2 * Copyright (c) 2008, 2009, 2010, 2011, 2012 Nicira Networks.
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.
18 #include "learning-switch.h"
22 #include <netinet/in.h>
26 #include "byte-order.h"
27 #include "classifier.h"
30 #include "mac-learning.h"
32 #include "ofp-errors.h"
33 #include "ofp-parse.h"
34 #include "ofp-print.h"
36 #include "openflow/openflow.h"
37 #include "poll-loop.h"
44 VLOG_DEFINE_THIS_MODULE(learning_switch
);
47 struct hmap_node hmap_node
; /* Hash node for port number. */
48 uint16_t port_no
; /* OpenFlow port number, in host byte order. */
49 uint32_t queue_id
; /* OpenFlow queue number. */
53 /* If nonnegative, the switch sets up flows that expire after the given
54 * number of seconds (or never expire, if the value is OFP_FLOW_PERMANENT).
55 * Otherwise, the switch processes every packet. */
58 unsigned long long int datapath_id
;
59 time_t last_features_request
;
60 struct mac_learning
*ml
; /* NULL to act as hub instead of switch. */
61 struct flow_wildcards wc
; /* Wildcards to apply to flows. */
62 bool action_normal
; /* Use OFPP_NORMAL? */
64 /* Queue distribution. */
65 uint32_t default_queue
; /* Default OpenFlow queue, or UINT32_MAX. */
66 struct hmap queue_numbers
; /* Map from port number to lswitch_port. */
67 struct shash queue_names
; /* Map from port name to lswitch_port. */
69 /* Number of outgoing queued packets on the rconn. */
70 struct rconn_packet_counter
*queued
;
73 /* The log messages here could actually be useful in debugging, so keep the
74 * rate limit relatively high. */
75 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(30, 300);
77 static void queue_tx(struct lswitch
*, struct rconn
*, struct ofpbuf
*);
78 static void send_features_request(struct lswitch
*, struct rconn
*);
80 static enum ofperr
process_switch_features(struct lswitch
*,
81 struct ofp_switch_features
*);
82 static void process_packet_in(struct lswitch
*, struct rconn
*,
83 const struct ofp_packet_in
*);
84 static void process_echo_request(struct lswitch
*, struct rconn
*,
85 const struct ofp_header
*);
87 /* Creates and returns a new learning switch whose configuration is given by
90 * 'rconn' is used to send out an OpenFlow features request. */
92 lswitch_create(struct rconn
*rconn
, const struct lswitch_config
*cfg
)
96 sw
= xzalloc(sizeof *sw
);
97 sw
->max_idle
= cfg
->max_idle
;
99 sw
->last_features_request
= time_now() - 1;
100 sw
->ml
= (cfg
->mode
== LSW_LEARN
101 ? mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME
)
103 sw
->action_normal
= cfg
->mode
== LSW_NORMAL
;
105 flow_wildcards_init_exact(&sw
->wc
);
106 if (cfg
->wildcards
) {
109 if (cfg
->wildcards
== UINT32_MAX
) {
110 /* Try to wildcard as many fields as possible, but we cannot
111 * wildcard all fields. We need in_port to detect moves. We need
112 * Ethernet source and dest and VLAN VID to do L2 learning. */
