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; /* See the NXT numbers in ofp-msgs.h. */
78 OFP_ASSERT(sizeof(struct nicira_header) == 16);
80 /* Header for Nicira vendor stats request and reply messages in OpenFlow
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. */
87 OFP_ASSERT(sizeof(struct nicira10_stats_msg) == 24);
89 /* Header for Nicira vendor stats request and reply messages in OpenFlow
91 struct nicira11_stats_msg {
92 struct ofp11_vendor_stats_msg vsm; /* Vendor NX_VENDOR_ID. */
93 ovs_be32 subtype; /* One of NXST_* below. */
95 OFP_ASSERT(sizeof(struct nicira11_stats_msg) == 24);
97 /* Fields to use when hashing flows. */
99 /* Ethernet source address (NXM_OF_ETH_SRC) only. */
100 NX_HASH_FIELDS_ETH_SRC,
102 /* L2 through L4, symmetric across src/dst. Specifically, each of the
103 * following fields, if present, is hashed (slashes separate symmetric
106 * - NXM_OF_ETH_DST / NXM_OF_ETH_SRC
108 * - The VID bits from NXM_OF_VLAN_TCI, ignoring PCP and CFI.
110 * - NXM_OF_IP_SRC / NXM_OF_IP_DST
111 * - NXM_OF_TCP_SRC / NXM_OF_TCP_DST
113 NX_HASH_FIELDS_SYMMETRIC_L4
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.
121 * When this feature is enabled, Open vSwitch treats struct ofp_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.
126 * The specific treatment of the table ID depends on the type of flow mod:
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
133 * OFPFMFC_ALL_TABLES_FULL error. (A controller can distinguish these
134 * cases by comparing the current and maximum number of entries reported
135 * in ofp_table_stats.)
137 * If the table ID is wildcarded, the switch picks an appropriate table
138 * itself. If an identical flow already exist in the selected flow table,
139 * then it is replaced. The choice of table might depend on the flows
140 * that are already in the switch; for example, if one table fills up then
141 * the switch might fall back to another one.
143 * - OFPFC_MODIFY, OFPFC_DELETE: Given a specific table ID, only flows
144 * within that table are matched and modified or deleted. If the table ID
145 * is wildcarded, flows within any table may be matched and modified or
148 * - OFPFC_MODIFY_STRICT, OFPFC_DELETE_STRICT: Given a specific table ID,
149 * only a flow within that table may be matched and modified or deleted.
150 * If the table ID is wildcarded and exactly one flow within any table
151 * matches, then it is modified or deleted; if flows in more than one
152 * table match, then none is modified or deleted.
154 struct nx_flow_mod_table_id {
155 uint8_t set; /* Nonzero to enable, zero to disable. */
158 OFP_ASSERT(sizeof(struct nx_flow_mod_table_id) == 8);
160 enum nx_packet_in_format {
161 NXPIF_OPENFLOW10 = 0, /* Standard OpenFlow 1.0 compatible. */
162 NXPIF_NXM = 1 /* Nicira Extended. */
165 /* NXT_SET_PACKET_IN_FORMAT request. */
166 struct nx_set_packet_in_format {
167 ovs_be32 format; /* One of NXPIF_*. */
169 OFP_ASSERT(sizeof(struct nx_set_packet_in_format) == 4);
171 /* NXT_PACKET_IN (analogous to OFPT_PACKET_IN).
173 * The NXT_PACKET_IN format is intended to model the OpenFlow-1.2 PACKET_IN
174 * with some minor tweaks. Most notably NXT_PACKET_IN includes the cookie of
175 * the rule which triggered the NXT_PACKET_IN message, and the match fields are
178 * The match fields in the NXT_PACKET_IN are intended to contain flow
179 * processing metadata collected at the time the NXT_PACKET_IN message was
180 * triggered. It is minimally required to contain the NXM_OF_IN_PORT of the
181 * packet, but may include other NXM headers such as flow registers. The match
182 * fields are allowed to contain non-metadata (e.g. NXM_OF_ETH_SRC etc).
183 * However, this information can typically be found in the packet directly, so
184 * it may be redundant.
186 * Whereas in most cases a controller can expect to only get back NXM fields
187 * that it set up itself (e.g. flow dumps will ordinarily report only NXM
188 * fields from flows that the controller added), NXT_PACKET_IN messages might
189 * contain fields that the controller does not understand, because the switch
190 * might support fields (new registers, new protocols, etc.) that the
191 * controller does not. The controller must prepared to tolerate these.
193 * The 'cookie' and 'table_id' fields have no meaning when 'reason' is
194 * OFPR_NO_MATCH. In this case they should be set to 0. */
195 struct nx_packet_in {
196 ovs_be32 buffer_id; /* ID assigned by datapath. */
197 ovs_be16 total_len; /* Full length of frame. */
198 uint8_t reason; /* Reason packet is sent (one of OFPR_*). */
199 uint8_t table_id; /* ID of the table that was looked up. */
200 ovs_be64 cookie; /* Cookie of the rule that was looked up. */
201 ovs_be16 match_len; /* Size of nx_match. */
202 uint8_t pad[6]; /* Align to 64-bits. */
204 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
205 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
206 * all-zero bytes, then
207 * - Exactly 2 all-zero padding bytes, then
208 * - An Ethernet frame whose length is inferred from nxh.header.length.
210 * The padding bytes preceding the Ethernet frame ensure that the IP
211 * header (if any) following the Ethernet header is 32-bit aligned. */
213 /* uint8_t nxm_fields[...]; */ /* Match. */
214 /* uint8_t pad[2]; */ /* Align to 64 bit + 16 bit. */
215 /* uint8_t data[0]; */ /* Ethernet frame. */
217 OFP_ASSERT(sizeof(struct nx_packet_in) == 24);
219 /* Configures the "role" of the sending controller. The default role is:
221 * - Other (NX_ROLE_OTHER), which allows the controller access to all
224 * The other possible roles are a related pair:
226 * - Master (NX_ROLE_MASTER) is equivalent to Other, except that there may
227 * be at most one Master controller at a time: when a controller
228 * configures itself as Master, any existing Master is demoted to the
231 * - Slave (NX_ROLE_SLAVE) allows the controller read-only access to
232 * OpenFlow features. In particular attempts to modify the flow table
233 * will be rejected with an OFPBRC_EPERM error.
235 * Slave controllers do not receive OFPT_PACKET_IN or OFPT_FLOW_REMOVED
236 * messages, but they do receive OFPT_PORT_STATUS messages.
238 struct nx_role_request {
239 ovs_be32 role; /* One of NX_ROLE_*. */
241 OFP_ASSERT(sizeof(struct nx_role_request) == 4);
244 NX_ROLE_OTHER, /* Default role, full access. */
245 NX_ROLE_MASTER, /* Full access, at most one. */
246 NX_ROLE_SLAVE /* Read-only access. */
249 /* NXT_SET_ASYNC_CONFIG.
251 * Sent by a controller, this message configures the asynchronous messages that
252 * the controller wants to receive. Element 0 in each array specifies messages
253 * of interest when the controller has an "other" or "master" role; element 1,
254 * when the controller has a "slave" role.
256 * Each array element is a bitmask in which a 0-bit disables receiving a
257 * particular message and a 1-bit enables receiving it. Each bit controls the
258 * message whose 'reason' corresponds to the bit index. For example, the bit
259 * with value 1<<2 == 4 in port_status_mask[1] determines whether the
260 * controller will receive OFPT_PORT_STATUS messages with reason OFPPR_MODIFY
261 * (value 2) when the controller has a "slave" role.
263 * As a side effect, for service controllers, this message changes the
264 * miss_send_len from default of zero to OFP_DEFAULT_MISS_SEND_LEN (128).
266 struct nx_async_config {
267 ovs_be32 packet_in_mask[2]; /* Bitmasks of OFPR_* values. */
268 ovs_be32 port_status_mask[2]; /* Bitmasks of OFPRR_* values. */
269 ovs_be32 flow_removed_mask[2]; /* Bitmasks of OFPPR_* values. */
271 OFP_ASSERT(sizeof(struct nx_async_config) == 24);
273 /* Nicira vendor flow actions. */
275 enum nx_action_subtype {
276 NXAST_SNAT__OBSOLETE, /* No longer used. */
277 NXAST_RESUBMIT, /* struct nx_action_resubmit */
278 NXAST_SET_TUNNEL, /* struct nx_action_set_tunnel */
279 NXAST_DROP_SPOOFED_ARP__OBSOLETE,
280 NXAST_SET_QUEUE, /* struct nx_action_set_queue */
281 NXAST_POP_QUEUE, /* struct nx_action_pop_queue */
282 NXAST_REG_MOVE, /* struct nx_action_reg_move */
283 NXAST_REG_LOAD, /* struct nx_action_reg_load */
284 NXAST_NOTE, /* struct nx_action_note */
285 NXAST_SET_TUNNEL64, /* struct nx_action_set_tunnel64 */
286 NXAST_MULTIPATH, /* struct nx_action_multipath */
287 NXAST_AUTOPATH, /* struct nx_action_autopath */
288 NXAST_BUNDLE, /* struct nx_action_bundle */
289 NXAST_BUNDLE_LOAD, /* struct nx_action_bundle */
290 NXAST_RESUBMIT_TABLE, /* struct nx_action_resubmit */
291 NXAST_OUTPUT_REG, /* struct nx_action_output_reg */
292 NXAST_LEARN, /* struct nx_action_learn */
293 NXAST_EXIT, /* struct nx_action_header */
294 NXAST_DEC_TTL, /* struct nx_action_header */
295 NXAST_FIN_TIMEOUT, /* struct nx_action_fin_timeout */
296 NXAST_CONTROLLER, /* struct nx_action_controller */
299 /* Header for Nicira-defined actions. */
300 struct nx_action_header {
301 ovs_be16 type; /* OFPAT_VENDOR. */
302 ovs_be16 len; /* Length is 16. */
303 ovs_be32 vendor; /* NX_VENDOR_ID. */
304 ovs_be16 subtype; /* NXAST_*. */
307 OFP_ASSERT(sizeof(struct nx_action_header) == 16);
309 /* Action structures for NXAST_RESUBMIT and NXAST_RESUBMIT_TABLE.
311 * These actions search one of the switch's flow tables:
313 * - For NXAST_RESUBMIT_TABLE only, if the 'table' member is not 255, then
314 * it specifies the table to search.
316 * - Otherwise (for NXAST_RESUBMIT_TABLE with a 'table' of 255, or for
317 * NXAST_RESUBMIT regardless of 'table'), it searches the current flow
318 * table, that is, the OpenFlow flow table that contains the flow from
319 * which this action was obtained. If this action did not come from a
320 * flow table (e.g. it came from an OFPT_PACKET_OUT message), then table 0
321 * is the current table.
323 * The flow table lookup uses a flow that may be slightly modified from the
326 * - For NXAST_RESUBMIT, the 'in_port' member of struct nx_action_resubmit
327 * is used as the flow's in_port.
329 * - For NXAST_RESUBMIT_TABLE, if the 'in_port' member is not OFPP_IN_PORT,
330 * then its value is used as the flow's in_port. Otherwise, the original
333 * - If actions that modify the flow (e.g. OFPAT_SET_VLAN_VID) precede the
334 * resubmit action, then the flow is updated with the new values.
336 * Following the lookup, the original in_port is restored.
338 * If the modified flow matched in the flow table, then the corresponding
339 * actions are executed. Afterward, actions following the resubmit in the
340 * original set of actions, if any, are executed; any changes made to the
341 * packet (e.g. changes to VLAN) by secondary actions persist when those
342 * actions are executed, although the original in_port is restored.
344 * Resubmit actions may be used any number of times within a set of actions.
346 * Resubmit actions may nest to an implementation-defined depth. Beyond this
347 * implementation-defined depth, further resubmit actions are simply ignored.
