTPUT (2 << 11) /* Add OFPAT_OUTPUT action. */
941 #define NX_LEARN_DST_RESERVED (3 << 11) /* Not yet defined. */
942 #define NX_LEARN_DST_MASK (3 << 11)
944 /* Action structure for NXAST_FIN_TIMEOUT.
946 * This action changes the idle timeout or hard timeout, or both, of this
947 * OpenFlow rule when the rule matches a TCP packet with the FIN or RST flag.
948 * When such a packet is observed, the action reduces the rule's idle timeout
949 * to 'fin_idle_timeout' and its hard timeout to 'fin_hard_timeout'. This
950 * action has no effect on an existing timeout that is already shorter than the
951 * one that the action specifies. A 'fin_idle_timeout' or 'fin_hard_timeout'
952 * of zero has no effect on the respective timeout.
954 * 'fin_idle_timeout' and 'fin_hard_timeout' are measured in seconds.
955 * 'fin_hard_timeout' specifies time since the flow's creation, not since the
956 * receipt of the FIN or RST.
958 * This is useful for quickly discarding learned TCP flows that otherwise will
959 * take a long time to expire.
961 * This action is intended for use with an OpenFlow rule that matches only a
962 * single TCP flow. If the rule matches multiple TCP flows (e.g. it wildcards
963 * all TCP traffic, or all TCP traffic to a particular port), then any FIN or
964 * RST in any of those flows will cause the entire OpenFlow rule to expire
965 * early, which is not normally desirable.
967 struct nx_action_fin_timeout {
968 ovs_be16 type; /* OFPAT_VENDOR. */
969 ovs_be16 len; /* 16. */
970 ovs_be32 vendor; /* NX_VENDOR_ID. */
971 ovs_be16 subtype; /* NXAST_FIN_TIMEOUT. */
972 ovs_be16 fin_idle_timeout; /* New idle timeout, if nonzero. */
973 ovs_be16 fin_hard_timeout; /* New hard timeout, if nonzero. */
974 ovs_be16 pad; /* Must be zero. */
976 OFP_ASSERT(sizeof(struct nx_action_fin_timeout) == 16);
978 /* Action structure for NXAST_AUTOPATH.
980 * This action performs the following steps in sequence:
982 * 1. Hashes the flow using an implementation-defined hash function.
984 * The hashed fields' values are drawn from the current state of the
985 * flow, including all modifications that have been made by actions up to
988 * 2. Selects an OpenFlow 'port'.
990 * 'port' is selected in an implementation-defined manner, taking into
991 * account 'id' and the hash value calculated in step 1.
993 * Generally a switch will have been configured with a set of ports that
994 * may be chosen given 'id'. The switch may take into account any number
995 * of factors when choosing 'port' from its configured set. Factors may
996 * include carrier, load, and the results of configuration protocols such
999 * 3. Stores 'port' in dst[ofs:ofs+n_bits].
1001 * The format and semantics of 'dst' and 'ofs_nbits' are similar to those
1002 * for the NXAST_REG_LOAD action.
1004 * The switch will reject actions in which ofs+n_bits is greater than the width
1005 * of 'dst', with error type OFPET_BAD_ACTION, code OFPBAC_BAD_ARGUMENT.
1007 struct nx_action_autopath {
1008 ovs_be16 type; /* OFPAT_VENDOR. */
1009 ovs_be16 len; /* Length is 24. */
1010 ovs_be32 vendor; /* NX_VENDOR_ID. */
1011 ovs_be16 subtype; /* NXAST_AUTOPATH. */
1013 /* Where to store the result. */
1014 ovs_be16 ofs_nbits; /* (ofs << 6) | (n_bits - 1). */
1015 ovs_be32 dst; /* Destination. */
1017 ovs_be32 id; /* Autopath ID. */
1020 OFP_ASSERT(sizeof(struct nx_action_autopath) == 24);
1022 /* Action structure for NXAST_BUNDLE and NXAST_BUNDLE_LOAD.
1024 * The bundle actions choose a slave from a supplied list of options.
1025 * NXAST_BUNDLE outputs to its selection. NXAST_BUNDLE_LOAD writes its
1026 * selection to a register.
1028 * The list of possible slaves follows the nx_action_bundle structure. The size
1029 * of each slave is governed by its type as indicated by the 'slave_type'
1030 * parameter. The list of slaves should be padded at its end with zeros to make
1031 * the total length of the action a multiple of 8.
1033 * Switches infer from the 'slave_type' parameter the size of each slave. All
1034 * implementations must support the NXM_OF_IN_PORT 'slave_type' which indicates
1035 * that the slaves are OpenFlow port numbers with NXM_LENGTH(NXM_OF_IN_PORT) ==
1036 * 2 byte width. Switches should reject actions which indicate unknown or
1037 * unsupported slave types.
1039 * Switches use a strategy dictated by the 'algorithm' parameter to choose a
1040 * slave. If the switch does not support the specified 'algorithm' parameter,
1041 * it should reject the action.
1043 * Several algorithms take into account liveness when selecting slaves. The
1044 * liveness of a slave is implementation defined (with one exception), but will
1045 * generally take into account things like its carrier status and the results
1046 * of any link monitoring protocols which happen to be running on it. In order
1047 * to give controllers a place-holder value, the OFPP_NONE port is always
1050 * Some slave selection strategies require the use of a hash function, in which
1051 * case the 'fields' and 'basis' parameters should be populated. The 'fields'
1052 * parameter (one of NX_HASH_FIELDS_*) designates which parts of the flow to
1053 * hash. Refer to the definition of "enum nx_hash_fields" for details. The
1054 * 'basis' parameter is used as a universal hash parameter. Different values
1055 * of 'basis' yield different hash results.
1057 * The 'zero' parameter at the end of the action structure is reserved for
1058 * future use. Switches are required to reject actions which have nonzero
1059 * bytes in the 'zero' field.
1061 * NXAST_BUNDLE actions should have 'ofs_nbits' and 'dst' zeroed. Switches
1062 * should reject actions which have nonzero bytes in either of these fields.
