]>
Commit | Line | Data |
---|---|---|
064af421 | 1 | /* |
c475ae67 | 2 | * Copyright (c) 2009, 2010 Nicira Networks. |
43253595 | 3 | * Copyright (c) 2010 Jean Tourrilhes - HP-Labs. |
064af421 | 4 | * |
a14bc59f BP |
5 | * Licensed under the Apache License, Version 2.0 (the "License"); |
6 | * you may not use this file except in compliance with the License. | |
7 | * You may obtain a copy of the License at: | |
064af421 | 8 | * |
a14bc59f BP |
9 | * http://www.apache.org/licenses/LICENSE-2.0 |
10 | * | |
11 | * Unless required by applicable law or agreed to in writing, software | |
12 | * distributed under the License is distributed on an "AS IS" BASIS, | |
13 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
14 | * See the License for the specific language governing permissions and | |
15 | * limitations under the License. | |
064af421 BP |
16 | */ |
17 | ||
18 | #include <config.h> | |
19 | #include "ofproto.h" | |
20 | #include <errno.h> | |
21 | #include <inttypes.h> | |
22 | #include <net/if.h> | |
23 | #include <netinet/in.h> | |
24 | #include <stdbool.h> | |
25 | #include <stdlib.h> | |
26 | #include "classifier.h" | |
27 | #include "coverage.h" | |
28 | #include "discovery.h" | |
29 | #include "dpif.h" | |
4f2cad2c | 30 | #include "dynamic-string.h" |
064af421 BP |
31 | #include "fail-open.h" |
32 | #include "in-band.h" | |
33 | #include "mac-learning.h" | |
34 | #include "netdev.h" | |
35 | #include "netflow.h" | |
36 | #include "odp-util.h" | |
37 | #include "ofp-print.h" | |
72b06300 | 38 | #include "ofproto-sflow.h" |
064af421 BP |
39 | #include "ofpbuf.h" |
40 | #include "openflow/nicira-ext.h" | |
41 | #include "openflow/openflow.h" | |
064af421 BP |
42 | #include "openvswitch/datapath-protocol.h" |
43 | #include "packets.h" | |
44 | #include "pinsched.h" | |
45 | #include "pktbuf.h" | |
46 | #include "poll-loop.h" | |
47 | #include "port-array.h" | |
48 | #include "rconn.h" | |
49 | #include "shash.h" | |
50 | #include "status.h" | |
51 | #include "stp.h" | |
fe55ad15 | 52 | #include "stream-ssl.h" |
064af421 BP |
53 | #include "svec.h" |
54 | #include "tag.h" | |
55 | #include "timeval.h" | |
4f2cad2c | 56 | #include "unixctl.h" |
064af421 | 57 | #include "vconn.h" |
064af421 BP |
58 | #include "xtoxll.h" |
59 | ||
60 | #define THIS_MODULE VLM_ofproto | |
61 | #include "vlog.h" | |
62 | ||
72b06300 | 63 | #include "sflow_api.h" |
064af421 BP |
64 | |
65 | enum { | |
66 | TABLEID_HASH = 0, | |
67 | TABLEID_CLASSIFIER = 1 | |
68 | }; | |
69 | ||
70 | struct ofport { | |
71 | struct netdev *netdev; | |
72 | struct ofp_phy_port opp; /* In host byte order. */ | |
73 | }; | |
74 | ||
75 | static void ofport_free(struct ofport *); | |
76 | static void hton_ofp_phy_port(struct ofp_phy_port *); | |
77 | ||
78 | static int xlate_actions(const union ofp_action *in, size_t n_in, | |
79 | const flow_t *flow, struct ofproto *ofproto, | |
80 | const struct ofpbuf *packet, | |
81 | struct odp_actions *out, tag_type *tags, | |
6a07af36 | 82 | bool *may_set_up_flow, uint16_t *nf_output_iface); |
064af421 BP |
83 | |
84 | struct rule { | |
85 | struct cls_rule cr; | |
86 | ||
39997502 JP |
87 | uint64_t flow_cookie; /* Controller-issued identifier. |
88 | (Kept in network-byte order.) */ | |
064af421 BP |
89 | uint16_t idle_timeout; /* In seconds from time of last use. */ |
90 | uint16_t hard_timeout; /* In seconds from time of creation. */ | |
ca069229 | 91 | bool send_flow_removed; /* Send a flow removed message? */ |
064af421 BP |
92 | long long int used; /* Last-used time (0 if never used). */ |
93 | long long int created; /* Creation time. */ | |
94 | uint64_t packet_count; /* Number of packets received. */ | |
95 | uint64_t byte_count; /* Number of bytes received. */ | |
96 | uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */ | |
064af421 | 97 | tag_type tags; /* Tags (set only by hooks). */ |
0193b2af | 98 | struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ |
064af421 BP |
99 | |
100 | /* If 'super' is non-NULL, this rule is a subrule, that is, it is an | |
101 | * exact-match rule (having cr.wc.wildcards of 0) generated from the | |
102 | * wildcard rule 'super'. In this case, 'list' is an element of the | |
103 | * super-rule's list. | |
104 | * | |
105 | * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of | |
106 | * a list of subrules. A super-rule with no wildcards (where | |
107 | * cr.wc.wildcards is 0) will never have any subrules. */ | |
108 | struct rule *super; | |
109 | struct list list; | |
110 | ||
111 | /* OpenFlow actions. | |
79eee1eb BP |
112 | * |
113 | * 'n_actions' is the number of elements in the 'actions' array. A single | |
114 | * action may take up more more than one element's worth of space. | |
064af421 BP |
115 | * |
116 | * A subrule has no actions (it uses the super-rule's actions). */ | |
117 | int n_actions; | |
118 | union ofp_action *actions; | |
119 | ||
120 | /* Datapath actions. | |
121 | * | |
122 | * A super-rule with wildcard fields never has ODP actions (since the | |
123 | * datapath only supports exact-match flows). */ | |
124 | bool installed; /* Installed in datapath? */ | |
125 | bool may_install; /* True ordinarily; false if actions must | |
126 | * be reassessed for every packet. */ | |
127 | int n_odp_actions; | |
128 | union odp_action *odp_actions; | |
129 | }; | |
130 | ||
131 | static inline bool | |
132 | rule_is_hidden(const struct rule *rule) | |
133 | { | |
134 | /* Subrules are merely an implementation detail, so hide them from the | |
135 | * controller. */ | |
136 | if (rule->super != NULL) { | |
137 | return true; | |
138 | } | |
139 | ||
8cd4882f | 140 | /* Rules with priority higher than UINT16_MAX are set up by ofproto itself |
064af421 BP |
141 | * (e.g. by in-band control) and are intentionally hidden from the |
142 | * controller. */ | |
143 | if (rule->cr.priority > UINT16_MAX) { | |
144 | return true; | |
145 | } | |
146 | ||
147 | return false; | |
148 | } | |
149 | ||
0193b2af JG |
150 | static struct rule *rule_create(struct ofproto *, struct rule *super, |
151 | const union ofp_action *, size_t n_actions, | |
ca069229 | 152 | uint16_t idle_timeout, uint16_t hard_timeout, |
39997502 | 153 | uint64_t flow_cookie, bool send_flow_removed); |
064af421 BP |
154 | static void rule_free(struct rule *); |
155 | static void rule_destroy(struct ofproto *, struct rule *); | |
156 | static struct rule *rule_from_cls_rule(const struct cls_rule *); | |
157 | static void rule_insert(struct ofproto *, struct rule *, | |
158 | struct ofpbuf *packet, uint16_t in_port); | |
159 | static void rule_remove(struct ofproto *, struct rule *); | |
160 | static bool rule_make_actions(struct ofproto *, struct rule *, | |
161 | const struct ofpbuf *packet); | |
162 | static void rule_install(struct ofproto *, struct rule *, | |
163 | struct rule *displaced_rule); | |
164 | static void rule_uninstall(struct ofproto *, struct rule *); | |
165 | static void rule_post_uninstall(struct ofproto *, struct rule *); | |
ca069229 JP |
166 | static void send_flow_removed(struct ofproto *p, struct rule *rule, |
167 | long long int now, uint8_t reason); | |
064af421 | 168 | |
76ce9432 BP |
169 | /* ofproto supports two kinds of OpenFlow connections: |
170 | * | |
171 | * - "Controller connections": Connections to ordinary OpenFlow controllers. | |
172 | * ofproto maintains persistent connections to these controllers and by | |
173 | * default sends them asynchronous messages such as packet-ins. | |
174 | * | |
175 | * - "Transient connections", e.g. from ovs-ofctl. When these connections | |
176 | * drop, it is the other side's responsibility to reconnect them if | |
177 | * necessary. ofproto does not send them asynchronous messages by default. | |
178 | */ | |
179 | enum ofconn_type { | |
180 | OFCONN_CONTROLLER, /* An OpenFlow controller. */ | |
181 | OFCONN_TRANSIENT /* A transient connection. */ | |
182 | }; | |
064af421 | 183 | |
76ce9432 BP |
184 | /* An OpenFlow connection. */ |
185 | struct ofconn { | |
186 | struct ofproto *ofproto; /* The ofproto that owns this connection. */ | |
187 | struct list node; /* In struct ofproto's "all_conns" list. */ | |
188 | struct rconn *rconn; /* OpenFlow connection. */ | |
189 | enum ofconn_type type; /* Type. */ | |
190 | ||
191 | /* OFPT_PACKET_IN related data. */ | |
192 | struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */ | |
193 | struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */ | |
194 | struct pktbuf *pktbuf; /* OpenFlow packet buffers. */ | |
195 | int miss_send_len; /* Bytes to send of buffered packets. */ | |
196 | ||
197 | /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow | |
198 | * requests, and the maximum number before we stop reading OpenFlow | |
199 | * requests. */ | |
064af421 BP |
200 | #define OFCONN_REPLY_MAX 100 |
201 | struct rconn_packet_counter *reply_counter; | |
76ce9432 BP |
202 | |
203 | /* type == OFCONN_CONTROLLER only. */ | |
9deba63b | 204 | enum nx_role role; /* Role. */ |
76ce9432 BP |
205 | struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */ |
206 | struct discovery *discovery; /* Controller discovery object, if enabled. */ | |
207 | struct status_category *ss; /* Switch status category. */ | |
d2ede7bc | 208 | enum ofproto_band band; /* In-band or out-of-band? */ |
064af421 BP |
209 | }; |
210 | ||
76ce9432 BP |
211 | /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's |
212 | * "schedulers" array. Their values are 0 and 1, and their meanings and values | |
213 | * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In | |
214 | * case anything ever changes, check their values here. */ | |
215 | #define N_SCHEDULERS 2 | |
216 | BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0); | |
217 | BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR); | |
218 | BUILD_ASSERT_DECL(OFPR_ACTION == 1); | |
219 | BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR); | |
220 | ||
221 | static struct ofconn *ofconn_create(struct ofproto *, struct rconn *, | |
222 | enum ofconn_type); | |
c475ae67 | 223 | static void ofconn_destroy(struct ofconn *); |
064af421 BP |
224 | static void ofconn_run(struct ofconn *, struct ofproto *); |
225 | static void ofconn_wait(struct ofconn *); | |
226 | static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn, | |
227 | struct rconn_packet_counter *counter); | |
228 | ||
76ce9432 BP |
229 | static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg); |
230 | static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn); | |
231 | ||
064af421 BP |
232 | struct ofproto { |
233 | /* Settings. */ | |
234 | uint64_t datapath_id; /* Datapath ID. */ | |
235 | uint64_t fallback_dpid; /* Datapath ID if no better choice found. */ | |
5a719c38 JP |
236 | char *mfr_desc; /* Manufacturer. */ |
237 | char *hw_desc; /* Hardware. */ | |
238 | char *sw_desc; /* Software version. */ | |
239 | char *serial_desc; /* Serial number. */ | |
8abc4ed7 | 240 | char *dp_desc; /* Datapath description. */ |
064af421 BP |
241 | |
242 | /* Datapath. */ | |
c228a364 | 243 | struct dpif *dpif; |
e9e28be3 | 244 | struct netdev_monitor *netdev_monitor; |
064af421 BP |
245 | struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is |
246 | * OFP port nr. */ | |
247 | struct shash port_by_name; | |
248 | uint32_t max_ports; | |
249 | ||
250 | /* Configuration. */ | |
251 | struct switch_status *switch_status; | |
064af421 | 252 | struct fail_open *fail_open; |
064af421 | 253 | struct netflow *netflow; |
72b06300 | 254 | struct ofproto_sflow *sflow; |
064af421 | 255 | |
d2ede7bc BP |
256 | /* In-band control. */ |
257 | struct in_band *in_band; | |
258 | long long int next_in_band_update; | |
917e50e1 BP |
259 | struct sockaddr_in *extra_in_band_remotes; |
260 | size_t n_extra_remotes; | |
261 | ||
064af421 BP |
262 | /* Flow table. */ |
263 | struct classifier cls; | |
264 | bool need_revalidate; | |
265 | long long int next_expiration; | |
266 | struct tag_set revalidate_set; | |
659586ef | 267 | bool tun_id_from_cookie; |
064af421 BP |
268 | |
269 | /* OpenFlow connections. */ | |
76ce9432 BP |
270 | struct hmap controllers; /* Controller "struct ofconn"s. */ |
271 | struct list all_conns; /* Contains "struct ofconn"s. */ | |
064af421 BP |
272 | struct pvconn **listeners; |
273 | size_t n_listeners; | |
274 | struct pvconn **snoops; | |
275 | size_t n_snoops; | |
276 | ||
277 | /* Hooks for ovs-vswitchd. */ | |
278 | const struct ofhooks *ofhooks; | |
279 | void *aux; | |
280 | ||
281 | /* Used by default ofhooks. */ | |
282 | struct mac_learning *ml; | |
283 | }; | |
284 | ||
285 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
286 | ||
287 | static const struct ofhooks default_ofhooks; | |
288 | ||
fa60c019 | 289 | static uint64_t pick_datapath_id(const struct ofproto *); |
064af421 | 290 | static uint64_t pick_fallback_dpid(void); |
76ce9432 | 291 | |
064af421 | 292 | static void update_used(struct ofproto *); |
0193b2af JG |
293 | static void update_stats(struct ofproto *, struct rule *, |
294 | const struct odp_flow_stats *); | |
064af421 | 295 | static void expire_rule(struct cls_rule *, void *ofproto); |
0193b2af | 296 | static void active_timeout(struct ofproto *ofproto, struct rule *rule); |
064af421 BP |
297 | static bool revalidate_rule(struct ofproto *p, struct rule *rule); |
298 | static void revalidate_cb(struct cls_rule *rule_, void *p_); | |
299 | ||
300 | static void handle_odp_msg(struct ofproto *, struct ofpbuf *); | |
301 | ||
302 | static void handle_openflow(struct ofconn *, struct ofproto *, | |
303 | struct ofpbuf *); | |
304 | ||
72b06300 BP |
305 | static void refresh_port_groups(struct ofproto *); |
306 | ||
064af421 BP |
307 | static void update_port(struct ofproto *, const char *devname); |
308 | static int init_ports(struct ofproto *); | |
309 | static void reinit_ports(struct ofproto *); | |
310 | ||
311 | int | |
1a6f1e2a JG |
312 | ofproto_create(const char *datapath, const char *datapath_type, |
313 | const struct ofhooks *ofhooks, void *aux, | |
064af421 BP |
314 | struct ofproto **ofprotop) |
315 | { | |
064af421 BP |
316 | struct odp_stats stats; |
317 | struct ofproto *p; | |
c228a364 | 318 | struct dpif *dpif; |
064af421 BP |
319 | int error; |
320 | ||
321 | *ofprotop = NULL; | |
322 | ||
323 | /* Connect to datapath and start listening for messages. */ | |
1a6f1e2a | 324 | error = dpif_open(datapath, datapath_type, &dpif); |
064af421 BP |
325 | if (error) { |
326 | VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error)); | |
327 | return error; | |
328 | } | |
c228a364 | 329 | error = dpif_get_dp_stats(dpif, &stats); |
064af421 BP |
330 | if (error) { |
331 | VLOG_ERR("failed to obtain stats for datapath %s: %s", | |
332 | datapath, strerror(error)); | |
c228a364 | 333 | dpif_close(dpif); |
064af421 BP |
334 | return error; |
335 | } | |
72b06300 | 336 | error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW); |
064af421 BP |
337 | if (error) { |
338 | VLOG_ERR("failed to listen on datapath %s: %s", | |
339 | datapath, strerror(error)); | |
c228a364 | 340 | dpif_close(dpif); |
064af421 BP |
341 | return error; |
342 | } | |
c228a364 | 343 | dpif_flow_flush(dpif); |
8f24562a | 344 | dpif_recv_purge(dpif); |
064af421 BP |
345 | |