349 * NXAST_RESUBMIT ignores 'table' and 'pad'. NXAST_RESUBMIT_TABLE requires
350 * 'pad' to be all-bits-zero.
352 * Open vSwitch 1.0.1 and earlier did not support recursion. Open vSwitch
353 * before 1.2.90 did not support NXAST_RESUBMIT_TABLE.
355 struct nx_action_resubmit {
356 ovs_be16 type; /* OFPAT_VENDOR. */
357 ovs_be16 len; /* Length is 16. */
358 ovs_be32 vendor; /* NX_VENDOR_ID. */
359 ovs_be16 subtype; /* NXAST_RESUBMIT. */
360 ovs_be16 in_port; /* New in_port for checking flow table. */
361 uint8_t table; /* NXAST_RESUBMIT_TABLE: table to use. */
364 OFP_ASSERT(sizeof(struct nx_action_resubmit) == 16);
366 /* Action structure for NXAST_SET_TUNNEL.
368 * Sets the encapsulating tunnel ID to a 32-bit value. The most-significant 32
369 * bits of the tunnel ID are set to 0. */
370 struct nx_action_set_tunnel {
371 ovs_be16 type; /* OFPAT_VENDOR. */
372 ovs_be16 len; /* Length is 16. */
373 ovs_be32 vendor; /* NX_VENDOR_ID. */
374 ovs_be16 subtype; /* NXAST_SET_TUNNEL. */
376 ovs_be32 tun_id; /* Tunnel ID. */
378 OFP_ASSERT(sizeof(struct nx_action_set_tunnel) == 16);
380 /* Action structure for NXAST_SET_TUNNEL64.
382 * Sets the encapsulating tunnel ID to a 64-bit value. */
383 struct nx_action_set_tunnel64 {
384 ovs_be16 type; /* OFPAT_VENDOR. */
385 ovs_be16 len; /* Length is 24. */
386 ovs_be32 vendor; /* NX_VENDOR_ID. */
387 ovs_be16 subtype; /* NXAST_SET_TUNNEL64. */
389 ovs_be64 tun_id; /* Tunnel ID. */
391 OFP_ASSERT(sizeof(struct nx_action_set_tunnel64) == 24);
393 /* Action structure for NXAST_SET_QUEUE.
395 * Set the queue that should be used when packets are output. This is similar
396 * to the OpenFlow OFPAT_ENQUEUE action, but does not take the output port as
397 * an argument. This allows the queue to be defined before the port is
399 struct nx_action_set_queue {
400 ovs_be16 type; /* OFPAT_VENDOR. */
401 ovs_be16 len; /* Length is 16. */
402 ovs_be32 vendor; /* NX_VENDOR_ID. */
403 ovs_be16 subtype; /* NXAST_SET_QUEUE. */
405 ovs_be32 queue_id; /* Where to enqueue packets. */
407 OFP_ASSERT(sizeof(struct nx_action_set_queue) == 16);
409 /* Action structure for NXAST_POP_QUEUE.
411 * Restores the queue to the value it was before any NXAST_SET_QUEUE actions
412 * were used. Only the original queue can be restored this way; no stack is
414 struct nx_action_pop_queue {
415 ovs_be16 type; /* OFPAT_VENDOR. */
416 ovs_be16 len; /* Length is 16. */
417 ovs_be32 vendor; /* NX_VENDOR_ID. */
418 ovs_be16 subtype; /* NXAST_POP_QUEUE. */
421 OFP_ASSERT(sizeof(struct nx_action_pop_queue) == 16);
423 /* Action structure for NXAST_REG_MOVE.
425 * Copies src[src_ofs:src_ofs+n_bits] to dst[dst_ofs:dst_ofs+n_bits], where
426 * a[b:c] denotes the bits within 'a' numbered 'b' through 'c' (not including
427 * bit 'c'). Bit numbering starts at 0 for the least-significant bit, 1 for
428 * the next most significant bit, and so on.
430 * 'src' and 'dst' are nxm_header values with nxm_hasmask=0. (It doesn't make
431 * sense to use nxm_hasmask=1 because the action does not do any kind of
432 * matching; it uses the actual value of a field.)
434 * The following nxm_header values are potentially acceptable as 'src':
457 * - NXM_NX_ICMPV6_TYPE
458 * - NXM_NX_ICMPV6_CODE
461 * - NXM_NX_REG(idx) for idx in the switch's accepted range.
463 * The following nxm_header values are potentially acceptable as 'dst':
474 * Modifying any of the above fields changes the corresponding packet
477 * - NXM_NX_REG(idx) for idx in the switch's accepted range.
479 * - NXM_OF_VLAN_TCI. Modifying this field's value has side effects on the
480 * packet's 802.1Q header. Setting a value with CFI=0 removes the 802.1Q
481 * header (if any), ignoring the other bits. Setting a value with CFI=1
482 * adds or modifies the 802.1Q header appropriately, setting the TCI field
483 * to the field's new value (with the CFI bit masked out).
485 * - NXM_NX_TUN_ID. Modifying this value modifies the tunnel ID used for the
486 * packet's next tunnel encapsulation.
488 * A given nxm_header value may be used as 'src' or 'dst' only on a flow whose
489 * nx_match satisfies its prerequisites. For example, NXM_OF_IP_TOS may be
490 * used only if the flow's nx_match includes an nxm_entry that specifies
491 * nxm_type=NXM_OF_ETH_TYPE, nxm_hasmask=0, and nxm_value=0x0800.
493 * The switch will reject actions for which src_ofs+n_bits is greater than the
494 * width of 'src' or dst_ofs+n_bits is greater than the width of 'dst' with
495 * error type OFPET_BAD_ACTION, code OFPBAC_BAD_ARGUMENT.
497 struct nx_action_reg_move {
498 ovs_be16 type; /* OFPAT_VENDOR. */
499 ovs_be16 len; /* Length is 24. */
500 ovs_be32 vendor; /* NX_VENDOR_ID. */
501 ovs_be16 subtype; /* NXAST_REG_MOVE. */
502 ovs_be16 n_bits; /* Number of bits. */
503 ovs_be16 src_ofs; /* Starting bit offset in source. */
504 ovs_be16 dst_ofs; /* Starting bit offset in destination. */
505 ovs_be32 src; /* Source register. */
506 ovs_be32 dst; /* Destination register. */
508 OFP_ASSERT(sizeof(struct nx_action_reg_move) == 24);
510 /* Action structure for NXAST_REG_LOAD.
512 * Copies value[0:n_bits] to dst[ofs:ofs+n_bits], where a[b:c] denotes the bits
513 * within 'a' numbered 'b' through 'c' (not including bit 'c'). Bit numbering
514 * starts at 0 for the least-significant bit, 1 for the next most significant
517 * 'dst' is an nxm_header with nxm_hasmask=0. See the documentation for
518 * NXAST_REG_MOVE, above, for the permitted fields and for the side effects of
521 * The 'ofs' and 'n_bits' fields are combined into a single 'ofs_nbits' field
522 * to avoid enlarging the structure by another 8 bytes. To allow 'n_bits' to
523 * take a value between 1 and 64 (inclusive) while taking up only 6 bits, it is
524 * also stored as one less than its true value:
527 * +------------------------------+------------------+
528 * | ofs | n_bits - 1 |
529 * +------------------------------+------------------+
531 * The switch will reject actions for which ofs+n_bits is greater than the
532 * width of 'dst', or in which any bits in 'value' with value 2**n_bits or
533 * greater are set to 1, with error type OFPET_BAD_ACTION, code
534 * OFPBAC_BAD_ARGUMENT.
536 struct nx_action_reg_load {
537 ovs_be16 type; /* OFPAT_VENDOR. */
538 ovs_be16 len; /* Length is 24. */
539 ovs_be32 vendor; /* NX_VENDOR_ID. */
540 ovs_be16 subtype; /* NXAST_REG_LOAD. */
541 ovs_be16 ofs_nbits; /* (ofs << 6) | (n_bits - 1). */
542 ovs_be32 dst; /* Destination register. */
543 ovs_be64 value; /* Immediate value. */
545 OFP_ASSERT(sizeof(struct nx_action_reg_load) == 24);
547 /* Action structure for NXAST_NOTE.
549 * This action has no effect. It is variable length. The switch does not
550 * attempt to interpret the user-defined 'note' data in any way. A controller
551 * can use this action to attach arbitrary metadata to a flow.
553 * This action might go away in the future.
555 struct nx_action_note {
556 ovs_be16 type; /* OFPAT_VENDOR. */
557 ovs_be16 len; /* A multiple of 8, but at least 16. */
558 ovs_be32 vendor; /* NX_VENDOR_ID. */
559 ovs_be16 subtype; /* NXAST_NOTE. */
560 uint8_t note[6]; /* Start of user-defined data. */
561 /* Possibly followed by additional user-defined data. */
563 OFP_ASSERT(sizeof(struct nx_action_note) == 16);
565 /* Action structure for NXAST_MULTIPATH.
567 * This action performs the following steps in sequence:
569 * 1. Hashes the fields designated by 'fields', one of NX_HASH_FIELDS_*.
570 * Refer to the definition of "enum nx_mp_fields" for details.
572 * The 'basis' value is used as a universal hash parameter, that is,
573 * different values of 'basis' yield different hash functions. The
574 * particular universal hash function used is implementation-defined.
576 * The hashed fields' values are drawn from the current state of the
577 * flow, including all modifications that have been made by actions up to
580 * 2. Applies the multipath link choice algorithm specified by 'algorithm',
581 * one of NX_MP_ALG_*. Refer to the definition of "enum nx_mp_algorithm"
584 * The output of the algorithm is 'link', an unsigned integer less than
585 * or equal to 'max_link'.
587 * Some algorithms use 'arg' as an additional argument.
589 * 3. Stores 'link' in dst[ofs:ofs+n_bits]. The format and semantics of
590 * 'dst' and 'ofs_nbits' are similar to those for the NXAST_REG_LOAD
593 * The switch will reject actions that have an unknown 'fields', or an unknown
594 * 'algorithm', or in which ofs+n_bits is greater than the width of 'dst', or
595 * in which 'max_link' is greater than or equal to 2**n_bits, with error type
596 * OFPET_BAD_ACTION, code OFPBAC_BAD_ARGUMENT.
598 struct nx_action_multipath {
599 ovs_be16 type; /* OFPAT_VENDOR. */
600 ovs_be16 len; /* Length is 32. */
601 ovs_be32 vendor; /* NX_VENDOR_ID. */
602 ovs_be16 subtype; /* NXAST_MULTIPATH. */
604 /* What fields to hash and how. */
605 ovs_be16 fields; /* One of NX_HASH_FIELDS_*. */
606 ovs_be16 basis; /* Universal hash parameter. */
609 /* Multipath link choice algorithm to apply to hash value. */
610 ovs_be16 algorithm; /* One of NX_MP_ALG_*. */
611 ovs_be16 max_link; /* Number of output links, minus 1. */
612 ovs_be32 arg; /* Algorithm-specific argument. */
615 /* Where to store the result. */
616 ovs_be16 ofs_nbits; /* (ofs << 6) | (n_bits - 1). */
617 ovs_be32 dst; /* Destination. */
619 OFP_ASSERT(sizeof(struct nx_action_multipath) == 32);
621 /* NXAST_MULTIPATH: Multipath link choice algorithm to apply.
623 * In the descriptions below, 'n_links' is max_link + 1. */
624 enum nx_mp_algorithm {
625 /* link = hash(flow) % n_links.
627 * Redistributes all traffic when n_links changes. O(1) performance. See
630 * Use UINT16_MAX for max_link to get a raw hash value. */
633 /* link = hash(flow) / (MAX_HASH / n_links).
635 * Redistributes between one-quarter and one-half of traffic when n_links
636 * changes. O(1) performance. See RFC 2992.