1064 * NXAST_BUNDLE_LOAD stores the OpenFlow port number of the selected slave in
1065 * dst[ofs:ofs+n_bits]. The format and semantics of 'dst' and 'ofs_nbits' are
1066 * similar to those for the NXAST_REG_LOAD action. */
1067 struct nx_action_bundle {
1068 ovs_be16 type; /* OFPAT_VENDOR. */
1069 ovs_be16 len; /* Length including slaves. */
1070 ovs_be32 vendor; /* NX_VENDOR_ID. */
1071 ovs_be16 subtype; /* NXAST_BUNDLE or NXAST_BUNDLE_LOAD. */
1073 /* Slave choice algorithm to apply to hash value. */
1074 ovs_be16 algorithm; /* One of NX_BD_ALG_*. */
1076 /* What fields to hash and how. */
1077 ovs_be16 fields; /* One of NX_HASH_FIELDS_*. */
1078 ovs_be16 basis; /* Universal hash parameter. */
1080 ovs_be32 slave_type; /* NXM_OF_IN_PORT. */
1081 ovs_be16 n_slaves; /* Number of slaves. */
1083 ovs_be16 ofs_nbits; /* (ofs << 6) | (n_bits - 1). */
1084 ovs_be32 dst; /* Destination. */
1086 uint8_t zero[4]; /* Reserved. Must be zero. */
1088 OFP_ASSERT(sizeof(struct nx_action_bundle) == 32);
1090 /* NXAST_BUNDLE: Bundle slave choice algorithm to apply.
1092 * In the descriptions below, 'slaves' is the list of possible slaves in the
1093 * order they appear in the OpenFlow action. */
1094 enum nx_bd_algorithm {
1095 /* Chooses the first live slave listed in the bundle.
1097 * O(n_slaves) performance. */
1098 NX_BD_ALG_ACTIVE_BACKUP,
1100 /* for i in [0,n_slaves):
1101 * weights[i] = hash(flow, i)
1102 * slave = { slaves[i] such that weights[i] >= weights[j] for all j != i }
1104 * Redistributes 1/n_slaves of traffic when a slave's liveness changes.
1105 * O(n_slaves) performance.
1107 * Uses the 'fields' and 'basis' parameters. */
1108 NX_BD_ALG_HRW /* Highest Random Weight. */
1111 /* Action structure for NXAST_OUTPUT_REG.
1113 * Outputs to the OpenFlow port number written to src[ofs:ofs+nbits].
1115 * The format and semantics of 'src' and 'ofs_nbits' are similar to those for
1116 * the NXAST_REG_LOAD action.
1118 * The acceptable nxm_header values for 'src' are the same as the acceptable
1119 * nxm_header values for the 'src' field of NXAST_REG_MOVE.
1121 * The 'max_len' field indicates the number of bytes to send when the chosen
1122 * port is OFPP_CONTROLLER. Its semantics are equivalent to the 'max_len'
1123 * field of OFPAT_OUTPUT.
1125 * The 'zero' field is required to be zeroed for forward compatibility. */
1126 struct nx_action_output_reg {
1127 ovs_be16 type; /* OFPAT_VENDOR. */
1128 ovs_be16 len; /* 24. */
1129 ovs_be32 vendor; /* NX_VENDOR_ID. */
1130 ovs_be16 subtype; /* NXAST_OUTPUT_REG. */
1132 ovs_be16 ofs_nbits; /* (ofs << 6) | (n_bits - 1). */
1133 ovs_be32 src; /* Source. */
1135 ovs_be16 max_len; /* Max length to send to controller. */
1137 uint8_t zero[6]; /* Reserved, must be zero. */
1139 OFP_ASSERT(sizeof(struct nx_action_output_reg) == 24);
1143 * Discontinues action processing.
1145 * The NXAST_EXIT action causes the switch to immediately halt processing
1146 * actions for the flow. Any actions which have already been processed are
1147 * executed by the switch. However, any further actions, including those which
1148 * may be in different tables, or different levels of the NXAST_RESUBMIT
1149 * hierarchy, will be ignored.
1151 * Uses the nx_action_header structure. */
1153 /* Flexible flow specifications (aka NXM = Nicira Extended Match).
1155 * OpenFlow 1.0 has "struct ofp10_match" for specifying flow matches. This
1156 * structure is fixed-length and hence difficult to extend. This section
1157 * describes a more flexible, variable-length flow match, called "nx_match" for
1158 * short, that is also supported by Open vSwitch. This section also defines a
1159 * replacement for each OpenFlow message that includes struct ofp10_match.
1165 * An nx_match is a sequence of zero or more "nxm_entry"s, which are
1166 * type-length-value (TLV) entries, each 5 to 259 (inclusive) bytes long.
1167 * "nxm_entry"s are not aligned on or padded to any multibyte boundary. The
1168 * first 4 bytes of an nxm_entry are its "header", followed by the entry's
1171 * An nxm_entry's header is interpreted as a 32-bit word in network byte order:
1173 * |<-------------------- nxm_type ------------------>|
1175 * |31 16 15 9| 8 7 0
1176 * +----------------------------------+---------------+--+------------------+
1177 * | nxm_vendor | nxm_field |hm| nxm_length |
1178 * +----------------------------------+---------------+--+------------------+
1180 * The most-significant 23 bits of the header are collectively "nxm_type".
1181 * Bits 16...31 are "nxm_vendor", one of the NXM_VENDOR_* values below. Bits
1182 * 9...15 are "nxm_field", which is a vendor-specific value. nxm_type normally
1183 * designates a protocol header, such as the Ethernet type, but it can also
1184 * refer to packet metadata, such as the switch port on which a packet arrived.
1186 * Bit 8 is "nxm_hasmask" (labeled "hm" above for space reasons). The meaning
1187 * of this bit is explained later.