346 | /* Initialize settings. */ | |
ec6fde61 | 347 | p = xzalloc(sizeof *p); |
064af421 | 348 | p->fallback_dpid = pick_fallback_dpid(); |
fa60c019 | 349 | p->datapath_id = p->fallback_dpid; |
5a719c38 JP |
350 | p->mfr_desc = xstrdup(DEFAULT_MFR_DESC); |
351 | p->hw_desc = xstrdup(DEFAULT_HW_DESC); | |
352 | p->sw_desc = xstrdup(DEFAULT_SW_DESC); | |
353 | p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC); | |
23ff2821 | 354 | p->dp_desc = xstrdup(DEFAULT_DP_DESC); |
064af421 BP |
355 | |
356 | /* Initialize datapath. */ | |
357 | p->dpif = dpif; | |
8b61709d | 358 | p->netdev_monitor = netdev_monitor_create(); |
064af421 BP |
359 | port_array_init(&p->ports); |
360 | shash_init(&p->port_by_name); | |
361 | p->max_ports = stats.max_ports; | |
362 | ||
363 | /* Initialize submodules. */ | |
364 | p->switch_status = switch_status_create(p); | |
365 | p->in_band = NULL; | |
064af421 | 366 | p->fail_open = NULL; |
064af421 | 367 | p->netflow = NULL; |
72b06300 | 368 | p->sflow = NULL; |
064af421 BP |
369 | |
370 | /* Initialize flow table. */ | |
371 | classifier_init(&p->cls); | |
372 | p->need_revalidate = false; | |
373 | p->next_expiration = time_msec() + 1000; | |
374 | tag_set_init(&p->revalidate_set); | |
375 | ||
376 | /* Initialize OpenFlow connections. */ | |
377 | list_init(&p->all_conns); | |
76ce9432 | 378 | hmap_init(&p->controllers); |
064af421 BP |
379 | p->listeners = NULL; |
380 | p->n_listeners = 0; | |
381 | p->snoops = NULL; | |
382 | p->n_snoops = 0; | |
383 | ||
384 | /* Initialize hooks. */ | |
385 | if (ofhooks) { | |
386 | p->ofhooks = ofhooks; | |
387 | p->aux = aux; | |
388 | p->ml = NULL; | |
389 | } else { | |
390 | p->ofhooks = &default_ofhooks; | |
391 | p->aux = p; | |
392 | p->ml = mac_learning_create(); | |
393 | } | |
394 | ||
fa60c019 BP |
395 | /* Pick final datapath ID. */ |
396 | p->datapath_id = pick_datapath_id(p); | |
b123cc3c | 397 | VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id); |
fa60c019 | 398 | |
064af421 BP |
399 | *ofprotop = p; |
400 | return 0; | |
401 | } | |
402 | ||
403 | void | |
404 | ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id) | |
405 | { | |
406 | uint64_t old_dpid = p->datapath_id; | |
fa60c019 | 407 | p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p); |
064af421 | 408 | if (p->datapath_id != old_dpid) { |
76ce9432 BP |
409 | struct ofconn *ofconn; |
410 | ||
b123cc3c | 411 | VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id); |
76ce9432 BP |
412 | |
413 | /* Force all active connections to reconnect, since there is no way to | |
414 | * notify a controller that the datapath ID has changed. */ | |
415 | LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) { | |
416 | rconn_reconnect(ofconn->rconn); | |
417 | } | |
064af421 BP |
418 | } |
419 | } | |
420 | ||
76ce9432 BP |
421 | static bool |
422 | is_discovery_controller(const struct ofproto_controller *c) | |
423 | { | |
424 | return !strcmp(c->target, "discover"); | |
425 | } | |
426 | ||
427 | static bool | |
428 | is_in_band_controller(const struct ofproto_controller *c) | |
429 | { | |
430 | return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND; | |
431 | } | |
432 | ||
433 | /* Creates a new controller in 'ofproto'. Some of the settings are initially | |
434 | * drawn from 'c', but update_controller() needs to be called later to finish | |
435 | * the new ofconn's configuration. */ | |
436 | static void | |
437 | add_controller(struct ofproto *ofproto, const struct ofproto_controller *c) | |
438 | { | |
439 | struct discovery *discovery; | |
440 | struct ofconn *ofconn; | |
441 | ||
442 | if (is_discovery_controller(c)) { | |
443 | int error = discovery_create(c->accept_re, c->update_resolv_conf, | |
444 | ofproto->dpif, ofproto->switch_status, | |
445 | &discovery); | |
446 | if (error) { | |
447 | return; | |
448 | } | |
449 | } else { | |
450 | discovery = NULL; | |
451 | } | |
452 | ||
453 | ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_CONTROLLER); | |
454 | ofconn->pktbuf = pktbuf_create(); | |
455 | ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN; | |
456 | if (discovery) { | |
457 | ofconn->discovery = discovery; | |
458 | } else { | |
459 | rconn_connect(ofconn->rconn, c->target); | |
460 | } | |
461 | hmap_insert(&ofproto->controllers, &ofconn->hmap_node, | |
462 | hash_string(c->target, 0)); | |
463 | } | |
464 | ||
465 | /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's | |
466 | * target or turn discovery on or off (these are done by creating new ofconns | |
467 | * and deleting old ones), but it can update the rest of an ofconn's | |
468 | * settings. */ | |
469 | static void | |
470 | update_controller(struct ofconn *ofconn, const struct ofproto_controller *c) | |
064af421 | 471 | { |
76ce9432 BP |
472 | struct ofproto *ofproto = ofconn->ofproto; |
473 | int probe_interval; | |
474 | int i; | |
79c9f2ee | 475 | |
d2ede7bc BP |
476 | ofconn->band = (is_in_band_controller(c) |
477 | ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND); | |
478 | ||
76ce9432 | 479 | rconn_set_max_backoff(ofconn->rconn, c->max_backoff); |
79c9f2ee | 480 | |
76ce9432 BP |
481 | probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0; |
482 | rconn_set_probe_interval(ofconn->rconn, probe_interval); | |
79c9f2ee | 483 | |
76ce9432 BP |
484 | if (ofconn->discovery) { |
485 | discovery_set_update_resolv_conf(ofconn->discovery, | |
486 | c->update_resolv_conf); | |
487 | discovery_set_accept_controller_re(ofconn->discovery, c->accept_re); | |
488 | } | |
79c9f2ee | 489 | |
76ce9432 BP |
490 | for (i = 0; i < N_SCHEDULERS; i++) { |
491 | struct pinsched **s = &ofconn->schedulers[i]; | |
79c9f2ee | 492 | |
76ce9432 BP |
493 | if (c->rate_limit > 0) { |
494 | if (!*s) { | |
495 | *s = pinsched_create(c->rate_limit, c->burst_limit, | |
496 | ofproto->switch_status); | |
79c9f2ee | 497 | } else { |
76ce9432 | 498 | pinsched_set_limits(*s, c->rate_limit, c->burst_limit); |
79c9f2ee | 499 | } |
76ce9432 BP |
500 | } else { |
501 | pinsched_destroy(*s); | |
502 | *s = NULL; | |
79c9f2ee | 503 | } |
76ce9432 BP |
504 | } |
505 | } | |
79c9f2ee | 506 | |
76ce9432 BP |
507 | static const char * |
508 | ofconn_get_target(const struct ofconn *ofconn) | |
509 | { | |
510 | return ofconn->discovery ? "discover" : rconn_get_name(ofconn->rconn); | |
511 | } | |
512 | ||
513 | static struct ofconn * | |
514 | find_controller_by_target(struct ofproto *ofproto, const char *target) | |
515 | { | |
516 | struct ofconn *ofconn; | |
517 | ||
518 | HMAP_FOR_EACH_WITH_HASH (ofconn, struct ofconn, hmap_node, | |
519 | hash_string(target, 0), &ofproto->controllers) { | |
520 | if (!strcmp(ofconn_get_target(ofconn), target)) { | |
521 | return ofconn; | |
79c9f2ee | 522 | } |
064af421 | 523 | } |
76ce9432 BP |
524 | return NULL; |
525 | } | |
064af421 | 526 | |
d2ede7bc BP |
527 | static void |
528 | update_in_band_remotes(struct ofproto *ofproto) | |
529 | { | |
530 | const struct ofconn *ofconn; | |
531 | struct sockaddr_in *addrs; | |
917e50e1 | 532 | size_t max_addrs, n_addrs; |
d2ede7bc | 533 | bool discovery; |
917e50e1 | 534 | size_t i; |
d2ede7bc | 535 | |
917e50e1 BP |
536 | /* Allocate enough memory for as many remotes as we could possibly have. */ |
537 | max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers); | |
538 | addrs = xmalloc(max_addrs * sizeof *addrs); | |
d2ede7bc BP |
539 | n_addrs = 0; |
540 | ||
541 | /* Add all the remotes. */ | |
542 | discovery = false; | |
543 | HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &ofproto->controllers) { | |
544 | struct sockaddr_in *sin = &addrs[n_addrs]; | |
545 | ||
546 | sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn); | |
547 | if (sin->sin_addr.s_addr) { | |
548 | sin->sin_port = rconn_get_remote_port(ofconn->rconn); | |
549 | n_addrs++; | |
550 | } | |
551 | if (ofconn->discovery) { | |
552 | discovery = true; | |
553 | } | |
554 | } | |
917e50e1 BP |
555 | for (i = 0; i < ofproto->n_extra_remotes; i++) { |
556 | addrs[n_addrs++] = ofproto->extra_in_band_remotes[i]; | |
557 | } | |
d2ede7bc BP |
558 | |
559 | /* Create or update or destroy in-band. | |
560 | * | |
561 | * Ordinarily we only enable in-band if there's at least one remote | |
562 | * address, but discovery needs the in-band rules for DHCP to be installed | |
563 | * even before we know any remote addresses. */ | |
564 | if (n_addrs || discovery) { | |
565 | if (!ofproto->in_band) { | |
566 | in_band_create(ofproto, ofproto->dpif, ofproto->switch_status, | |
567 | &ofproto->in_band); | |
568 | } | |
569 | in_band_set_remotes(ofproto->in_band, addrs, n_addrs); | |
570 | ofproto->next_in_band_update = time_msec() + 1000; | |
571 | } else { | |
572 | in_band_destroy(ofproto->in_band); | |
573 | ofproto->in_band = NULL; | |
574 | } | |
575 | ||
576 | /* Clean up. */ | |
577 | free(addrs); | |
578 | } | |
579 | ||
76ce9432 BP |
580 | void |
581 | ofproto_set_controllers(struct ofproto *p, | |
582 | const struct ofproto_controller *controllers, | |
583 | size_t n_controllers) | |
584 | { | |
585 | struct shash new_controllers; | |
76ce9432 BP |
586 | enum ofproto_fail_mode fail_mode; |
587 | struct ofconn *ofconn, *next; | |
588 | bool ss_exists; | |
76ce9432 | 589 | size_t i; |
79c9f2ee | 590 | |
76ce9432 BP |
591 | shash_init(&new_controllers); |
592 | for (i = 0; i < n_controllers; i++) { | |
593 | const struct ofproto_controller *c = &controllers[i]; | |
594 | ||
595 | shash_add_once(&new_controllers, c->target, &controllers[i]); | |
596 | if (!find_controller_by_target(p, c->target)) { | |
597 | add_controller(p, c); | |
598 | } | |
599 | } | |
600 | ||
76ce9432 BP |
601 | fail_mode = OFPROTO_FAIL_STANDALONE; |
602 | ss_exists = false; | |
603 | HMAP_FOR_EACH_SAFE (ofconn, next, struct ofconn, hmap_node, | |
604 | &p->controllers) { | |
605 | struct ofproto_controller *c; | |
606 | ||
607 | c = shash_find_data(&new_controllers, ofconn_get_target(ofconn)); | |
608 | if (!c) { | |
609 | ofconn_destroy(ofconn); | |
79c9f2ee | 610 | } else { |
76ce9432 | 611 | update_controller(ofconn, c); |
76ce9432 BP |
612 | if (ofconn->ss) { |
613 | ss_exists = true; | |
614 | } | |
76ce9432 BP |
615 | if (c->fail == OFPROTO_FAIL_SECURE) { |
616 | fail_mode = OFPROTO_FAIL_SECURE; | |
617 | } | |
618 | } | |
619 | } | |
620 | shash_destroy(&new_controllers); | |
621 | ||
d2ede7bc | 622 | update_in_band_remotes(p); |
76ce9432 BP |
623 | |
624 | if (!hmap_is_empty(&p->controllers) | |
625 | && fail_mode == OFPROTO_FAIL_STANDALONE) { | |
626 | struct rconn **rconns; | |
627 | size_t n; | |
79c9f2ee | 628 | |
79c9f2ee | 629 | if (!p->fail_open) { |
76ce9432 BP |
630 | p->fail_open = fail_open_create(p, p->switch_status); |
631 | } | |
632 | ||
633 | n = 0; | |
634 | rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns); | |
635 | HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &p->controllers) { | |
636 | rconns[n++] = ofconn->rconn; | |
79c9f2ee | 637 | } |
76ce9432 BP |
638 | |
639 | fail_open_set_controllers(p->fail_open, rconns, n); | |
640 | /* p->fail_open takes ownership of 'rconns'. */ | |
79c9f2ee BP |
641 | } else { |
642 | fail_open_destroy(p->fail_open); | |
643 | p->fail_open = NULL; | |
644 | } | |
645 | ||
76ce9432 BP |
646 | if (!hmap_is_empty(&p->controllers) && !ss_exists) { |
647 | ofconn = CONTAINER_OF(hmap_first(&p->controllers), | |
648 | struct ofconn, hmap_node); | |
649 | ofconn->ss = switch_status_register(p->switch_status, "remote", | |
650 | rconn_status_cb, ofconn->rconn); | |
79c9f2ee | 651 | } |
064af421 BP |
652 | } |
653 | ||
917e50e1 BP |
654 | static bool |
655 | any_extras_changed(const struct ofproto *ofproto, | |
656 | const struct sockaddr_in *extras, size_t n) | |
657 | { | |
658 | size_t i; | |
659 | ||
660 | if (n != ofproto->n_extra_remotes) { | |
661 | return true; | |
662 | } | |
663 | ||
664 | for (i = 0; i < n; i++) { | |
665 | const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i]; | |
666 | const struct sockaddr_in *new = &extras[i]; | |
667 | ||
668 | if (old->sin_addr.s_addr != new->sin_addr.s_addr || | |
669 | old->sin_port != new->sin_port) { | |
670 | return true; | |
671 | } | |
672 | } | |
673 | ||
674 | return false; | |
675 | } | |
676 | ||
677 | /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s | |
678 | * in-band control should guarantee access, in the same way that in-band | |
679 | * control guarantees access to OpenFlow controllers. */ | |
680 | void | |
681 | ofproto_set_extra_in_band_remotes(struct ofproto *ofproto, | |
682 | const struct sockaddr_in *extras, size_t n) | |
683 | { | |
684 | if (!any_extras_changed(ofproto, extras, n)) { | |
685 | return; | |
686 | } | |
687 | ||
688 | free(ofproto->extra_in_band_remotes); | |
689 | ofproto->n_extra_remotes = n; | |
690 | ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras); | |
691 | ||
692 | update_in_band_remotes(ofproto); | |
693 | } | |
694 | ||
064af421 BP |
695 | void |
696 | ofproto_set_desc(struct ofproto *p, | |
5a719c38 JP |
697 | const char *mfr_desc, const char *hw_desc, |
698 | const char *sw_desc, const char *serial_desc, | |
8abc4ed7 | 699 | const char *dp_desc) |
064af421 | 700 | { |
5a719c38 JP |
701 | struct ofp_desc_stats *ods; |
702 | ||
703 | if (mfr_desc) { | |
704 | if (strlen(mfr_desc) >= sizeof ods->mfr_desc) { | |
705 | VLOG_WARN("truncating mfr_desc, must be less than %zu characters", | |
706 | sizeof ods->mfr_desc); | |
707 | } | |
708 | free(p->mfr_desc); | |
709 | p->mfr_desc = xstrdup(mfr_desc); | |
064af421 | 710 | } |
5a719c38 JP |
711 | if (hw_desc) { |
712 | if (strlen(hw_desc) >= sizeof ods->hw_desc) { | |
713 | VLOG_WARN("truncating hw_desc, must be less than %zu characters", | |
714 | sizeof ods->hw_desc); | |
715 | } | |
716 | free(p->hw_desc); | |
717 | p->hw_desc = xstrdup(hw_desc); | |
064af421 | 718 | } |
5a719c38 JP |
719 | if (sw_desc) { |
720 | if (strlen(sw_desc) >= sizeof ods->sw_desc) { | |
721 | VLOG_WARN("truncating sw_desc, must be less than %zu characters", | |
722 | sizeof ods->sw_desc); | |
723 | } | |
724 | free(p->sw_desc); | |
725 | p->sw_desc = xstrdup(sw_desc); | |
726 | } | |
727 | if (serial_desc) { | |
728 | if (strlen(serial_desc) >= sizeof ods->serial_num) { | |
729 | VLOG_WARN("truncating serial_desc, must be less than %zu " | |
730 | "characters", | |
731 | sizeof ods->serial_num); | |
732 | } | |
733 | free(p->serial_desc); | |
734 | p->serial_desc = xstrdup(serial_desc); | |
064af421 | 735 | } |
8abc4ed7 | 736 | if (dp_desc) { |
5a719c38 JP |
737 | if (strlen(dp_desc) >= sizeof ods->dp_desc) { |
738 | VLOG_WARN("truncating dp_desc, must be less than %zu characters", | |
739 | sizeof ods->dp_desc); | |
740 | } | |
8abc4ed7 JP |
741 | free(p->dp_desc); |
742 | p->dp_desc = xstrdup(dp_desc); | |
743 | } | |
064af421 BP |
744 | } |
745 | ||
064af421 BP |
746 | static int |
747 | set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp, | |
748 | const struct svec *svec) | |
749 | { | |
750 | struct pvconn **pvconns = *pvconnsp; | |
751 | size_t n_pvconns = *n_pvconnsp; | |
752 | int retval = 0; | |
753 | size_t i; | |
754 | ||
755 | for (i = 0; i < n_pvconns; i++) { | |
756 | pvconn_close(pvconns[i]); | |
757 | } | |
758 | free(pvconns); | |
759 | ||
760 | pvconns = xmalloc(svec->n * sizeof *pvconns); | |