638 NX_MP_ALG_HASH_THRESHOLD,
640 /* for i in [0,n_links):
641 * weights[i] = hash(flow, i)
642 * link = { i such that weights[i] >= weights[j] for all j != i }
644 * Redistributes 1/n_links of traffic when n_links changes. O(n_links)
645 * performance. If n_links is greater than a threshold (currently 64, but
646 * subject to change), Open vSwitch will substitute another algorithm
647 * automatically. See RFC 2992. */
648 NX_MP_ALG_HRW, /* Highest Random Weight. */
653 * link = hash(flow, i) % arg
654 * while link > max_link
656 * Redistributes 1/n_links of traffic when n_links changes. O(1)
657 * performance when arg/max_link is bounded by a constant.
659 * Redistributes all traffic when arg changes.
661 * arg must be greater than max_link and for best performance should be no
662 * more than approximately max_link * 2. If arg is outside the acceptable
663 * range, Open vSwitch will automatically substitute the least power of 2
664 * greater than max_link.
666 * This algorithm is specific to Open vSwitch.
668 NX_MP_ALG_ITER_HASH /* Iterative Hash. */
671 /* Action structure for NXAST_LEARN.
673 * This action adds or modifies a flow in an OpenFlow table, similar to
674 * OFPT_FLOW_MOD with OFPFC_MODIFY_STRICT as 'command'. The new flow has the
675 * specified idle timeout, hard timeout, priority, cookie, and flags. The new
676 * flow's match criteria and actions are built by applying each of the series
677 * of flow_mod_spec elements included as part of the action.
679 * A flow_mod_spec starts with a 16-bit header. A header that is all-bits-0 is
680 * a no-op used for padding the action as a whole to a multiple of 8 bytes in
681 * length. Otherwise, the flow_mod_spec can be thought of as copying 'n_bits'
682 * bits from a source to a destination. In this case, the header contains
685 * 15 14 13 12 11 10 0
686 * +------+---+------+---------------------------------+
687 * | 0 |src| dst | n_bits |
688 * +------+---+------+---------------------------------+
690 * The meaning and format of a flow_mod_spec depends on 'src' and 'dst'. The
691 * following table summarizes the meaning of each possible combination.
692 * Details follow the table:
695 * --- --- ----------------------------------------------------------
696 * 0 0 Add match criteria based on value in a field.
697 * 1 0 Add match criteria based on an immediate value.
698 * 0 1 Add NXAST_REG_LOAD action to copy field into a different field.
699 * 1 1 Add NXAST_REG_LOAD action to load immediate value into a field.
700 * 0 2 Add OFPAT_OUTPUT action to output to port from specified field.
701 * All other combinations are undefined and not allowed.
703 * The flow_mod_spec header is followed by a source specification and a
704 * destination specification. The format and meaning of the source
705 * specification depends on 'src':
707 * - If 'src' is 0, the source bits are taken from a field in the flow to
708 * which this action is attached. (This should be a wildcarded field. If
709 * its value is fully specified then the source bits being copied have
712 * The source specification is an ovs_be32 'field' and an ovs_be16 'ofs'.
713 * 'field' is an nxm_header with nxm_hasmask=0, and 'ofs' the starting bit
714 * offset within that field. The source bits are field[ofs:ofs+n_bits-1].
715 * 'field' and 'ofs' are subject to the same restrictions as the source
716 * field in NXAST_REG_MOVE.
718 * - If 'src' is 1, the source bits are a constant value. The source
719 * specification is (n_bits+15)/16*2 bytes long. Taking those bytes as a
720 * number in network order, the source bits are the 'n_bits'
721 * least-significant bits. The switch will report an error if other bits
722 * in the constant are nonzero.
724 * The flow_mod_spec destination specification, for 'dst' of 0 or 1, is an
725 * ovs_be32 'field' and an ovs_be16 'ofs'. 'field' is an nxm_header with
726 * nxm_hasmask=0 and 'ofs' is a starting bit offset within that field. The
727 * meaning of the flow_mod_spec depends on 'dst':
729 * - If 'dst' is 0, the flow_mod_spec specifies match criteria for the new
730 * flow. The new flow matches only if bits field[ofs:ofs+n_bits-1] in a
731 * packet equal the source bits. 'field' may be any nxm_header with
732 * nxm_hasmask=0 that is allowed in NXT_FLOW_MOD.
734 * Order is significant. Earlier flow_mod_specs must satisfy any
735 * prerequisites for matching fields specified later, by copying constant
736 * values into prerequisite fields.
738 * The switch will reject flow_mod_specs that do not satisfy NXM masking
741 * - If 'dst' is 1, the flow_mod_spec specifies an NXAST_REG_LOAD action for
742 * the new flow. The new flow copies the source bits into
743 * field[ofs:ofs+n_bits-1]. Actions are executed in the same order as the
746 * A single NXAST_REG_LOAD action writes no more than 64 bits, so n_bits
747 * greater than 64 yields multiple NXAST_REG_LOAD actions.
749 * The flow_mod_spec destination spec for 'dst' of 2 (when 'src' is 0) is
750 * empty. It has the following meaning:
752 * - The flow_mod_spec specifies an OFPAT_OUTPUT action for the new flow.
753 * The new flow outputs to the OpenFlow port specified by the source field.
754 * Of the special output ports with value OFPP_MAX or larger, OFPP_IN_PORT,
755 * OFPP_FLOOD, OFPP_LOCAL, and OFPP_ALL are supported. Other special ports
758 * Resource Management
759 * -------------------
761 * A switch has a finite amount of flow table space available for learning.
762 * When this space is exhausted, no new learning table entries will be learned
763 * until some existing flow table entries expire. The controller should be
764 * prepared to handle this by flooding (which can be implemented as a
765 * low-priority flow).
767 * If a learned flow matches a single TCP stream with a relatively long
768 * timeout, one may make the best of resource constraints by setting
769 * 'fin_idle_timeout' or 'fin_hard_timeout' (both measured in seconds), or
770 * both, to shorter timeouts. When either of these is specified as a nonzero
771 * value, OVS adds a NXAST_FIN_TIMEOUT action, with the specified timeouts, to
777 * The following examples give a prose description of the flow_mod_specs along
778 * with informal notation for how those would be represented and a hex dump of
779 * the bytes that would be required.
781 * These examples could work with various nx_action_learn parameters. Typical
782 * values would be idle_timeout=OFP_FLOW_PERMANENT, hard_timeout=60,
783 * priority=OFP_DEFAULT_PRIORITY, flags=0, table_id=10.
785 * 1. Learn input port based on the source MAC, with lookup into
786 * NXM_NX_REG1[16:31] by resubmit to in_port=99:
788 * Match on in_port=99:
789 * ovs_be16(src=1, dst=0, n_bits=16), 20 10
790 * ovs_be16(99), 00 63
791 * ovs_be32(NXM_OF_IN_PORT), ovs_be16(0) 00 00 00 02 00 00
793 * Match Ethernet destination on Ethernet source from packet:
794 * ovs_be16(src=0, dst=0, n_bits=48), 00 30
795 * ovs_be32(NXM_OF_ETH_SRC), ovs_be16(0) 00 00 04 06 00 00
796 * ovs_be32(NXM_OF_ETH_DST), ovs_be16(0) 00 00 02 06 00 00
798 * Set NXM_NX_REG1[16:31] to the packet's input port:
799 * ovs_be16(src=0, dst=1, n_bits=16), 08 10
800 * ovs_be32(NXM_OF_IN_PORT), ovs_be16(0) 00 00 00 02 00 00
801 * ovs_be32(NXM_NX_REG1), ovs_be16(16) 00 01 02 04 00 10
803 * Given a packet that arrived on port A with Ethernet source address B,
804 * this would set up the flow "in_port=99, dl_dst=B,
805 * actions=load:A->NXM_NX_REG1[16..31]".
807 * In syntax accepted by ovs-ofctl, this action is: learn(in_port=99,
808 * NXM_OF_ETH_DST[]=NXM_OF_ETH_SRC[],
809 * load:NXM_OF_IN_PORT[]->NXM_NX_REG1[16..31])
811 * 2. Output to input port based on the source MAC and VLAN VID, with lookup
812 * into NXM_NX_REG1[16:31]:
814 * Match on same VLAN ID as packet:
815 * ovs_be16(src=0, dst=0, n_bits=12), 00 0c
816 * ovs_be32(NXM_OF_VLAN_TCI), ovs_be16(0) 00 00 08 02 00 00
817 * ovs_be32(NXM_OF_VLAN_TCI), ovs_be16(0) 00 00 08 02 00 00
819 * Match Ethernet destination on Ethernet source from packet:
820 * ovs_be16(src=0, dst=0, n_bits=48), 00 30
821 * ovs_be32(NXM_OF_ETH_SRC), ovs_be16(0) 00 00 04 06 00 00
822 * ovs_be32(NXM_OF_ETH_DST), ovs_be16(0) 00 00 02 06 00 00
824 * Output to the packet's input port:
825 * ovs_be16(src=0, dst=2, n_bits=16), 10 10
826 * ovs_be32(NXM_OF_IN_PORT), ovs_be16(0) 00 00 00 02 00 00
828 * Given a packet that arrived on port A with Ethernet source address B in
829 * VLAN C, this would set up the flow "dl_dst=B, vlan_vid=C,
832 * In syntax accepted by ovs-ofctl, this action is:
833 * learn(NXM_OF_VLAN_TCI[0..11], NXM_OF_ETH_DST[]=NXM_OF_ETH_SRC[],
834 * output:NXM_OF_IN_PORT[])
836 * 3. Here's a recipe for a very simple-minded MAC learning switch. It uses a
837 * 10-second MAC expiration time to make it easier to see what's going on
839 * ovs-vsctl del-controller br0
840 * ovs-ofctl del-flows br0
841 * ovs-ofctl add-flow br0 "table=0 actions=learn(table=1, \
842 hard_timeout=10, NXM_OF_VLAN_TCI[0..11], \
843 NXM_OF_ETH_DST[]=NXM_OF_ETH_SRC[], \
844 output:NXM_OF_IN_PORT[]), resubmit(,1)"
845 * ovs-ofctl add-flow br0 "table=1 priority=0 actions=flood"
847 * You can then dump the MAC learning table with:
849 * ovs-ofctl dump-flows br0 table=1
854 * For best performance, segregate learned flows into a table that is not used
855 * for any other flows except possibly for a lowest-priority "catch-all" flow
856 * (a flow with no match criteria). If different learning actions specify
857 * different match criteria, use different tables for the learned flows.
859 * The meaning of 'hard_timeout' and 'idle_timeout' can be counterintuitive.
860 * These timeouts apply to the flow that is added, which means that a flow with
861 * an idle timeout will expire when no traffic has been sent *to* the learned
862 * address. This is not usually the intent in MAC learning; instead, we want
863 * the MAC learn entry to expire when no traffic has been sent *from* the
864 * learned address. Use a hard timeout for that.