1189 * The least-significant 8 bits are "nxm_length", a positive integer. The
1190 * length of the nxm_entry, including the header, is exactly 4 + nxm_length
1193 * For a given nxm_vendor, nxm_field, and nxm_hasmask value, nxm_length is a
1194 * constant. It is included only to allow software to minimally parse
1195 * "nxm_entry"s of unknown types. (Similarly, for a given nxm_vendor,
1196 * nxm_field, and nxm_length, nxm_hasmask is a constant.)
1202 * A zero-length nx_match (one with no "nxm_entry"s) matches every packet.
1204 * An nxm_entry places a constraint on the packets matched by the nx_match:
1206 * - If nxm_hasmask is 0, the nxm_entry's body contains a value for the
1207 * field, called "nxm_value". The nx_match matches only packets in which
1208 * the field equals nxm_value.
1210 * - If nxm_hasmask is 1, then the nxm_entry's body contains a value for the
1211 * field (nxm_value), followed by a bitmask of the same length as the
1212 * value, called "nxm_mask". For each 1-bit in position J in nxm_mask, the
1213 * nx_match matches only packets for which bit J in the given field's value
1214 * matches bit J in nxm_value. A 0-bit in nxm_mask causes the
1215 * corresponding bits in nxm_value and the field's value to be ignored.
1216 * (The sense of the nxm_mask bits is the opposite of that used by the
1217 * "wildcards" member of struct ofp10_match.)
1219 * When nxm_hasmask is 1, nxm_length is always even.
1221 * An all-zero-bits nxm_mask is equivalent to omitting the nxm_entry
1222 * entirely. An all-one-bits nxm_mask is equivalent to specifying 0 for
1225 * When there are multiple "nxm_entry"s, all of the constraints must be met.
1231 * Masks may be restricted:
1233 * - Some nxm_types may not support masked wildcards, that is, nxm_hasmask
1234 * must always be 0 when these fields are specified. For example, the
1235 * field that identifies the port on which a packet was received may not be
1238 * - Some nxm_types that do support masked wildcards may only support certain
1239 * nxm_mask patterns. For example, fields that have IPv4 address values
1240 * may be restricted to CIDR masks.
1242 * These restrictions should be noted in specifications for individual fields.
1243 * A switch may accept an nxm_hasmask or nxm_mask value that the specification
1244 * disallows, if the switch correctly implements support for that nxm_hasmask
1245 * or nxm_mask value. A switch must reject an attempt to set up a flow that
1246 * contains a nxm_hasmask or nxm_mask value that it does not support.
1249 * Prerequisite Restrictions
1250 * =========================
1252 * The presence of an nxm_entry with a given nxm_type may be restricted based
1253 * on the presence of or values of other "nxm_entry"s. For example:
1255 * - An nxm_entry for nxm_type=NXM_OF_IP_TOS is allowed only if it is
1256 * preceded by another entry with nxm_type=NXM_OF_ETH_TYPE, nxm_hasmask=0,
1257 * and nxm_value=0x0800. That is, matching on the IP source address is
1258 * allowed only if the Ethernet type is explicitly set to IP.
1260 * - An nxm_entry for nxm_type=NXM_OF_TCP_SRC is allowed only if it is
1261 * preceded by an entry with nxm_type=NXM_OF_ETH_TYPE, nxm_hasmask=0, and
1262 * nxm_value either 0x0800 or 0x86dd, and another with
1263 * nxm_type=NXM_OF_IP_PROTO, nxm_hasmask=0, nxm_value=6, in that order.
1264 * That is, matching on the TCP source port is allowed only if the Ethernet
1265 * type is IP or IPv6 and the IP protocol is TCP.
1267 * These restrictions should be noted in specifications for individual fields.
1268 * A switch may implement relaxed versions of these restrictions. A switch
1269 * must reject an attempt to set up a flow that violates its restrictions.
1272 * Ordering Restrictions
1273 * =====================
1275 * An nxm_entry that has prerequisite restrictions must appear after the
1276 * "nxm_entry"s for its prerequisites. Ordering of "nxm_entry"s within an
1277 * nx_match is not otherwise constrained.
1279 * Any given nxm_type may appear in an nx_match at most once.
1282 * nxm_entry Examples
1283 * ==================
1285 * These examples show the format of a single nxm_entry with particular
1286 * nxm_hasmask and nxm_length values. The diagrams are labeled with field
1287 * numbers and byte indexes.
1290 * 8-bit nxm_value, nxm_hasmask=1, nxm_length=2:
1293 * +------------+---+---+
1294 * | header | v | m |
1295 * +------------+---+---+
1298 * 16-bit nxm_value, nxm_hasmask=0, nxm_length=2:
1301 * +------------+------+
1303 * +------------+------+
1306 * 32-bit nxm_value, nxm_hasmask=0, nxm_length=4:
1309 * +------------+-------------+
1310 * | header | nxm_value |
1311 * +------------+-------------+
1314 * 48-bit nxm_value, nxm_hasmask=0, nxm_length=6:
1317 * +------------+------------------+
1318 * | header | nxm_value |
1319 * +------------+------------------+
1322 * 48-bit nxm_value, nxm_hasmask=1, nxm_length=12:
1325 * +------------+------------------+------------------+
1326 * | header | nxm_value | nxm_mask |
1327 * +------------+------------------+------------------+
1333 * A switch should report an error in an nx_match using error type
1334 * OFPET_BAD_REQUEST and one of the NXBRC_NXM_* codes. Ideally the switch
1335 * should report a specific error code, if one is assigned for the particular
1336 * problem, but NXBRC_NXM_INVALID is also available to report a generic
1340 #define NXM_HEADER__(VENDOR, FIELD, HASMASK, LENGTH) \
1341 (((VENDOR) << 16) | ((FIELD) << 9) | ((HASMASK) << 8) | (LENGTH))
1342 #define NXM_HEADER(VENDOR, FIELD, LENGTH) \
1343 NXM_HEADER__(VENDOR, FIELD, 0, LENGTH)
1344 #define NXM_HEADER_W(VENDOR, FIELD, LENGTH) \
1345 NXM_HEADER__(VENDOR, FIELD, 1, (LENGTH) * 2)
1346 #define NXM_VENDOR(HEADER) ((HEADER) >> 16)
1347 #define NXM_FIELD(HEADER) (((HEADER) >> 9) & 0x7f)
1348 #define NXM_TYPE(HEADER) (((HEADER) >> 9) & 0x7fffff)
1349 #define NXM_HASMASK(HEADER) (((HEADER) >> 8) & 1)
1350 #define NXM_LENGTH(HEADER) ((HEADER) & 0xff)
1352 #define NXM_MAKE_WILD_HEADER(HEADER) \
1353 NXM_HEADER_W(NXM_VENDOR(HEADER), NXM_FIELD(HEADER), NXM_LENGTH(HEADER))
1355 /* ## ------------------------------- ## */
1356 /* ## OpenFlow 1.0-compatible fields. ## */
1357 /* ## ------------------------------- ## */
1359 /* Physical or virtual port on which the packet was received.
