2 * Copyright (c) 2008, 2009, 2010, 2011 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-parse.h"
33 #include "ofp-print.h"
35 #include "openflow/openflow.h"
36 #include "poll-loop.h"
43 VLOG_DEFINE_THIS_MODULE(learning_switch
);
46 struct hmap_node hmap_node
; /* Hash node for port number. */
47 uint16_t port_no
; /* OpenFlow port number, in host byte order. */
48 uint32_t queue_id
; /* OpenFlow queue number. */
52 /* If nonnegative, the switch sets up flows that expire after the given
53 * number of seconds (or never expire, if the value is OFP_FLOW_PERMANENT).
54 * Otherwise, the switch processes every packet. */
57 unsigned long long int datapath_id
;
58 time_t last_features_request
;
59 struct mac_learning
*ml
; /* NULL to act as hub instead of switch. */
60 struct flow_wildcards wc
; /* Wildcards to apply to flows. */
61 bool action_normal
; /* Use OFPP_NORMAL? */
63 /* Queue distribution. */
64 uint32_t default_queue
; /* Default OpenFlow queue, or UINT32_MAX. */
65 struct hmap queue_numbers
; /* Map from port number to lswitch_port. */
66 struct shash queue_names
; /* Map from port name to lswitch_port. */
68 /* Number of outgoing queued packets on the rconn. */
69 struct rconn_packet_counter
*queued
;
72 /* The log messages here could actually be useful in debugging, so keep the
73 * rate limit relatively high. */
74 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(30, 300);
76 static void queue_tx(struct lswitch
*, struct rconn
*, struct ofpbuf
*);
77 static void send_features_request(struct lswitch
*, struct rconn
*);
79 static void process_switch_features(struct lswitch
*,
80 struct ofp_switch_features
*);
81 static void process_packet_in(struct lswitch
*, struct rconn
*,
82 const struct ofp_packet_in
*);
83 static void process_echo_request(struct lswitch
*, struct rconn
*,
84 const struct ofp_header
*);
86 /* Creates and returns a new learning switch whose configuration is given by
89 * 'rconn' is used to send out an OpenFlow features request. */
91 lswitch_create(struct rconn
*rconn
, const struct lswitch_config
*cfg
)
95 sw
= xzalloc(sizeof *sw
);
96 sw
->max_idle
= cfg
->max_idle
;
98 sw
->last_features_request
= time_now() - 1;
99 sw
->ml
= cfg
->mode
== LSW_LEARN
? mac_learning_create() : NULL
;
100 sw
->action_normal
= cfg
->mode
== LSW_NORMAL
;
102 flow_wildcards_init_exact(&sw
->wc
);
103 if (cfg
->wildcards
) {
106 if (cfg
->wildcards
== UINT32_MAX
) {
107 /* Try to wildcard as many fields as possible, but we cannot
108 * wildcard all fields. We need in_port to detect moves. We need
109 * Ethernet source and dest and VLAN VID to do L2 learning. */
110 ofpfw
= (OFPFW_DL_TYPE
| OFPFW_DL_VLAN_PCP
111 | OFPFW_NW_SRC_ALL
| OFPFW_NW_DST_ALL
112 | OFPFW_NW_TOS
| OFPFW_NW_PROTO
113 | OFPFW_TP_SRC
| OFPFW_TP_DST
);
115 ofpfw
= cfg
->wildcards
;
118 ofputil_wildcard_from_openflow(ofpfw
, &sw
->wc
);
121 sw
->default_queue
= cfg
->default_queue
;
122 hmap_init(&sw
->queue_numbers
);
123 shash_init(&sw
->queue_names
);
124 if (cfg
->port_queues
) {
125 struct shash_node
*node
;
127 SHASH_FOR_EACH (node
, cfg
->port_queues
) {
128 struct lswitch_port
*port
