]>
Commit | Line | Data |
---|---|---|
064af421 | 1 | /* |
50f96b10 | 2 | * Copyright (c) 2008-2017 Nicira, Inc. |
064af421 | 3 | * |
a14bc59f BP |
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: | |
064af421 | 7 | * |
a14bc59f BP |
8 | * http://www.apache.org/licenses/LICENSE-2.0 |
9 | * | |
10 | * Unless required by applicable law or agreed to in writing, software | |
11 | * distributed under the License is distributed on an "AS IS" BASIS, | |
12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
13 | * See the License for the specific language governing permissions and | |
14 | * limitations under the License. | |
064af421 BP |
15 | */ |
16 | ||
17 | #include <config.h> | |
18 | #include "learning-switch.h" | |
19 | ||
20 | #include <errno.h> | |
21 | #include <inttypes.h> | |
b2befd5b | 22 | #include <sys/types.h> |
064af421 BP |
23 | #include <netinet/in.h> |
24 | #include <stdlib.h> | |
25 | #include <time.h> | |
26 | ||
10a24935 | 27 | #include "byte-order.h" |
daa68e9f | 28 | #include "classifier.h" |
cf62fa4c | 29 | #include "dp-packet.h" |
064af421 | 30 | #include "flow.h" |
ee89ea7b | 31 | #include "openvswitch/hmap.h" |
064af421 | 32 | #include "mac-learning.h" |
064af421 | 33 | #include "openflow/openflow.h" |
b598f214 | 34 | #include "openvswitch/ofp-actions.h" |
0d71302e | 35 | #include "openvswitch/ofp-connection.h" |
d271907f | 36 | #include "openvswitch/ofp-errors.h" |
0d71302e BP |
37 | #include "openvswitch/ofp-flow.h" |
38 | #include "openvswitch/ofp-match.h" | |
d271907f | 39 | #include "openvswitch/ofp-msgs.h" |
25d436fb | 40 | #include "openvswitch/ofp-print.h" |
d271907f | 41 | #include "openvswitch/ofp-util.h" |
0d71302e BP |
42 | #include "openvswitch/ofp-packet.h" |
43 | #include "openvswitch/ofp-port.h" | |
44 | #include "openvswitch/ofp-switch.h" | |
d271907f BW |
45 | #include "openvswitch/ofpbuf.h" |
46 | #include "openvswitch/vconn.h" | |
47 | #include "openvswitch/vlog.h" | |
fd016ae3 | 48 | #include "openvswitch/poll-loop.h" |
dc02e1eb | 49 | #include "openvswitch/rconn.h" |
ee89ea7b | 50 | #include "openvswitch/shash.h" |
44bac24b | 51 | #include "simap.h" |
064af421 | 52 | #include "timeval.h" |
064af421 | 53 | |
d98e6007 | 54 | VLOG_DEFINE_THIS_MODULE(learning_switch); |
064af421 | 55 | |
d4cdc6b4 BP |
56 | struct lswitch_port { |
57 | struct hmap_node hmap_node; /* Hash node for port number. */ | |
4e022ec0 | 58 | ofp_port_t port_no; /* OpenFlow port number. */ |
d4cdc6b4 BP |
59 | uint32_t queue_id; /* OpenFlow queue number. */ |
60 | }; | |
61 | ||
f702893a BP |
62 | enum lswitch_state { |
63 | S_CONNECTING, /* Waiting for connection to complete. */ | |
64 | S_FEATURES_REPLY, /* Waiting for features reply. */ | |
65 | S_SWITCHING, /* Switching flows. */ | |
66 | }; | |
67 | ||
064af421 | 68 | struct lswitch { |
002c3f17 | 69 | struct rconn *rconn; |
f702893a | 70 | enum lswitch_state state; |
002c3f17 | 71 | |
064af421 BP |
72 | /* If nonnegative, the switch sets up flows that expire after the given |
73 | * number of seconds (or never expire, if the value is OFP_FLOW_PERMANENT). | |
74 | * Otherwise, the switch processes every packet. */ | |
75 | int max_idle; | |
76 | ||
f25d0cf3 | 77 | enum ofputil_protocol protocol; |
064af421 | 78 | unsigned long long int datapath_id; |