1363 * Format: 16-bit integer in network byte order.
1365 * Masking: Not maskable. */
1366 #define NXM_OF_IN_PORT NXM_HEADER (0x0000, 0, 2)
1368 /* Source or destination address in Ethernet header.
1372 * Format: 48-bit Ethernet MAC address.
1374 * Masking: Fully maskable, in versions 1.8 and later. Earlier versions only
1375 * supported the following masks for NXM_OF_ETH_DST_W: 00:00:00:00:00:00,
1376 * fe:ff:ff:ff:ff:ff, 01:00:00:00:00:00, ff:ff:ff:ff:ff:ff. */
1377 #define NXM_OF_ETH_DST NXM_HEADER (0x0000, 1, 6)
1378 #define NXM_OF_ETH_DST_W NXM_HEADER_W(0x0000, 1, 6)
1379 #define NXM_OF_ETH_SRC NXM_HEADER (0x0000, 2, 6)
1380 #define NXM_OF_ETH_SRC_W NXM_HEADER_W(0x0000, 2, 6)
1382 /* Packet's Ethernet type.
1384 * For an Ethernet II packet this is taken from the Ethernet header. For an
1385 * 802.2 LLC+SNAP header with OUI 00-00-00 this is taken from the SNAP header.
1386 * A packet that has neither format has value 0x05ff
1387 * (OFP_DL_TYPE_NOT_ETH_TYPE).
1389 * For a packet with an 802.1Q header, this is the type of the encapsulated
1394 * Format: 16-bit integer in network byte order.
1396 * Masking: Not maskable. */
1397 #define NXM_OF_ETH_TYPE NXM_HEADER (0x0000, 3, 2)
1401 * For a packet with an 802.1Q header, this is the Tag Control Information
1402 * (TCI) field, with the CFI bit forced to 1. For a packet with no 802.1Q
1403 * header, this has value 0.
1407 * Format: 16-bit integer in network byte order.
1409 * Masking: Arbitrary masks.
1411 * This field can be used in various ways:
1413 * - If it is not constrained at all, the nx_match matches packets without
1414 * an 802.1Q header or with an 802.1Q header that has any TCI value.
1416 * - Testing for an exact match with 0 matches only packets without an
1419 * - Testing for an exact match with a TCI value with CFI=1 matches packets
1420 * that have an 802.1Q header with a specified VID and PCP.
1422 * - Testing for an exact match with a nonzero TCI value with CFI=0 does
1423 * not make sense. The switch may reject this combination.
1425 * - Testing with a specific VID and CFI=1, with nxm_mask=0x1fff, matches
1426 * packets that have an 802.1Q header with that VID (and any PCP).
1428 * - Testing with a specific PCP and CFI=1, with nxm_mask=0xf000, matches
1429 * packets that have an 802.1Q header with that PCP (and any VID).
1431 * - Testing with nxm_value=0, nxm_mask=0x0fff matches packets with no 802.1Q
1432 * header or with an 802.1Q header with a VID of 0.
1434 * - Testing with nxm_value=0, nxm_mask=0xe000 matches packets with no 802.1Q
1435 * header or with an 802.1Q header with a PCP of 0.
1437 * - Testing with nxm_value=0, nxm_mask=0xefff matches packets with no 802.1Q
1438 * header or with an 802.1Q header with both VID and PCP of 0.
1440 #define NXM_OF_VLAN_TCI NXM_HEADER (0x0000, 4, 2)
1441 #define NXM_OF_VLAN_TCI_W NXM_HEADER_W(0x0000, 4, 2)
1443 /* The "type of service" byte of the IP header, with the ECN bits forced to 0.
1445 * Prereqs: NXM_OF_ETH_TYPE must be either 0x0800 or 0x86dd.
1447 * Format: 8-bit integer with 2 least-significant bits forced to 0.
1449 * Masking: Not maskable. */
1450 #define NXM_OF_IP_TOS NXM_HEADER (0x0000, 5, 1)
1452 /* The "protocol" byte in the IP header.
1454 * Prereqs: NXM_OF_ETH_TYPE must be either 0x0800 or 0x86dd.
1456 * Format: 8-bit integer.
1458 * Masking: Not maskable. */
1459 #define NXM_OF_IP_PROTO NXM_HEADER (0x0000, 6, 1)
1461 /* The source or destination address in the IP header.
1463 * Prereqs: NXM_OF_ETH_TYPE must match 0x0800 exactly.
1465 * Format: 32-bit integer in network byte order.
1467 * Masking: Fully maskable, in Open vSwitch 1.8 and later. In earlier
1468 * versions, only CIDR masks are allowed, that is, masks that consist of N
1469 * high-order bits set to 1 and the other 32-N bits set to 0. */
1470 #define NXM_OF_IP_SRC NXM_HEADER (0x0000, 7, 4)
1471 #define NXM_OF_IP_SRC_W NXM_HEADER_W(0x0000, 7, 4)
1472 #define NXM_OF_IP_DST NXM_HEADER (0x0000, 8, 4)
1473 #define NXM_OF_IP_DST_W NXM_HEADER_W(0x0000, 8, 4)
1475 /* The source or destination port in the TCP header.