= xmalloc(sizeof *port
);
129 hmap_node_nullify(&port
->hmap_node
);
130 port
->queue_id
= (uintptr_t) node
->data
;
131 shash_add(&sw
->queue_names
, node
->name
, port
);
135 sw
->queued
= rconn_packet_counter_create();
136 send_features_request(sw
, rconn
);
138 if (cfg
->default_flows
) {
139 const struct ofpbuf
*b
;
141 LIST_FOR_EACH (b
, list_node
, cfg
->default_flows
) {
142 struct ofpbuf
*copy
= ofpbuf_clone(b
);
143 int error
= rconn_send(rconn
, copy
, NULL
);
145 VLOG_INFO_RL(&rl
, "%s: failed to queue default flows (%s)",
146 rconn_get_name(rconn
), strerror(error
));
158 lswitch_destroy(struct lswitch
*sw
)
161 struct lswitch_port
*node
, *next
;
163 HMAP_FOR_EACH_SAFE (node
, next
, hmap_node
, &sw
->queue_numbers
) {
164 hmap_remove(&sw
->queue_numbers
, &node
->hmap_node
);
167 shash_destroy(&sw
->queue_names
);
168 mac_learning_destroy(sw
->ml
);
169 rconn_packet_counter_destroy(sw
->queued
);
174 /* Takes care of necessary 'sw' activity, except for receiving packets (which
175 * the caller must do). */
177 lswitch_run(struct lswitch
*sw
)
180 mac_learning_run(sw
->ml
, NULL
);
185 lswitch_wait(struct lswitch
*sw
)
188 mac_learning_wait(sw
->ml
);
192 /* Processes 'msg', which should be an OpenFlow received on 'rconn', according
193 * to the learning switch state in 'sw'. The most likely result of processing
194 * is that flow-setup and packet-out OpenFlow messages will be sent out on
197 lswitch_process_packet(struct lswitch
*sw
, struct rconn
*rconn
,
198 const struct ofpbuf
*msg
)
200 const struct ofp_header
*oh
= msg
->data
;
201 const struct ofputil_msg_type
*type
;
203 if (sw
->datapath_id
== 0
204 && oh
->type
!= OFPT_ECHO_REQUEST
205 && oh
->type
!= OFPT_FEATURES_REPLY
) {
206 send_features_request(sw
, rconn
);
210 ofputil_decode_msg_type(oh
, &type
);
211 switch (ofputil_msg_type_code(type
)) {
212 case OFPUTIL_OFPT_ECHO_REQUEST
:
213 process_echo_request(sw
, rconn
, msg
->data
);
216 case OFPUTIL_OFPT_FEATURES_REPLY
:
217 process_switch_features(sw
, msg
->data
);
220 case OFPUTIL_OFPT_PACKET_IN
:
221 process_packet_in(sw
, rconn
, msg
->data
);
224 case OFPUTIL_OFPT_FLOW_REMOVED
:
228 case OFPUTIL_MSG_INVALID
:
229 case OFPUTIL_OFPT_HELLO
:
230 case OFPUTIL_OFPT_ERROR
:
231 case OFPUTIL_OFPT_ECHO_REPLY
:
232 case OFPUTIL_OFPT_FEATURES_REQUEST
:
233 case OFPUTIL_OFPT_GET_CONFIG_REQUEST
:
234 case OFPUTIL_OFPT_GET_CONFIG_REPLY
:
235 case OFPUTIL_OFPT_SET_CONFIG
:
236 case OFPUTIL_OFPT_PORT_STATUS
:
237 case OFPUTIL_OFPT_PACKET_OUT
:
238 case OFPUTIL_OFPT_FLOW_MOD
:
239 case OFPUTIL_OFPT_PORT_MOD
:
240 case OFPUTIL_OFPT_BARRIER_REQUEST
:
241 case OFPUTIL_OFPT_BARRIER_REPLY
:
242 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST
:
243 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY
:
244 case OFPUTIL_OFPST_DESC_REQUEST
:
245 case OFPUTIL_OFPST_FLOW_REQUEST
:
246 case OFPUTIL_OFPST_AGGREGATE_REQUEST
:
247 case OFPUTIL_OFPST_TABLE_REQUEST
:
248 case OFPUTIL_OFPST_PORT_REQUEST
:
249 case OFPUTIL_OFPST_QUEUE_REQUEST
:
250 case OFPUTIL_OFPST_DESC_REPLY
:
251 case OFPUTIL_OFPST_FLOW_REPLY
:
252 case OFPUTIL_OFPST_QUEUE_REPLY
:
253 case OFPUTIL_OFPST_PORT_REPLY