064af421 | 79 | struct mac_learning *ml; /* NULL to act as hub instead of switch. */ |
daa68e9f | 80 | struct flow_wildcards wc; /* Wildcards to apply to flows. */ |
9af9e2e8 | 81 | bool action_normal; /* Use OFPP_NORMAL? */ |
d4cdc6b4 BP |
82 | |
83 | /* Queue distribution. */ | |
84 | uint32_t default_queue; /* Default OpenFlow queue, or UINT32_MAX. */ | |
85 | struct hmap queue_numbers; /* Map from port number to lswitch_port. */ | |
86 | struct shash queue_names; /* Map from port name to lswitch_port. */ | |
064af421 BP |
87 | |
88 | /* Number of outgoing queued packets on the rconn. */ | |
89 | struct rconn_packet_counter *queued; | |
002c3f17 BP |
90 | |
91 | /* If true, do not reply to any messages from the switch (for debugging | |
92 | * fail-open mode). */ | |
93 | bool mute; | |
f702893a BP |
94 | |
95 | /* Optional "flow mod" requests to send to the switch at connection time, | |
96 | * to set up the flow table. */ | |
97 | const struct ofputil_flow_mod *default_flows; | |
98 | size_t n_default_flows; | |
db0b6c29 | 99 | enum ofputil_protocol usable_protocols; |
064af421 BP |
100 | }; |
101 | ||
102 | /* The log messages here could actually be useful in debugging, so keep the | |
103 | * rate limit relatively high. */ | |
104 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); | |
105 | ||
002c3f17 BP |
106 | static void queue_tx(struct lswitch *, struct ofpbuf *); |
107 | static void send_features_request(struct lswitch *); | |
064af421 | 108 | |
002c3f17 | 109 | static void lswitch_process_packet(struct lswitch *, const struct ofpbuf *); |
9e1fd49b | 110 | static enum ofperr process_switch_features(struct lswitch *, |
982697a4 | 111 | struct ofp_header *); |
002c3f17 BP |
112 | static void process_packet_in(struct lswitch *, const struct ofp_header *); |
113 | static void process_echo_request(struct lswitch *, const struct ofp_header *); | |
064af421 | 114 | |
9d078ec2 BP |
115 | static ofp_port_t get_mac_entry_ofp_port(const struct mac_learning *ml, |
116 | const struct mac_entry *) | |
117 | OVS_REQ_RDLOCK(ml->rwlock); | |
118 | static void set_mac_entry_ofp_port(struct mac_learning *ml, | |
119 | struct mac_entry *, ofp_port_t) | |
120 | OVS_REQ_WRLOCK(ml->rwlock); | |
121 | ||
ad67e568 BP |
122 | /* Creates and returns a new learning switch whose configuration is given by |
123 | * 'cfg'. | |
aaaa7553 | 124 | * |
064af421 BP |
125 | * 'rconn' is used to send out an OpenFlow features request. */ |
126 | struct lswitch * | |
ad67e568 | 127 | lswitch_create(struct rconn *rconn, const struct lswitch_config *cfg) |
064af421 BP |
128 | { |
129 | struct lswitch *sw; | |
3cbd9931 | 130 | uint32_t ofpfw; |
064af421 | 131 | |
ec6fde61 | 132 | sw = xzalloc(sizeof *sw); |
002c3f17 | 133 | sw->rconn = rconn; |
f702893a | 134 | sw->state = S_CONNECTING; |
ad67e568 | 135 | sw->max_idle = cfg->max_idle; |
064af421 | 136 | sw->datapath_id = 0; |
e764773c BP |
137 | sw->ml = (cfg->mode == LSW_LEARN |
138 | ? mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME) | |
139 | : NULL); | |
ad67e568 | 140 | sw->action_normal = cfg->mode == LSW_NORMAL; |
daa68e9f | 141 | |
3cbd9931 BP |
142 | switch (cfg->wildcards) { |
143 | case 0: | |
144 | ofpfw = 0; | |
145 | break; | |
146 | ||
147 | case UINT32_MAX: | |