1478 * NXM_OF_ETH_TYPE must be either 0x0800 or 0x86dd.
1479 * NXM_OF_IP_PROTO must match 6 exactly.
1481 * Format: 16-bit integer in network byte order.
1483 * Masking: Fully maskable, in Open vSwitch 1.6 and later. Not maskable, in
1484 * earlier versions. */
1485 #define NXM_OF_TCP_SRC NXM_HEADER (0x0000, 9, 2)
1486 #define NXM_OF_TCP_SRC_W NXM_HEADER_W(0x0000, 9, 2)
1487 #define NXM_OF_TCP_DST NXM_HEADER (0x0000, 10, 2)
1488 #define NXM_OF_TCP_DST_W NXM_HEADER_W(0x0000, 10, 2)
1490 /* The source or destination port in the UDP header.
1493 * NXM_OF_ETH_TYPE must match either 0x0800 or 0x86dd.
1494 * NXM_OF_IP_PROTO must match 17 exactly.
1496 * Format: 16-bit integer in network byte order.
1498 * Masking: Fully maskable, in Open vSwitch 1.6 and later. Not maskable, in
1499 * earlier versions. */
1500 #define NXM_OF_UDP_SRC NXM_HEADER (0x0000, 11, 2)
1501 #define NXM_OF_UDP_SRC_W NXM_HEADER_W(0x0000, 11, 2)
1502 #define NXM_OF_UDP_DST NXM_HEADER (0x0000, 12, 2)
1503 #define NXM_OF_UDP_DST_W NXM_HEADER_W(0x0000, 12, 2)
1505 /* The type or code in the ICMP header.
1508 * NXM_OF_ETH_TYPE must match 0x0800 exactly.
1509 * NXM_OF_IP_PROTO must match 1 exactly.
1511 * Format: 8-bit integer.
1513 * Masking: Not maskable. */
1514 #define NXM_OF_ICMP_TYPE NXM_HEADER (0x0000, 13, 1)
1515 #define NXM_OF_ICMP_CODE NXM_HEADER (0x0000, 14, 1)
1519 * For an Ethernet+IP ARP packet, the opcode in the ARP header. Always 0
1520 * otherwise. Only ARP opcodes between 1 and 255 should be specified for
1523 * Prereqs: NXM_OF_ETH_TYPE must match 0x0806 exactly.
1525 * Format: 16-bit integer in network byte order.
1527 * Masking: Not maskable. */
1528 #define NXM_OF_ARP_OP NXM_HEADER (0x0000, 15, 2)
1530 /* For an Ethernet+IP ARP packet, the source or target protocol address
1531 * in the ARP header. Always 0 otherwise.
1533 * Prereqs: NXM_OF_ETH_TYPE must match 0x0806 exactly.
1535 * Format: 32-bit integer in network byte order.
1537 * Masking: Fully maskable, in Open vSwitch 1.8 and later. In earlier
1538 * versions, only CIDR masks are allowed, that is, masks that consist of N
1539 * high-order bits set to 1 and the other 32-N bits set to 0. */
1540 #define NXM_OF_ARP_SPA NXM_HEADER (0x0000, 16, 4)
1541 #define NXM_OF_ARP_SPA_W NXM_HEADER_W(0x0000, 16, 4)
1542 #define NXM_OF_ARP_TPA NXM_HEADER (0x0000, 17, 4)
1543 #define NXM_OF_ARP_TPA_W NXM_HEADER_W(0x0000, 17, 4)
1545 /* ## ------------------------ ## */
1546 /* ## Nicira match extensions. ## */
1547 /* ## ------------------------ ## */
1549 /* Metadata registers.
1551 * Registers initially have value 0. Actions allow register values to be
1556 * Format: Array of 32-bit integer registers. Space is reserved for up to
1557 * NXM_NX_MAX_REGS registers, but switches may implement fewer.
1559 * Masking: Arbitrary masks. */
1560 #define NXM_NX_MAX_REGS 16
1561 #define NXM_NX_REG(IDX) NXM_HEADER (0x0001, IDX, 4)
1562 #define NXM_NX_REG_W(IDX) NXM_HEADER_W(0x0001, IDX, 4)
1563 #define NXM_NX_REG_IDX(HEADER) NXM_FIELD(HEADER)
1564 #define NXM_IS_NX_REG(HEADER) (!((((HEADER) ^ NXM_NX_REG0)) & 0xffffe1ff))
1565 #define NXM_IS_NX_REG_W(HEADER) (!((((HEADER) ^ NXM_NX_REG0_W)) & 0xffffe1ff))
1566 #define NXM_NX_REG0 NXM_HEADER (0x0001, 0, 4)
1567 #define NXM_NX_REG0_W NXM_HEADER_W(0x0001, 0, 4)
1568 #define NXM_NX_REG1 NXM_HEADER (0x0001, 1, 4)
1569 #define NXM_NX_REG1_W NXM_HEADER_W(0x0001, 1, 4)
1570 #define NXM_NX_REG2 NXM_HEADER (0x0001, 2, 4)
1571 #define NXM_NX_REG2_W NXM_HEADER_W(0x0001, 2, 4)
1572 #define NXM_NX_REG3 NXM_HEADER (0x0001, 3, 4)
1573 #define NXM_NX_REG3_W NXM_HEADER_W(0x0001, 3, 4)
1574 #define NXM_NX_REG4 NXM_HEADER (0x0001, 4, 4)
1575 #define NXM_NX_REG4_W NXM_HEADER_W(0x0001, 4, 4)
1576 #define NXM_NX_REG5 NXM_HEADER (0x0001, 5, 4)
1577 #define NXM_NX_REG5_W NXM_HEADER_W(0x0001, 5, 4)
1578 #define NXM_NX_REG6 NXM_HEADER (0x0001, 6, 4)
1579 #define NXM_NX_REG6_W NXM_HEADER_W(0x0001, 6, 4)
1580 #define NXM_NX_REG7 NXM_HEADER (0x0001, 7, 4)
1581 #define NXM_NX_REG7_W NXM_HEADER_W(0x0001, 7, 4)
1585 * For a packet received via GRE tunnel including a (32-bit) key, the key is
1586 * stored in the low 32-bits and the high bits are zeroed. For other packets,
1591 * Format: 64-bit integer in network byte order.