113 ofpfw
= (OFPFW_DL_TYPE
| OFPFW_DL_VLAN_PCP
114 | OFPFW_NW_SRC_ALL
| OFPFW_NW_DST_ALL
115 | OFPFW_NW_TOS
| OFPFW_NW_PROTO
116 | OFPFW_TP_SRC
| OFPFW_TP_DST
);
118 ofpfw
= cfg
->wildcards
;
121 ofputil_wildcard_from_openflow(ofpfw
, &sw
->wc
);
124 sw
->default_queue
= cfg
->default_queue
;
125 hmap_init(&sw
->queue_numbers
);
126 shash_init(&sw
->queue_names
);
127 if (cfg
->port_queues
) {
128 struct shash_node
*node
;
130 SHASH_FOR_EACH (node
, cfg
->port_queues
) {
131 struct lswitch_port
*port
= xmalloc(sizeof *port
);
132 hmap_node_nullify(&port
->hmap_node
);
133 port
->queue_id
= (uintptr_t) node
->data
;
134 shash_add(&sw
->queue_names
, node
->name
, port
);
138 sw
->queued
= rconn_packet_counter_create();
139 send_features_request(sw
, rconn
);
141 if (cfg
->default_flows
) {
142 enum ofputil_protocol usable_protocols
;
143 enum ofputil_protocol protocol
;
144 struct ofpbuf
*msg
= NULL
;
149 /* Figure out the initial protocol on the connection. */
150 ofp_version
= rconn_get_version(rconn
);
151 protocol
= ofputil_protocol_from_ofp_version(ofp_version
);
153 /* If the initial protocol isn't good enough for default_flows, then
154 * pick one that will work and encode messages to set up that
157 * This could be improved by actually negotiating a mutually acceptable
158 * flow format with the switch, but that would require an asynchronous
159 * state machine. This version ought to work fine in practice. */
160 usable_protocols
= ofputil_flow_mod_usable_protocols(
161 cfg
->default_flows
, cfg
->n_default_flows
);
162 if (!(protocol
& usable_protocols
)) {
163 enum ofputil_protocol want
= rightmost_1bit(usable_protocols
);
165 msg
= ofputil_encode_set_protocol(protocol
, want
, &protocol
);
169 error
= rconn_send(rconn
, msg
, NULL
);
173 for (i
= 0; !error
&& i
< cfg
->n_default_flows
; i
++) {
174 msg
= ofputil_encode_flow_mod(&cfg
->default_flows
[i
], protocol
);
175 error
= rconn_send(rconn
, msg
, NULL
);
179 VLOG_INFO_RL(&rl
, "%s: failed to queue default flows (%s)",
180 rconn_get_name(rconn
), strerror(error
));
189 lswitch_destroy(struct lswitch
*sw
)
192 struct lswitch_port
*node
, *next
;
194 HMAP_FOR_EACH_SAFE (node
, next
, hmap_node
, &sw
->queue_numbers
) {
195 hmap_remove(&sw
->queue_numbers
, &node
->hmap_node
);
198 shash_destroy(&sw
->queue_names
);
199 mac_learning_destroy(sw
->ml
);
200 rconn_packet_counter_destroy(sw
->queued
);
205 /* Takes care of necessary 'sw' activity, except for receiving packets (which
206 * the caller must do). */
208 lswitch_run(struct lswitch
*sw
)
211 mac_learning_run(sw
->ml
, NULL
);
216 lswitch_wait(struct lswitch
*sw
)
219 mac_learning_wait(sw
->ml
);
223 /* Processes 'msg', which should be an OpenFlow received on 'rconn', according
224 * to the learning switch state in 'sw'. The most likely result of processing