148 | /* Try to wildcard as many fields as possible, but we cannot | |
149 | * wildcard all fields. We need in_port to detect moves. We need | |
150 | * Ethernet source and dest and VLAN VID to do L2 learning. */ | |
151 | ofpfw = (OFPFW10_DL_TYPE | OFPFW10_DL_VLAN_PCP | |
152 | | OFPFW10_NW_SRC_ALL | OFPFW10_NW_DST_ALL | |
153 | | OFPFW10_NW_TOS | OFPFW10_NW_PROTO | |
154 | | OFPFW10_TP_SRC | OFPFW10_TP_DST); | |
155 | break; | |
156 | ||
157 | default: | |
158 | ofpfw = cfg->wildcards; | |
159 | break; | |
52f4c6fa | 160 | } |
3cbd9931 | 161 | ofputil_wildcard_from_ofpfw10(ofpfw, &sw->wc); |
d4cdc6b4 BP |
162 | |
163 | sw->default_queue = cfg->default_queue; | |
164 | hmap_init(&sw->queue_numbers); | |
165 | shash_init(&sw->queue_names); | |
166 | if (cfg->port_queues) { | |
44bac24b | 167 | struct simap_node *node; |
d4cdc6b4 | 168 | |
44bac24b | 169 | SIMAP_FOR_EACH (node, cfg->port_queues) { |
d4cdc6b4 BP |
170 | struct lswitch_port *port = xmalloc(sizeof *port); |
171 | hmap_node_nullify(&port->hmap_node); | |
44bac24b | 172 | port->queue_id = node->data; |
d4cdc6b4 BP |
173 | shash_add(&sw->queue_names, node->name, port); |
174 | } | |
175 | } | |
176 | ||
f702893a BP |
177 | sw->default_flows = cfg->default_flows; |
178 | sw->n_default_flows = cfg->n_default_flows; | |
db0b6c29 | 179 | sw->usable_protocols = cfg->usable_protocols; |
f702893a | 180 | |
064af421 | 181 | sw->queued = rconn_packet_counter_create(); |
f702893a BP |
182 | |
183 | return sw; | |
184 | } | |
185 | ||
186 | static void | |
187 | lswitch_handshake(struct lswitch *sw) | |
188 | { | |
189 | enum ofputil_protocol protocol; | |
76fb6ae4 | 190 | enum ofp_version version; |
f702893a | 191 | |
002c3f17 | 192 | send_features_request(sw); |
09913dfd | 193 | |
76fb6ae4 BP |
194 | version = rconn_get_version(sw->rconn); |
195 | protocol = ofputil_protocol_from_ofp_version(version); | |
196 | if (version >= OFP13_VERSION) { | |
197 | /* OpenFlow 1.3 and later by default drop packets that miss in the flow | |
198 | * table. Set up a flow to send packets to the controller by | |
199 | * default. */ | |
76fb6ae4 BP |
200 | struct ofpact_output output; |
201 | struct ofpbuf *msg; | |
202 | int error; | |
203 | ||
204 | ofpact_init_OUTPUT(&output); | |
205 | output.port = OFPP_CONTROLLER; | |
206 | output.max_len = OFP_DEFAULT_MISS_SEND_LEN; | |
207 | ||
39cc5c4a | 208 | struct ofputil_flow_mod fm = { |
39cc5c4a BP |
209 | .priority = 0, |
210 | .table_id = 0, | |
211 | .command = OFPFC_ADD, | |
212 | .buffer_id = UINT32_MAX, | |
213 | .out_port = OFPP_NONE, | |
214 | .out_group = OFPG_ANY, | |
215 | .ofpacts = &output.ofpact, | |
216 | .ofpacts_len = sizeof output, | |
217 | }; | |
6a6b7060 | 218 | minimatch_init_catchall(&fm.match); |
76fb6ae4 | 219 | msg = ofputil_encode_flow_mod(&fm, protocol); |
6a6b7060 BP |
220 | minimatch_destroy(&fm.match); |
221 | ||
76fb6ae4 BP |
222 | error = rconn_send(sw->rconn, msg, NULL); |
223 | if (error) { | |
224 | VLOG_INFO_RL(&rl, "%s: failed to add default flow (%s)", | |
225 | rconn_get_name(sw->rconn), ovs_strerror(error)); | |
226 | } | |
227 | } | |
f702893a | 228 | if (sw->default_flows) { |
27527aa0 | 229 | struct ofpbuf *msg = NULL; |
27527aa0 BP |
230 | int error = 0; |
231 | size_t i; | |
232 | ||
27527aa0 BP |