1593 * Masking: Arbitrary masks. */
1594 #define NXM_NX_TUN_ID NXM_HEADER (0x0001, 16, 8)
1595 #define NXM_NX_TUN_ID_W NXM_HEADER_W(0x0001, 16, 8)
1597 /* For an Ethernet+IP ARP packet, the source or target hardware address
1598 * in the ARP header. Always 0 otherwise.
1600 * Prereqs: NXM_OF_ETH_TYPE must match 0x0806 exactly.
1602 * Format: 48-bit Ethernet MAC address.
1604 * Masking: Not maskable. */
1605 #define NXM_NX_ARP_SHA NXM_HEADER (0x0001, 17, 6)
1606 #define NXM_NX_ARP_THA NXM_HEADER (0x0001, 18, 6)
1608 /* The source or destination address in the IPv6 header.
1610 * Prereqs: NXM_OF_ETH_TYPE must match 0x86dd exactly.
1612 * Format: 128-bit IPv6 address.
1614 * Masking: Fully maskable, in Open vSwitch 1.8 and later. In previous
1615 * versions, only CIDR masks are allowed, that is, masks that consist of N
1616 * high-order bits set to 1 and the other 128-N bits set to 0. */
1617 #define NXM_NX_IPV6_SRC NXM_HEADER (0x0001, 19, 16)
1618 #define NXM_NX_IPV6_SRC_W NXM_HEADER_W(0x0001, 19, 16)
1619 #define NXM_NX_IPV6_DST NXM_HEADER (0x0001, 20, 16)
1620 #define NXM_NX_IPV6_DST_W NXM_HEADER_W(0x0001, 20, 16)
1622 /* The type or code in the ICMPv6 header.
1625 * NXM_OF_ETH_TYPE must match 0x86dd exactly.
1626 * NXM_OF_IP_PROTO must match 58 exactly.
1628 * Format: 8-bit integer.
1630 * Masking: Not maskable. */
1631 #define NXM_NX_ICMPV6_TYPE NXM_HEADER (0x0001, 21, 1)
1632 #define NXM_NX_ICMPV6_CODE NXM_HEADER (0x0001, 22, 1)
1634 /* The target address in an IPv6 Neighbor Discovery message.
1637 * NXM_OF_ETH_TYPE must match 0x86dd exactly.
1638 * NXM_OF_IP_PROTO must match 58 exactly.
1639 * NXM_OF_ICMPV6_TYPE must be either 135 or 136.
1641 * Format: 128-bit IPv6 address.
1643 * Masking: Fully maskable, in Open vSwitch 1.8 and later. In previous
1644 * versions, only CIDR masks are allowed, that is, masks that consist of N
1645 * high-order bits set to 1 and the other 128-N bits set to 0. */
1646 #define NXM_NX_ND_TARGET NXM_HEADER (0x0001, 23, 16)
1647 #define NXM_NX_ND_TARGET_W NXM_HEADER_W (0x0001, 23, 16)
1649 /* The source link-layer address option in an IPv6 Neighbor Discovery
1653 * NXM_OF_ETH_TYPE must match 0x86dd exactly.
1654 * NXM_OF_IP_PROTO must match 58 exactly.
1655 * NXM_OF_ICMPV6_TYPE must be exactly 135.
1657 * Format: 48-bit Ethernet MAC address.
1659 * Masking: Not maskable. */
1660 #define NXM_NX_ND_SLL NXM_HEADER (0x0001, 24, 6)
1662 /* The target link-layer address option in an IPv6 Neighbor Discovery
1666 * NXM_OF_ETH_TYPE must match 0x86dd exactly.
1667 * NXM_OF_IP_PROTO must match 58 exactly.
1668 * NXM_OF_ICMPV6_TYPE must be exactly 136.
1670 * Format: 48-bit Ethernet MAC address.
1672 * Masking: Not maskable. */
1673 #define NXM_NX_ND_TLL NXM_HEADER (0x0001, 25, 6)
1675 /* IP fragment information.
1678 * NXM_OF_ETH_TYPE must be either 0x0800 or 0x86dd.
1680 * Format: 8-bit value with one of the values 0, 1, or 3, as described below.
1682 * Masking: Fully maskable.
1684 * This field has three possible values:
1686 * - A packet that is not an IP fragment has value 0.
1688 * - A packet that is an IP fragment with offset 0 (the first fragment) has
1689 * bit 0 set and thus value 1.
1691 * - A packet that is an IP fragment with nonzero offset has bits 0 and 1 set
1694 * NX_IP_FRAG_ANY and NX_IP_FRAG_LATER are declared to symbolically represent
1695 * the meanings of bits 0 and 1.
1697 * The switch may reject matches against values that can never appear.
1699 * It is important to understand how this field interacts with the OpenFlow IP
1700 * fragment handling mode:
1702 * - In OFPC_FRAG_DROP mode, the OpenFlow switch drops all IP fragments
1703 * before they reach the flow table, so every packet that is available for
1704 * matching will have value 0 in this field.
1706 * - Open vSwitch does not implement OFPC_FRAG_REASM mode, but if it did then
1707 * IP fragments would be reassembled before they reached the flow table and
1708 * again every packet available for matching would always have value 0.
1710 * - In OFPC_FRAG_NORMAL mode, all three values are possible, but OpenFlow
1711 * 1.0 says that fragments' transport ports are always 0, even for the
1712 * first fragment, so this does not provide much extra information.
1714 * - In OFPC_FRAG_NX_MATCH mode, all three values are possible. For
1715 * fragments with offset 0, Open vSwitch makes L4 header information
1718 #define NXM_NX_IP_FRAG NXM_HEADER (0x0001, 26, 1)
1719 #define NXM_NX_IP_FRAG_W NXM_HEADER_W(0x0001, 26, 1)
1721 /* Bits in the value of NXM_NX_IP_FRAG. */
1722 #define NX_IP_FRAG_ANY (1 << 0) /* Is this a fragment? */
1723 #define NX_IP_FRAG_LATER (1 << 1) /* Is this a fragment with nonzero offset? */
1725 /* The flow label in the IPv6 header.