761 | n_pvconns = 0; | |
762 | for (i = 0; i < svec->n; i++) { | |
763 | const char *name = svec->names[i]; | |
764 | struct pvconn *pvconn; | |
765 | int error; | |
766 | ||
767 | error = pvconn_open(name, &pvconn); | |
768 | if (!error) { | |
769 | pvconns[n_pvconns++] = pvconn; | |
770 | } else { | |
771 | VLOG_ERR("failed to listen on %s: %s", name, strerror(error)); | |
772 | if (!retval) { | |
773 | retval = error; | |
774 | } | |
775 | } | |
776 | } | |
777 | ||
778 | *pvconnsp = pvconns; | |
779 | *n_pvconnsp = n_pvconns; | |
780 | ||
781 | return retval; | |
782 | } | |
783 | ||
784 | int | |
785 | ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners) | |
786 | { | |
787 | return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners); | |
788 | } | |
789 | ||
790 | int | |
791 | ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops) | |
792 | { | |
793 | return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops); | |
794 | } | |
795 | ||
796 | int | |
0193b2af JG |
797 | ofproto_set_netflow(struct ofproto *ofproto, |
798 | const struct netflow_options *nf_options) | |
064af421 | 799 | { |
76343538 | 800 | if (nf_options && nf_options->collectors.n) { |
064af421 BP |
801 | if (!ofproto->netflow) { |
802 | ofproto->netflow = netflow_create(); | |
803 | } | |
0193b2af | 804 | return netflow_set_options(ofproto->netflow, nf_options); |
064af421 BP |
805 | } else { |
806 | netflow_destroy(ofproto->netflow); | |
807 | ofproto->netflow = NULL; | |
808 | return 0; | |
809 | } | |
810 | } | |
811 | ||
72b06300 BP |
812 | void |
813 | ofproto_set_sflow(struct ofproto *ofproto, | |
814 | const struct ofproto_sflow_options *oso) | |
815 | { | |
816 | struct ofproto_sflow *os = ofproto->sflow; | |
817 | if (oso) { | |
818 | if (!os) { | |
819 | struct ofport *ofport; | |
820 | unsigned int odp_port; | |
821 | ||
822 | os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif); | |
823 | refresh_port_groups(ofproto); | |
824 | PORT_ARRAY_FOR_EACH (ofport, &ofproto->ports, odp_port) { | |
825 | ofproto_sflow_add_port(os, odp_port, | |
826 | netdev_get_name(ofport->netdev)); | |
827 | } | |
828 | } | |
829 | ofproto_sflow_set_options(os, oso); | |
830 | } else { | |
831 | ofproto_sflow_destroy(os); | |
832 | ofproto->sflow = NULL; | |
833 | } | |
834 | } | |
835 | ||
064af421 | 836 | int |
67a4917b | 837 | ofproto_set_stp(struct ofproto *ofproto OVS_UNUSED, bool enable_stp) |
064af421 BP |
838 | { |
839 | /* XXX */ | |
840 | if (enable_stp) { | |
841 | VLOG_WARN("STP is not yet implemented"); | |
842 | return EINVAL; | |
843 | } else { | |
844 | return 0; | |
845 | } | |
846 | } | |
847 | ||
064af421 BP |
848 | uint64_t |
849 | ofproto_get_datapath_id(const struct ofproto *ofproto) | |
850 | { | |
851 | return ofproto->datapath_id; | |
852 | } | |
853 | ||
76ce9432 BP |
854 | bool |
855 | ofproto_has_controller(const struct ofproto *ofproto) | |
064af421 | 856 | { |
76ce9432 | 857 | return !hmap_is_empty(&ofproto->controllers); |
064af421 BP |
858 | } |
859 | ||
860 | void | |
861 | ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners) | |
862 | { | |
863 | size_t i; | |
864 | ||
865 | for (i = 0; i < ofproto->n_listeners; i++) { | |
866 | svec_add(listeners, pvconn_get_name(ofproto->listeners[i])); | |
867 | } | |
868 | } | |
869 | ||
870 | void | |
871 | ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops) | |
872 | { | |
873 | size_t i; | |
874 | ||
875 | for (i = 0; i < ofproto->n_snoops; i++) { | |
876 | svec_add(snoops, pvconn_get_name(ofproto->snoops[i])); | |
877 | } | |
878 | } | |
879 | ||
880 | void | |
881 | ofproto_destroy(struct ofproto *p) | |
882 | { | |
883 | struct ofconn *ofconn, *next_ofconn; | |
884 | struct ofport *ofport; | |
885 | unsigned int port_no; | |
886 | size_t i; | |
887 | ||
888 | if (!p) { | |
889 | return; | |
890 | } | |
891 | ||
f7de2cdf | 892 | /* Destroy fail-open and in-band early, since they touch the classifier. */ |
79c9f2ee BP |
893 | fail_open_destroy(p->fail_open); |
894 | p->fail_open = NULL; | |
895 | ||
896 | in_band_destroy(p->in_band); | |
897 | p->in_band = NULL; | |
917e50e1 | 898 | free(p->extra_in_band_remotes); |
2f6d3445 | 899 | |
064af421 BP |
900 | ofproto_flush_flows(p); |
901 | classifier_destroy(&p->cls); | |
902 | ||
903 | LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node, | |
904 | &p->all_conns) { | |
c475ae67 | 905 | ofconn_destroy(ofconn); |
064af421 | 906 | } |
76ce9432 | 907 | hmap_destroy(&p->controllers); |
064af421 | 908 | |
c228a364 | 909 | dpif_close(p->dpif); |
e9e28be3 | 910 | netdev_monitor_destroy(p->netdev_monitor); |
064af421 BP |
911 | PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) { |
912 | ofport_free(ofport); | |
913 | } | |
914 | shash_destroy(&p->port_by_name); | |
915 | ||
916 | switch_status_destroy(p->switch_status); | |
064af421 | 917 | netflow_destroy(p->netflow); |
72b06300 | 918 | ofproto_sflow_destroy(p->sflow); |
064af421 | 919 | |
064af421 BP |
920 | for (i = 0; i < p->n_listeners; i++) { |
921 | pvconn_close(p->listeners[i]); | |
922 | } | |
923 | free(p->listeners); | |
924 | ||
925 | for (i = 0; i < p->n_snoops; i++) { | |
926 | pvconn_close(p->snoops[i]); | |
927 | } | |
928 | free(p->snoops); | |
929 | ||
930 | mac_learning_destroy(p->ml); | |
931 | ||
5a719c38 JP |
932 | free(p->mfr_desc); |
933 | free(p->hw_desc); | |
934 | free(p->sw_desc); | |
935 | free(p->serial_desc); | |
cb871ae0 JP |
936 | free(p->dp_desc); |
937 | ||
3b917492 BP |
938 | port_array_destroy(&p->ports); |
939 | ||
064af421 BP |
940 | free(p); |
941 | } | |
942 | ||
943 | int | |
944 | ofproto_run(struct ofproto *p) | |
945 | { | |
946 | int error = ofproto_run1(p); | |
947 | if (!error) { | |
948 | error = ofproto_run2(p, false); | |
949 | } | |
950 | return error; | |
951 | } | |
952 | ||
e9e28be3 BP |
953 | static void |
954 | process_port_change(struct ofproto *ofproto, int error, char *devname) | |
955 | { | |
956 | if (error == ENOBUFS) { | |
957 | reinit_ports(ofproto); | |
958 | } else if (!error) { | |
959 | update_port(ofproto, devname); | |
960 | free(devname); | |
961 | } | |
962 | } | |
963 | ||
76ce9432 BP |
964 | /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'. |
965 | * Connects this vconn to a controller. */ | |
966 | static void | |
967 | add_snooper(struct ofproto *ofproto, struct vconn *vconn) | |
968 | { | |
969 | struct ofconn *ofconn; | |
970 | ||
971 | /* Arbitrarily pick the first controller in the list for monitoring. We | |
972 | * could do something smarter or more flexible later, if it ever proves | |
973 | * useful. */ | |
974 | LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) { | |
975 | if (ofconn->type == OFCONN_CONTROLLER) { | |
976 | rconn_add_monitor(ofconn->rconn, vconn); | |
977 | return; | |
978 | } | |
979 | ||
980 | } | |
981 | VLOG_INFO_RL(&rl, "no controller connection to monitor"); | |
982 | vconn_close(vconn); | |
983 | } | |
984 | ||
064af421 BP |
985 | int |
986 | ofproto_run1(struct ofproto *p) | |
987 | { | |
988 | struct ofconn *ofconn, *next_ofconn; | |
989 | char *devname; | |
990 | int error; | |
991 | int i; | |
992 | ||
149f577a JG |
993 | if (shash_is_empty(&p->port_by_name)) { |
994 | init_ports(p); | |
995 | } | |
996 | ||
064af421 BP |
997 | for (i = 0; i < 50; i++) { |
998 | struct ofpbuf *buf; | |
999 | int error; | |
1000 | ||
c228a364 | 1001 | error = dpif_recv(p->dpif, &buf); |
064af421 BP |
1002 | if (error) { |
1003 | if (error == ENODEV) { | |
1004 | /* Someone destroyed the datapath behind our back. The caller | |
1005 | * better destroy us and give up, because we're just going to | |
1006 | * spin from here on out. */ | |
39a559f2 BP |
1007 | static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5); |
1008 | VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally", | |
c228a364 | 1009 | dpif_name(p->dpif)); |
064af421 BP |
1010 | return ENODEV; |
1011 | } | |
1012 | break; | |
1013 | } | |
1014 | ||
1015 | handle_odp_msg(p, buf); | |
1016 | } | |
1017 | ||
e9e28be3 BP |
1018 | while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) { |
1019 | process_port_change(p, error, devname); | |
1020 | } | |
1021 | while ((error = netdev_monitor_poll(p->netdev_monitor, | |
1022 | &devname)) != EAGAIN) { | |
1023 | process_port_change(p, error, devname); | |
064af421 BP |
1024 | } |
1025 | ||
1026 | if (p->in_band) { | |
d2ede7bc BP |
1027 | if (time_msec() >= p->next_in_band_update) { |
1028 | update_in_band_remotes(p); | |
1029 | } | |
064af421 BP |
1030 | in_band_run(p->in_band); |
1031 | } | |
064af421 BP |
1032 | |
1033 | LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node, | |
1034 | &p->all_conns) { | |
1035 | ofconn_run(ofconn, p); | |
1036 | } | |
1037 | ||
7778bd15 BP |
1038 | /* Fail-open maintenance. Do this after processing the ofconns since |
1039 | * fail-open checks the status of the controller rconn. */ | |
1040 | if (p->fail_open) { | |
1041 | fail_open_run(p->fail_open); | |
1042 | } | |
1043 | ||
064af421 BP |
1044 | for (i = 0; i < p->n_listeners; i++) { |
1045 | struct vconn *vconn; | |
1046 | int retval; | |
1047 | ||
1048 | retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn); | |
1049 | if (!retval) { | |
76ce9432 BP |
1050 | ofconn_create(p, rconn_new_from_vconn("passive", vconn), |
1051 | OFCONN_TRANSIENT); | |
064af421 BP |
1052 | } else if (retval != EAGAIN) { |
1053 | VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval)); | |
1054 | } | |
1055 | } | |
1056 | ||
1057 | for (i = 0; i < p->n_snoops; i++) { | |
1058 | struct vconn *vconn; | |
1059 | int retval; | |
1060 | ||
1061 | retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn); | |
1062 | if (!retval) { | |
76ce9432 | 1063 | add_snooper(p, vconn); |
064af421 BP |
1064 | } else if (retval != EAGAIN) { |
1065 | VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval)); | |
1066 | } | |
1067 | } | |
1068 | ||
1069 | if (time_msec() >= p->next_expiration) { | |
1070 | COVERAGE_INC(ofproto_expiration); | |
1071 | p->next_expiration = time_msec() + 1000; | |
1072 | update_used(p); | |
1073 | ||
1074 | classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p); | |
1075 | ||
1076 | /* Let the hook know that we're at a stable point: all outstanding data | |
1077 | * in existing flows has been accounted to the account_cb. Thus, the | |
1078 | * hook can now reasonably do operations that depend on having accurate | |
1079 | * flow volume accounting (currently, that's just bond rebalancing). */ | |
1080 | if (p->ofhooks->account_checkpoint_cb) { | |
1081 | p->ofhooks->account_checkpoint_cb(p->aux); | |
1082 | } | |
1083 | } | |
1084 | ||
1085 | if (p->netflow) { | |
1086 | netflow_run(p->netflow); | |
1087 | } | |
72b06300 BP |
1088 | if (p->sflow) { |
1089 | ofproto_sflow_run(p->sflow); | |
1090 | } | |
064af421 BP |
1091 | |
1092 | return 0; | |
1093 | } | |
1094 | ||
1095 | struct revalidate_cbdata { | |
1096 | struct ofproto *ofproto; | |
1097 | bool revalidate_all; /* Revalidate all exact-match rules? */ | |
1098 | bool revalidate_subrules; /* Revalidate all exact-match subrules? */ | |
1099 | struct tag_set revalidate_set; /* Set of tags to revalidate. */ | |
1100 | }; | |
1101 | ||
1102 | int | |
1103 | ofproto_run2(struct ofproto *p, bool revalidate_all) | |
1104 | { | |
1105 | if (p->need_revalidate || revalidate_all | |
1106 | || !tag_set_is_empty(&p->revalidate_set)) { | |
1107 | struct revalidate_cbdata cbdata; | |
1108 | cbdata.ofproto = p; | |
1109 | cbdata.revalidate_all = revalidate_all; | |
1110 | cbdata.revalidate_subrules = p->need_revalidate; | |
1111 | cbdata.revalidate_set = p->revalidate_set; | |
1112 | tag_set_init(&p->revalidate_set); | |
1113 | COVERAGE_INC(ofproto_revalidate); | |
1114 | classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata); | |
1115 | p->need_revalidate = false; | |
1116 | } | |
1117 | ||
1118 | return 0; | |
1119 | } | |
1120 | ||
1121 | void | |
1122 | ofproto_wait(struct ofproto *p) | |
1123 | { | |
1124 | struct ofconn *ofconn; | |
1125 | size_t i; | |
1126 | ||
c228a364 | 1127 | dpif_recv_wait(p->dpif); |
e9e28be3 BP |
1128 | dpif_port_poll_wait(p->dpif); |
1129 | netdev_monitor_poll_wait(p->netdev_monitor); | |
064af421 BP |
1130 | LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) { |
1131 | ofconn_wait(ofconn); | |
1132 | } | |
1133 | if (p->in_band) { | |
d2ede7bc | 1134 | poll_timer_wait(p->next_in_band_update - time_msec()); |
064af421 BP |
1135 | in_band_wait(p->in_band); |
1136 | } | |
064af421 BP |
1137 | if (p->fail_open) { |
1138 | fail_open_wait(p->fail_open); | |
1139 | } | |
72b06300 BP |
1140 | if (p->sflow) { |
1141 | ofproto_sflow_wait(p->sflow); | |
1142 | } | |
064af421 BP |
1143 | if (!tag_set_is_empty(&p->revalidate_set)) { |
1144 | poll_immediate_wake(); | |
1145 | } | |
1146 | if (p->need_revalidate) { | |
1147 | /* Shouldn't happen, but if it does just go around again. */ | |
1148 | VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); | |
1149 | poll_immediate_wake(); | |
1150 | } else if (p->next_expiration != LLONG_MAX) { | |
1151 | poll_timer_wait(p->next_expiration - time_msec()); | |
1152 | } | |
1153 | for (i = 0; i < p->n_listeners; i++) { | |
1154 | pvconn_wait(p->listeners[i]); | |
1155 | } | |
1156 | for (i = 0; i < p->n_snoops; i++) { | |
1157 | pvconn_wait(p->snoops[i]); | |
1158 | } | |
1159 | } | |
1160 | ||
1161 | void | |
1162 | ofproto_revalidate(struct ofproto *ofproto, tag_type tag) | |
1163 | { | |
1164 | tag_set_add(&ofproto->revalidate_set, tag); | |
1165 | } | |
1166 | ||
1167 | struct tag_set * | |
1168 | ofproto_get_revalidate_set(struct ofproto *ofproto) | |
1169 | { | |
1170 | return &ofproto->revalidate_set; | |
1171 | } | |
1172 | ||
1173 | bool | |
1174 | ofproto_is_alive(const struct ofproto *p) | |
1175 | { | |
76ce9432 | 1176 | return !hmap_is_empty(&p->controllers); |
064af421 BP |
1177 | } |
1178 | ||
1179 | int | |
1180 | ofproto_send_packet(struct ofproto *p, const flow_t *flow, | |
1181 | const union ofp_action *actions, size_t n_actions, | |
1182 | const struct ofpbuf *packet) | |
1183 | { | |
1184 | struct odp_actions odp_actions; | |
1185 | int error; | |
1186 | ||
1187 | error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions, | |
6a07af36 | 1188 | NULL, NULL, NULL); |
064af421 BP |
1189 | if (error) { |
1190 | return error; | |
1191 | } | |
1192 | ||
1193 | /* XXX Should we translate the dpif_execute() errno value into an OpenFlow | |
1194 | * error code? */ | |
c228a364 | 1195 | dpif_execute(p->dpif, flow->in_port, odp_actions.actions, |
064af421 BP |
1196 | odp_actions.n_actions, packet); |
1197 | return 0; | |
1198 | } | |
1199 | ||
1200 | void | |
1201 | ofproto_add_flow(struct ofproto *p, | |
1202 | const flow_t *flow, uint32_t wildcards, unsigned int priority, | |
1203 | const union ofp_action *actions, size_t n_actions, | |
1204 | int idle_timeout) | |
1205 | { | |
1206 | struct rule *rule; | |
0193b2af | 1207 | rule = rule_create(p, NULL, actions, n_actions, |
ca069229 | 1208 | idle_timeout >= 0 ? idle_timeout : 5 /* XXX */, |
39997502 | 1209 | 0, 0, false); |
659586ef | 1210 | cls_rule_from_flow(flow, wildcards, priority, &rule->cr); |
064af421 BP |
1211 | rule_insert(p, rule, NULL, 0); |
1212 | } | |