233 | /* If the initial protocol isn't good enough for default_flows, then |
234 | * pick one that will work and encode messages to set up that | |
235 | * protocol. | |
236 | * | |
237 | * This could be improved by actually negotiating a mutually acceptable | |
238 | * flow format with the switch, but that would require an asynchronous | |
239 | * state machine. This version ought to work fine in practice. */ | |
db0b6c29 JR |
240 | if (!(protocol & sw->usable_protocols)) { |
241 | enum ofputil_protocol want = rightmost_1bit(sw->usable_protocols); | |
27527aa0 BP |
242 | while (!error) { |
243 | msg = ofputil_encode_set_protocol(protocol, want, &protocol); | |
244 | if (!msg) { | |
245 | break; | |
246 | } | |
f702893a | 247 | error = rconn_send(sw->rconn, msg, NULL); |
b85145b5 | 248 | } |
b3907fbc | 249 | } |
db0b6c29 | 250 | if (protocol & sw->usable_protocols) { |
e43928f2 BP |
251 | for (i = 0; !error && i < sw->n_default_flows; i++) { |
252 | msg = ofputil_encode_flow_mod(&sw->default_flows[i], protocol); | |
253 | error = rconn_send(sw->rconn, msg, NULL); | |
254 | } | |
27527aa0 | 255 | |
e43928f2 BP |
256 | if (error) { |
257 | VLOG_INFO_RL(&rl, "%s: failed to queue default flows (%s)", | |
10a89ef0 | 258 | rconn_get_name(sw->rconn), ovs_strerror(error)); |
e43928f2 BP |
259 | } |
260 | } else { | |
261 | VLOG_INFO_RL(&rl, "%s: failed to set usable protocol", | |
262 | rconn_get_name(sw->rconn)); | |
27527aa0 | 263 | } |
b3907fbc | 264 | } |
f25d0cf3 | 265 | sw->protocol = protocol; |
064af421 BP |
266 | } |
267 | ||
002c3f17 BP |
268 | bool |
269 | lswitch_is_alive(const struct lswitch *sw) | |
270 | { | |
271 | return rconn_is_alive(sw->rconn); | |
272 | } | |
273 | ||
064af421 BP |
274 | /* Destroys 'sw'. */ |
275 | void | |
276 | lswitch_destroy(struct lswitch *sw) | |
277 | { | |
278 | if (sw) { | |
4ec3d7c7 | 279 | struct lswitch_port *node; |
d4cdc6b4 | 280 | |
002c3f17 | 281 | rconn_destroy(sw->rconn); |
4ec3d7c7 | 282 | HMAP_FOR_EACH_POP (node, hmap_node, &sw->queue_numbers) { |
d4cdc6b4 BP |
283 | free(node); |
284 | } | |
285 | shash_destroy(&sw->queue_names); | |
5d989517 | 286 | mac_learning_unref(sw->ml); |
064af421 BP |
287 | rconn_packet_counter_destroy(sw->queued); |
288 | free(sw); | |
289 | } | |
290 | } | |
291 | ||
292 | /* Takes care of necessary 'sw' activity, except for receiving packets (which | |
293 | * the caller must do). */ | |
294 | void | |
ba186119 | 295 | lswitch_run(struct lswitch *sw) |
064af421 | 296 | { |
002c3f17 BP |
297 | int i; |
298 | ||
064af421 | 299 | if (sw->ml) { |
509c0149 | 300 | ovs_rwlock_wrlock(&sw->ml->rwlock); |
30618594 | 301 | mac_learning_run(sw->ml); |
509c0149 | 302 | ovs_rwlock_unlock(&sw->ml->rwlock); |
064af421 | 303 | } |
002c3f17 BP |
304 | |
305 | rconn_run(sw->rconn); | |
306 | ||
f702893a | 307 | if (sw->state == S_CONNECTING) { |
476d2551 | 308 | if (rconn_is_connected(sw->rconn)) { |
f702893a BP |
309 | lswitch_handshake(sw); |
310 | sw->state = S_FEATURES_REPLY; | |
311 | } | |
312 | return; | |
313 | } | |
314 | ||
002c3f17 BP |
315 | for (i = 0; i < 50; i++) { |
316 | struct ofpbuf *msg; | |
317 | ||
318 | msg = rconn_recv(sw->rconn); | |
319 | if (!msg) { | |
320 | break; | |
321 | } | |
322 | ||
323 | if (!sw->mute) { | |
324 | lswitch_process_packet(sw, msg); | |