1727 * Prereqs: NXM_OF_ETH_TYPE must match 0x86dd exactly.
1729 * Format: 20-bit IPv6 flow label in least-significant bits.
1731 * Masking: Not maskable. */
1732 #define NXM_NX_IPV6_LABEL NXM_HEADER (0x0001, 27, 4)
1734 /* The ECN of the IP header.
1736 * Prereqs: NXM_OF_ETH_TYPE must be either 0x0800 or 0x86dd.
1738 * Format: ECN in the low-order 2 bits.
1740 * Masking: Not maskable. */
1741 #define NXM_NX_IP_ECN NXM_HEADER (0x0001, 28, 1)
1743 /* The time-to-live/hop limit of the IP header.
1745 * Prereqs: NXM_OF_ETH_TYPE must be either 0x0800 or 0x86dd.
1747 * Format: 8-bit integer.
1749 * Masking: Not maskable. */
1750 #define NXM_NX_IP_TTL NXM_HEADER (0x0001, 29, 1)
1754 * This may be used to gain the OpenFlow 1.1-like ability to restrict
1755 * certain NXM-based Flow Mod and Flow Stats Request messages to flows
1756 * with specific cookies. See the "nx_flow_mod" and "nx_flow_stats_request"
1757 * structure definitions for more details. This match is otherwise not
1762 * Format: 64-bit integer in network byte order.
1764 * Masking: Arbitrary masks. */
1765 #define NXM_NX_COOKIE NXM_HEADER (0x0001, 30, 8)
1766 #define NXM_NX_COOKIE_W NXM_HEADER_W(0x0001, 30, 8)
1768 /* ## --------------------- ## */
1769 /* ## Requests and replies. ## */
1770 /* ## --------------------- ## */
1772 enum nx_flow_format {
1773 NXFF_OPENFLOW10 = 0, /* Standard OpenFlow 1.0 compatible. */
1774 NXFF_NXM = 2 /* Nicira extended match. */
1777 /* NXT_SET_FLOW_FORMAT request. */
1778 struct nx_set_flow_format {
1779 struct nicira_header nxh;
1780 ovs_be32 format; /* One of NXFF_*. */
1782 OFP_ASSERT(sizeof(struct nx_set_flow_format) == 20);
1784 /* NXT_FLOW_MOD (analogous to OFPT_FLOW_MOD).
1786 * It is possible to limit flow deletions and modifications to certain
1787 * cookies by using the NXM_NX_COOKIE(_W) matches. The "cookie" field
1788 * is used only to add or modify flow cookies.
1790 struct nx_flow_mod {
1791 struct nicira_header nxh;
1792 ovs_be64 cookie; /* Opaque controller-issued identifier. */
1793 ovs_be16 command; /* One of OFPFC_*. */
1794 ovs_be16 idle_timeout; /* Idle time before discarding (seconds). */
1795 ovs_be16 hard_timeout; /* Max time before discarding (seconds). */
1796 ovs_be16 priority; /* Priority level of flow entry. */
1797 ovs_be32 buffer_id; /* Buffered packet to apply to (or -1).
1798 Not meaningful for OFPFC_DELETE*. */
1799 ovs_be16 out_port; /* For OFPFC_DELETE* commands, require
1800 matching entries to include this as an
1801 output port. A value of OFPP_NONE
1802 indicates no restriction. */
1803 ovs_be16 flags; /* One of OFPFF_*. */
1804 ovs_be16 match_len; /* Size of nx_match. */
1805 uint8_t pad[6]; /* Align to 64-bits. */
1807 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
1808 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
1809 * all-zero bytes, then
1810 * - Actions to fill out the remainder of the message length (always a
1814 OFP_ASSERT(sizeof(struct nx_flow_mod) == 48);
1816 /* NXT_FLOW_REMOVED (analogous to OFPT_FLOW_REMOVED). */
1817 struct nx_flow_removed {
1818 struct nicira_header nxh;
1819 ovs_be64 cookie; /* Opaque controller-issued identifier. */
1820 ovs_be16 priority; /* Priority level of flow entry. */
1821 uint8_t reason; /* One of OFPRR_*. */
1822 uint8_t pad[1]; /* Align to 32-bits. */
1823 ovs_be32 duration_sec; /* Time flow was alive in seconds. */
1824 ovs_be32 duration_nsec; /* Time flow was alive in nanoseconds beyond
1826 ovs_be16 idle_timeout; /* Idle timeout from original flow mod. */
1827 ovs_be16 match_len; /* Size of nx_match. */
1828 ovs_be64 packet_count;
1829 ovs_be64 byte_count;
1831 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
1832 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
1833 * all-zero bytes. */
1835 OFP_ASSERT(sizeof(struct nx_flow_removed) == 56);
1837 /* Nicira vendor stats request of type NXST_FLOW (analogous to OFPST_FLOW
1840 * It is possible to limit matches to certain cookies by using the
1841 * NXM_NX_COOKIE and NXM_NX_COOKIE_W matches.
1843 struct nx_flow_stats_request {
1844 struct nicira_stats_msg nsm;
1845 ovs_be16 out_port; /* Require matching entries to include this
1846 as an output port. A value of OFPP_NONE
1847 indicates no restriction. */
1848 ovs_be16 match_len; /* Length of nx_match. */
1849 uint8_t table_id; /* ID of table to read (from ofp_table_stats)
1850 or 0xff for all tables. */
1851 uint8_t pad[3]; /* Align to 64 bits. */
1853 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
1854 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
1855 * all-zero bytes, which must also exactly fill out the length of the
1859 OFP_ASSERT(sizeof(struct nx_flow_stats_request) == 32);
1861 /* Body for Nicira vendor stats reply of type NXST_FLOW (analogous to
1862 * OFPST_FLOW reply).