866 struct nx_action_learn {
867 ovs_be16 type; /* OFPAT_VENDOR. */
868 ovs_be16 len; /* At least 24. */
869 ovs_be32 vendor; /* NX_VENDOR_ID. */
870 ovs_be16 subtype; /* NXAST_LEARN. */
871 ovs_be16 idle_timeout; /* Idle time before discarding (seconds). */
872 ovs_be16 hard_timeout; /* Max time before discarding (seconds). */
873 ovs_be16 priority; /* Priority level of flow entry. */
874 ovs_be64 cookie; /* Cookie for new flow. */
875 ovs_be16 flags; /* Either 0 or OFPFF_SEND_FLOW_REM. */
876 uint8_t table_id; /* Table to insert flow entry. */
877 uint8_t pad; /* Must be zero. */
878 ovs_be16 fin_idle_timeout; /* Idle timeout after FIN, if nonzero. */
879 ovs_be16 fin_hard_timeout; /* Hard timeout after FIN, if nonzero. */
880 /* Followed by a sequence of flow_mod_spec elements, as described above,
881 * until the end of the action is reached. */
883 OFP_ASSERT(sizeof(struct nx_action_learn) == 32);
885 #define NX_LEARN_N_BITS_MASK 0x3ff
887 #define NX_LEARN_SRC_FIELD (0 << 13) /* Copy from field. */
888 #define NX_LEARN_SRC_IMMEDIATE (1 << 13) /* Copy from immediate value. */
889 #define NX_LEARN_SRC_MASK (1 << 13)
891 #define NX_LEARN_DST_MATCH (0 << 11) /* Add match criterion. */
892 #define NX_LEARN_DST_LOAD (1 << 11) /* Add NXAST_REG_LOAD action. */
893 #define NX_LEARN_DST_OUTPUT (2 << 11) /* Add OFPAT_OUTPUT action. */
894 #define NX_LEARN_DST_RESERVED (3 << 11) /* Not yet defined. */
895 #define NX_LEARN_DST_MASK (3 << 11)
897 /* Action structure for NXAST_FIN_TIMEOUT.
899 * This action changes the idle timeout or hard timeout, or both, of this
900 * OpenFlow rule when the rule matches a TCP packet with the FIN or RST flag.
901 * When such a packet is observed, the action reduces the rule's idle timeout
902 * to 'fin_idle_timeout' and its hard timeout to 'fin_hard_timeout'. This
903 * action has no effect on an existing timeout that is already shorter than the
904 * one that the action specifies. A 'fin_idle_timeout' or 'fin_hard_timeout'
905 * of zero has no effect on the respective timeout.
907 * 'fin_idle_timeout' and 'fin_hard_timeout' are measured in seconds.
908 * 'fin_hard_timeout' specifies time since the flow's creation, not since the
909 * receipt of the FIN or RST.
911 * This is useful for quickly discarding learned TCP flows that otherwise will
912 * take a long time to expire.
914 * This action is intended for use with an OpenFlow rule that matches only a
915 * single TCP flow. If the rule matches multiple TCP flows (e.g. it wildcards
916 * all TCP traffic, or all TCP traffic to a particular port), then any FIN or
917 * RST in any of those flows will cause the entire OpenFlow rule to expire
918 * early, which is not normally desirable.
920 struct nx_action_fin_timeout {
921 ovs_be16 type; /* OFPAT_VENDOR. */
922 ovs_be16 len; /* 16. */
923 ovs_be32 vendor; /* NX_VENDOR_ID. */
924 ovs_be16 subtype; /* NXAST_FIN_TIMEOUT. */
925 ovs_be16 fin_idle_timeout; /* New idle timeout, if nonzero. */
926 ovs_be16 fin_hard_timeout; /* New hard timeout, if nonzero. */
927 ovs_be16 pad; /* Must be zero. */
929 OFP_ASSERT(sizeof(struct nx_action_fin_timeout) == 16);
931 /* Action structure for NXAST_AUTOPATH.
933 * This action performs the following steps in sequence:
935 * 1. Hashes the flow using an implementation-defined hash function.
937 * The hashed fields' values are drawn from the current state of the
938 * flow, including all modifications that have been made by actions up to
941 * 2. Selects an OpenFlow 'port'.
943 * 'port' is selected in an implementation-defined manner, taking into
944 * account 'id' and the hash value calculated in step 1.
946 * Generally a switch will have been configured with a set of ports that
947 * may be chosen given 'id'. The switch may take into account any number
948 * of factors when choosing 'port' from its configured set. Factors may
949 * include carrier, load, and the results of configuration protocols such
952 * 3. Stores 'port' in dst[ofs:ofs+n_bits].
954 * The format and semantics of 'dst' and 'ofs_nbits' are similar to those
955 * for the NXAST_REG_LOAD action.
957 * The switch will reject actions in which ofs+n_bits is greater than the width
958 * of 'dst', with error type OFPET_BAD_ACTION, code OFPBAC_BAD_ARGUMENT.
960 struct nx_action_autopath {
961 ovs_be16 type; /* OFPAT_VENDOR. */
962 ovs_be16 len; /* Length is 24. */
963 ovs_be32 vendor; /* NX_VENDOR_ID. */
964 ovs_be16 subtype; /* NXAST_AUTOPATH. */
966 /* Where to store the result. */
967 ovs_be16 ofs_nbits; /* (ofs << 6) | (n_bits - 1). */
968 ovs_be32 dst; /* Destination. */
970 ovs_be32 id; /* Autopath ID. */
973 OFP_ASSERT(sizeof(struct nx_action_autopath) == 24);
975 /* Action structure for NXAST_BUNDLE and NXAST_BUNDLE_LOAD.
977 * The bundle actions choose a slave from a supplied list of options.
978 * NXAST_BUNDLE outputs to its selection. NXAST_BUNDLE_LOAD writes its
979 * selection to a register.
981 * The list of possible slaves follows the nx_action_bundle structure. The size
982 * of each slave is governed by its type as indicated by the 'slave_type'
983 * parameter. The list of slaves should be padded at its end with zeros to make
984 * the total length of the action a multiple of 8.
986 * Switches infer from the 'slave_type' parameter the size of each slave. All
987 * implementations must support the NXM_OF_IN_PORT 'slave_type' which indicates
988 * that the slaves are OpenFlow port numbers with NXM_LENGTH(NXM_OF_IN_PORT) ==
989 * 2 byte width. Switches should reject actions which indicate unknown or
990 * unsupported slave types.
992 * Switches use a strategy dictated by the 'algorithm' parameter to choose a
993 * slave. If the switch does not support the specified 'algorithm' parameter,
994 * it should reject the action.
996 * Several algorithms take into account liveness when selecting slaves. The
997 * liveness of a slave is implementation defined (with one exception), but will
998 * generally take into account things like its carrier status and the results
999 * of any link monitoring protocols which happen to be running on it. In order
1000 * to give controllers a place-holder value, the OFPP_NONE port is always
1003 * Some slave selection strategies require the use of a hash function, in which
1004 * case the 'fields' and 'basis' parameters should be populated. The 'fields'
1005 * parameter (one of NX_HASH_FIELDS_*) designates which parts of the flow to
1006 * hash. Refer to the definition of "enum nx_hash_fields" for details. The
1007 * 'basis' parameter is used as a universal hash parameter. Different values
1008 * of 'basis' yield different hash results.
1010 * The 'zero' parameter at the end of the action structure is reserved for
1011 * future use. Switches are required to reject actions which have nonzero
1012 * bytes in the 'zero' field.
1014 * NXAST_BUNDLE actions should have 'ofs_nbits' and 'dst' zeroed. Switches
1015 * should reject actions which have nonzero bytes in either of these fields.
1017 * NXAST_BUNDLE_LOAD stores the OpenFlow port number of the selected slave in
1018 * dst[ofs:ofs+n_bits]. The format and semantics of 'dst' and 'ofs_nbits' are
1019 * similar to those for the NXAST_REG_LOAD action. */
1020 struct nx_action_bundle {
1021 ovs_be16 type; /* OFPAT_VENDOR. */
1022 ovs_be16 len; /* Length including slaves. */
1023 ovs_be32 vendor; /* NX_VENDOR_ID. */
1024 ovs_be16 subtype; /* NXAST_BUNDLE or NXAST_BUNDLE_LOAD. */
1026 /* Slave choice algorithm to apply to hash value. */
1027 ovs_be16 algorithm; /* One of NX_BD_ALG_*. */
1029 /* What fields to hash and how. */
1030 ovs_be16 fields; /* One of NX_HASH_FIELDS_*. */
1031 ovs_be16 basis; /* Universal hash parameter. */
1033 ovs_be32 slave_type; /* NXM_OF_IN_PORT. */
1034 ovs_be16 n_slaves; /* Number of slaves. */
1036 ovs_be16 ofs_nbits; /* (ofs << 6) | (n_bits - 1). */
1037 ovs_be32 dst; /* Destination. */
1039 uint8_t zero[4]; /* Reserved. Must be zero. */
1041 OFP_ASSERT(sizeof(struct nx_action_bundle) == 32);
1043 /* NXAST_BUNDLE: Bundle slave choice algorithm to apply.
1045 * In the descriptions below, 'slaves' is the list of possible slaves in the
1046 * order they appear in the OpenFlow action. */
1047 enum nx_bd_algorithm {
1048 /* Chooses the first live slave listed in the bundle.
1050 * O(n_slaves) performance. */
1051 NX_BD_ALG_ACTIVE_BACKUP,
1053 /* for i in [0,n_slaves):
1054 * weights[i] = hash(flow, i)
1055 * slave = { slaves[i] such that weights[i] >= weights[j] for all j != i }
1057 * Redistributes 1/n_slaves of traffic when a slave's liveness changes.
1058 * O(n_slaves) performance.
1060 * Uses the 'fields' and 'basis' parameters. */
1061 NX_BD_ALG_HRW /* Highest Random Weight. */
1064 /* Action structure for NXAST_OUTPUT_REG.
1066 * Outputs to the OpenFlow port number written to src[ofs:ofs+nbits].
1068 * The format and semantics of 'src' and 'ofs_nbits' are similar to those for
1069 * the NXAST_REG_LOAD action.
1071 * The acceptable nxm_header values for 'src' are the same as the acceptable
1072 * nxm_header values for the 'src' field of NXAST_REG_MOVE.
1074 * The 'max_len' field indicates the number of bytes to send when the chosen
1075 * port is OFPP_CONTROLLER. Its semantics are equivalent to the 'max_len'
1076 * field of OFPAT_OUTPUT.
1078 * The 'zero' field is required to be zeroed for forward compatibility. */
1079 struct nx_action_output_reg {
1080 ovs_be16 type; /* OFPAT_VENDOR. */
1081 ovs_be16 len; /* 24. */
1082 ovs_be32 vendor; /* NX_VENDOR_ID. */
1083 ovs_be16 subtype; /* NXAST_OUTPUT_REG. */
1085 ovs_be16 ofs_nbits; /* (ofs << 6) | (n_bits - 1). */
1086 ovs_be32 src; /* Source. */
1088 ovs_be16 max_len; /* Max length to send to controller. */
1090 uint8_t zero[6]; /* Reserved, must be zero. */
1092 OFP_ASSERT(sizeof(struct nx_action_output_reg) == 24);
1096 * Discontinues action processing.
1098 * The NXAST_EXIT action causes the switch to immediately halt processing
1099 * actions for the flow. Any actions which have already been processed are
1100 * executed by the switch. However, any further actions, including those which
1101 * may be in different tables, or different levels of the NXAST_RESUBMIT
1102 * hierarchy, will be ignored.
1104 * Uses the nx_action_header structure. */
1106 /* Flexible flow specifications (aka NXM = Nicira Extended Match).
1108 * OpenFlow 1.0 has "struct ofp10_match" for specifying flow matches. This
1109 * structure is fixed-length and hence difficult to extend. This section
1110 * describes a more flexible, variable-length flow match, called "nx_match" for
1111 * short, that is also supported by Open vSwitch. This section also defines a
1112 * replacement for each OpenFlow message that includes struct ofp10_match.
1118 * An nx_match is a sequence of zero or more "nxm_entry"s, which are
1119 * type-length-value (TLV) entries, each 5 to 259 (inclusive) bytes long.
1120 * "nxm_entry"s are not aligned on or padded to any multibyte boundary. The
1121 * first 4 bytes of an nxm_entry are its "header", followed by the entry's
1124 * An nxm_entry's header is interpreted as a 32-bit word in network byte order:
1126 * |<-------------------- nxm_type ------------------>|
1128 * |31 16 15 9| 8 7 0
1129 * +----------------------------------+---------------+--+------------------+
1130 * | nxm_vendor | nxm_field |hm| nxm_length |
1131 * +----------------------------------+---------------+--+------------------+
1133 * The most-significant 23 bits of the header are collectively "nxm_type".