1213 | ||
1214 | void | |
1215 | ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow, | |
1216 | uint32_t wildcards, unsigned int priority) | |
1217 | { | |
1218 | struct rule *rule; | |
1219 | ||
1220 | rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls, | |
1221 | flow, wildcards, | |
1222 | priority)); | |
1223 | if (rule) { | |
1224 | rule_remove(ofproto, rule); | |
1225 | } | |
1226 | } | |
1227 | ||
1228 | static void | |
1229 | destroy_rule(struct cls_rule *rule_, void *ofproto_) | |
1230 | { | |
1231 | struct rule *rule = rule_from_cls_rule(rule_); | |
1232 | struct ofproto *ofproto = ofproto_; | |
1233 | ||
1234 | /* Mark the flow as not installed, even though it might really be | |
1235 | * installed, so that rule_remove() doesn't bother trying to uninstall it. | |
1236 | * There is no point in uninstalling it individually since we are about to | |
1237 | * blow away all the flows with dpif_flow_flush(). */ | |
1238 | rule->installed = false; | |
1239 | ||
1240 | rule_remove(ofproto, rule); | |
1241 | } | |
1242 | ||
1243 | void | |
1244 | ofproto_flush_flows(struct ofproto *ofproto) | |
1245 | { | |
1246 | COVERAGE_INC(ofproto_flush); | |
1247 | classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto); | |
c228a364 | 1248 | dpif_flow_flush(ofproto->dpif); |
064af421 BP |
1249 | if (ofproto->in_band) { |
1250 | in_band_flushed(ofproto->in_band); | |
1251 | } | |
1252 | if (ofproto->fail_open) { | |
1253 | fail_open_flushed(ofproto->fail_open); | |
1254 | } | |
1255 | } | |
1256 | \f | |
1257 | static void | |
1258 | reinit_ports(struct ofproto *p) | |
1259 | { | |
1260 | struct svec devnames; | |
1261 | struct ofport *ofport; | |
1262 | unsigned int port_no; | |
1263 | struct odp_port *odp_ports; | |
1264 | size_t n_odp_ports; | |
1265 | size_t i; | |
1266 | ||
1267 | svec_init(&devnames); | |
1268 | PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) { | |
1269 | svec_add (&devnames, (char *) ofport->opp.name); | |
1270 | } | |
c228a364 | 1271 | dpif_port_list(p->dpif, &odp_ports, &n_odp_ports); |
064af421 BP |
1272 | for (i = 0; i < n_odp_ports; i++) { |
1273 | svec_add (&devnames, odp_ports[i].devname); | |
1274 | } | |
1275 | free(odp_ports); | |
1276 | ||
1277 | svec_sort_unique(&devnames); | |
1278 | for (i = 0; i < devnames.n; i++) { | |
1279 | update_port(p, devnames.names[i]); | |
1280 | } | |
1281 | svec_destroy(&devnames); | |
1282 | } | |
1283 | ||
72b06300 | 1284 | static size_t |
064af421 BP |
1285 | refresh_port_group(struct ofproto *p, unsigned int group) |
1286 | { | |
1287 | uint16_t *ports; | |
1288 | size_t n_ports; | |
1289 | struct ofport *port; | |
1290 | unsigned int port_no; | |
1291 | ||
1292 | assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD); | |
1293 | ||
1294 | ports = xmalloc(port_array_count(&p->ports) * sizeof *ports); | |
1295 | n_ports = 0; | |
1296 | PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) { | |
1297 | if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) { | |
1298 | ports[n_ports++] = port_no; | |
1299 | } | |
1300 | } | |
c228a364 | 1301 | dpif_port_group_set(p->dpif, group, ports, n_ports); |
064af421 | 1302 | free(ports); |
72b06300 BP |
1303 | |
1304 | return n_ports; | |
064af421 BP |
1305 | } |
1306 | ||
1307 | static void | |
1308 | refresh_port_groups(struct ofproto *p) | |
1309 | { | |
72b06300 BP |
1310 | size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD); |
1311 | size_t n_all = refresh_port_group(p, DP_GROUP_ALL); | |
1312 | if (p->sflow) { | |
1313 | ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all); | |
1314 | } | |
064af421 BP |
1315 | } |
1316 | ||
1317 | static struct ofport * | |
1318 | make_ofport(const struct odp_port *odp_port) | |
1319 | { | |
149f577a | 1320 | struct netdev_options netdev_options; |
064af421 BP |
1321 | enum netdev_flags flags; |
1322 | struct ofport *ofport; | |
1323 | struct netdev *netdev; | |
1324 | bool carrier; | |
1325 | int error; | |
1326 | ||
149f577a JG |
1327 | memset(&netdev_options, 0, sizeof netdev_options); |
1328 | netdev_options.name = odp_port->devname; | |
1329 | netdev_options.ethertype = NETDEV_ETH_TYPE_NONE; | |
1330 | netdev_options.may_open = true; | |
1331 | ||
1332 | error = netdev_open(&netdev_options, &netdev); | |
064af421 BP |
1333 | if (error) { |
1334 | VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s " | |
1335 | "cannot be opened (%s)", | |
1336 | odp_port->devname, odp_port->port, | |
1337 | odp_port->devname, strerror(error)); | |
1338 | return NULL; | |
1339 | } | |
1340 | ||
1341 | ofport = xmalloc(sizeof *ofport); | |
1342 | ofport->netdev = netdev; | |
1343 | ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port); | |
80992a35 | 1344 | netdev_get_etheraddr(netdev, ofport->opp.hw_addr); |
064af421 BP |
1345 | memcpy(ofport->opp.name, odp_port->devname, |
1346 | MIN(sizeof ofport->opp.name, sizeof odp_port->devname)); | |
1347 | ofport->opp.name[sizeof ofport->opp.name - 1] = '\0'; | |
1348 | ||
1349 | netdev_get_flags(netdev, &flags); | |
1350 | ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN; | |
1351 | ||
1352 | netdev_get_carrier(netdev, &carrier); | |
1353 | ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN; | |
1354 | ||
1355 | netdev_get_features(netdev, | |
1356 | &ofport->opp.curr, &ofport->opp.advertised, | |
1357 | &ofport->opp.supported, &ofport->opp.peer); | |
1358 | return ofport; | |
1359 | } | |
1360 | ||
1361 | static bool | |
1362 | ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port) | |
1363 | { | |
1364 | if (port_array_get(&p->ports, odp_port->port)) { | |
1365 | VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath", | |
1366 | odp_port->port); | |
1367 | return true; | |
1368 | } else if (shash_find(&p->port_by_name, odp_port->devname)) { | |
1369 | VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath", | |
1370 | odp_port->devname); | |
1371 | return true; | |
1372 | } else { | |
1373 | return false; | |
1374 | } | |
1375 | } | |
1376 | ||
1377 | static int | |
1378 | ofport_equal(const struct ofport *a_, const struct ofport *b_) | |
1379 | { | |
1380 | const struct ofp_phy_port *a = &a_->opp; | |
1381 | const struct ofp_phy_port *b = &b_->opp; | |
1382 | ||
1383 | BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */ | |
1384 | return (a->port_no == b->port_no | |
1385 | && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr) | |
1386 | && !strcmp((char *) a->name, (char *) b->name) | |
1387 | && a->state == b->state | |
1388 | && a->config == b->config | |
1389 | && a->curr == b->curr | |
1390 | && a->advertised == b->advertised | |
1391 | && a->supported == b->supported | |
1392 | && a->peer == b->peer); | |
1393 | } | |
1394 | ||
1395 | static void | |
1396 | send_port_status(struct ofproto *p, const struct ofport *ofport, | |
1397 | uint8_t reason) | |
1398 | { | |
1399 | /* XXX Should limit the number of queued port status change messages. */ | |
1400 | struct ofconn *ofconn; | |
1401 | LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) { | |
1402 | struct ofp_port_status *ops; | |
1403 | struct ofpbuf *b; | |
1404 | ||
9deba63b BP |
1405 | if (ofconn->role == NX_ROLE_SLAVE) { |
1406 | continue; | |
1407 | } | |
1408 | ||
064af421 BP |
1409 | ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b); |
1410 | ops->reason = reason; | |
1411 | ops->desc = ofport->opp; | |
1412 | hton_ofp_phy_port(&ops->desc); | |
1413 | queue_tx(b, ofconn, NULL); | |
1414 | } | |
1415 | if (p->ofhooks->port_changed_cb) { | |
1416 | p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux); | |
1417 | } | |
1418 | } | |
1419 | ||
1420 | static void | |
1421 | ofport_install(struct ofproto *p, struct ofport *ofport) | |
1422 | { | |
72b06300 BP |
1423 | uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no); |
1424 | const char *netdev_name = (const char *) ofport->opp.name; | |
1425 | ||
e9e28be3 | 1426 | netdev_monitor_add(p->netdev_monitor, ofport->netdev); |
72b06300 BP |
1427 | port_array_set(&p->ports, odp_port, ofport); |
1428 | shash_add(&p->port_by_name, netdev_name, ofport); | |
1429 | if (p->sflow) { | |
1430 | ofproto_sflow_add_port(p->sflow, odp_port, netdev_name); | |
1431 | } | |
064af421 BP |
1432 | } |
1433 | ||
1434 | static void | |
1435 | ofport_remove(struct ofproto *p, struct ofport *ofport) | |
1436 | { | |
72b06300 BP |
1437 | uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no); |
1438 | ||
e9e28be3 | 1439 | netdev_monitor_remove(p->netdev_monitor, ofport->netdev); |
72b06300 | 1440 | port_array_set(&p->ports, odp_port, NULL); |
064af421 BP |
1441 | shash_delete(&p->port_by_name, |
1442 | shash_find(&p->port_by_name, (char *) ofport->opp.name)); | |
72b06300 BP |
1443 | if (p->sflow) { |
1444 | ofproto_sflow_del_port(p->sflow, odp_port); | |
1445 | } | |
064af421 BP |
1446 | } |
1447 | ||
1448 | static void | |
1449 | ofport_free(struct ofport *ofport) | |
1450 | { | |
1451 | if (ofport) { | |
1452 | netdev_close(ofport->netdev); | |
1453 | free(ofport); | |
1454 | } | |
1455 | } | |
1456 | ||
1457 | static void | |
1458 | update_port(struct ofproto *p, const char *devname) | |
1459 | { | |
1460 | struct odp_port odp_port; | |
c874dc6d BP |
1461 | struct ofport *old_ofport; |
1462 | struct ofport *new_ofport; | |
064af421 BP |
1463 | int error; |
1464 | ||
1465 | COVERAGE_INC(ofproto_update_port); | |
c874dc6d BP |
1466 | |
1467 | /* Query the datapath for port information. */ | |
c228a364 | 1468 | error = dpif_port_query_by_name(p->dpif, devname, &odp_port); |
064af421 | 1469 | |
c874dc6d BP |
1470 | /* Find the old ofport. */ |
1471 | old_ofport = shash_find_data(&p->port_by_name, devname); | |
1472 | if (!error) { | |
1473 | if (!old_ofport) { | |
1474 | /* There's no port named 'devname' but there might be a port with | |
1475 | * the same port number. This could happen if a port is deleted | |
1476 | * and then a new one added in its place very quickly, or if a port | |
1477 | * is renamed. In the former case we want to send an OFPPR_DELETE | |
1478 | * and an OFPPR_ADD, and in the latter case we want to send a | |
1479 | * single OFPPR_MODIFY. We can distinguish the cases by comparing | |
1480 | * the old port's ifindex against the new port, or perhaps less | |
1481 | * reliably but more portably by comparing the old port's MAC | |
1482 | * against the new port's MAC. However, this code isn't that smart | |
1483 | * and always sends an OFPPR_MODIFY (XXX). */ | |
1484 | old_ofport = port_array_get(&p->ports, odp_port.port); | |
064af421 | 1485 | } |
c874dc6d | 1486 | } else if (error != ENOENT && error != ENODEV) { |
064af421 BP |
1487 | VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error " |
1488 | "%s", strerror(error)); | |
1489 | return; | |
1490 | } | |
c874dc6d BP |
1491 | |
1492 | /* Create a new ofport. */ | |
1493 | new_ofport = !error ? make_ofport(&odp_port) : NULL; | |
1494 | ||
1495 | /* Eliminate a few pathological cases. */ | |
1496 | if (!old_ofport && !new_ofport) { | |
1497 | return; | |
1498 | } else if (old_ofport && new_ofport) { | |
1499 | /* Most of the 'config' bits are OpenFlow soft state, but | |
1500 | * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow | |
1501 | * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and | |
1502 | * leaves the other bits 0.) */ | |
1503 | new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN; | |
1504 | ||
1505 | if (ofport_equal(old_ofport, new_ofport)) { | |
1506 | /* False alarm--no change. */ | |
1507 | ofport_free(new_ofport); | |
1508 | return; | |
1509 | } | |
1510 | } | |
1511 | ||
1512 | /* Now deal with the normal cases. */ | |
1513 | if (old_ofport) { | |
1514 | ofport_remove(p, old_ofport); | |
1515 | } | |
1516 | if (new_ofport) { | |
1517 | ofport_install(p, new_ofport); | |
1518 | } | |
1519 | send_port_status(p, new_ofport ? new_ofport : old_ofport, | |
1520 | (!old_ofport ? OFPPR_ADD | |
1521 | : !new_ofport ? OFPPR_DELETE | |
1522 | : OFPPR_MODIFY)); | |
1523 | ofport_free(old_ofport); | |
1524 | ||
1525 | /* Update port groups. */ | |
064af421 BP |
1526 | refresh_port_groups(p); |
1527 | } | |
1528 | ||
1529 | static int | |
1530 | init_ports(struct ofproto *p) | |
1531 | { | |
1532 | struct odp_port *ports; | |
1533 | size_t n_ports; | |
1534 | size_t i; | |
1535 | int error; | |
1536 | ||
c228a364 | 1537 | error = dpif_port_list(p->dpif, &ports, &n_ports); |
064af421 BP |
1538 | if (error) { |
1539 | return error; | |
1540 | } | |
1541 | ||
1542 | for (i = 0; i < n_ports; i++) { | |
1543 | const struct odp_port *odp_port = &ports[i]; | |
1544 | if (!ofport_conflicts(p, odp_port)) { | |
1545 | struct ofport *ofport = make_ofport(odp_port); | |
1546 | if (ofport) { | |
1547 | ofport_install(p, ofport); | |
1548 | } | |
1549 | } | |
1550 | } | |
1551 | free(ports); | |
1552 | refresh_port_groups(p); | |
1553 | return 0; | |
1554 | } | |
1555 | \f | |
1556 | static struct ofconn * | |
76ce9432 | 1557 | ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type) |
064af421 | 1558 | { |
76ce9432 BP |
1559 | struct ofconn *ofconn = xzalloc(sizeof *ofconn); |
1560 | ofconn->ofproto = p; | |
064af421 BP |
1561 | list_push_back(&p->all_conns, &ofconn->node); |
1562 | ofconn->rconn = rconn; | |
76ce9432 | 1563 | ofconn->type = type; |
9deba63b | 1564 | ofconn->role = NX_ROLE_OTHER; |
76ce9432 | 1565 | ofconn->packet_in_counter = rconn_packet_counter_create (); |
064af421 | 1566 | ofconn->pktbuf = NULL; |
064af421 | 1567 | ofconn->miss_send_len = 0; |
064af421 BP |
1568 | ofconn->reply_counter = rconn_packet_counter_create (); |
1569 | return ofconn; | |
1570 | } | |
1571 | ||
1572 | static void | |
c475ae67 | 1573 | ofconn_destroy(struct ofconn *ofconn) |
064af421 | 1574 | { |
76ce9432 BP |
1575 | if (ofconn->type == OFCONN_CONTROLLER) { |
1576 | hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node); | |
1577 | } | |
1578 | discovery_destroy(ofconn->discovery); | |
1579 | ||
064af421 | 1580 | list_remove(&ofconn->node); |
76ce9432 | 1581 | switch_status_unregister(ofconn->ss); |
064af421 BP |
1582 | rconn_destroy(ofconn->rconn); |
1583 | rconn_packet_counter_destroy(ofconn->packet_in_counter); | |
1584 | rconn_packet_counter_destroy(ofconn->reply_counter); | |
1585 | pktbuf_destroy(ofconn->pktbuf); | |
1586 | free(ofconn); | |
1587 | } | |
1588 | ||
1589 | static void | |
1590 | ofconn_run(struct ofconn *ofconn, struct ofproto *p) | |
1591 | { | |
1592 | int iteration; | |
76ce9432 BP |
1593 | size_t i; |
1594 | ||
1595 | if (ofconn->discovery) { | |
1596 | char *controller_name; | |
1597 | if (rconn_is_connectivity_questionable(ofconn->rconn)) { | |
1598 | discovery_question_connectivity(ofconn->discovery); | |
1599 | } | |
1600 | if (discovery_run(ofconn->discovery, &controller_name)) { | |
1601 | if (controller_name) { | |
1602 | rconn_connect(ofconn->rconn, controller_name); | |
1603 | } else { | |
1604 | rconn_disconnect(ofconn->rconn); | |
1605 | } | |
1606 | } | |
1607 | } | |
1608 | ||
1609 | for (i = 0; i < N_SCHEDULERS; i++) { | |
1610 | pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn); | |
1611 | } | |
064af421 BP |
1612 | |
1613 | rconn_run(ofconn->rconn); | |
1614 | ||
1615 | if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) { | |
1616 | /* Limit the number of iterations to prevent other tasks from | |