:
254 case OFPUTIL_OFPST_TABLE_REPLY
:
255 case OFPUTIL_OFPST_AGGREGATE_REPLY
:
256 case OFPUTIL_NXT_ROLE_REQUEST
:
257 case OFPUTIL_NXT_ROLE_REPLY
:
258 case OFPUTIL_NXT_FLOW_MOD_TABLE_ID
:
259 case OFPUTIL_NXT_SET_FLOW_FORMAT
:
260 case OFPUTIL_NXT_FLOW_MOD
:
261 case OFPUTIL_NXT_FLOW_REMOVED
:
262 case OFPUTIL_NXST_FLOW_REQUEST
:
263 case OFPUTIL_NXST_AGGREGATE_REQUEST
:
264 case OFPUTIL_NXST_FLOW_REPLY
:
265 case OFPUTIL_NXST_AGGREGATE_REPLY
:
267 if (VLOG_IS_DBG_ENABLED()) {
268 char *s
= ofp_to_string(msg
->data
, msg
->size
, 2);
269 VLOG_DBG_RL(&rl
, "%016llx: OpenFlow packet ignored: %s",
277 send_features_request(struct lswitch
*sw
, struct rconn
*rconn
)
279 time_t now
= time_now();
280 if (now
>= sw
->last_features_request
+ 1) {
282 struct ofp_switch_config
*osc
;
284 /* Send OFPT_FEATURES_REQUEST. */
285 make_openflow(sizeof(struct ofp_header
), OFPT_FEATURES_REQUEST
, &b
);
286 queue_tx(sw
, rconn
, b
);
288 /* Send OFPT_SET_CONFIG. */
289 osc
= make_openflow(sizeof *osc
, OFPT_SET_CONFIG
, &b
);
290 osc
->miss_send_len
= htons(OFP_DEFAULT_MISS_SEND_LEN
);
291 queue_tx(sw
, rconn
, b
);
293 sw
->last_features_request
= now
;
298 queue_tx(struct lswitch
*sw
, struct rconn
*rconn
, struct ofpbuf
*b
)
300 int retval
= rconn_send_with_limit(rconn
, b
, sw
->queued
, 10);
301 if (retval
&& retval
!= ENOTCONN
) {
302 if (retval
== EAGAIN
) {
303 VLOG_INFO_RL(&rl
, "%016llx: %s: tx queue overflow",
304 sw
->datapath_id
, rconn_get_name(rconn
));
306 VLOG_WARN_RL(&rl
, "%016llx: %s: send: %s",
307 sw
->datapath_id
, rconn_get_name(rconn
),
314 process_switch_features(struct lswitch
*sw
, struct ofp_switch_features
*osf
)
319 sw
->datapath_id
= ntohll(osf
->datapath_id
);
321 n_ports
= (ntohs(osf
->header
.length
) - sizeof *osf
) / sizeof *osf
->ports
;
322 for (i
= 0; i
< n_ports
; i
++) {
323 struct ofp_phy_port
*opp
= &osf
->ports
[i
];
324 struct lswitch_port
*lp
;
326 opp
->name
[OFP_MAX_PORT_NAME_LEN
- 1] = '\0';
327 lp
= shash_find_data(&sw
->queue_names
, opp
->name
);
328 if (lp
&& hmap_node_is_null(&lp
->hmap_node
)) {
329 lp
->port_no
= ntohs(opp
->port_no
);
330 hmap_insert(&sw
->queue_numbers
, &lp
->hmap_node
,
331 hash_int(lp
->port_no
, 0));
337 lswitch_choose_destination(struct lswitch
*sw
, const struct flow
*flow
)
341 /* Learn the source MAC. */
342 if (mac_learning_may_learn(sw
->ml
, flow
->dl_src
, 0)) {
343 struct mac_entry
*mac
= mac_learning_insert(sw
->ml
, flow
->dl_src
, 0);
344 if (mac_entry_is_new(mac
) || mac
->port
.i
!= flow
->in_port
) {
345 VLOG_DBG_RL(&rl
, "%016llx: learned that "ETH_ADDR_FMT
" is on "
346 "port %"PRIu16
, sw
->datapath_id
,
347 ETH_ADDR_ARGS(flow
->dl_src
), flow
->in_port
);
349 mac
->port
.i
= flow
->in_port
;
350 mac_learning_changed(sw
->ml
, mac
);
354 /* Drop frames for reserved multicast addresses. */
355 if (eth_addr_is_reserved(flow
->dl_dst
)) {
359 out_port
= OFPP_FLOOD
;
361 struct mac_entry
*mac
;
363 mac
= mac_learning_lookup(sw
->ml
, flow
->dl_dst
, 0, NULL
);
365 out_port
= mac
->port
.i
;
366 if (out_port
== flow
->in_port
) {
367 /* Don't send a packet back out its input port. */