225 * is that flow-setup and packet-out OpenFlow messages will be sent out on
228 lswitch_process_packet(struct lswitch
*sw
, struct rconn
*rconn
,
229 const struct ofpbuf
*msg
)
231 const struct ofp_header
*oh
= msg
->data
;
232 const struct ofputil_msg_type
*type
;
234 if (sw
->datapath_id
== 0
235 && oh
->type
!= OFPT_ECHO_REQUEST
236 && oh
->type
!= OFPT_FEATURES_REPLY
) {
237 send_features_request(sw
, rconn
);
241 ofputil_decode_msg_type(oh
, &type
);
242 switch (ofputil_msg_type_code(type
)) {
243 case OFPUTIL_OFPT_ECHO_REQUEST
:
244 process_echo_request(sw
, rconn
, msg
->data
);
247 case OFPUTIL_OFPT_FEATURES_REPLY
:
248 process_switch_features(sw
, msg
->data
);
251 case OFPUTIL_OFPT_PACKET_IN
:
252 process_packet_in(sw
, rconn
, msg
->data
);
255 case OFPUTIL_OFPT_FLOW_REMOVED
:
259 case OFPUTIL_MSG_INVALID
:
260 case OFPUTIL_OFPT_HELLO
:
261 case OFPUTIL_OFPT_ERROR
:
262 case OFPUTIL_OFPT_ECHO_REPLY
:
263 case OFPUTIL_OFPT_FEATURES_REQUEST
:
264 case OFPUTIL_OFPT_GET_CONFIG_REQUEST
:
265 case OFPUTIL_OFPT_GET_CONFIG_REPLY
:
266 case OFPUTIL_OFPT_SET_CONFIG
:
267 case OFPUTIL_OFPT_PORT_STATUS
:
268 case OFPUTIL_OFPT_PACKET_OUT
:
269 case OFPUTIL_OFPT_FLOW_MOD
:
270 case OFPUTIL_OFPT_PORT_MOD
:
271 case OFPUTIL_OFPT_BARRIER_REQUEST
:
272 case OFPUTIL_OFPT_BARRIER_REPLY
:
273 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST
:
274 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY
:
275 case OFPUTIL_OFPST_DESC_REQUEST
:
276 case OFPUTIL_OFPST_FLOW_REQUEST
:
277 case OFPUTIL_OFPST_AGGREGATE_REQUEST
:
278 case OFPUTIL_OFPST_TABLE_REQUEST
:
279 case OFPUTIL_OFPST_PORT_REQUEST
:
280 case OFPUTIL_OFPST_QUEUE_REQUEST
:
281 case OFPUTIL_OFPST_DESC_REPLY
:
282 case OFPUTIL_OFPST_FLOW_REPLY
:
283 case OFPUTIL_OFPST_QUEUE_REPLY
:
284 case OFPUTIL_OFPST_PORT_REPLY
:
285 case OFPUTIL_OFPST_TABLE_REPLY
:
286 case OFPUTIL_OFPST_AGGREGATE_REPLY
:
287 case OFPUTIL_NXT_ROLE_REQUEST
:
288 case OFPUTIL_NXT_ROLE_REPLY
:
289 case OFPUTIL_NXT_FLOW_MOD_TABLE_ID
:
290 case OFPUTIL_NXT_SET_FLOW_FORMAT
:
291 case OFPUTIL_NXT_SET_PACKET_IN_FORMAT
:
292 case OFPUTIL_NXT_PACKET_IN
:
293 case OFPUTIL_NXT_FLOW_MOD
:
294 case OFPUTIL_NXT_FLOW_REMOVED
:
295 case OFPUTIL_NXT_FLOW_AGE
:
296 case OFPUTIL_NXT_SET_ASYNC_CONFIG
:
297 case OFPUTIL_NXT_SET_CONTROLLER_ID
:
298 case OFPUTIL_NXST_FLOW_REQUEST
:
299 case OFPUTIL_NXST_AGGREGATE_REQUEST
:
300 case OFPUTIL_NXST_FLOW_REPLY
:
301 case OFPUTIL_NXST_AGGREGATE_REPLY
:
303 if (VLOG_IS_DBG_ENABLED()) {
304 char *s
= ofp_to_string(msg
->data
, msg
->size
, 2);
305 VLOG_DBG_RL(&rl
, "%016llx: OpenFlow packet ignored: %s",
313 send_features_request(struct lswitch
*sw
, struct rconn
*rconn
)
315 time_t now
= time_now();
316 if (now
>= sw
->last_features_request
+ 1) {
318 struct ofp_switch_config
*osc
;
320 /* Send OFPT_FEATURES_REQUEST. */
321 make_openflow(sizeof(struct ofp_header
), OFPT_FEATURES_REQUEST