1617 | * starving. */ | |
1618 | for (iteration = 0; iteration < 50; iteration++) { | |
1619 | struct ofpbuf *of_msg = rconn_recv(ofconn->rconn); | |
1620 | if (!of_msg) { | |
1621 | break; | |
1622 | } | |
7778bd15 BP |
1623 | if (p->fail_open) { |
1624 | fail_open_maybe_recover(p->fail_open); | |
1625 | } | |
064af421 BP |
1626 | handle_openflow(ofconn, p, of_msg); |
1627 | ofpbuf_delete(of_msg); | |
1628 | } | |
1629 | } | |
1630 | ||
76ce9432 | 1631 | if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) { |
c475ae67 | 1632 | ofconn_destroy(ofconn); |
064af421 BP |
1633 | } |
1634 | } | |
1635 | ||
1636 | static void | |
1637 | ofconn_wait(struct ofconn *ofconn) | |
1638 | { | |
76ce9432 BP |
1639 | int i; |
1640 | ||
1641 | if (ofconn->discovery) { | |
1642 | discovery_wait(ofconn->discovery); | |
1643 | } | |
1644 | for (i = 0; i < N_SCHEDULERS; i++) { | |
1645 | pinsched_wait(ofconn->schedulers[i]); | |
1646 | } | |
064af421 BP |
1647 | rconn_run_wait(ofconn->rconn); |
1648 | if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) { | |
1649 | rconn_recv_wait(ofconn->rconn); | |
1650 | } else { | |
1651 | COVERAGE_INC(ofproto_ofconn_stuck); | |
1652 | } | |
1653 | } | |
1654 | \f | |
1655 | /* Caller is responsible for initializing the 'cr' member of the returned | |
1656 | * rule. */ | |
1657 | static struct rule * | |
0193b2af | 1658 | rule_create(struct ofproto *ofproto, struct rule *super, |
064af421 | 1659 | const union ofp_action *actions, size_t n_actions, |
ca069229 | 1660 | uint16_t idle_timeout, uint16_t hard_timeout, |
39997502 | 1661 | uint64_t flow_cookie, bool send_flow_removed) |
064af421 | 1662 | { |
ec6fde61 | 1663 | struct rule *rule = xzalloc(sizeof *rule); |
064af421 BP |
1664 | rule->idle_timeout = idle_timeout; |
1665 | rule->hard_timeout = hard_timeout; | |
39997502 | 1666 | rule->flow_cookie = flow_cookie; |
064af421 | 1667 | rule->used = rule->created = time_msec(); |
ca069229 | 1668 | rule->send_flow_removed = send_flow_removed; |
064af421 BP |
1669 | rule->super = super; |
1670 | if (super) { | |
1671 | list_push_back(&super->list, &rule->list); | |
1672 | } else { | |
1673 | list_init(&rule->list); | |
1674 | } | |
1675 | rule->n_actions = n_actions; | |
1676 | rule->actions = xmemdup(actions, n_actions * sizeof *actions); | |
0193b2af JG |
1677 | netflow_flow_clear(&rule->nf_flow); |
1678 | netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created); | |
1679 | ||
064af421 BP |
1680 | return rule; |
1681 | } | |
1682 | ||
1683 | static struct rule * | |
1684 | rule_from_cls_rule(const struct cls_rule *cls_rule) | |
1685 | { | |
1686 | return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL; | |
1687 | } | |
1688 | ||
1689 | static void | |
1690 | rule_free(struct rule *rule) | |
1691 | { | |
1692 | free(rule->actions); | |
1693 | free(rule->odp_actions); | |
1694 | free(rule); | |
1695 | } | |
1696 | ||
1697 | /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its | |
1698 | * super-rule's list of subrules. If 'rule' is a super-rule, also iterates | |
1699 | * through all of its subrules and revalidates them, destroying any that no | |
1700 | * longer has a super-rule (which is probably all of them). | |
1701 | * | |
1702 | * Before calling this function, the caller must make have removed 'rule' from | |
1703 | * the classifier. If 'rule' is an exact-match rule, the caller is also | |
1704 | * responsible for ensuring that it has been uninstalled from the datapath. */ | |
1705 | static void | |
1706 | rule_destroy(struct ofproto *ofproto, struct rule *rule) | |
1707 | { | |
1708 | if (!rule->super) { | |
1709 | struct rule *subrule, *next; | |
1710 | LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) { | |
1711 | revalidate_rule(ofproto, subrule); | |
1712 | } | |
1713 | } else { | |
1714 | list_remove(&rule->list); | |
1715 | } | |
1716 | rule_free(rule); | |
1717 | } | |
1718 | ||
1719 | static bool | |
1720 | rule_has_out_port(const struct rule *rule, uint16_t out_port) | |
1721 | { | |
1722 | const union ofp_action *oa; | |
1723 | struct actions_iterator i; | |
1724 | ||
1725 | if (out_port == htons(OFPP_NONE)) { | |
1726 | return true; | |
1727 | } | |
1728 | for (oa = actions_first(&i, rule->actions, rule->n_actions); oa; | |
1729 | oa = actions_next(&i)) { | |
1730 | if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) { | |
1731 | return true; | |
1732 | } | |
1733 | } | |
1734 | return false; | |
1735 | } | |
1736 | ||
1737 | /* Executes the actions indicated by 'rule' on 'packet', which is in flow | |
1738 | * 'flow' and is considered to have arrived on ODP port 'in_port'. | |
1739 | * | |
1740 | * The flow that 'packet' actually contains does not need to actually match | |
1741 | * 'rule'; the actions in 'rule' will be applied to it either way. Likewise, | |
1742 | * the packet and byte counters for 'rule' will be credited for the packet sent | |
1743 | * out whether or not the packet actually matches 'rule'. | |
1744 | * | |
1745 | * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow, | |
1746 | * the caller must already have accurately composed ODP actions for it given | |
1747 | * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if | |
1748 | * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this | |
1749 | * function will compose a set of ODP actions based on 'rule''s OpenFlow | |
1750 | * actions and apply them to 'packet'. */ | |
1751 | static void | |
1752 | rule_execute(struct ofproto *ofproto, struct rule *rule, | |
1753 | struct ofpbuf *packet, const flow_t *flow) | |
1754 | { | |
1755 | const union odp_action *actions; | |
1756 | size_t n_actions; | |
1757 | struct odp_actions a; | |
1758 | ||
1759 | /* Grab or compose the ODP actions. | |
1760 | * | |
1761 | * The special case for an exact-match 'rule' where 'flow' is not the | |
1762 | * rule's flow is important to avoid, e.g., sending a packet out its input | |
1763 | * port simply because the ODP actions were composed for the wrong | |
1764 | * scenario. */ | |
1765 | if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) { | |
1766 | struct rule *super = rule->super ? rule->super : rule; | |
1767 | if (xlate_actions(super->actions, super->n_actions, flow, ofproto, | |
6a07af36 | 1768 | packet, &a, NULL, 0, NULL)) { |
064af421 BP |
1769 | return; |
1770 | } | |
1771 | actions = a.actions; | |
1772 | n_actions = a.n_actions; | |
1773 | } else { | |
1774 | actions = rule->odp_actions; | |
1775 | n_actions = rule->n_odp_actions; | |
1776 | } | |
1777 | ||
1778 | /* Execute the ODP actions. */ | |
c228a364 | 1779 | if (!dpif_execute(ofproto->dpif, flow->in_port, |
064af421 BP |
1780 | actions, n_actions, packet)) { |
1781 | struct odp_flow_stats stats; | |
1782 | flow_extract_stats(flow, packet, &stats); | |
0193b2af | 1783 | update_stats(ofproto, rule, &stats); |
064af421 | 1784 | rule->used = time_msec(); |
0193b2af | 1785 | netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used); |
064af421 BP |
1786 | } |
1787 | } | |
1788 | ||
1789 | static void | |
1790 | rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet, | |
1791 | uint16_t in_port) | |
1792 | { | |
1793 | struct rule *displaced_rule; | |
1794 | ||
1795 | /* Insert the rule in the classifier. */ | |
1796 | displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr)); | |
1797 | if (!rule->cr.wc.wildcards) { | |
1798 | rule_make_actions(p, rule, packet); | |
1799 | } | |
1800 | ||
1801 | /* Send the packet and credit it to the rule. */ | |
1802 | if (packet) { | |
1803 | flow_t flow; | |
659586ef | 1804 | flow_extract(packet, 0, in_port, &flow); |
064af421 BP |
1805 | rule_execute(p, rule, packet, &flow); |
1806 | } | |
1807 | ||
1808 | /* Install the rule in the datapath only after sending the packet, to | |
1809 | * avoid packet reordering. */ | |
1810 | if (rule->cr.wc.wildcards) { | |
1811 | COVERAGE_INC(ofproto_add_wc_flow); | |
1812 | p->need_revalidate = true; | |
1813 | } else { | |
1814 | rule_install(p, rule, displaced_rule); | |
1815 | } | |
1816 | ||
1817 | /* Free the rule that was displaced, if any. */ | |
1818 | if (displaced_rule) { | |
1819 | rule_destroy(p, displaced_rule); | |
1820 | } | |
1821 | } | |
1822 | ||
1823 | static struct rule * | |
1824 | rule_create_subrule(struct ofproto *ofproto, struct rule *rule, | |
1825 | const flow_t *flow) | |
1826 | { | |
0193b2af | 1827 | struct rule *subrule = rule_create(ofproto, rule, NULL, 0, |
ca069229 | 1828 | rule->idle_timeout, rule->hard_timeout, |
39997502 | 1829 | 0, false); |
064af421 | 1830 | COVERAGE_INC(ofproto_subrule_create); |
659586ef JG |
1831 | cls_rule_from_flow(flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX |
1832 | : rule->cr.priority), &subrule->cr); | |
064af421 BP |
1833 | classifier_insert_exact(&ofproto->cls, &subrule->cr); |
1834 | ||
1835 | return subrule; | |
1836 | } | |
1837 | ||
1838 | static void | |
1839 | rule_remove(struct ofproto *ofproto, struct rule *rule) | |
1840 | { | |
1841 | if (rule->cr.wc.wildcards) { | |
1842 | COVERAGE_INC(ofproto_del_wc_flow); | |
1843 | ofproto->need_revalidate = true; | |
1844 | } else { | |
1845 | rule_uninstall(ofproto, rule); | |
1846 | } | |
1847 | classifier_remove(&ofproto->cls, &rule->cr); | |
1848 | rule_destroy(ofproto, rule); | |
1849 | } | |
1850 | ||
1851 | /* Returns true if the actions changed, false otherwise. */ | |
1852 | static bool | |
1853 | rule_make_actions(struct ofproto *p, struct rule *rule, | |
1854 | const struct ofpbuf *packet) | |
1855 | { | |
1856 | const struct rule *super; | |
1857 | struct odp_actions a; | |
1858 | size_t actions_len; | |
1859 | ||
1860 | assert(!rule->cr.wc.wildcards); | |
1861 | ||
1862 | super = rule->super ? rule->super : rule; | |
1863 | rule->tags = 0; | |
1864 | xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p, | |
6a07af36 | 1865 | packet, &a, &rule->tags, &rule->may_install, |
0193b2af | 1866 | &rule->nf_flow.output_iface); |
064af421 BP |
1867 | |
1868 | actions_len = a.n_actions * sizeof *a.actions; | |
1869 | if (rule->n_odp_actions != a.n_actions | |
1870 | || memcmp(rule->odp_actions, a.actions, actions_len)) { | |
1871 | COVERAGE_INC(ofproto_odp_unchanged); | |
1872 | free(rule->odp_actions); | |
1873 | rule->n_odp_actions = a.n_actions; | |
1874 | rule->odp_actions = xmemdup(a.actions, actions_len); | |
1875 | return true; | |
1876 | } else { | |
1877 | return false; | |
1878 | } | |
1879 | } | |
1880 | ||
1881 | static int | |
1882 | do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags, | |
1883 | struct odp_flow_put *put) | |
1884 | { | |
1885 | memset(&put->flow.stats, 0, sizeof put->flow.stats); | |
1886 | put->flow.key = rule->cr.flow; | |
1887 | put->flow.actions = rule->odp_actions; | |
1888 | put->flow.n_actions = rule->n_odp_actions; | |
ab48643b | 1889 | put->flow.flags = 0; |
064af421 | 1890 | put->flags = flags; |
c228a364 | 1891 | return dpif_flow_put(ofproto->dpif, put); |
064af421 BP |
1892 | } |
1893 | ||
1894 | static void | |
1895 | rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule) | |
1896 | { | |
1897 | assert(!rule->cr.wc.wildcards); | |
1898 | ||
1899 | if (rule->may_install) { | |
1900 | struct odp_flow_put put; | |
1901 | if (!do_put_flow(p, rule, | |
1902 | ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS, | |
1903 | &put)) { | |
1904 | rule->installed = true; | |
1905 | if (displaced_rule) { | |
14986b31 | 1906 | update_stats(p, displaced_rule, &put.flow.stats); |
064af421 BP |
1907 | rule_post_uninstall(p, displaced_rule); |
1908 | } | |
1909 | } | |
1910 | } else if (displaced_rule) { | |
1911 | rule_uninstall(p, displaced_rule); | |
1912 | } | |
1913 | } | |
1914 | ||
1915 | static void | |
1916 | rule_reinstall(struct ofproto *ofproto, struct rule *rule) | |
1917 | { | |
1918 | if (rule->installed) { | |
1919 | struct odp_flow_put put; | |
1920 | COVERAGE_INC(ofproto_dp_missed); | |
1921 | do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put); | |
1922 | } else { | |
1923 | rule_install(ofproto, rule, NULL); | |
1924 | } | |
1925 | } | |
1926 | ||
1927 | static void | |
1928 | rule_update_actions(struct ofproto *ofproto, struct rule *rule) | |
1929 | { | |
42c3641c JG |
1930 | bool actions_changed; |
1931 | uint16_t new_out_iface, old_out_iface; | |
1932 | ||
1933 | old_out_iface = rule->nf_flow.output_iface; | |
1934 | actions_changed = rule_make_actions(ofproto, rule, NULL); | |
1935 | ||
064af421 BP |
1936 | if (rule->may_install) { |
1937 | if (rule->installed) { | |
1938 | if (actions_changed) { | |
064af421 | 1939 | struct odp_flow_put put; |
42c3641c JG |
1940 | do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY |
1941 | | ODPPF_ZERO_STATS, &put); | |
1942 | update_stats(ofproto, rule, &put.flow.stats); | |
1943 | ||
1944 | /* Temporarily set the old output iface so that NetFlow | |
1945 | * messages have the correct output interface for the old | |
1946 | * stats. */ | |
1947 | new_out_iface = rule->nf_flow.output_iface; | |
1948 | rule->nf_flow.output_iface = old_out_iface; | |
1949 | rule_post_uninstall(ofproto, rule); | |
1950 | rule->nf_flow.output_iface = new_out_iface; | |
064af421 BP |
1951 | } |
1952 | } else { | |
1953 | rule_install(ofproto, rule, NULL); | |
1954 | } | |
1955 | } else { | |
1956 | rule_uninstall(ofproto, rule); | |
1957 | } | |
1958 | } | |
1959 | ||
1960 | static void | |
1961 | rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes) | |
1962 | { | |
1963 | uint64_t total_bytes = rule->byte_count + extra_bytes; | |
1964 | ||
1965 | if (ofproto->ofhooks->account_flow_cb | |
1966 | && total_bytes > rule->accounted_bytes) | |
1967 | { | |
1968 | ofproto->ofhooks->account_flow_cb( | |
1969 | &rule->cr.flow, rule->odp_actions, rule->n_odp_actions, | |
1970 | total_bytes - rule->accounted_bytes, ofproto->aux); | |
1971 | rule->accounted_bytes = total_bytes; | |
1972 | } | |
1973 | } | |
1974 | ||
1975 | static void | |
1976 | rule_uninstall(struct ofproto *p, struct rule *rule) | |
1977 | { | |
1978 | assert(!rule->cr.wc.wildcards); | |
1979 | if (rule->installed) { | |
1980 | struct odp_flow odp_flow; | |
1981 | ||
1982 | odp_flow.key = rule->cr.flow; | |
1983 | odp_flow.actions = NULL; | |
1984 | odp_flow.n_actions = 0; | |
ab48643b | 1985 | odp_flow.flags = 0; |
c228a364 | 1986 | if (!dpif_flow_del(p->dpif, &odp_flow)) { |
0193b2af | 1987 | update_stats(p, rule, &odp_flow.stats); |
064af421 BP |
1988 | } |
1989 | rule->installed = false; | |
1990 | ||
1991 | rule_post_uninstall(p, rule); | |
1992 | } | |
1993 | } | |
1994 | ||
0193b2af JG |
1995 | static bool |
1996 | is_controller_rule(struct rule *rule) | |
1997 | { | |
1998 | /* If the only action is send to the controller then don't report | |
1999 | * NetFlow expiration messages since it is just part of the control | |
2000 | * logic for the network and not real traffic. */ | |
2001 | ||
2002 | if (rule && rule->super) { | |
2003 | struct rule *super = rule->super; | |
2004 | ||
2005 | return super->n_actions == 1 && | |
2006 | super->actions[0].type == htons(OFPAT_OUTPUT) && | |
2007 | super->actions[0].output.port == htons(OFPP_CONTROLLER); | |
2008 | } | |
2009 | ||
2010 | return false; | |
2011 | } | |
2012 | ||
064af421 BP |
2013 | static void |