325 | } | |
326 | ofpbuf_delete(msg); | |
327 | } | |
064af421 BP |
328 | } |
329 | ||
330 | void | |
331 | lswitch_wait(struct lswitch *sw) | |
332 | { | |
333 | if (sw->ml) { | |
509c0149 | 334 | ovs_rwlock_rdlock(&sw->ml->rwlock); |
064af421 | 335 | mac_learning_wait(sw->ml); |
509c0149 | 336 | ovs_rwlock_unlock(&sw->ml->rwlock); |
064af421 | 337 | } |
f702893a | 338 | rconn_run_wait(sw->rconn); |
002c3f17 | 339 | rconn_recv_wait(sw->rconn); |
064af421 BP |
340 | } |
341 | ||
342 | /* Processes 'msg', which should be an OpenFlow received on 'rconn', according | |
343 | * to the learning switch state in 'sw'. The most likely result of processing | |
344 | * is that flow-setup and packet-out OpenFlow messages will be sent out on | |
345 | * 'rconn'. */ | |
002c3f17 BP |
346 | static void |
347 | lswitch_process_packet(struct lswitch *sw, const struct ofpbuf *msg) | |
064af421 | 348 | { |
982697a4 BP |
349 | enum ofptype type; |
350 | struct ofpbuf b; | |
351 | ||
352 | b = *msg; | |
353 | if (ofptype_pull(&type, &b)) { | |
354 | return; | |
355 | } | |
d1e2cf21 | 356 | |
f702893a | 357 | if (sw->state == S_FEATURES_REPLY |
982697a4 BP |
358 | && type != OFPTYPE_ECHO_REQUEST |
359 | && type != OFPTYPE_FEATURES_REPLY) { | |
064af421 BP |
360 | return; |
361 | } | |
362 | ||
33fa5fe5 | 363 | if (type == OFPTYPE_ECHO_REQUEST) { |
6fd6ed71 | 364 | process_echo_request(sw, msg->data); |
33fa5fe5 | 365 | } else if (type == OFPTYPE_FEATURES_REPLY) { |
f702893a | 366 | if (sw->state == S_FEATURES_REPLY) { |
6fd6ed71 | 367 | if (!process_switch_features(sw, msg->data)) { |
f702893a BP |
368 | sw->state = S_SWITCHING; |
369 | } else { | |
370 | rconn_disconnect(sw->rconn); | |
371 | } | |
372 | } | |
33fa5fe5 | 373 | } else if (type == OFPTYPE_PACKET_IN) { |
6fd6ed71 | 374 | process_packet_in(sw, msg->data); |
33fa5fe5 | 375 | } else if (type == OFPTYPE_FLOW_REMOVED) { |
d1e2cf21 | 376 | /* Nothing to do. */ |
33fa5fe5 | 377 | } else if (VLOG_IS_DBG_ENABLED()) { |
4bc938cc | 378 | char *s = ofp_to_string(msg->data, msg->size, NULL, NULL, 2); |
33fa5fe5 BP |
379 | VLOG_DBG_RL(&rl, "%016llx: OpenFlow packet ignored: %s", |
380 | sw->datapath_id, s); | |
381 | free(s); | |
064af421 | 382 | } |
064af421 BP |
383 | } |
384 | \f | |
385 | static void | |
002c3f17 | 386 | send_features_request(struct lswitch *sw) |
064af421 | 387 | { |
f702893a | 388 | struct ofpbuf *b; |
f702893a | 389 | int ofp_version = rconn_get_version(sw->rconn); |
064af421 | 390 | |
cb22974d | 391 | ovs_assert(ofp_version > 0 && ofp_version < 0xff); |
064af421 | 392 | |
f702893a BP |
393 | /* Send OFPT_FEATURES_REQUEST. */ |
394 | b = ofpraw_alloc(OFPRAW_OFPT_FEATURES_REQUEST, ofp_version, 0); | |
395 | queue_tx(sw, b); | |
064af421 | 396 | |
f702893a | 397 | /* Send OFPT_SET_CONFIG. */ |
ad99e2ed BP |
398 | struct ofputil_switch_config config = { |
399 | .miss_send_len = OFP_DEFAULT_MISS_SEND_LEN | |
400 | }; | |
401 | queue_tx(sw, ofputil_encode_set_config(&config, ofp_version)); | |
064af421 BP |
402 | } |
403 | ||
404 | static void | |
002c3f17 | 405 | queue_tx(struct lswitch *sw, struct ofpbuf *b) |
064af421 | 406 | { |
002c3f17 | 407 | int retval = rconn_send_with_limit(sw->rconn, b, sw->queued, 10); |
064af421 BP |
408 | if (retval && retval != ENOTCONN) { |