1134 * Bits 16...31 are "nxm_vendor", one of the NXM_VENDOR_* values below. Bits
1135 * 9...15 are "nxm_field", which is a vendor-specific value. nxm_type normally
1136 * designates a protocol header, such as the Ethernet type, but it can also
1137 * refer to packet metadata, such as the switch port on which a packet arrived.
1139 * Bit 8 is "nxm_hasmask" (labeled "hm" above for space reasons). The meaning
1140 * of this bit is explained later.
1142 * The least-significant 8 bits are "nxm_length", a positive integer. The
1143 * length of the nxm_entry, including the header, is exactly 4 + nxm_length
1146 * For a given nxm_vendor, nxm_field, and nxm_hasmask value, nxm_length is a
1147 * constant. It is included only to allow software to minimally parse
1148 * "nxm_entry"s of unknown types. (Similarly, for a given nxm_vendor,
1149 * nxm_field, and nxm_length, nxm_hasmask is a constant.)
1155 * A zero-length nx_match (one with no "nxm_entry"s) matches every packet.
1157 * An nxm_entry places a constraint on the packets matched by the nx_match:
1159 * - If nxm_hasmask is 0, the nxm_entry's body contains a value for the
1160 * field, called "nxm_value". The nx_match matches only packets in which
1161 * the field equals nxm_value.
1163 * - If nxm_hasmask is 1, then the nxm_entry's body contains a value for the
1164 * field (nxm_value), followed by a bitmask of the same length as the
1165 * value, called "nxm_mask". For each 1-bit in position J in nxm_mask, the
1166 * nx_match matches only packets for which bit J in the given field's value
1167 * matches bit J in nxm_value. A 0-bit in nxm_mask causes the
1168 * corresponding bits in nxm_value and the field's value to be ignored.
1169 * (The sense of the nxm_mask bits is the opposite of that used by the
1170 * "wildcards" member of struct ofp10_match.)
1172 * When nxm_hasmask is 1, nxm_length is always even.
1174 * An all-zero-bits nxm_mask is equivalent to omitting the nxm_entry
1175 * entirely. An all-one-bits nxm_mask is equivalent to specifying 0 for
1178 * When there are multiple "nxm_entry"s, all of the constraints must be met.
1184 * Masks may be restricted:
1186 * - Some nxm_types may not support masked wildcards, that is, nxm_hasmask
1187 * must always be 0 when these fields are specified. For example, the
1188 * field that identifies the port on which a packet was received may not be
1191 * - Some nxm_types that do support masked wildcards may only support certain
1192 * nxm_mask patterns. For example, fields that have IPv4 address values
1193 * may be restricted to CIDR masks.
1195 * These restrictions should be noted in specifications for individual fields.
1196 * A switch may accept an nxm_hasmask or nxm_mask value that the specification
1197 * disallows, if the switch correctly implements support for that nxm_hasmask
1198 * or nxm_mask value. A switch must reject an attempt to set up a flow that
1199 * contains a nxm_hasmask or nxm_mask value that it does not support.
1202 * Prerequisite Restrictions
1203 * =========================
1205 * The presence of an nxm_entry with a given nxm_type may be restricted based
1206 * on the presence of or values of other "nxm_entry"s. For example:
1208 * - An nxm_entry for nxm_type=NXM_OF_IP_TOS is allowed only if it is
1209 * preceded by another entry with nxm_type=NXM_OF_ETH_TYPE, nxm_hasmask=0,
1210 * and nxm_value=0x0800. That is, matching on the IP source address is
1211 * allowed only if the Ethernet type is explicitly set to IP.
1213 * - An nxm_entry for nxm_type=NXM_OF_TCP_SRC is allowed only if it is
1214 * preceded by an entry with nxm_type=NXM_OF_ETH_TYPE, nxm_hasmask=0, and
1215 * nxm_value either 0x0800 or 0x86dd, and another with
1216 * nxm_type=NXM_OF_IP_PROTO, nxm_hasmask=0, nxm_value=6, in that order.
1217 * That is, matching on the TCP source port is allowed only if the Ethernet
1218 * type is IP or IPv6 and the IP protocol is TCP.
1220 * These restrictions should be noted in specifications for individual fields.
1221 * A switch may implement relaxed versions of these restrictions. A switch
1222 * must reject an attempt to set up a flow that violates its restrictions.
1225 * Ordering Restrictions
1226 * =====================
1228 * An nxm_entry that has prerequisite restrictions must appear after the
1229 * "nxm_entry"s for its prerequisites. Ordering of "nxm_entry"s within an
1230 * nx_match is not otherwise constrained.
1232 * Any given nxm_type may appear in an nx_match at most once.
1235 * nxm_entry Examples
1236 * ==================
1238 * These examples show the format of a single nxm_entry with particular
1239 * nxm_hasmask and nxm_length values. The diagrams are labeled with field
1240 * numbers and byte indexes.
1243 * 8-bit nxm_value, nxm_hasmask=1, nxm_length=2:
1246 * +------------+---+---+
1247 * | header | v | m |
1248 * +------------+---+---+
1251 * 16-bit nxm_value, nxm_hasmask=0, nxm_length=2:
1254 * +------------+------+
1256 * +------------+------+
1259 * 32-bit nxm_value, nxm_hasmask=0, nxm_length=4:
1262 * +------------+-------------+
1263 * | header | nxm_value |
1264 * +------------+-------------+
1267 * 48-bit nxm_value, nxm_hasmask=0, nxm_length=6:
1270 * +------------+------------------+
1271 * | header | nxm_value |
1272 * +------------+------------------+
1275 * 48-bit nxm_value, nxm_hasmask=1, nxm_length=12:
1278 * +------------+------------------+------------------+
1279 * | header | nxm_value | nxm_mask |
1280 * +------------+------------------+------------------+
1286 * A switch should report an error in an nx_match using error type
1287 * OFPET_BAD_REQUEST and one of the NXBRC_NXM_* codes. Ideally the switch
1288 * should report a specific error code, if one is assigned for the particular
1289 * problem, but NXBRC_NXM_INVALID is also available to report a generic
1293 #define NXM_HEADER__(VENDOR, FIELD, HASMASK, LENGTH) \
1294 (((VENDOR) << 16) | ((FIELD) << 9) | ((HASMASK) << 8) | (LENGTH))
1295 #define NXM_HEADER(VENDOR, FIELD, LENGTH) \
1296 NXM_HEADER__(VENDOR, FIELD, 0, LENGTH)
1297 #define NXM_HEADER_W(VENDOR, FIELD, LENGTH) \
1298 NXM_HEADER__(VENDOR, FIELD, 1, (LENGTH) * 2)
1299 #define NXM_VENDOR(HEADER) ((HEADER) >> 16)
1300 #define NXM_FIELD(HEADER) (((HEADER) >> 9) & 0x7f)
1301 #define NXM_TYPE(HEADER) (((HEADER) >> 9) & 0x7fffff)
1302 #define NXM_HASMASK(HEADER) (((HEADER) >> 8) & 1)
1303 #define NXM_LENGTH(HEADER) ((HEADER) & 0xff)
1305 #define NXM_MAKE_WILD_HEADER(HEADER) \
1306 NXM_HEADER_W(NXM_VENDOR(HEADER), NXM_FIELD(HEADER), NXM_LENGTH(HEADER))
1308 /* ## ------------------------------- ## */
1309 /* ## OpenFlow 1.0-compatible fields. ## */
1310 /* ## ------------------------------- ## */
1312 /* Physical or virtual port on which the packet was received.
1316 * Format: 16-bit integer in network byte order.
1318 * Masking: Not maskable. */
1319 #define NXM_OF_IN_PORT NXM_HEADER (0x0000, 0, 2)
1321 /* Source or destination address in Ethernet header.
1325 * Format: 48-bit Ethernet MAC address.
1327 * Masking: Fully maskable, in versions 1.8 and later. Earlier versions only
1328 * supported the following masks for NXM_OF_ETH_DST_W: 00:00:00:00:00:00,
1329 * fe:ff:ff:ff:ff:ff, 01:00:00:00:00:00, ff:ff:ff:ff:ff:ff. */
1330 #define NXM_OF_ETH_DST NXM_HEADER (0x0000, 1, 6)
1331 #define NXM_OF_ETH_DST_W NXM_HEADER_W(0x0000, 1, 6)
1332 #define NXM_OF_ETH_SRC NXM_HEADER (0x0000, 2, 6)
1333 #define NXM_OF_ETH_SRC_W NXM_HEADER_W(0x0000, 2, 6)
1335 /* Packet's Ethernet type.
1337 * For an Ethernet II packet this is taken from the Ethernet header. For an
1338 * 802.2 LLC+SNAP header with OUI 00-00-00 this is taken from the SNAP header.
1339 * A packet that has neither format has value 0x05ff
1340 * (OFP_DL_TYPE_NOT_ETH_TYPE).
1342 * For a packet with an 802.1Q header, this is the type of the encapsulated
1347 * Format: 16-bit integer in network byte order.
1349 * Masking: Not maskable. */
1350 #define NXM_OF_ETH_TYPE NXM_HEADER (0x0000, 3, 2)
1354 * For a packet with an 802.1Q header, this is the Tag Control Information
1355 * (TCI) field, with the CFI bit forced to 1. For a packet with no 802.1Q
1356 * header, this has value 0.
1360 * Format: 16-bit integer in network byte order.
1362 * Masking: Arbitrary masks.
1364 * This field can be used in various ways:
1366 * - If it is not constrained at all, the nx_match matches packets without
1367 * an 802.1Q header or with an 802.1Q header that has any TCI value.
1369 * - Testing for an exact match with 0 matches only packets without an
1372 * - Testing for an exact match with a TCI value with CFI=1 matches packets
1373 * that have an 802.1Q header with a specified VID and PCP.
1375 * - Testing for an exact match with a nonzero TCI value with CFI=0 does
1376 * not make sense. The switch may reject this combination.
1378 * - Testing with a specific VID and CFI=1, with nxm_mask=0x1fff, matches
1379 * packets that have an 802.1Q header with that VID (and any PCP).
1381 * - Testing with a specific PCP and CFI=1, with nxm_mask=0xf000, matches
1382 * packets that have an 802.1Q header with that PCP (and any VID).
1384 * - Testing with nxm_value=0, nxm_mask=0x0fff matches packets with no 802.1Q
1385 * header or with an 802.1Q header with a VID of 0.
1387 * - Testing with nxm_value=0, nxm_mask=0xe000 matches packets with no 802.1Q
1388 * header or with an 802.1Q header with a PCP of 0.
1390 * - Testing with nxm_value=0, nxm_mask=0xefff matches packets with no 802.1Q
1391 * header or with an 802.1Q header with both VID and PCP of 0.
1393 #define NXM_OF_VLAN_TCI NXM_HEADER (0x0000, 4, 2)
1394 #define NXM_OF_VLAN_TCI_W NXM_HEADER_W(0x0000, 4, 2)
1396 /* The "type of service" byte of the IP header, with the ECN bits forced to 0.
1398 * Prereqs: NXM_OF_ETH_TYPE must be either 0x0800 or 0x86dd.
1400 * Format: 8-bit integer with 2 least-significant bits forced to 0.
1402 * Masking: Not maskable. */
1403 #define NXM_OF_IP_TOS NXM_HEADER (0x0000, 5, 1)
1405 /* The "protocol" byte in the IP header.
1407 * Prereqs: NXM_OF_ETH_TYPE must be either 0x0800 or 0x86dd.
1409 * Format: 8-bit integer.