2014 | rule_post_uninstall(struct ofproto *ofproto, struct rule *rule) | |
2015 | { | |
2016 | struct rule *super = rule->super; | |
2017 | ||
2018 | rule_account(ofproto, rule, 0); | |
6a07af36 | 2019 | |
0193b2af | 2020 | if (ofproto->netflow && !is_controller_rule(rule)) { |
064af421 BP |
2021 | struct ofexpired expired; |
2022 | expired.flow = rule->cr.flow; | |
2023 | expired.packet_count = rule->packet_count; | |
2024 | expired.byte_count = rule->byte_count; | |
2025 | expired.used = rule->used; | |
0193b2af | 2026 | netflow_expire(ofproto->netflow, &rule->nf_flow, &expired); |
064af421 BP |
2027 | } |
2028 | if (super) { | |
2029 | super->packet_count += rule->packet_count; | |
2030 | super->byte_count += rule->byte_count; | |
064af421 | 2031 | |
0c0afbec JG |
2032 | /* Reset counters to prevent double counting if the rule ever gets |
2033 | * reinstalled. */ | |
2034 | rule->packet_count = 0; | |
2035 | rule->byte_count = 0; | |
2036 | rule->accounted_bytes = 0; | |
0193b2af JG |
2037 | |
2038 | netflow_flow_clear(&rule->nf_flow); | |
0c0afbec | 2039 | } |
064af421 BP |
2040 | } |
2041 | \f | |
2042 | static void | |
2043 | queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn, | |
2044 | struct rconn_packet_counter *counter) | |
2045 | { | |
2046 | update_openflow_length(msg); | |
2047 | if (rconn_send(ofconn->rconn, msg, counter)) { | |
2048 | ofpbuf_delete(msg); | |
2049 | } | |
2050 | } | |
2051 | ||
2052 | static void | |
2053 | send_error(const struct ofconn *ofconn, const struct ofp_header *oh, | |
2054 | int error, const void *data, size_t len) | |
2055 | { | |
2056 | struct ofpbuf *buf; | |
2057 | struct ofp_error_msg *oem; | |
2058 | ||
2059 | if (!(error >> 16)) { | |
2060 | VLOG_WARN_RL(&rl, "not sending bad error code %d to controller", | |
2061 | error); | |
2062 | return; | |
2063 | } | |
2064 | ||
2065 | COVERAGE_INC(ofproto_error); | |
2066 | oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR, | |
2067 | oh ? oh->xid : 0, &buf); | |
2068 | oem->type = htons((unsigned int) error >> 16); | |
2069 | oem->code = htons(error & 0xffff); | |
2070 | memcpy(oem->data, data, len); | |
2071 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
2072 | } | |
2073 | ||
2074 | static void | |
2075 | send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh, | |
2076 | int error) | |
2077 | { | |
2078 | size_t oh_length = ntohs(oh->length); | |
2079 | send_error(ofconn, oh, error, oh, MIN(oh_length, 64)); | |
2080 | } | |
2081 | ||
2082 | static void | |
2083 | hton_ofp_phy_port(struct ofp_phy_port *opp) | |
2084 | { | |
2085 | opp->port_no = htons(opp->port_no); | |
2086 | opp->config = htonl(opp->config); | |
2087 | opp->state = htonl(opp->state); | |
2088 | opp->curr = htonl(opp->curr); | |
2089 | opp->advertised = htonl(opp->advertised); | |
2090 | opp->supported = htonl(opp->supported); | |
2091 | opp->peer = htonl(opp->peer); | |
2092 | } | |
2093 | ||
2094 | static int | |
2095 | handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh) | |
2096 | { | |
2097 | struct ofp_header *rq = oh; | |
2098 | queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter); | |
2099 | return 0; | |
2100 | } | |
2101 | ||
2102 | static int | |
2103 | handle_features_request(struct ofproto *p, struct ofconn *ofconn, | |
2104 | struct ofp_header *oh) | |
2105 | { | |
2106 | struct ofp_switch_features *osf; | |
2107 | struct ofpbuf *buf; | |
2108 | unsigned int port_no; | |
2109 | struct ofport *port; | |
2110 | ||
2111 | osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf); | |
2112 | osf->datapath_id = htonll(p->datapath_id); | |
2113 | osf->n_buffers = htonl(pktbuf_capacity()); | |
2114 | osf->n_tables = 2; | |
2115 | osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS | | |
0254ae23 | 2116 | OFPC_PORT_STATS | OFPC_ARP_MATCH_IP); |
064af421 BP |
2117 | osf->actions = htonl((1u << OFPAT_OUTPUT) | |
2118 | (1u << OFPAT_SET_VLAN_VID) | | |
2119 | (1u << OFPAT_SET_VLAN_PCP) | | |
2120 | (1u << OFPAT_STRIP_VLAN) | | |
2121 | (1u << OFPAT_SET_DL_SRC) | | |
2122 | (1u << OFPAT_SET_DL_DST) | | |
2123 | (1u << OFPAT_SET_NW_SRC) | | |
2124 | (1u << OFPAT_SET_NW_DST) | | |
959a2ecd | 2125 | (1u << OFPAT_SET_NW_TOS) | |
064af421 BP |
2126 | (1u << OFPAT_SET_TP_SRC) | |
2127 | (1u << OFPAT_SET_TP_DST)); | |
2128 | ||
2129 | PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) { | |
2130 | hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp)); | |
2131 | } | |
2132 | ||
2133 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
2134 | return 0; | |
2135 | } | |
2136 | ||
2137 | static int | |
2138 | handle_get_config_request(struct ofproto *p, struct ofconn *ofconn, | |
2139 | struct ofp_header *oh) | |
2140 | { | |
2141 | struct ofpbuf *buf; | |
2142 | struct ofp_switch_config *osc; | |
2143 | uint16_t flags; | |
2144 | bool drop_frags; | |
2145 | ||
2146 | /* Figure out flags. */ | |
c228a364 | 2147 | dpif_get_drop_frags(p->dpif, &drop_frags); |
064af421 | 2148 | flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL; |
064af421 BP |
2149 | |
2150 | /* Send reply. */ | |
2151 | osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf); | |
2152 | osc->flags = htons(flags); | |
2153 | osc->miss_send_len = htons(ofconn->miss_send_len); | |
2154 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
2155 | ||
2156 | return 0; | |
2157 | } | |
2158 | ||
2159 | static int | |
2160 | handle_set_config(struct ofproto *p, struct ofconn *ofconn, | |
2161 | struct ofp_switch_config *osc) | |
2162 | { | |
2163 | uint16_t flags; | |
2164 | int error; | |
2165 | ||
2166 | error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc); | |
2167 | if (error) { | |
2168 | return error; | |
2169 | } | |
2170 | flags = ntohs(osc->flags); | |
2171 | ||
9deba63b | 2172 | if (ofconn->type == OFCONN_CONTROLLER && ofconn->role != NX_ROLE_SLAVE) { |
064af421 BP |
2173 | switch (flags & OFPC_FRAG_MASK) { |
2174 | case OFPC_FRAG_NORMAL: | |
c228a364 | 2175 | dpif_set_drop_frags(p->dpif, false); |
064af421 BP |
2176 | break; |
2177 | case OFPC_FRAG_DROP: | |
c228a364 | 2178 | dpif_set_drop_frags(p->dpif, true); |
064af421 BP |
2179 | break; |
2180 | default: | |
2181 | VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")", | |
2182 | osc->flags); | |
2183 | break; | |
2184 | } | |
2185 | } | |
2186 | ||
064af421 BP |
2187 | ofconn->miss_send_len = ntohs(osc->miss_send_len); |
2188 | ||
2189 | return 0; | |
2190 | } | |
2191 | ||
2192 | static void | |
6a07af36 JG |
2193 | add_output_group_action(struct odp_actions *actions, uint16_t group, |
2194 | uint16_t *nf_output_iface) | |
064af421 BP |
2195 | { |
2196 | odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group; | |
6a07af36 JG |
2197 | |
2198 | if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) { | |
2199 | *nf_output_iface = NF_OUT_FLOOD; | |
2200 | } | |
064af421 BP |
2201 | } |
2202 | ||
2203 | static void | |
2204 | add_controller_action(struct odp_actions *actions, | |
2205 | const struct ofp_action_output *oao) | |
2206 | { | |
2207 | union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER); | |
30ea5d93 | 2208 | a->controller.arg = ntohs(oao->max_len); |
064af421 BP |
2209 | } |
2210 | ||
2211 | struct action_xlate_ctx { | |
2212 | /* Input. */ | |
e18fe8a2 | 2213 | flow_t flow; /* Flow to which these actions correspond. */ |
064af421 BP |
2214 | int recurse; /* Recursion level, via xlate_table_action. */ |
2215 | struct ofproto *ofproto; | |
2216 | const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a | |
2217 | * null pointer if we are revalidating | |
2218 | * without a packet to refer to. */ | |
2219 | ||
2220 | /* Output. */ | |
2221 | struct odp_actions *out; /* Datapath actions. */ | |
2222 | tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */ | |
d6fbec6d | 2223 | bool may_set_up_flow; /* True ordinarily; false if the actions must |
064af421 | 2224 | * be reassessed for every packet. */ |
6a07af36 | 2225 | uint16_t nf_output_iface; /* Output interface index for NetFlow. */ |
064af421 BP |
2226 | }; |
2227 | ||
2228 | static void do_xlate_actions(const union ofp_action *in, size_t n_in, | |
2229 | struct action_xlate_ctx *ctx); | |
2230 | ||
2231 | static void | |
2232 | add_output_action(struct action_xlate_ctx *ctx, uint16_t port) | |
2233 | { | |
2234 | const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port); | |
6cfaf517 BP |
2235 | |
2236 | if (ofport) { | |
2237 | if (ofport->opp.config & OFPPC_NO_FWD) { | |
2238 | /* Forwarding disabled on port. */ | |
2239 | return; | |
2240 | } | |
2241 | } else { | |
2242 | /* | |
2243 | * We don't have an ofport record for this port, but it doesn't hurt to | |
2244 | * allow forwarding to it anyhow. Maybe such a port will appear later | |
2245 | * and we're pre-populating the flow table. | |
2246 | */ | |
064af421 | 2247 | } |
6cfaf517 BP |
2248 | |
2249 | odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port; | |
6a07af36 | 2250 | ctx->nf_output_iface = port; |
064af421 BP |
2251 | } |
2252 | ||
2253 | static struct rule * | |
2254 | lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow) | |
2255 | { | |
2256 | struct rule *rule; | |
2257 | rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow)); | |
2258 | ||
2259 | /* The rule we found might not be valid, since we could be in need of | |
2260 | * revalidation. If it is not valid, don't return it. */ | |
2261 | if (rule | |
2262 | && rule->super | |
2263 | && ofproto->need_revalidate | |
2264 | && !revalidate_rule(ofproto, rule)) { | |
2265 | COVERAGE_INC(ofproto_invalidated); | |
2266 | return NULL; | |
2267 | } | |
2268 | ||
2269 | return rule; | |
2270 | } | |
2271 | ||
2272 | static void | |
2273 | xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port) | |
2274 | { | |
2275 | if (!ctx->recurse) { | |
2c5d1389 | 2276 | uint16_t old_in_port; |
064af421 | 2277 | struct rule *rule; |
064af421 | 2278 | |
2c5d1389 BP |
2279 | /* Look up a flow with 'in_port' as the input port. Then restore the |
2280 | * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will | |
2281 | * have surprising behavior). */ | |
2282 | old_in_port = ctx->flow.in_port; | |
e18fe8a2 BP |
2283 | ctx->flow.in_port = in_port; |
2284 | rule = lookup_valid_rule(ctx->ofproto, &ctx->flow); | |
2c5d1389 BP |
2285 | ctx->flow.in_port = old_in_port; |
2286 | ||
064af421 BP |
2287 | if (rule) { |
2288 | if (rule->super) { | |
2289 | rule = rule->super; | |
2290 | } | |
2291 | ||
2292 | ctx->recurse++; | |
2293 | do_xlate_actions(rule->actions, rule->n_actions, ctx); | |
2294 | ctx->recurse--; | |
2295 | } | |
2296 | } | |
2297 | } | |
2298 | ||
2299 | static void | |
2300 | xlate_output_action(struct action_xlate_ctx *ctx, | |
2301 | const struct ofp_action_output *oao) | |
2302 | { | |
2303 | uint16_t odp_port; | |
6a07af36 JG |
2304 | uint16_t prev_nf_output_iface = ctx->nf_output_iface; |
2305 | ||
2306 | ctx->nf_output_iface = NF_OUT_DROP; | |
064af421 BP |
2307 | |
2308 | switch (ntohs(oao->port)) { | |
2309 | case OFPP_IN_PORT: | |
e18fe8a2 | 2310 | add_output_action(ctx, ctx->flow.in_port); |
064af421 BP |
2311 | break; |
2312 | case OFPP_TABLE: | |
e18fe8a2 | 2313 | xlate_table_action(ctx, ctx->flow.in_port); |
064af421 BP |
2314 | break; |
2315 | case OFPP_NORMAL: | |
e18fe8a2 | 2316 | if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet, |
064af421 | 2317 | ctx->out, ctx->tags, |
6a07af36 | 2318 | &ctx->nf_output_iface, |
064af421 BP |
2319 | ctx->ofproto->aux)) { |
2320 | COVERAGE_INC(ofproto_uninstallable); | |
d6fbec6d | 2321 | ctx->may_set_up_flow = false; |
064af421 BP |
2322 | } |
2323 | break; | |
2324 | case OFPP_FLOOD: | |
6a07af36 JG |
2325 | add_output_group_action(ctx->out, DP_GROUP_FLOOD, |
2326 | &ctx->nf_output_iface); | |
064af421 BP |
2327 | break; |
2328 | case OFPP_ALL: | |
6a07af36 | 2329 | add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface); |
064af421 BP |
2330 | break; |
2331 | case OFPP_CONTROLLER: | |
2332 | add_controller_action(ctx->out, oao); | |
2333 | break; | |
2334 | case OFPP_LOCAL: | |
2335 | add_output_action(ctx, ODPP_LOCAL); | |
2336 | break; | |
2337 | default: | |
2338 | odp_port = ofp_port_to_odp_port(ntohs(oao->port)); | |
e18fe8a2 | 2339 | if (odp_port != ctx->flow.in_port) { |
064af421 BP |
2340 | add_output_action(ctx, odp_port); |
2341 | } | |
2342 | break; | |
2343 | } | |
6a07af36 JG |
2344 | |
2345 | if (prev_nf_output_iface == NF_OUT_FLOOD) { | |
2346 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
2347 | } else if (ctx->nf_output_iface == NF_OUT_DROP) { | |
2348 | ctx->nf_output_iface = prev_nf_output_iface; | |
2349 | } else if (prev_nf_output_iface != NF_OUT_DROP && | |
2350 | ctx->nf_output_iface != NF_OUT_FLOOD) { | |
2351 | ctx->nf_output_iface = NF_OUT_MULTI; | |
2352 | } | |
064af421 BP |
2353 | } |
2354 | ||
2355 | static void | |
2356 | xlate_nicira_action(struct action_xlate_ctx *ctx, | |
2357 | const struct nx_action_header *nah) | |
2358 | { | |
2359 | const struct nx_action_resubmit *nar; | |
659586ef JG |
2360 | const struct nx_action_set_tunnel *nast; |
2361 | union odp_action *oa; | |
064af421 BP |
2362 | int subtype = ntohs(nah->subtype); |
2363 | ||
2364 | assert(nah->vendor == htonl(NX_VENDOR_ID)); | |
2365 | switch (subtype) { | |
2366 | case NXAST_RESUBMIT: | |
2367 | nar = (const struct nx_action_resubmit *) nah; | |
2368 | xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port))); | |
2369 | break; | |
2370 | ||
659586ef JG |
2371 | case NXAST_SET_TUNNEL: |
2372 | nast = (const struct nx_action_set_tunnel *) nah; | |
2373 | oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL); | |
2374 | ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id; | |
2375 | break; | |
2376 | ||
999f0d45 BP |
2377 | /* If you add a new action here that modifies flow data, don't forget to |
2378 | * update the flow key in ctx->flow in the same key. */ | |
2379 | ||
064af421 BP |
2380 | default: |
2381 | VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype); | |
2382 | break; | |
2383 | } | |
2384 | } | |
2385 | ||
2386 | static void | |
2387 | do_xlate_actions(const union ofp_action *in, size_t n_in, | |
2388 | struct action_xlate_ctx *ctx) | |
2389 | { | |
2390 | struct actions_iterator iter; | |
2391 | const union ofp_action *ia; | |
2392 | const struct ofport *port; | |
2393 | ||
e18fe8a2 | 2394 | port = port_array_get(&ctx->ofproto->ports, ctx->flow.in_port); |
064af421 | 2395 | if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) && |
e18fe8a2 | 2396 | port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, stp_eth_addr) |
064af421 BP |
2397 | ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) { |
2398 | /* Drop this flow. */ | |
2399 | return; | |
2400 | } | |
2401 | ||
2402 | for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) { | |
2403 | uint16_t type = ntohs(ia->type); | |
2404 | union odp_action *oa; | |
2405 | ||
2406 | switch (type) { | |
2407 | case OFPAT_OUTPUT: | |
2408 | xlate_output_action(ctx, &ia->output); | |
2409 | break; | |
2410 | ||
2411 | case OFPAT_SET_VLAN_VID: | |
2412 | oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID); | |
999f0d45 | 2413 | ctx->flow.dl_vlan = oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid; |
064af421 BP |
2414 | break; |
2415 | ||