409 | if (retval == EAGAIN) { | |
b123cc3c | 410 | VLOG_INFO_RL(&rl, "%016llx: %s: tx queue overflow", |
002c3f17 | 411 | sw->datapath_id, rconn_get_name(sw->rconn)); |
064af421 | 412 | } else { |
b123cc3c | 413 | VLOG_WARN_RL(&rl, "%016llx: %s: send: %s", |
002c3f17 | 414 | sw->datapath_id, rconn_get_name(sw->rconn), |
10a89ef0 | 415 | ovs_strerror(retval)); |
064af421 BP |
416 | } |
417 | } | |
418 | } | |
419 | ||
9e1fd49b | 420 | static enum ofperr |
982697a4 | 421 | process_switch_features(struct lswitch *sw, struct ofp_header *oh) |
064af421 | 422 | { |
9e1fd49b BP |
423 | struct ofputil_switch_features features; |
424 | struct ofputil_phy_port port; | |
9e1fd49b | 425 | |
0a2869d5 BP |
426 | struct ofpbuf b = ofpbuf_const_initializer(oh, ntohs(oh->length)); |
427 | enum ofperr error = ofputil_pull_switch_features(&b, &features); | |
9e1fd49b BP |
428 | if (error) { |
429 | VLOG_ERR("received invalid switch feature reply (%s)", | |
430 | ofperr_to_string(error)); | |
431 | return error; | |
432 | } | |
d4cdc6b4 | 433 | |
9e1fd49b | 434 | sw->datapath_id = features.datapath_id; |
d4cdc6b4 | 435 | |
982697a4 | 436 | while (!ofputil_pull_phy_port(oh->version, &b, &port)) { |
9e1fd49b | 437 | struct lswitch_port *lp = shash_find_data(&sw->queue_names, port.name); |
d4cdc6b4 | 438 | if (lp && hmap_node_is_null(&lp->hmap_node)) { |
9e1fd49b | 439 | lp->port_no = port.port_no; |
d4cdc6b4 | 440 | hmap_insert(&sw->queue_numbers, &lp->hmap_node, |
f9c0c3ec | 441 | hash_ofp_port(lp->port_no)); |
d4cdc6b4 BP |
442 | } |
443 | } | |
9e1fd49b | 444 | return 0; |
064af421 BP |
445 | } |
446 | ||
4e022ec0 | 447 | static ofp_port_t |
ae412e7d | 448 | lswitch_choose_destination(struct lswitch *sw, const struct flow *flow) |
064af421 | 449 | { |
4e022ec0 | 450 | ofp_port_t out_port; |
064af421 | 451 | |
81f3cad4 | 452 | /* Learn the source MAC. */ |
cbd577d6 Z |
453 | if (sw->ml) { |
454 | ovs_rwlock_wrlock(&sw->ml->rwlock); | |
455 | if (mac_learning_may_learn(sw->ml, flow->dl_src, 0)) { | |
456 | struct mac_entry *mac = mac_learning_insert(sw->ml, flow->dl_src, | |
457 | 0); | |
9d078ec2 BP |
458 | if (get_mac_entry_ofp_port(sw->ml, mac) |
459 | != flow->in_port.ofp_port) { | |
cbd577d6 | 460 | VLOG_DBG_RL(&rl, "%016llx: learned that "ETH_ADDR_FMT" is on " |
94783c7c | 461 | "port %"PRIu32, sw->datapath_id, |
cbd577d6 Z |
462 | ETH_ADDR_ARGS(flow->dl_src), |
463 | flow->in_port.ofp_port); | |
464 | ||
9d078ec2 | 465 | set_mac_entry_ofp_port(sw->ml, mac, flow->in_port.ofp_port); |
cbd577d6 | 466 | } |
064af421 | 467 | } |
cbd577d6 | 468 | ovs_rwlock_unlock(&sw->ml->rwlock); |
064af421 BP |
469 | } |
470 | ||
5a003f60 | 471 | /* Drop frames for reserved multicast addresses. */ |
81f3cad4 BP |
472 | if (eth_addr_is_reserved(flow->dl_dst)) { |
473 | return OFPP_NONE; | |
064af421 BP |
474 | } |
475 | ||
81f3cad4 | 476 | out_port = OFPP_FLOOD; |
064af421 | 477 | if (sw->ml) { |
db8077c3 BP |
478 | struct mac_entry *mac; |
479 | ||
509c0149 | 480 | ovs_rwlock_rdlock(&sw->ml->rwlock); |
30618594 | 481 | mac = mac_learning_lookup(sw->ml, flow->dl_dst, 0); |
db8077c3 | 482 | if (mac) { |
9d078ec2 | 483 | out_port = get_mac_entry_ofp_port(sw->ml, mac); |
4e022ec0 | 484 | if (out_port == flow->in_port.ofp_port) { |