, &b
);
322 queue_tx(sw
, rconn
, b
);
324 /* Send OFPT_SET_CONFIG. */
325 osc
= make_openflow(sizeof *osc
, OFPT_SET_CONFIG
, &b
);
326 osc
->miss_send_len
= htons(OFP_DEFAULT_MISS_SEND_LEN
);
327 queue_tx(sw
, rconn
, b
);
329 sw
->last_features_request
= now
;
334 queue_tx(struct lswitch
*sw
, struct rconn
*rconn
, struct ofpbuf
*b
)
336 int retval
= rconn_send_with_limit(rconn
, b
, sw
->queued
, 10);
337 if (retval
&& retval
!= ENOTCONN
) {
338 if (retval
== EAGAIN
) {
339 VLOG_INFO_RL(&rl
, "%016llx: %s: tx queue overflow",
340 sw
->datapath_id
, rconn_get_name(rconn
));
342 VLOG_WARN_RL(&rl
, "%016llx: %s: send: %s",
343 sw
->datapath_id
, rconn_get_name(rconn
),
350 process_switch_features(struct lswitch
*sw
, struct ofp_switch_features
*osf
)
352 struct ofputil_switch_features features
;
353 struct ofputil_phy_port port
;
357 error
= ofputil_decode_switch_features(osf
, &features
, &b
);
359 VLOG_ERR("received invalid switch feature reply (%s)",
360 ofperr_to_string(error
));
364 sw
->datapath_id
= features
.datapath_id
;
366 while (!ofputil_pull_switch_features_port(&b
, &port
)) {
367 struct lswitch_port
*lp
= shash_find_data(&sw
->queue_names
, port
.name
);
368 if (lp
&& hmap_node_is_null(&lp
->hmap_node
)) {
369 lp
->port_no
= port
.port_no
;
370 hmap_insert(&sw
->queue_numbers
, &lp
->hmap_node
,
371 hash_int(lp
->port_no
, 0));
378 lswitch_choose_destination(struct lswitch
*sw
, const struct flow
*flow
)
382 /* Learn the source MAC. */
383 if (mac_learning_may_learn(sw
->ml
, flow
->dl_src
, 0)) {
384 struct mac_entry
*mac
= mac_learning_insert(sw
->ml
, flow
->dl_src
, 0);
385 if (mac_entry_is_new(mac
) || mac
->port
.i
!= flow
->in_port
) {
386 VLOG_DBG_RL(&rl
, "%016llx: learned that "ETH_ADDR_FMT
" is on "
387 "port %"PRIu16
, sw
->datapath_id
,
388 ETH_ADDR_ARGS(flow
->dl_src
), flow
->in_port
);
390 mac
->port
.i
= flow
->in_port
;
391 mac_learning_changed(sw
->ml
, mac
);
395 /* Drop frames for reserved multicast addresses. */
396 if (eth_addr_is_reserved(flow
->dl_dst
)) {
400 out_port
= OFPP_FLOOD
;
402 struct mac_entry
*mac
;
404 mac
= mac_learning_lookup(sw
->ml
, flow
->dl_dst
, 0, NULL
);
406 out_port
= mac
->port
.i
;
407 if (out_port
== flow
->in_port
) {
408 /* Don't send a packet back out its input port. */
414 /* Check if we need to use "NORMAL" action. */
415 if (sw
->action_normal
&& out_port
!= OFPP_FLOOD
) {
423 get_queue_id(const struct lswitch
*sw
, uint16_t in_port
)
425 const struct lswitch_port
*port
;
427 HMAP_FOR_EACH_WITH_HASH (port
, hmap_node
, hash_int(in_port
, 0),
428 &sw
->queue_numbers
) {
429 if (port
->port_no
== in_port
) {
430 return port
->queue_id
;
434 return sw
->default_queue
;
438 process_packet_in(struct lswitch
*sw
, struct rconn
*rconn
,
439 const struct ofp_packet_in
*opi
)
441 uint16_t in_port
= ntohs(opi
->in_port
);