2416 | case OFPAT_SET_VLAN_PCP: | |
2417 | oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP); | |
999f0d45 | 2418 | ctx->flow.dl_vlan_pcp = oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp; |
064af421 BP |
2419 | break; |
2420 | ||
2421 | case OFPAT_STRIP_VLAN: | |
2422 | odp_actions_add(ctx->out, ODPAT_STRIP_VLAN); | |
999f0d45 BP |
2423 | ctx->flow.dl_vlan = OFP_VLAN_NONE; |
2424 | ctx->flow.dl_vlan_pcp = 0; | |
064af421 BP |
2425 | break; |
2426 | ||
2427 | case OFPAT_SET_DL_SRC: | |
2428 | oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC); | |
2429 | memcpy(oa->dl_addr.dl_addr, | |
2430 | ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); | |
999f0d45 BP |
2431 | memcpy(ctx->flow.dl_src, |
2432 | ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); | |
064af421 BP |
2433 | break; |
2434 | ||
2435 | case OFPAT_SET_DL_DST: | |
2436 | oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST); | |
2437 | memcpy(oa->dl_addr.dl_addr, | |
2438 | ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); | |
999f0d45 BP |
2439 | memcpy(ctx->flow.dl_dst, |
2440 | ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); | |
064af421 BP |
2441 | break; |
2442 | ||
2443 | case OFPAT_SET_NW_SRC: | |
2444 | oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC); | |
999f0d45 | 2445 | ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr; |
064af421 BP |
2446 | break; |
2447 | ||
2d70a31a JP |
2448 | case OFPAT_SET_NW_DST: |
2449 | oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST); | |
999f0d45 | 2450 | ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr; |
2d38e234 | 2451 | break; |
959a2ecd JP |
2452 | |
2453 | case OFPAT_SET_NW_TOS: | |
2454 | oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS); | |
999f0d45 | 2455 | ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos; |
2d70a31a JP |
2456 | break; |
2457 | ||
064af421 BP |
2458 | case OFPAT_SET_TP_SRC: |
2459 | oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC); | |
999f0d45 | 2460 | ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port; |
064af421 BP |
2461 | break; |
2462 | ||
2d70a31a JP |
2463 | case OFPAT_SET_TP_DST: |
2464 | oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST); | |
999f0d45 | 2465 | ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port; |
2d70a31a JP |
2466 | break; |
2467 | ||
064af421 BP |
2468 | case OFPAT_VENDOR: |
2469 | xlate_nicira_action(ctx, (const struct nx_action_header *) ia); | |
2470 | break; | |
2471 | ||
2472 | default: | |
2473 | VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type); | |
2474 | break; | |
2475 | } | |
2476 | } | |
2477 | } | |
2478 | ||
2479 | static int | |
2480 | xlate_actions(const union ofp_action *in, size_t n_in, | |
2481 | const flow_t *flow, struct ofproto *ofproto, | |
2482 | const struct ofpbuf *packet, | |
6a07af36 JG |
2483 | struct odp_actions *out, tag_type *tags, bool *may_set_up_flow, |
2484 | uint16_t *nf_output_iface) | |
064af421 BP |
2485 | { |
2486 | tag_type no_tags = 0; | |
2487 | struct action_xlate_ctx ctx; | |
2488 | COVERAGE_INC(ofproto_ofp2odp); | |
2489 | odp_actions_init(out); | |
e18fe8a2 | 2490 | ctx.flow = *flow; |
064af421 BP |
2491 | ctx.recurse = 0; |
2492 | ctx.ofproto = ofproto; | |
2493 | ctx.packet = packet; | |
2494 | ctx.out = out; | |
2495 | ctx.tags = tags ? tags : &no_tags; | |
d6fbec6d | 2496 | ctx.may_set_up_flow = true; |
6a07af36 | 2497 | ctx.nf_output_iface = NF_OUT_DROP; |
064af421 | 2498 | do_xlate_actions(in, n_in, &ctx); |
0ad9b732 | 2499 | |
d6fbec6d | 2500 | /* Check with in-band control to see if we're allowed to set up this |
0ad9b732 JP |
2501 | * flow. */ |
2502 | if (!in_band_rule_check(ofproto->in_band, flow, out)) { | |
d6fbec6d | 2503 | ctx.may_set_up_flow = false; |
0ad9b732 JP |
2504 | } |
2505 | ||
d6fbec6d BP |
2506 | if (may_set_up_flow) { |
2507 | *may_set_up_flow = ctx.may_set_up_flow; | |
064af421 | 2508 | } |
6a07af36 JG |
2509 | if (nf_output_iface) { |
2510 | *nf_output_iface = ctx.nf_output_iface; | |
064af421 BP |
2511 | } |
2512 | if (odp_actions_overflow(out)) { | |
2513 | odp_actions_init(out); | |
2514 | return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY); | |
2515 | } | |
2516 | return 0; | |
2517 | } | |
2518 | ||
9deba63b BP |
2519 | /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow |
2520 | * error message code (composed with ofp_mkerr()) for the caller to propagate | |
2521 | * upward. Otherwise, returns 0. | |
2522 | * | |
2523 | * 'oh' is used to make log messages more informative. */ | |
2524 | static int | |
2525 | reject_slave_controller(struct ofconn *ofconn, const struct ofp_header *oh) | |
2526 | { | |
2527 | if (ofconn->type == OFCONN_CONTROLLER && ofconn->role == NX_ROLE_SLAVE) { | |
2528 | static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
2529 | char *type_name; | |
2530 | ||
2531 | type_name = ofp_message_type_to_string(oh->type); | |
2532 | VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller", | |
2533 | type_name); | |
2534 | free(type_name); | |
2535 | ||
2536 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM); | |
2537 | } else { | |
2538 | return 0; | |
2539 | } | |
2540 | } | |
2541 | ||
064af421 BP |
2542 | static int |
2543 | handle_packet_out(struct ofproto *p, struct ofconn *ofconn, | |
2544 | struct ofp_header *oh) | |
2545 | { | |
2546 | struct ofp_packet_out *opo; | |
2547 | struct ofpbuf payload, *buffer; | |
2548 | struct odp_actions actions; | |
2549 | int n_actions; | |
2550 | uint16_t in_port; | |
2551 | flow_t flow; | |
2552 | int error; | |
2553 | ||
9deba63b BP |
2554 | error = reject_slave_controller(ofconn, oh); |
2555 | if (error) { | |
2556 | return error; | |
2557 | } | |
2558 | ||
064af421 BP |
2559 | error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports); |
2560 | if (error) { | |
2561 | return error; | |
2562 | } | |
2563 | opo = (struct ofp_packet_out *) oh; | |
2564 | ||
2565 | COVERAGE_INC(ofproto_packet_out); | |
2566 | if (opo->buffer_id != htonl(UINT32_MAX)) { | |
2567 | error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id), | |
2568 | &buffer, &in_port); | |
7778bd15 | 2569 | if (error || !buffer) { |
064af421 BP |
2570 | return error; |
2571 | } | |
2572 | payload = *buffer; | |
2573 | } else { | |
2574 | buffer = NULL; | |
2575 | } | |
2576 | ||
659586ef | 2577 | flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow); |
064af421 | 2578 | error = xlate_actions((const union ofp_action *) opo->actions, n_actions, |
6a07af36 | 2579 | &flow, p, &payload, &actions, NULL, NULL, NULL); |
064af421 BP |
2580 | if (error) { |
2581 | return error; | |
2582 | } | |
2583 | ||
c228a364 | 2584 | dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions, |
064af421 BP |
2585 | &payload); |
2586 | ofpbuf_delete(buffer); | |
2587 | ||
2588 | return 0; | |
2589 | } | |
2590 | ||
2591 | static void | |
2592 | update_port_config(struct ofproto *p, struct ofport *port, | |
2593 | uint32_t config, uint32_t mask) | |
2594 | { | |
2595 | mask &= config ^ port->opp.config; | |
2596 | if (mask & OFPPC_PORT_DOWN) { | |
2597 | if (config & OFPPC_PORT_DOWN) { | |
2598 | netdev_turn_flags_off(port->netdev, NETDEV_UP, true); | |
2599 | } else { | |
2600 | netdev_turn_flags_on(port->netdev, NETDEV_UP, true); | |
2601 | } | |
2602 | } | |
2603 | #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD) | |
2604 | if (mask & REVALIDATE_BITS) { | |
2605 | COVERAGE_INC(ofproto_costly_flags); | |
2606 | port->opp.config ^= mask & REVALIDATE_BITS; | |
2607 | p->need_revalidate = true; | |
2608 | } | |
2609 | #undef REVALIDATE_BITS | |
2610 | if (mask & OFPPC_NO_FLOOD) { | |
2611 | port->opp.config ^= OFPPC_NO_FLOOD; | |
72b06300 | 2612 | refresh_port_groups(p); |
064af421 BP |
2613 | } |
2614 | if (mask & OFPPC_NO_PACKET_IN) { | |
2615 | port->opp.config ^= OFPPC_NO_PACKET_IN; | |
2616 | } | |
2617 | } | |
2618 | ||
2619 | static int | |
9deba63b BP |
2620 | handle_port_mod(struct ofproto *p, struct ofconn *ofconn, |
2621 | struct ofp_header *oh) | |
064af421 BP |
2622 | { |
2623 | const struct ofp_port_mod *opm; | |
2624 | struct ofport *port; | |
2625 | int error; | |
2626 | ||
9deba63b BP |
2627 | error = reject_slave_controller(ofconn, oh); |
2628 | if (error) { | |
2629 | return error; | |
2630 | } | |
064af421 BP |
2631 | error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm); |
2632 | if (error) { | |
2633 | return error; | |
2634 | } | |
2635 | opm = (struct ofp_port_mod *) oh; | |
2636 | ||
2637 | port = port_array_get(&p->ports, | |
2638 | ofp_port_to_odp_port(ntohs(opm->port_no))); | |
2639 | if (!port) { | |
2640 | return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT); | |
2641 | } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) { | |
2642 | return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR); | |
2643 | } else { | |
2644 | update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask)); | |
2645 | if (opm->advertise) { | |
2646 | netdev_set_advertisements(port->netdev, ntohl(opm->advertise)); | |
2647 | } | |
2648 | } | |
2649 | return 0; | |
2650 | } | |
2651 | ||
2652 | static struct ofpbuf * | |
2653 | make_stats_reply(uint32_t xid, uint16_t type, size_t body_len) | |
2654 | { | |
2655 | struct ofp_stats_reply *osr; | |
2656 | struct ofpbuf *msg; | |
2657 | ||
2658 | msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX)); | |
2659 | osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg); | |
2660 | osr->type = type; | |
2661 | osr->flags = htons(0); | |
2662 | return msg; | |
2663 | } | |
2664 | ||
2665 | static struct ofpbuf * | |
2666 | start_stats_reply(const struct ofp_stats_request *request, size_t body_len) | |
2667 | { | |
2668 | return make_stats_reply(request->header.xid, request->type, body_len); | |
2669 | } | |
2670 | ||
2671 | static void * | |
2672 | append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp) | |
2673 | { | |
2674 | struct ofpbuf *msg = *msgp; | |
2675 | assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply)); | |
2676 | if (nbytes + msg->size > UINT16_MAX) { | |
2677 | struct ofp_stats_reply *reply = msg->data; | |
2678 | reply->flags = htons(OFPSF_REPLY_MORE); | |
2679 | *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes); | |
2680 | queue_tx(msg, ofconn, ofconn->reply_counter); | |
2681 | } | |
2682 | return ofpbuf_put_uninit(*msgp, nbytes); | |
2683 | } | |
2684 | ||
2685 | static int | |
2686 | handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn, | |
2687 | struct ofp_stats_request *request) | |
2688 | { | |
2689 | struct ofp_desc_stats *ods; | |
2690 | struct ofpbuf *msg; | |
2691 | ||
2692 | msg = start_stats_reply(request, sizeof *ods); | |
2693 | ods = append_stats_reply(sizeof *ods, ofconn, &msg); | |
5a719c38 JP |
2694 | memset(ods, 0, sizeof *ods); |
2695 | ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc); | |
2696 | ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc); | |
2697 | ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc); | |
2698 | ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num); | |
2699 | ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc); | |
064af421 BP |
2700 | queue_tx(msg, ofconn, ofconn->reply_counter); |
2701 | ||
2702 | return 0; | |
2703 | } | |
2704 | ||
2705 | static void | |
2706 | count_subrules(struct cls_rule *cls_rule, void *n_subrules_) | |
2707 | { | |
2708 | struct rule *rule = rule_from_cls_rule(cls_rule); | |
2709 | int *n_subrules = n_subrules_; | |
2710 | ||
2711 | if (rule->super) { | |
2712 | (*n_subrules)++; | |
2713 | } | |
2714 | } | |
2715 | ||
2716 | static int | |
2717 | handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn, | |
2718 | struct ofp_stats_request *request) | |
2719 | { | |
2720 | struct ofp_table_stats *ots; | |
2721 | struct ofpbuf *msg; | |
2722 | struct odp_stats dpstats; | |
2723 | int n_exact, n_subrules, n_wild; | |
2724 | ||
2725 | msg = start_stats_reply(request, sizeof *ots * 2); | |
2726 | ||
2727 | /* Count rules of various kinds. */ | |
2728 | n_subrules = 0; | |
2729 | classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules); | |
2730 | n_exact = classifier_count_exact(&p->cls) - n_subrules; | |
2731 | n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls); | |
2732 | ||
2733 | /* Hash table. */ | |
c228a364 | 2734 | dpif_get_dp_stats(p->dpif, &dpstats); |
064af421 BP |
2735 | ots = append_stats_reply(sizeof *ots, ofconn, &msg); |
2736 | memset(ots, 0, sizeof *ots); | |
2737 | ots->table_id = TABLEID_HASH; | |
2738 | strcpy(ots->name, "hash"); | |
2739 | ots->wildcards = htonl(0); | |
2740 | ots->max_entries = htonl(dpstats.max_capacity); | |
2741 | ots->active_count = htonl(n_exact); | |
2742 | ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit + | |
2743 | dpstats.n_missed); | |
2744 | ots->matched_count = htonll(dpstats.n_hit); /* XXX */ | |
2745 | ||
2746 | /* Classifier table. */ | |
2747 | ots = append_stats_reply(sizeof *ots, ofconn, &msg); | |
2748 | memset(ots, 0, sizeof *ots); | |
2749 | ots->table_id = TABLEID_CLASSIFIER; | |
2750 | strcpy(ots->name, "classifier"); | |
659586ef JG |
2751 | ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL) |
2752 | : htonl(OFPFW_ALL); | |
064af421 BP |
2753 | ots->max_entries = htonl(65536); |
2754 | ots->active_count = htonl(n_wild); | |
2755 | ots->lookup_count = htonll(0); /* XXX */ | |
2756 | ots->matched_count = htonll(0); /* XXX */ | |
2757 | ||
2758 | queue_tx(msg, ofconn, ofconn->reply_counter); | |
2759 | return 0; | |
2760 | } | |
2761 | ||
abaad8cf JP |
2762 | static void |
2763 | append_port_stat(struct ofport *port, uint16_t port_no, struct ofconn *ofconn, | |
a4948b95 | 2764 | struct ofpbuf **msgp) |
abaad8cf JP |
2765 | { |
2766 | struct netdev_stats stats; | |
2767 | struct ofp_port_stats *ops; | |
2768 | ||
2769 | /* Intentionally ignore return value, since errors will set | |
2770 | * 'stats' to all-1s, which is correct for OpenFlow, and | |
2771 | * netdev_get_stats() will log errors. */ | |
2772 | netdev_get_stats(port->netdev, &stats); | |
2773 | ||
a4948b95 | 2774 | ops = append_stats_reply(sizeof *ops, ofconn, msgp); |
abaad8cf JP |
2775 | ops->port_no = htons(odp_port_to_ofp_port(port_no)); |
2776 | memset(ops->pad, 0, sizeof ops->pad); | |
2777 | ops->rx_packets = htonll(stats.rx_packets); | |
2778 | ops->tx_packets = htonll(stats.tx_packets); | |
2779 | ops->rx_bytes = htonll(stats.rx_bytes); | |
2780 | ops->tx_bytes = htonll(stats.tx_bytes); | |
2781 | ops->rx_dropped = htonll(stats.rx_dropped); | |
2782 | ops->tx_dropped = htonll(stats.tx_dropped); | |
2783 | ops->rx_errors = htonll(stats.rx_errors); | |
2784 | ops->tx_errors = htonll(stats.tx_errors); | |
2785 | ops->rx_frame_err = htonll(stats.rx_frame_errors); | |
2786 | ops->rx_over_err = htonll(stats.rx_over_errors); | |
2787 | ops->rx_crc_err = htonll(stats.rx_crc_errors); | |
2788 | ops->collisions = htonll(stats.collisions); | |
2789 | } | |
2790 | ||
064af421 BP |
2791 | static int |
2792 | handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn, | |
abaad8cf JP |
2793 | struct ofp_stats_request *osr, |
2794 | size_t arg_size) | |
064af421 | 2795 | { |
abaad8cf | 2796 | struct ofp_port_stats_request *psr; |
064af421 BP |
2797 | struct ofp_port_stats *ops; |