373 /* Check if we need to use "NORMAL" action. */
374 if (sw
->action_normal
&& out_port
!= OFPP_FLOOD
) {
382 get_queue_id(const struct lswitch
*sw
, uint16_t in_port
)
384 const struct lswitch_port
*port
;
386 HMAP_FOR_EACH_WITH_HASH (port
, hmap_node
, hash_int(in_port
, 0),
387 &sw
->queue_numbers
) {
388 if (port
->port_no
== in_port
) {
389 return port
->queue_id
;
393 return sw
->default_queue
;
397 process_packet_in(struct lswitch
*sw
, struct rconn
*rconn
,
398 const struct ofp_packet_in
*opi
)
400 uint16_t in_port
= ntohs(opi
->in_port
);
404 struct ofp_action_header actions
[2];
407 size_t pkt_ofs
, pkt_len
;
411 /* Ignore packets sent via output to OFPP_CONTROLLER. This library never
412 * uses such an action. You never know what experiments might be going on,
413 * though, and it seems best not to interfere with them. */
414 if (opi
->reason
!= OFPR_NO_MATCH
) {
418 /* Extract flow data from 'opi' into 'flow'. */
419 pkt_ofs
= offsetof(struct ofp_packet_in
, data
);
420 pkt_len
= ntohs(opi
->header
.length
) - pkt_ofs
;
421 ofpbuf_use_const(&pkt
, opi
->data
, pkt_len
);
422 flow_extract(&pkt
, 0, 0, in_port
, &flow
);
424 /* Choose output port. */
425 out_port
= lswitch_choose_destination(sw
, &flow
);
428 queue_id
= get_queue_id(sw
, in_port
);
429 if (out_port
== OFPP_NONE
) {
431 } else if (queue_id
== UINT32_MAX
|| out_port
>= OFPP_MAX
) {
432 struct ofp_action_output oao
;
434 memset(&oao
, 0, sizeof oao
);
435 oao
.type
= htons(OFPAT_OUTPUT
);
436 oao
.len
= htons(sizeof oao
);
437 oao
.port
= htons(out_port
);
439 memcpy(actions
, &oao
, sizeof oao
);
440 actions_len
= sizeof oao
;
442 struct ofp_action_enqueue oae
;
444 memset(&oae
, 0, sizeof oae
);
445 oae
.type
= htons(OFPAT_ENQUEUE
);
446 oae
.len
= htons(sizeof oae
);
447 oae
.port
= htons(out_port
);
448 oae
.queue_id
= htonl(queue_id
);
450 memcpy(actions
, &oae
, sizeof oae
);
451 actions_len
= sizeof oae
;
453 assert(actions_len
<= sizeof actions
);
455 /* Send the packet, and possibly the whole flow, to the output port. */
456 if (sw
->max_idle
>= 0 && (!sw
->ml
|| out_port
!= OFPP_FLOOD
)) {
457 struct ofpbuf
*buffer
;
458 struct cls_rule rule
;
460 /* The output port is known, or we always flood everything, so add a
462 cls_rule_init(&flow
, &sw
->wc
, 0, &rule
);
463 buffer
= make_add_flow(&rule
, ntohl(opi
->buffer_id
),
464 sw
->max_idle
, actions_len
);
465 ofpbuf_put(buffer
, actions
, actions_len
);
466 queue_tx(sw
, rconn
, buffer
);
468 /* If the switch didn't buffer the packet, we need to send a copy. */
469 if (ntohl(opi
->buffer_id
) == UINT32_MAX
&& actions_len
> 0) {
471 make_packet_out(&pkt
, UINT32_MAX
, in_port
,
472 actions
, actions_len
/ sizeof *actions
));
475 /* We don't know that MAC, or we don't set up flows. Send along the
476 * packet without setting up a flow. */
477 if (ntohl(opi
->buffer_id
) != UINT32_MAX
|| actions_len
> 0) {
479 make_packet_out(&pkt
, ntohl(opi
->buffer_id
), in_port
,
480 actions
, actions_len
/ sizeof *actions
));
486 process_echo_request(struct lswitch
*sw
, struct rconn
*rconn
,
487 const struct ofp_header
*rq
)
489 queue_tx(sw
, rconn
, make_echo_reply(rq
));