1411 * Masking: Not maskable. */
1412 #define NXM_OF_IP_PROTO NXM_HEADER (0x0000, 6, 1)
1414 /* The source or destination address in the IP header.
1416 * Prereqs: NXM_OF_ETH_TYPE must match 0x0800 exactly.
1418 * Format: 32-bit integer in network byte order.
1420 * Masking: Fully maskable, in Open vSwitch 1.8 and later. In earlier
1421 * versions, only CIDR masks are allowed, that is, masks that consist of N
1422 * high-order bits set to 1 and the other 32-N bits set to 0. */
1423 #define NXM_OF_IP_SRC NXM_HEADER (0x0000, 7, 4)
1424 #define NXM_OF_IP_SRC_W NXM_HEADER_W(0x0000, 7, 4)
1425 #define NXM_OF_IP_DST NXM_HEADER (0x0000, 8, 4)
1426 #define NXM_OF_IP_DST_W NXM_HEADER_W(0x0000, 8, 4)
1428 /* The source or destination port in the TCP header.
1431 * NXM_OF_ETH_TYPE must be either 0x0800 or 0x86dd.
1432 * NXM_OF_IP_PROTO must match 6 exactly.
1434 * Format: 16-bit integer in network byte order.
1436 * Masking: Fully maskable, in Open vSwitch 1.6 and later. Not maskable, in
1437 * earlier versions. */
1438 #define NXM_OF_TCP_SRC NXM_HEADER (0x0000, 9, 2)
1439 #define NXM_OF_TCP_SRC_W NXM_HEADER_W(0x0000, 9, 2)
1440 #define NXM_OF_TCP_DST NXM_HEADER (0x0000, 10, 2)
1441 #define NXM_OF_TCP_DST_W NXM_HEADER_W(0x0000, 10, 2)
1443 /* The source or destination port in the UDP header.
1446 * NXM_OF_ETH_TYPE must match either 0x0800 or 0x86dd.
1447 * NXM_OF_IP_PROTO must match 17 exactly.
1449 * Format: 16-bit integer in network byte order.
1451 * Masking: Fully maskable, in Open vSwitch 1.6 and later. Not maskable, in
1452 * earlier versions. */
1453 #define NXM_OF_UDP_SRC NXM_HEADER (0x0000, 11, 2)
1454 #define NXM_OF_UDP_SRC_W NXM_HEADER_W(0x0000, 11, 2)
1455 #define NXM_OF_UDP_DST NXM_HEADER (0x0000, 12, 2)
1456 #define NXM_OF_UDP_DST_W NXM_HEADER_W(0x0000, 12, 2)
1458 /* The type or code in the ICMP header.
1461 * NXM_OF_ETH_TYPE must match 0x0800 exactly.
1462 * NXM_OF_IP_PROTO must match 1 exactly.
1464 * Format: 8-bit integer.
1466 * Masking: Not maskable. */
1467 #define NXM_OF_ICMP_TYPE NXM_HEADER (0x0000, 13, 1)
1468 #define NXM_OF_ICMP_CODE NXM_HEADER (0x0000, 14, 1)
1472 * For an Ethernet+IP ARP packet, the opcode in the ARP header. Always 0
1473 * otherwise. Only ARP opcodes between 1 and 255 should be specified for
1476 * Prereqs: NXM_OF_ETH_TYPE must match 0x0806 exactly.
1478 * Format: 16-bit integer in network byte order.
1480 * Masking: Not maskable. */
1481 #define NXM_OF_ARP_OP NXM_HEADER (0x0000, 15, 2)
1483 /* For an Ethernet+IP ARP packet, the source or target protocol address
1484 * in the ARP header. Always 0 otherwise.
1486 * Prereqs: NXM_OF_ETH_TYPE must match 0x0806 exactly.
1488 * Format: 32-bit integer in network byte order.
1490 * Masking: Fully maskable, in Open vSwitch 1.8 and later. In earlier
1491 * versions, only CIDR masks are allowed, that is, masks that consist of N
1492 * high-order bits set to 1 and the other 32-N bits set to 0. */
1493 #define NXM_OF_ARP_SPA NXM_HEADER (0x0000, 16, 4)
1494 #define NXM_OF_ARP_SPA_W NXM_HEADER_W(0x0000, 16, 4)
1495 #define NXM_OF_ARP_TPA NXM_HEADER (0x0000, 17, 4)
1496 #define NXM_OF_ARP_TPA_W NXM_HEADER_W(0x0000, 17, 4)
1498 /* ## ------------------------ ## */
1499 /* ## Nicira match extensions. ## */
1500 /* ## ------------------------ ## */
1502 /* Metadata registers.
1504 * Registers initially have value 0. Actions allow register values to be
1509 * Format: Array of 32-bit integer registers. Space is reserved for up to
1510 * NXM_NX_MAX_REGS registers, but switches may implement fewer.
1512 * Masking: Arbitrary masks. */
1513 #define NXM_NX_MAX_REGS 16
1514 #define NXM_NX_REG(IDX) NXM_HEADER (0x0001, IDX, 4)
1515 #define NXM_NX_REG_W(IDX) NXM_HEADER_W(0x0001, IDX, 4)
1516 #define NXM_NX_REG_IDX(HEADER) NXM_FIELD(HEADER)
1517 #define NXM_IS_NX_REG(HEADER) (!((((HEADER) ^ NXM_NX_REG0)) & 0xffffe1ff))
1518 #define NXM_IS_NX_REG_W(HEADER) (!((((HEADER) ^ NXM_NX_REG0_W)) & 0xffffe1ff))
1519 #define NXM_NX_REG0 NXM_HEADER (0x0001, 0, 4)
1520 #define NXM_NX_REG0_W NXM_HEADER_W(0x0001, 0, 4)
1521 #define NXM_NX_REG1 NXM_HEADER (0x0001, 1, 4)
1522 #define NXM_NX_REG1_W NXM_HEADER_W(0x0001, 1, 4)
1523 #define NXM_NX_REG2 NXM_HEADER (0x0001, 2, 4)
1524 #define NXM_NX_REG2_W NXM_HEADER_W(0x0001, 2, 4)
1525 #define NXM_NX_REG3 NXM_HEADER (0x0001, 3, 4)
1526 #define NXM_NX_REG3_W NXM_HEADER_W(0x0001, 3, 4)
1527 #define NXM_NX_REG4 NXM_HEADER (0x0001, 4, 4)
1528 #define NXM_NX_REG4_W NXM_HEADER_W(0x0001, 4, 4)
1529 #define NXM_NX_REG5 NXM_HEADER (0x0001, 5, 4)
1530 #define NXM_NX_REG5_W NXM_HEADER_W(0x0001, 5, 4)
1531 #define NXM_NX_REG6 NXM_HEADER (0x0001, 6, 4)
1532 #define NXM_NX_REG6_W NXM_HEADER_W(0x0001, 6, 4)
1533 #define NXM_NX_REG7 NXM_HEADER (0x0001, 7, 4)
1534 #define NXM_NX_REG7_W NXM_HEADER_W(0x0001, 7, 4)
1538 * For a packet received via GRE tunnel including a (32-bit) key, the key is
1539 * stored in the low 32-bits and the high bits are zeroed. For other packets,
1544 * Format: 64-bit integer in network byte order.
1546 * Masking: Arbitrary masks. */
1547 #define NXM_NX_TUN_ID NXM_HEADER (0x0001, 16, 8)
1548 #define NXM_NX_TUN_ID_W NXM_HEADER_W(0x0001, 16, 8)
1550 /* For an Ethernet+IP ARP packet, the source or target hardware address
1551 * in the ARP header. Always 0 otherwise.
1553 * Prereqs: NXM_OF_ETH_TYPE must match 0x0806 exactly.
1555 * Format: 48-bit Ethernet MAC address.
1557 * Masking: Not maskable. */
1558 #define NXM_NX_ARP_SHA NXM_HEADER (0x0001, 17, 6)
1559 #define NXM_NX_ARP_THA NXM_HEADER (0x0001, 18, 6)
1561 /* The source or destination address in the IPv6 header.
1563 * Prereqs: NXM_OF_ETH_TYPE must match 0x86dd exactly.
1565 * Format: 128-bit IPv6 address.
1567 * Masking: Fully maskable, in Open vSwitch 1.8 and later. In previous
1568 * versions, only CIDR masks are allowed, that is, masks that consist of N
1569 * high-order bits set to 1 and the other 128-N bits set to 0. */
1570 #define NXM_NX_IPV6_SRC NXM_HEADER (0x0001, 19, 16)
1571 #define NXM_NX_IPV6_SRC_W NXM_HEADER_W(0x0001, 19, 16)
1572 #define NXM_NX_IPV6_DST NXM_HEADER (0x0001, 20, 16)
1573 #define NXM_NX_IPV6_DST_W NXM_HEADER_W(0x0001, 20, 16)
1575 /* The type or code in the ICMPv6 header.
1578 * NXM_OF_ETH_TYPE must match 0x86dd exactly.
1579 * NXM_OF_IP_PROTO must match 58 exactly.
1581 * Format: 8-bit integer.
1583 * Masking: Not maskable. */
1584 #define NXM_NX_ICMPV6_TYPE NXM_HEADER (0x0001, 21, 1)
1585 #define NXM_NX_ICMPV6_CODE NXM_HEADER (0x0001, 22, 1)
1587 /* The target address in an IPv6 Neighbor Discovery message.
1590 * NXM_OF_ETH_TYPE must match 0x86dd exactly.
1591 * NXM_OF_IP_PROTO must match 58 exactly.
1592 * NXM_OF_ICMPV6_TYPE must be either 135 or 136.
1594 * Format: 128-bit IPv6 address.
1596 * Masking: Fully maskable, in Open vSwitch 1.8 and later. In previous
1597 * versions, only CIDR masks are allowed, that is, masks that consist of N
1598 * high-order bits set to 1 and the other 128-N bits set to 0. */
1599 #define NXM_NX_ND_TARGET NXM_HEADER (0x0001, 23, 16)
1600 #define NXM_NX_ND_TARGET_W NXM_HEADER_W (0x0001, 23, 16)
1602 /* The source link-layer address option in an IPv6 Neighbor Discovery
1606 * NXM_OF_ETH_TYPE must match 0x86dd exactly.
1607 * NXM_OF_IP_PROTO must match 58 exactly.
1608 * NXM_OF_ICMPV6_TYPE must be exactly 135.
1610 * Format: 48-bit Ethernet MAC address.
1612 * Masking: Not maskable. */
1613 #define NXM_NX_ND_SLL NXM_HEADER (0x0001, 24, 6)
1615 /* The target link-layer address option in an IPv6 Neighbor Discovery
1619 * NXM_OF_ETH_TYPE must match 0x86dd exactly.
1620 * NXM_OF_IP_PROTO must match 58 exactly.
1621 * NXM_OF_ICMPV6_TYPE must be exactly 136.
1623 * Format: 48-bit Ethernet MAC address.
1625 * Masking: Not maskable. */
1626 #define NXM_NX_ND_TLL NXM_HEADER (0x0001, 25, 6)
1628 /* IP fragment information.
1631 * NXM_OF_ETH_TYPE must be either 0x0800 or 0x86dd.
1633 * Format: 8-bit value with one of the values 0, 1, or 3, as described below.
1635 * Masking: Fully maskable.
1637 * This field has three possible values:
1639 * - A packet that is not an IP fragment has value 0.
1641 * - A packet that is an IP fragment with offset 0 (the first fragment) has
1642 * bit 0 set and thus value 1.
1644 * - A packet that is an IP fragment with nonzero offset has bits 0 and 1 set
1647 * NX_IP_FRAG_ANY and NX_IP_FRAG_LATER are declared to symbolically represent
1648 * the meanings of bits 0 and 1.
1650 * The switch may reject matches against values that can never appear.