81f3cad4 | 485 | /* Don't send a packet back out its input port. */ |
509c0149 | 486 | ovs_rwlock_unlock(&sw->ml->rwlock); |
81f3cad4 BP |
487 | return OFPP_NONE; |
488 | } | |
064af421 | 489 | } |
509c0149 | 490 | ovs_rwlock_unlock(&sw->ml->rwlock); |
064af421 BP |
491 | } |
492 | ||
81f3cad4 BP |
493 | /* Check if we need to use "NORMAL" action. */ |
494 | if (sw->action_normal && out_port != OFPP_FLOOD) { | |
495 | return OFPP_NORMAL; | |
496 | } | |
497 | ||
498 | return out_port; | |
499 | } | |
500 | ||
d4cdc6b4 | 501 | static uint32_t |
4e022ec0 | 502 | get_queue_id(const struct lswitch *sw, ofp_port_t in_port) |
d4cdc6b4 BP |
503 | { |
504 | const struct lswitch_port *port; | |
505 | ||
f9c0c3ec | 506 | HMAP_FOR_EACH_WITH_HASH (port, hmap_node, hash_ofp_port(in_port), |
d4cdc6b4 BP |
507 | &sw->queue_numbers) { |
508 | if (port->port_no == in_port) { | |
509 | return port->queue_id; | |
510 | } | |
511 | } | |
512 | ||
513 | return sw->default_queue; | |
514 | } | |
515 | ||
81f3cad4 | 516 | static void |
002c3f17 | 517 | process_packet_in(struct lswitch *sw, const struct ofp_header *oh) |
81f3cad4 | 518 | { |
f25d0cf3 | 519 | struct ofputil_packet_in pi; |
9bfe9334 | 520 | uint32_t buffer_id; |
d4cdc6b4 | 521 | uint32_t queue_id; |
4e022ec0 | 522 | ofp_port_t out_port; |
81f3cad4 | 523 | |
f25d0cf3 BP |
524 | uint64_t ofpacts_stub[64 / 8]; |
525 | struct ofpbuf ofpacts; | |
c71270b7 | 526 | |
c6a93eb7 | 527 | struct ofputil_packet_out po; |
f25d0cf3 | 528 | enum ofperr error; |
c6a93eb7 | 529 | |
cf62fa4c | 530 | struct dp_packet pkt; |
ae412e7d | 531 | struct flow flow; |
81f3cad4 | 532 | |
3cddeff0 | 533 | error = ofputil_decode_packet_in(oh, true, NULL, NULL, &pi, NULL, |
8d8ab6c2 | 534 | &buffer_id, NULL); |
f25d0cf3 BP |
535 | if (error) { |
536 | VLOG_WARN_RL(&rl, "failed to decode packet-in: %s", | |
537 | ofperr_to_string(error)); | |
538 | return; | |
539 | } | |
540 | ||
6699af68 BP |
541 | /* Ignore packets sent via output to OFPP_CONTROLLER. This library never |
542 | * uses such an action. You never know what experiments might be going on, | |
543 | * though, and it seems best not to interfere with them. */ | |
f25d0cf3 | 544 | if (pi.reason != OFPR_NO_MATCH) { |
6699af68 BP |
545 | return; |
546 | } | |
547 | ||
9bfe9334 | 548 | /* Extract flow data from 'pi' into 'flow'. */ |
4adaf182 | 549 | dp_packet_use_const(&pkt, pi.packet, pi.packet_len); |
cf62fa4c | 550 | flow_extract(&pkt, &flow); |
50dcbd8e JG |
551 | flow.in_port.ofp_port = pi.flow_metadata.flow.in_port.ofp_port; |
552 | flow.tunnel.tun_id = pi.flow_metadata.flow.tunnel.tun_id; | |
81f3cad4 BP |
553 | |
554 | /* Choose output port. */ | |
555 | out_port = lswitch_choose_destination(sw, &flow); | |
556 | ||
c71270b7 | 557 | /* Make actions. */ |
50dcbd8e | 558 | queue_id = get_queue_id(sw, pi.flow_metadata.flow.in_port.ofp_port); |
f25d0cf3 | 559 | ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub); |
c71270b7 | 560 | if (out_port == OFPP_NONE) { |
f25d0cf3 | 561 | /* No actions. */ |
4e022ec0 AW |
562 | } else if (queue_id == UINT32_MAX |
563 | || ofp_to_u16(out_port) >= ofp_to_u16(OFPP_MAX)) { | |
f25d0cf3 | 564 | ofpact_put_OUTPUT(&ofpacts)->port = out_port; |
c71270b7 | 565 | } else { |
f25d0cf3 BP |
566 | struct ofpact_enqueue *enqueue = ofpact_put_ENQUEUE(&ofpacts); |