445 struct ofp_action_header actions
[2];
448 struct ofputil_packet_out po
;
450 size_t pkt_ofs
, pkt_len
;
454 /* Ignore packets sent via output to OFPP_CONTROLLER. This library never
455 * uses such an action. You never know what experiments might be going on,
456 * though, and it seems best not to interfere with them. */
457 if (opi
->reason
!= OFPR_NO_MATCH
) {
461 /* Extract flow data from 'opi' into 'flow'. */
462 pkt_ofs
= offsetof(struct ofp_packet_in
, data
);
463 pkt_len
= ntohs(opi
->header
.length
) - pkt_ofs
;
464 ofpbuf_use_const(&pkt
, opi
->data
, pkt_len
);
465 flow_extract(&pkt
, 0, 0, in_port
, &flow
);
467 /* Choose output port. */
468 out_port
= lswitch_choose_destination(sw
, &flow
);
471 queue_id
= get_queue_id(sw
, in_port
);
472 if (out_port
== OFPP_NONE
) {
474 } else if (queue_id
== UINT32_MAX
|| out_port
>= OFPP_MAX
) {
475 struct ofp_action_output oao
;
477 memset(&oao
, 0, sizeof oao
);
478 oao
.type
= htons(OFPAT10_OUTPUT
);
479 oao
.len
= htons(sizeof oao
);
480 oao
.port
= htons(out_port
);
482 memcpy(actions
, &oao
, sizeof oao
);
483 actions_len
= sizeof oao
;
485 struct ofp_action_enqueue oae
;
487 memset(&oae
, 0, sizeof oae
);
488 oae
.type
= htons(OFPAT10_ENQUEUE
);
489 oae
.len
= htons(sizeof oae
);
490 oae
.port
= htons(out_port
);
491 oae
.queue_id
= htonl(queue_id
);
493 memcpy(actions
, &oae
, sizeof oae
);
494 actions_len
= sizeof oae
;
496 assert(actions_len
<= sizeof actions
);
498 /* Prepare packet_out in case we need one. */
499 po
.buffer_id
= ntohl(opi
->buffer_id
);
500 if (po
.buffer_id
== UINT32_MAX
) {
501 po
.packet
= pkt
.data
;
502 po
.packet_len
= pkt
.size
;
507 po
.in_port
= in_port
;
508 po
.actions
= (union ofp_action
*) actions
;
509 po
.n_actions
= actions_len
/ sizeof *actions
;
511 /* Send the packet, and possibly the whole flow, to the output port. */
512 if (sw
->max_idle
>= 0 && (!sw
->ml
|| out_port
!= OFPP_FLOOD
)) {
513 struct ofpbuf
*buffer
;
514 struct cls_rule rule
;
516 /* The output port is known, or we always flood everything, so add a
518 cls_rule_init(&flow
, &sw
->wc
, 0, &rule
);
519 buffer
= make_add_flow(&rule
, ntohl(opi
->buffer_id
),
520 sw
->max_idle
, actions_len
);
521 ofpbuf_put(buffer
, actions
, actions_len
);
522 queue_tx(sw
, rconn
, buffer
);
524 /* If the switch didn't buffer the packet, we need to send a copy. */
525 if (ntohl(opi
->buffer_id
) == UINT32_MAX
&& actions_len
> 0) {
526 queue_tx(sw
, rconn
, ofputil_encode_packet_out(&po
));
529 /* We don't know that MAC, or we don't set up flows. Send along the
530 * packet without setting up a flow. */
531 if (ntohl(opi
->buffer_id
) != UINT32_MAX
|| actions_len
> 0) {
532 queue_tx(sw
, rconn
, ofputil_encode_packet_out(&po
));
538 process_echo_request(struct lswitch
*sw
, struct rconn
*rconn
,
539 const struct ofp_header
*rq
)
541 queue_tx(sw
, rconn
, make_echo_reply(rq
));