1864 * The values of 'idle_age' and 'hard_age' are only meaningful when talking to
1865 * a switch that implements the NXT_FLOW_AGE extension. Zero means that the
1866 * true value is unknown, perhaps because hardware does not track the value.
1867 * (Zero is also the value that one should ordinarily expect to see talking to
1868 * a switch that does not implement NXT_FLOW_AGE, since those switches zero the
1869 * padding bytes that these fields replaced.) A nonzero value X represents X-1
1870 * seconds. A value of 65535 represents 65534 or more seconds.
1872 * 'idle_age' is the number of seconds that the flow has been idle, that is,
1873 * the number of seconds since a packet passed through the flow. 'hard_age' is
1874 * the number of seconds since the flow was last modified (e.g. OFPFC_MODIFY or
1875 * OFPFC_MODIFY_STRICT). (The 'duration_*' fields are the elapsed time since
1876 * the flow was added, regardless of subsequent modifications.)
1878 * For a flow with an idle or hard timeout, 'idle_age' or 'hard_age',
1879 * respectively, will ordinarily be smaller than the timeout, but flow
1880 * expiration times are only approximate and so one must be prepared to
1881 * tolerate expirations that occur somewhat early or late.
1883 struct nx_flow_stats {
1884 ovs_be16 length; /* Length of this entry. */
1885 uint8_t table_id; /* ID of table flow came from. */
1887 ovs_be32 duration_sec; /* Time flow has been alive in seconds. */
1888 ovs_be32 duration_nsec; /* Time flow has been alive in nanoseconds
1889 beyond duration_sec. */
1890 ovs_be16 priority; /* Priority of the entry. Only meaningful
1891 when this is not an exact-match entry. */
1892 ovs_be16 idle_timeout; /* Number of seconds idle before expiration. */
1893 ovs_be16 hard_timeout; /* Number of seconds before expiration. */
1894 ovs_be16 match_len; /* Length of nx_match. */
1895 ovs_be16 idle_age; /* Seconds since last packet, plus one. */
1896 ovs_be16 hard_age; /* Seconds since last modification, plus one. */
1897 ovs_be64 cookie; /* Opaque controller-issued identifier. */
1898 ovs_be64 packet_count; /* Number of packets, UINT64_MAX if unknown. */
1899 ovs_be64 byte_count; /* Number of bytes, UINT64_MAX if unknown. */
1901 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
1902 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
1903 * all-zero bytes, then
1904 * - Actions to fill out the remainder 'length' bytes (always a multiple
1908 OFP_ASSERT(sizeof(struct nx_flow_stats) == 48);
1910 /* Nicira vendor stats request of type NXST_AGGREGATE (analogous to
1911 * OFPST_AGGREGATE request). */
1912 struct nx_aggregate_stats_request {
1913 struct nicira_stats_msg nsm;
1914 ovs_be16 out_port; /* Require matching entries to include this
1915 as an output port. A value of OFPP_NONE
1916 indicates no restriction. */
1917 ovs_be16 match_len; /* Length of nx_match. */
1918 uint8_t table_id; /* ID of table to read (from ofp_table_stats)
1919 or 0xff for all tables. */
1920 uint8_t pad[3]; /* Align to 64 bits. */
1922 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
1923 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
1924 * all-zero bytes, which must also exactly fill out the length of the
1928 OFP_ASSERT(sizeof(struct nx_aggregate_stats_request) == 32);
1930 /* Body for nicira_stats_msg reply of type NXST_AGGREGATE (analogous to
1931 * OFPST_AGGREGATE reply). */
1932 struct nx_aggregate_stats_reply {
1933 struct nicira_stats_msg nsm;
1934 ovs_be64 packet_count; /* Number of packets, UINT64_MAX if unknown. */
1935 ovs_be64 byte_count; /* Number of bytes, UINT64_MAX if unknown. */
1936 ovs_be32 flow_count; /* Number of flows. */
1937 uint8_t pad[4]; /* Align to 64 bits. */
1939 OFP_ASSERT(sizeof(struct nx_aggregate_stats_reply) == 48);
1941 /* NXT_SET_CONTROLLER_ID.
1943 * Each OpenFlow controller connection has a 16-bit identifier that is
1944 * initially 0. This message changes the connection's ID to 'id'.
1946 * Controller connection IDs need not be unique.
1948 * The NXAST_CONTROLLER action is the only current user of controller
1949 * connection IDs. */
1950 struct nx_controller_id {
1951 struct nicira_header nxh;
1952 uint8_t zero[6]; /* Must be zero. */
1953 ovs_be16 controller_id; /* New controller connection ID. */
1955 OFP_ASSERT(sizeof(struct nx_controller_id) == 24);
1957 /* Action structure for NXAST_CONTROLLER.
1959 * This generalizes using OFPAT_OUTPUT to send a packet to OFPP_CONTROLLER. In
1960 * addition to the 'max_len' that OFPAT_OUTPUT supports, it also allows
1963 * - 'reason': The reason code to use in the ofp_packet_in or nx_packet_in.
1965 * - 'controller_id': The ID of the controller connection to which the
1966 * ofp_packet_in should be sent. The ofp_packet_in or nx_packet_in is
1967 * sent only to controllers that have the specified controller connection
1968 * ID. See "struct nx_controller_id" for more information. */
1969 struct nx_action_controller {
1970 ovs_be16 type; /* OFPAT_VENDOR. */
1971 ovs_be16 len; /* Length is 16. */
1972 ovs_be32 vendor; /* NX_VENDOR_ID. */
1973 ovs_be16 subtype; /* NXAST_CONTROLLER. */
1974 ovs_be16 max_len; /* Maximum length to send to controller. */
1975 ovs_be16 controller_id; /* Controller ID to send packet-in. */
1976 uint8_t reason; /* enum ofp_packet_in_reason (OFPR_*). */
1977 uint8_t zero; /* Must be zero. */
1979 OFP_ASSERT(sizeof(struct nx_action_controller) == 16);
1981 #endif /* openflow/nicira-ext.h */