1652 * It is important to understand how this field interacts with the OpenFlow IP
1653 * fragment handling mode:
1655 * - In OFPC_FRAG_DROP mode, the OpenFlow switch drops all IP fragments
1656 * before they reach the flow table, so every packet that is available for
1657 * matching will have value 0 in this field.
1659 * - Open vSwitch does not implement OFPC_FRAG_REASM mode, but if it did then
1660 * IP fragments would be reassembled before they reached the flow table and
1661 * again every packet available for matching would always have value 0.
1663 * - In OFPC_FRAG_NORMAL mode, all three values are possible, but OpenFlow
1664 * 1.0 says that fragments' transport ports are always 0, even for the
1665 * first fragment, so this does not provide much extra information.
1667 * - In OFPC_FRAG_NX_MATCH mode, all three values are possible. For
1668 * fragments with offset 0, Open vSwitch makes L4 header information
1671 #define NXM_NX_IP_FRAG NXM_HEADER (0x0001, 26, 1)
1672 #define NXM_NX_IP_FRAG_W NXM_HEADER_W(0x0001, 26, 1)
1674 /* Bits in the value of NXM_NX_IP_FRAG. */
1675 #define NX_IP_FRAG_ANY (1 << 0) /* Is this a fragment? */
1676 #define NX_IP_FRAG_LATER (1 << 1) /* Is this a fragment with nonzero offset? */
1678 /* The flow label in the IPv6 header.
1680 * Prereqs: NXM_OF_ETH_TYPE must match 0x86dd exactly.
1682 * Format: 20-bit IPv6 flow label in least-significant bits.
1684 * Masking: Not maskable. */
1685 #define NXM_NX_IPV6_LABEL NXM_HEADER (0x0001, 27, 4)
1687 /* The ECN of the IP header.
1689 * Prereqs: NXM_OF_ETH_TYPE must be either 0x0800 or 0x86dd.
1691 * Format: ECN in the low-order 2 bits.
1693 * Masking: Not maskable. */
1694 #define NXM_NX_IP_ECN NXM_HEADER (0x0001, 28, 1)
1696 /* The time-to-live/hop limit of the IP header.
1698 * Prereqs: NXM_OF_ETH_TYPE must be either 0x0800 or 0x86dd.
1700 * Format: 8-bit integer.
1702 * Masking: Not maskable. */
1703 #define NXM_NX_IP_TTL NXM_HEADER (0x0001, 29, 1)
1707 * This may be used to gain the OpenFlow 1.1-like ability to restrict
1708 * certain NXM-based Flow Mod and Flow Stats Request messages to flows
1709 * with specific cookies. See the "nx_flow_mod" and "nx_flow_stats_request"
1710 * structure definitions for more details. This match is otherwise not
1715 * Format: 64-bit integer in network byte order.
1717 * Masking: Arbitrary masks. */
1718 #define NXM_NX_COOKIE NXM_HEADER (0x0001, 30, 8)
1719 #define NXM_NX_COOKIE_W NXM_HEADER_W(0x0001, 30, 8)
1721 /* ## --------------------- ## */
1722 /* ## Requests and replies. ## */
1723 /* ## --------------------- ## */
1725 enum nx_flow_format {
1726 NXFF_OPENFLOW10 = 0, /* Standard OpenFlow 1.0 compatible. */
1727 NXFF_NXM = 2, /* Nicira extended match. */
1728 NXFF_OPENFLOW12 = 3 /* OpenFlow 1.2 format. */
1731 /* NXT_SET_FLOW_FORMAT request. */
1732 struct nx_set_flow_format {
1733 ovs_be32 format; /* One of NXFF_*. */
1735 OFP_ASSERT(sizeof(struct nx_set_flow_format) == 4);
1737 /* NXT_FLOW_MOD (analogous to OFPT_FLOW_MOD).
1739 * It is possible to limit flow deletions and modifications to certain
1740 * cookies by using the NXM_NX_COOKIE(_W) matches. The "cookie" field
1741 * is used only to add or modify flow cookies.
1743 struct nx_flow_mod {
1744 ovs_be64 cookie; /* Opaque controller-issued identifier. */
1745 ovs_be16 command; /* One of OFPFC_*. */
1746 ovs_be16 idle_timeout; /* Idle time before discarding (seconds). */
1747 ovs_be16 hard_timeout; /* Max time before discarding (seconds). */
1748 ovs_be16 priority; /* Priority level of flow entry. */
1749 ovs_be32 buffer_id; /* Buffered packet to apply to (or -1).
1750 Not meaningful for OFPFC_DELETE*. */
1751 ovs_be16 out_port; /* For OFPFC_DELETE* commands, require
1752 matching entries to include this as an
1753 output port. A value of OFPP_NONE
1754 indicates no restriction. */
1755 ovs_be16 flags; /* One of OFPFF_*. */
1756 ovs_be16 match_len; /* Size of nx_match. */
1757 uint8_t pad[6]; /* Align to 64-bits. */
1759 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
1760 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
1761 * all-zero bytes, then
1762 * - Actions to fill out the remainder of the message length (always a
1766 OFP_ASSERT(sizeof(struct nx_flow_mod) == 32);
1768 /* NXT_FLOW_REMOVED (analogous to OFPT_FLOW_REMOVED). */
1769 struct nx_flow_removed {
1770 ovs_be64 cookie; /* Opaque controller-issued identifier. */
1771 ovs_be16 priority; /* Priority level of flow entry. */
1772 uint8_t reason; /* One of OFPRR_*. */
1773 uint8_t pad[1]; /* Align to 32-bits. */
1774 ovs_be32 duration_sec; /* Time flow was alive in seconds. */
1775 ovs_be32 duration_nsec; /* Time flow was alive in nanoseconds beyond
1777 ovs_be16 idle_timeout; /* Idle timeout from original flow mod. */
1778 ovs_be16 match_len; /* Size of nx_match. */
1779 ovs_be64 packet_count;
1780 ovs_be64 byte_count;
1782 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
1783 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
1784 * all-zero bytes. */
1786 OFP_ASSERT(sizeof(struct nx_flow_removed) == 40);
1788 /* Nicira vendor stats request of type NXST_FLOW (analogous to OFPST_FLOW
1791 * It is possible to limit matches to certain cookies by using the
1792 * NXM_NX_COOKIE and NXM_NX_COOKIE_W matches.
1794 struct nx_flow_stats_request {
1795 ovs_be16 out_port; /* Require matching entries to include this
1796 as an output port. A value of OFPP_NONE
1797 indicates no restriction. */
1798 ovs_be16 match_len; /* Length of nx_match. */
1799 uint8_t table_id; /* ID of table to read (from ofp_table_stats)
1800 or 0xff for all tables. */
1801 uint8_t pad[3]; /* Align to 64 bits. */
1803 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
1804 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
1805 * all-zero bytes, which must also exactly fill out the length of the
1809 OFP_ASSERT(sizeof(struct nx_flow_stats_request) == 8);
1811 /* Body for Nicira vendor stats reply of type NXST_FLOW (analogous to
1812 * OFPST_FLOW reply).
1814 * The values of 'idle_age' and 'hard_age' are only meaningful when talking to
1815 * a switch that implements the NXT_FLOW_AGE extension. Zero means that the
1816 * true value is unknown, perhaps because hardware does not track the value.
1817 * (Zero is also the value that one should ordinarily expect to see talking to
1818 * a switch that does not implement NXT_FLOW_AGE, since those switches zero the
1819 * padding bytes that these fields replaced.) A nonzero value X represents X-1
1820 * seconds. A value of 65535 represents 65534 or more seconds.
1822 * 'idle_age' is the number of seconds that the flow has been idle, that is,
1823 * the number of seconds since a packet passed through the flow. 'hard_age' is
1824 * the number of seconds since the flow was last modified (e.g. OFPFC_MODIFY or
1825 * OFPFC_MODIFY_STRICT). (The 'duration_*' fields are the elapsed time since
1826 * the flow was added, regardless of subsequent modifications.)
1828 * For a flow with an idle or hard timeout, 'idle_age' or 'hard_age',
1829 * respectively, will ordinarily be smaller than the timeout, but flow
1830 * expiration times are only approximate and so one must be prepared to
1831 * tolerate expirations that occur somewhat early or late.
1833 struct nx_flow_stats {
1834 ovs_be16 length; /* Length of this entry. */
1835 uint8_t table_id; /* ID of table flow came from. */
1837 ovs_be32 duration_sec; /* Time flow has been alive in seconds. */
1838 ovs_be32 duration_nsec; /* Time flow has been alive in nanoseconds
1839 beyond duration_sec. */
1840 ovs_be16 priority; /* Priority of the entry. */
1841 ovs_be16 idle_timeout; /* Number of seconds idle before expiration. */
1842 ovs_be16 hard_timeout; /* Number of seconds before expiration. */
1843 ovs_be16 match_len; /* Length of nx_match. */
1844 ovs_be16 idle_age; /* Seconds since last packet, plus one. */
1845 ovs_be16 hard_age; /* Seconds since last modification, plus one. */
1846 ovs_be64 cookie; /* Opaque controller-issued identifier. */
1847 ovs_be64 packet_count; /* Number of packets, UINT64_MAX if unknown. */
1848 ovs_be64 byte_count; /* Number of bytes, UINT64_MAX if unknown. */
1850 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
1851 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
1852 * all-zero bytes, then
1853 * - Actions to fill out the remainder 'length' bytes (always a multiple
1857 OFP_ASSERT(sizeof(struct nx_flow_stats) == 48);
1859 /* Nicira vendor stats request of type NXST_AGGREGATE (analogous to
1860 * OFPST_AGGREGATE request).
1862 * The reply format is identical to the reply format for OFPST_AGGREGATE,
1863 * except for the header. */
1864 struct nx_aggregate_stats_request {
1865 ovs_be16 out_port; /* Require matching entries to include this
1866 as an output port. A value of OFPP_NONE
1867 indicates no restriction. */
1868 ovs_be16 match_len; /* Length of nx_match. */
1869 uint8_t table_id; /* ID of table to read (from ofp_table_stats)
1870 or 0xff for all tables. */
1871 uint8_t pad[3]; /* Align to 64 bits. */
1873 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
1874 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
1875 * all-zero bytes, which must also exactly fill out the length of the
1879 OFP_ASSERT(sizeof(struct nx_aggregate_stats_request) == 8);
1881 /* NXT_SET_CONTROLLER_ID.
1883 * Each OpenFlow controller connection has a 16-bit identifier that is
1884 * initially 0. This message changes the connection's ID to 'id'.
1886 * Controller connection IDs need not be unique.
1888 * The NXAST_CONTROLLER action is the only current user of controller
1889 * connection IDs. */
1890 struct nx_controller_id {
1891 uint8_t zero[6]; /* Must be zero. */
1892 ovs_be16 controller_id; /* New controller connection ID. */
1894 OFP_ASSERT(sizeof(struct nx_controller_id) == 8);
1896 /* Action structure for NXAST_CONTROLLER.
1898 * This generalizes using OFPAT_OUTPUT to send a packet to OFPP_CONTROLLER. In
1899 * addition to the 'max_len' that OFPAT_OUTPUT supports, it also allows
1902 * - 'reason': The reason code to use in the ofp_packet_in or nx_packet_in.