567 | enqueue->port = out_port; | |
568 | enqueue->queue = queue_id; | |
c71270b7 | 569 | } |
c71270b7 | 570 | |
c6a93eb7 | 571 | /* Prepare packet_out in case we need one. */ |
9bfe9334 BP |
572 | po.buffer_id = buffer_id; |
573 | if (buffer_id == UINT32_MAX) { | |
cf62fa4c PS |
574 | po.packet = dp_packet_data(&pkt); |
575 | po.packet_len = dp_packet_size(&pkt); | |
c6a93eb7 BP |
576 | } else { |
577 | po.packet = NULL; | |
578 | po.packet_len = 0; | |
579 | } | |
35eb6326 YHW |
580 | match_set_in_port(&po.flow_metadata, |
581 | pi.flow_metadata.flow.in_port.ofp_port); | |
6fd6ed71 PS |
582 | po.ofpacts = ofpacts.data; |
583 | po.ofpacts_len = ofpacts.size; | |
c6a93eb7 | 584 | |
81f3cad4 BP |
585 | /* Send the packet, and possibly the whole flow, to the output port. */ |
586 | if (sw->max_idle >= 0 && (!sw->ml || out_port != OFPP_FLOOD)) { | |
064af421 BP |
587 | /* The output port is known, or we always flood everything, so add a |
588 | * new flow. */ | |
39cc5c4a BP |
589 | struct ofputil_flow_mod fm = { |
590 | .priority = 1, /* Must be > 0 because of table-miss flow entry. */ | |
591 | .table_id = 0xff, | |
592 | .command = OFPFC_ADD, | |
593 | .idle_timeout = sw->max_idle, | |
9bfe9334 | 594 | .buffer_id = buffer_id, |
39cc5c4a BP |
595 | .out_port = OFPP_NONE, |
596 | .ofpacts = ofpacts.data, | |
597 | .ofpacts_len = ofpacts.size, | |
598 | }; | |
6a6b7060 BP |
599 | |
600 | struct match match; | |
601 | match_init(&match, &flow, &sw->wc); | |
602 | ofputil_normalize_match_quiet(&match); | |
603 | minimatch_init(&fm.match, &match); | |
39cc5c4a BP |
604 | |
605 | struct ofpbuf *buffer = ofputil_encode_flow_mod(&fm, sw->protocol); | |
f25d0cf3 | 606 | |
6a6b7060 BP |
607 | minimatch_destroy(&fm.match); |
608 | ||
002c3f17 | 609 | queue_tx(sw, buffer); |
064af421 BP |
610 | |
611 | /* If the switch didn't buffer the packet, we need to send a copy. */ | |
9bfe9334 | 612 | if (buffer_id == UINT32_MAX && out_port != OFPP_NONE) { |
de0f3156 | 613 | queue_tx(sw, ofputil_encode_packet_out(&po, sw->protocol)); |
064af421 BP |
614 | } |
615 | } else { | |
616 | /* We don't know that MAC, or we don't set up flows. Send along the | |
617 | * packet without setting up a flow. */ | |
9bfe9334 | 618 | if (buffer_id != UINT32_MAX || out_port != OFPP_NONE) { |
de0f3156 | 619 | queue_tx(sw, ofputil_encode_packet_out(&po, sw->protocol)); |
064af421 | 620 | } |
064af421 | 621 | } |
064af421 BP |
622 | } |
623 | ||
624 | static void | |
002c3f17 | 625 | process_echo_request(struct lswitch *sw, const struct ofp_header *rq) |
064af421 | 626 | { |
d9cea8f5 | 627 | queue_tx(sw, ofputil_encode_echo_reply(rq)); |
064af421 | 628 | } |
9d078ec2 BP |
629 | |
630 | static ofp_port_t | |
631 | get_mac_entry_ofp_port(const struct mac_learning *ml, | |
632 | const struct mac_entry *e) | |
633 | OVS_REQ_RDLOCK(ml->rwlock) | |
634 | { | |
635 | void *port = mac_entry_get_port(ml, e); | |
636 | return (OVS_FORCE ofp_port_t) (uintptr_t) port; | |
637 | } | |
638 | ||
639 | static void | |
640 | set_mac_entry_ofp_port(struct mac_learning *ml, | |
641 | struct mac_entry *e, ofp_port_t ofp_port) | |
642 | OVS_REQ_WRLOCK(ml->rwlock) | |
643 | { | |
644 | mac_entry_set_port(ml, e, (void *) (OVS_FORCE uintptr_t) ofp_port); | |
645 | } |