1904 * - 'controller_id': The ID of the controller connection to which the
1905 * ofp_packet_in should be sent. The ofp_packet_in or nx_packet_in is
1906 * sent only to controllers that have the specified controller connection
1907 * ID. See "struct nx_controller_id" for more information. */
1908 struct nx_action_controller {
1909 ovs_be16 type; /* OFPAT_VENDOR. */
1910 ovs_be16 len; /* Length is 16. */
1911 ovs_be32 vendor; /* NX_VENDOR_ID. */
1912 ovs_be16 subtype; /* NXAST_CONTROLLER. */
1913 ovs_be16 max_len; /* Maximum length to send to controller. */
1914 ovs_be16 controller_id; /* Controller ID to send packet-in. */
1915 uint8_t reason; /* enum ofp_packet_in_reason (OFPR_*). */
1916 uint8_t zero; /* Must be zero. */
1918 OFP_ASSERT(sizeof(struct nx_action_controller) == 16);
1920 /* Flow Table Monitoring
1921 * =====================
1923 * NXST_FLOW_MONITOR allows a controller to keep track of changes to OpenFlow
1924 * flow table(s) or subsets of them, with the following workflow:
1926 * 1. The controller sends an NXST_FLOW_MONITOR request to begin monitoring
1927 * flows. The 'id' in the request must be unique among all monitors that
1928 * the controller has started and not yet canceled on this OpenFlow
1931 * 2. The switch responds with an NXST_FLOW_MONITOR reply. If the request's
1932 * 'flags' included NXFMF_INITIAL, the reply includes all the flows that
1933 * matched the request at the time of the request (with event NXFME_ADDED).
1934 * If 'flags' did not include NXFMF_INITIAL, the reply is empty.
1936 * The reply uses the xid of the request (as do all replies to OpenFlow
1939 * 3. Whenever a change to a flow table entry matches some outstanding monitor
1940 * request's criteria and flags, the switch sends a notification to the
1941 * controller as an additional NXST_FLOW_MONITOR reply with xid 0.
1943 * When multiple outstanding monitors match a single change, only a single
1944 * notification is sent. This merged notification includes the information
1945 * requested in any of the individual monitors. That is, if any of the
1946 * matching monitors requests actions (NXFMF_ACTIONS), the notification
1947 * includes actions, and if any of the monitors request full changes for the
1948 * controller's own changes (NXFMF_OWN), the controller's own changes will
1949 * be included in full.
1951 * 4. The controller may cancel a monitor with NXT_FLOW_MONITOR_CANCEL. No
1952 * further notifications will be sent on the basis of the canceled monitor
1959 * OpenFlow messages for flow monitor notifications can overflow the buffer
1960 * space available to the switch, either temporarily (e.g. due to network
1961 * conditions slowing OpenFlow traffic) or more permanently (e.g. the sustained
1962 * rate of flow table change exceeds the network bandwidth between switch and
1965 * When Open vSwitch's notification buffer space reaches a limiting threshold,
1966 * OVS reacts as follows:
1968 * 1. OVS sends an NXT_FLOW_MONITOR_PAUSED message to the controller, following
1969 * all the already queued notifications. After it receives this message,
1970 * the controller knows that its view of the flow table, as represented by
1971 * flow monitor notifications, is incomplete.
1973 * 2. As long as the notification buffer is not empty:
1975 * - NXMFE_ADD and NXFME_MODIFIED notifications will not be sent.
1977 * - NXFME_DELETED notifications will still be sent, but only for flows
1978 * that existed before OVS sent NXT_FLOW_MONITOR_PAUSED.
1980 * - NXFME_ABBREV notifications will not be sent. They are treated as
1981 * the expanded version (and therefore only the NXFME_DELETED
1982 * components, if any, are sent).
1984 * 3. When the notification buffer empties, OVS sends NXFME_ADD notifications
1985 * for flows added since the buffer reached its limit and NXFME_MODIFIED
1986 * notifications for flows that existed before the limit was reached and
1987 * changed after the limit was reached.
1989 * 4. OVS sends an NXT_FLOW_MONITOR_RESUMED message to the controller. After
1990 * it receives this message, the controller knows that its view of the flow
1991 * table, as represented by flow monitor notifications, is again complete.
1993 * This allows the maximum buffer space requirement for notifications to be
1994 * bounded by the limit plus the maximum number of supported flows.
1997 * "Flow Removed" messages
1998 * =======================
2000 * The flow monitor mechanism is independent of OFPT_FLOW_REMOVED and
2001 * NXT_FLOW_REMOVED. Flow monitor updates for deletion are sent if
2002 * NXFMF_DELETE is set on a monitor, regardless of whether the
2003 * OFPFF_SEND_FLOW_REM flag was set when the flow was added. */
2005 /* NXST_FLOW_MONITOR request.
2007 * The NXST_FLOW_MONITOR request's body consists of an array of zero or more
2008 * instances of this structure. The request arranges to monitor the flows
2009 * that match the specified criteria, which are interpreted in the same way as
2012 * 'id' identifies a particular monitor for the purpose of allowing it to be
2013 * canceled later with NXT_FLOW_MONITOR_CANCEL. 'id' must be unique among
2014 * existing monitors that have not already been canceled.
2016 * The reply includes the initial flow matches for monitors that have the
2017 * NXFMF_INITIAL flag set. No single flow will be included in the reply more
2018 * than once, even if more than one requested monitor matches that flow. The
2019 * reply will be empty if none of the monitors has NXFMF_INITIAL set or if none
2020 * of the monitors initially matches any flows.
2022 * For NXFMF_ADD, an event will be reported if 'out_port' matches against the
2023 * actions of the flow being added or, for a flow that is replacing an existing
2024 * flow, if 'out_port' matches against the actions of the flow being replaced.
2025 * For NXFMF_DELETE, 'out_port' matches against the actions of a flow being
2026 * deleted. For NXFMF_MODIFY, an event will be reported if 'out_port' matches
2027 * either the old or the new actions. */
2028 struct nx_flow_monitor_request {
2029 ovs_be32 id; /* Controller-assigned ID for this monitor. */
2030 ovs_be16 flags; /* NXFMF_*. */
2031 ovs_be16 out_port; /* Required output port, if not OFPP_NONE. */
2032 ovs_be16 match_len; /* Length of nx_match. */
2033 uint8_t table_id; /* One table's ID or 0xff for all tables. */
2034 uint8_t zeros[5]; /* Align to 64 bits (must be zero). */
2036 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
2037 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
2038 * all-zero bytes. */
2040 OFP_ASSERT(sizeof(struct nx_flow_monitor_request) == 16);
2042 /* 'flags' bits in struct nx_flow_monitor_request. */
2043 enum nx_flow_monitor_flags {
2044 /* When to send updates. */
2045 NXFMF_INITIAL = 1 << 0, /* Initially matching flows. */
2046 NXFMF_ADD = 1 << 1, /* New matching flows as they are added. */
2047 NXFMF_DELETE = 1 << 2, /* Old matching flows as they are removed. */
2048 NXFMF_MODIFY = 1 << 3, /* Matching flows as they are changed. */
2050 /* What to include in updates. */
2051 NXFMF_ACTIONS = 1 << 4, /* If set, actions are included. */
2052 NXFMF_OWN = 1 << 5, /* If set, include own changes in full. */
2055 /* NXST_FLOW_MONITOR reply header.
2057 * The body of an NXST_FLOW_MONITOR reply is an array of variable-length
2058 * structures, each of which begins with this header. The 'length' member may
2059 * be used to traverse the array, and the 'event' member may be used to
2060 * determine the particular structure.
2062 * Every instance is a multiple of 8 bytes long. */
2063 struct nx_flow_update_header {
2064 ovs_be16 length; /* Length of this entry. */
2065 ovs_be16 event; /* One of NXFME_*. */
2066 /* ...other data depending on 'event'... */
2068 OFP_ASSERT(sizeof(struct nx_flow_update_header) == 4);
2070 /* 'event' values in struct nx_flow_update_header. */
2071 enum nx_flow_update_event {
2072 /* struct nx_flow_update_full. */
2073 NXFME_ADDED = 0, /* Flow was added. */
2074 NXFME_DELETED = 1, /* Flow was deleted. */
2075 NXFME_MODIFIED = 2, /* Flow (generally its actions) was changed. */
2077 /* struct nx_flow_update_abbrev. */
2078 NXFME_ABBREV = 3, /* Abbreviated reply. */
2081 /* NXST_FLOW_MONITOR reply for NXFME_ADDED, NXFME_DELETED, and
2082 * NXFME_MODIFIED. */
2083 struct nx_flow_update_full {
2084 ovs_be16 length; /* Length is 24. */
2085 ovs_be16 event; /* One of NXFME_*. */
2086 ovs_be16 reason; /* OFPRR_* for NXFME_DELETED, else zero. */
2087 ovs_be16 priority; /* Priority of the entry. */
2088 ovs_be16 idle_timeout; /* Number of seconds idle before expiration. */
2089 ovs_be16 hard_timeout; /* Number of seconds before expiration. */
2090 ovs_be16 match_len; /* Length of nx_match. */
2091 uint8_t table_id; /* ID of flow's table. */
2092 uint8_t pad; /* Reserved, currently zeroed. */
2093 ovs_be64 cookie; /* Opaque controller-issued identifier. */
2095 * - Exactly match_len (possibly 0) bytes containing the nx_match, then
2096 * - Exactly (match_len + 7)/8*8 - match_len (between 0 and 7) bytes of
2097 * all-zero bytes, then
2098 * - Actions to fill out the remainder 'length' bytes (always a multiple
2099 * of 8). If NXFMF_ACTIONS was not specified, or 'event' is
2100 * NXFME_DELETED, no actions are included.
2103 OFP_ASSERT(sizeof(struct nx_flow_update_full) == 24);
2105 /* NXST_FLOW_MONITOR reply for NXFME_ABBREV.
2107 * When the controller does not specify NXFMF_OWN in a monitor request, any
2108 * flow tables changes due to the controller's own requests (on the same
2109 * OpenFlow channel) will be abbreviated, when possible, to this form, which
2110 * simply specifies the 'xid' of the OpenFlow request (e.g. an OFPT_FLOW_MOD or
2111 * NXT_FLOW_MOD) that caused the change.
2113 * Some changes cannot be abbreviated and will be sent in full:
2115 * - Changes that only partially succeed. This can happen if, for example,
2116 * a flow_mod with type OFPFC_MODIFY affects multiple flows, but only some
2117 * of those modifications succeed (e.g. due to hardware limitations).
2119 * This cannot occur with the current implementation of the Open vSwitch
2120 * software datapath. It could happen with other datapath implementations.
2122 * - Changes that race with conflicting changes made by other controllers or
2123 * other flow_mods (not separated by barriers) by the same controller.
2125 * This cannot occur with the current Open vSwitch implementation
2126 * (regardless of datapath) because Open vSwitch internally serializes
2127 * potentially conflicting changes.
2129 * A flow_mod that does not change the flow table will not trigger any
2130 * notification, even an abbreviated one. For example, a "modify" or "delete"
2131 * flow_mod that does not match any flows will not trigger a notification.
2132 * Whether an "add" or "modify" that specifies all the same parameters that a
2133 * flow already has triggers a notification is unspecified and subject to
2134 * change in future versions of Open vSwitch.
2136 * OVS will always send the notifications for a given flow table change before
2137 * the reply to a OFPT_BARRIER_REQUEST request that precedes the flow table
2138 * change. Thus, if the controller does not receive an abbreviated
2139 * notification for a flow_mod before the next OFPT_BARRIER_REPLY, it will
2140 * never receive one. */
2141 struct nx_flow_update_abbrev {
2142 ovs_be16 length; /* Length is 8. */
2143 ovs_be16 event; /* NXFME_ABBREV. */
2144 ovs_be32 xid; /* Controller-specified xid from flow_mod. */
2146 OFP_ASSERT(sizeof(struct nx_flow_update_abbrev) == 8);
2148 /* NXT_FLOW_MONITOR_CANCEL.
2150 * Used by a controller to cancel an outstanding monitor. */
2151 struct nx_flow_monitor_cancel {
2152 ovs_be32 id; /* 'id' from nx_flow_monitor_request. */
2154 OFP_ASSERT(sizeof(struct nx_flow_monitor_cancel) == 4);
2156 #endif /* openflow/nicira-ext.h */