2798 | struct ofpbuf *msg; | |
2799 | struct ofport *port; | |
2800 | unsigned int port_no; | |
2801 | ||
abaad8cf JP |
2802 | if (arg_size != sizeof *psr) { |
2803 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
2804 | } | |
2805 | psr = (struct ofp_port_stats_request *) osr->body; | |
2806 | ||
2807 | msg = start_stats_reply(osr, sizeof *ops * 16); | |
2808 | if (psr->port_no != htons(OFPP_NONE)) { | |
2809 | port = port_array_get(&p->ports, | |
2810 | ofp_port_to_odp_port(ntohs(psr->port_no))); | |
2811 | if (port) { | |
a4948b95 | 2812 | append_port_stat(port, ntohs(psr->port_no), ofconn, &msg); |
abaad8cf JP |
2813 | } |
2814 | } else { | |
2815 | PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) { | |
a4948b95 | 2816 | append_port_stat(port, port_no, ofconn, &msg); |
abaad8cf | 2817 | } |
064af421 BP |
2818 | } |
2819 | ||
2820 | queue_tx(msg, ofconn, ofconn->reply_counter); | |
2821 | return 0; | |
2822 | } | |
2823 | ||
2824 | struct flow_stats_cbdata { | |
2825 | struct ofproto *ofproto; | |
2826 | struct ofconn *ofconn; | |
2827 | uint16_t out_port; | |
2828 | struct ofpbuf *msg; | |
2829 | }; | |
2830 | ||
01149cfd BP |
2831 | /* Obtains statistic counters for 'rule' within 'p' and stores them into |
2832 | * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the | |
2833 | * returned statistic include statistics for all of 'rule''s subrules. */ | |
064af421 BP |
2834 | static void |
2835 | query_stats(struct ofproto *p, struct rule *rule, | |
2836 | uint64_t *packet_countp, uint64_t *byte_countp) | |
2837 | { | |
2838 | uint64_t packet_count, byte_count; | |
2839 | struct rule *subrule; | |
2840 | struct odp_flow *odp_flows; | |
2841 | size_t n_odp_flows; | |
2842 | ||
01149cfd BP |
2843 | /* Start from historical data for 'rule' itself that are no longer tracked |
2844 | * by the datapath. This counts, for example, subrules that have | |
2845 | * expired. */ | |
b3137fe8 JG |
2846 | packet_count = rule->packet_count; |
2847 | byte_count = rule->byte_count; | |
2848 | ||
01149cfd BP |
2849 | /* Prepare to ask the datapath for statistics on 'rule', or if it is |
2850 | * wildcarded then on all of its subrules. | |
2851 | * | |
2852 | * Also, add any statistics that are not tracked by the datapath for each | |
2853 | * subrule. This includes, for example, statistics for packets that were | |
2854 | * executed "by hand" by ofproto via dpif_execute() but must be accounted | |
2855 | * to a flow. */ | |
064af421 | 2856 | n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1; |
ec6fde61 | 2857 | odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows); |
064af421 BP |
2858 | if (rule->cr.wc.wildcards) { |
2859 | size_t i = 0; | |
2860 | LIST_FOR_EACH (subrule, struct rule, list, &rule->list) { | |
2861 | odp_flows[i++].key = subrule->cr.flow; | |
b3137fe8 JG |
2862 | packet_count += subrule->packet_count; |
2863 | byte_count += subrule->byte_count; | |
064af421 BP |
2864 | } |
2865 | } else { | |
2866 | odp_flows[0].key = rule->cr.flow; | |
2867 | } | |
2868 | ||
28998b22 | 2869 | /* Fetch up-to-date statistics from the datapath and add them in. */ |
c228a364 | 2870 | if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) { |
064af421 BP |
2871 | size_t i; |
2872 | for (i = 0; i < n_odp_flows; i++) { | |
2873 | struct odp_flow *odp_flow = &odp_flows[i]; | |
2874 | packet_count += odp_flow->stats.n_packets; | |
2875 | byte_count += odp_flow->stats.n_bytes; | |
2876 | } | |
2877 | } | |
2878 | free(odp_flows); | |
2879 | ||
01149cfd | 2880 | /* Return the stats to the caller. */ |
064af421 BP |
2881 | *packet_countp = packet_count; |
2882 | *byte_countp = byte_count; | |
2883 | } | |
2884 | ||
2885 | static void | |
2886 | flow_stats_cb(struct cls_rule *rule_, void *cbdata_) | |
2887 | { | |
2888 | struct rule *rule = rule_from_cls_rule(rule_); | |
2889 | struct flow_stats_cbdata *cbdata = cbdata_; | |
2890 | struct ofp_flow_stats *ofs; | |
2891 | uint64_t packet_count, byte_count; | |
2892 | size_t act_len, len; | |
26c3f94a JP |
2893 | long long int tdiff = time_msec() - rule->created; |
2894 | uint32_t sec = tdiff / 1000; | |
2895 | uint32_t msec = tdiff - (sec * 1000); | |
064af421 BP |
2896 | |
2897 | if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) { | |
2898 | return; | |
2899 | } | |
2900 | ||
2901 | act_len = sizeof *rule->actions * rule->n_actions; | |
2902 | len = offsetof(struct ofp_flow_stats, actions) + act_len; | |
2903 | ||
2904 | query_stats(cbdata->ofproto, rule, &packet_count, &byte_count); | |
2905 | ||
2906 | ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg); | |
2907 | ofs->length = htons(len); | |
2908 | ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH; | |
2909 | ofs->pad = 0; | |
659586ef JG |
2910 | flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, |
2911 | cbdata->ofproto->tun_id_from_cookie, &ofs->match); | |
26c3f94a JP |
2912 | ofs->duration_sec = htonl(sec); |
2913 | ofs->duration_nsec = htonl(msec * 1000000); | |
39997502 | 2914 | ofs->cookie = rule->flow_cookie; |
064af421 BP |
2915 | ofs->priority = htons(rule->cr.priority); |
2916 | ofs->idle_timeout = htons(rule->idle_timeout); | |
2917 | ofs->hard_timeout = htons(rule->hard_timeout); | |
39997502 | 2918 | memset(ofs->pad2, 0, sizeof ofs->pad2); |
064af421 BP |
2919 | ofs->packet_count = htonll(packet_count); |
2920 | ofs->byte_count = htonll(byte_count); | |
2921 | memcpy(ofs->actions, rule->actions, act_len); | |
2922 | } | |
2923 | ||
2924 | static int | |
2925 | table_id_to_include(uint8_t table_id) | |
2926 | { | |
2927 | return (table_id == TABLEID_HASH ? CLS_INC_EXACT | |
2928 | : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD | |
2929 | : table_id == 0xff ? CLS_INC_ALL | |
2930 | : 0); | |
2931 | } | |
2932 | ||
2933 | static int | |
2934 | handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn, | |
2935 | const struct ofp_stats_request *osr, | |
2936 | size_t arg_size) | |
2937 | { | |
2938 | struct ofp_flow_stats_request *fsr; | |
2939 | struct flow_stats_cbdata cbdata; | |
2940 | struct cls_rule target; | |
2941 | ||
2942 | if (arg_size != sizeof *fsr) { | |
49bdc010 | 2943 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); |
064af421 BP |
2944 | } |
2945 | fsr = (struct ofp_flow_stats_request *) osr->body; | |
2946 | ||
2947 | COVERAGE_INC(ofproto_flows_req); | |
2948 | cbdata.ofproto = p; | |
2949 | cbdata.ofconn = ofconn; | |
2950 | cbdata.out_port = fsr->out_port; | |
2951 | cbdata.msg = start_stats_reply(osr, 1024); | |
659586ef | 2952 | cls_rule_from_match(&fsr->match, 0, false, 0, &target); |
064af421 BP |
2953 | classifier_for_each_match(&p->cls, &target, |
2954 | table_id_to_include(fsr->table_id), | |
2955 | flow_stats_cb, &cbdata); | |
2956 | queue_tx(cbdata.msg, ofconn, ofconn->reply_counter); | |
2957 | return 0; | |
2958 | } | |
2959 | ||
4f2cad2c JP |
2960 | struct flow_stats_ds_cbdata { |
2961 | struct ofproto *ofproto; | |
2962 | struct ds *results; | |
2963 | }; | |
2964 | ||
2965 | static void | |
2966 | flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_) | |
2967 | { | |
2968 | struct rule *rule = rule_from_cls_rule(rule_); | |
2969 | struct flow_stats_ds_cbdata *cbdata = cbdata_; | |
2970 | struct ds *results = cbdata->results; | |
2971 | struct ofp_match match; | |
2972 | uint64_t packet_count, byte_count; | |
2973 | size_t act_len = sizeof *rule->actions * rule->n_actions; | |
2974 | ||
2975 | /* Don't report on subrules. */ | |
2976 | if (rule->super != NULL) { | |
2977 | return; | |
2978 | } | |
2979 | ||
2980 | query_stats(cbdata->ofproto, rule, &packet_count, &byte_count); | |
659586ef JG |
2981 | flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, |
2982 | cbdata->ofproto->tun_id_from_cookie, &match); | |
4f2cad2c JP |
2983 | |
2984 | ds_put_format(results, "duration=%llds, ", | |
2985 | (time_msec() - rule->created) / 1000); | |
52ae00b3 | 2986 | ds_put_format(results, "priority=%u, ", rule->cr.priority); |
4f2cad2c JP |
2987 | ds_put_format(results, "n_packets=%"PRIu64", ", packet_count); |
2988 | ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count); | |
2989 | ofp_print_match(results, &match, true); | |
2990 | ofp_print_actions(results, &rule->actions->header, act_len); | |
2991 | ds_put_cstr(results, "\n"); | |
2992 | } | |
2993 | ||
2994 | /* Adds a pretty-printed description of all flows to 'results', including | |
2995 | * those marked hidden by secchan (e.g., by in-band control). */ | |
2996 | void | |
2997 | ofproto_get_all_flows(struct ofproto *p, struct ds *results) | |
2998 | { | |
2999 | struct ofp_match match; | |
3000 | struct cls_rule target; | |
3001 | struct flow_stats_ds_cbdata cbdata; | |
3002 | ||
3003 | memset(&match, 0, sizeof match); | |
659586ef | 3004 | match.wildcards = htonl(OVSFW_ALL); |
4f2cad2c JP |
3005 | |
3006 | cbdata.ofproto = p; | |
3007 | cbdata.results = results; | |
3008 | ||
659586ef | 3009 | cls_rule_from_match(&match, 0, false, 0, &target); |
4f2cad2c JP |
3010 | classifier_for_each_match(&p->cls, &target, CLS_INC_ALL, |
3011 | flow_stats_ds_cb, &cbdata); | |
3012 | } | |
3013 | ||
064af421 BP |
3014 | struct aggregate_stats_cbdata { |
3015 | struct ofproto *ofproto; | |
3016 | uint16_t out_port; | |
3017 | uint64_t packet_count; | |
3018 | uint64_t byte_count; | |
3019 | uint32_t n_flows; | |
3020 | }; | |
3021 | ||
3022 | static void | |
3023 | aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_) | |
3024 | { | |
3025 | struct rule *rule = rule_from_cls_rule(rule_); | |
3026 | struct aggregate_stats_cbdata *cbdata = cbdata_; | |
3027 | uint64_t packet_count, byte_count; | |
3028 | ||
3029 | if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) { | |
3030 | return; | |
3031 | } | |
3032 | ||
3033 | query_stats(cbdata->ofproto, rule, &packet_count, &byte_count); | |
3034 | ||
3035 | cbdata->packet_count += packet_count; | |
3036 | cbdata->byte_count += byte_count; | |
3037 | cbdata->n_flows++; | |
3038 | } | |
3039 | ||
3040 | static int | |
3041 | handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn, | |
3042 | const struct ofp_stats_request *osr, | |
3043 | size_t arg_size) | |
3044 | { | |
3045 | struct ofp_aggregate_stats_request *asr; | |
3046 | struct ofp_aggregate_stats_reply *reply; | |
3047 | struct aggregate_stats_cbdata cbdata; | |
3048 | struct cls_rule target; | |
3049 | struct ofpbuf *msg; | |
3050 | ||
3051 | if (arg_size != sizeof *asr) { | |
49bdc010 | 3052 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); |
064af421 BP |
3053 | } |
3054 | asr = (struct ofp_aggregate_stats_request *) osr->body; | |
3055 | ||
3056 | COVERAGE_INC(ofproto_agg_request); | |
3057 | cbdata.ofproto = p; | |
3058 | cbdata.out_port = asr->out_port; | |
3059 | cbdata.packet_count = 0; | |
3060 | cbdata.byte_count = 0; | |
3061 | cbdata.n_flows = 0; | |
659586ef | 3062 | cls_rule_from_match(&asr->match, 0, false, 0, &target); |
064af421 BP |
3063 | classifier_for_each_match(&p->cls, &target, |
3064 | table_id_to_include(asr->table_id), | |
3065 | aggregate_stats_cb, &cbdata); | |
3066 | ||
3067 | msg = start_stats_reply(osr, sizeof *reply); | |
3068 | reply = append_stats_reply(sizeof *reply, ofconn, &msg); | |
3069 | reply->flow_count = htonl(cbdata.n_flows); | |
3070 | reply->packet_count = htonll(cbdata.packet_count); | |
3071 | reply->byte_count = htonll(cbdata.byte_count); | |
3072 | queue_tx(msg, ofconn, ofconn->reply_counter); | |
3073 | return 0; | |
3074 | } | |
3075 | ||
3076 | static int | |
3077 | handle_stats_request(struct ofproto *p, struct ofconn *ofconn, | |
3078 | struct ofp_header *oh) | |
3079 | { | |
3080 | struct ofp_stats_request *osr; | |
3081 | size_t arg_size; | |
3082 | int error; | |
3083 | ||
3084 | error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr, | |
3085 | 1, &arg_size); | |
3086 | if (error) { | |
3087 | return error; | |
3088 | } | |
3089 | osr = (struct ofp_stats_request *) oh; | |
3090 | ||
3091 | switch (ntohs(osr->type)) { | |
3092 | case OFPST_DESC: | |
3093 | return handle_desc_stats_request(p, ofconn, osr); | |
3094 | ||
3095 | case OFPST_FLOW: | |
3096 | return handle_flow_stats_request(p, ofconn, osr, arg_size); | |
3097 | ||
3098 | case OFPST_AGGREGATE: | |
3099 | return handle_aggregate_stats_request(p, ofconn, osr, arg_size); | |
3100 | ||
3101 | case OFPST_TABLE: | |
3102 | return handle_table_stats_request(p, ofconn, osr); | |
3103 | ||
3104 | case OFPST_PORT: | |
abaad8cf | 3105 | return handle_port_stats_request(p, ofconn, osr, arg_size); |
064af421 BP |
3106 | |
3107 | case OFPST_VENDOR: | |
3108 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR); | |
3109 | ||
3110 | default: | |
3111 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT); | |
3112 | } | |
3113 | } | |
3114 | ||
3115 | static long long int | |
3116 | msec_from_nsec(uint64_t sec, uint32_t nsec) | |
3117 | { | |
3118 | return !sec ? 0 : sec * 1000 + nsec / 1000000; | |
3119 | } | |
3120 | ||
3121 | static void | |
0193b2af JG |
3122 | update_time(struct ofproto *ofproto, struct rule *rule, |
3123 | const struct odp_flow_stats *stats) | |
064af421 BP |
3124 | { |
3125 | long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec); | |
3126 | if (used > rule->used) { | |
3127 | rule->used = used; | |
4836f9f2 JP |
3128 | if (rule->super && used > rule->super->used) { |
3129 | rule->super->used = used; | |
3130 | } | |
0193b2af | 3131 | netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used); |
064af421 BP |
3132 | } |
3133 | } | |
3134 | ||
3135 | static void | |
0193b2af JG |
3136 | update_stats(struct ofproto *ofproto, struct rule *rule, |
3137 | const struct odp_flow_stats *stats) | |
064af421 | 3138 | { |
064af421 | 3139 | if (stats->n_packets) { |
0193b2af JG |
3140 | update_time(ofproto, rule, stats); |
3141 | rule->packet_count += stats->n_packets; | |
3142 | rule->byte_count += stats->n_bytes; | |
3143 | netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos, | |
3144 | stats->tcp_flags); | |
064af421 BP |
3145 | } |
3146 | } | |
3147 | ||
79eee1eb BP |
3148 | /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT |
3149 | * in which no matching flow already exists in the flow table. | |
3150 | * | |
3151 | * Adds the flow specified by 'ofm', which is followed by 'n_actions' | |
3152 | * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error | |
3153 | * code as encoded by ofp_mkerr() on failure. | |
3154 | * | |
3155 | * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, | |
3156 | * if any. */ | |
064af421 BP |
3157 | static int |
3158 | add_flow(struct ofproto *p, struct ofconn *ofconn, | |
79eee1eb | 3159 | const struct ofp_flow_mod *ofm, size_t n_actions) |
064af421 BP |
3160 | { |
3161 | struct ofpbuf *packet; | |
3162 | struct rule *rule; | |
3163 | uint16_t in_port; | |
3164 | int error; | |
3165 | ||
49bdc010 JP |
3166 | if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) { |
3167 | flow_t flow; | |
3168 | uint32_t wildcards; | |
3169 | ||
659586ef JG |
3170 | flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie, |
3171 | &flow, &wildcards); | |
49bdc010 JP |
3172 | if (classifier_rule_overlaps(&p->cls, &flow, wildcards, |
3173 | ntohs(ofm->priority))) { | |
3174 | return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP); | |
3175 | } | |
3176 | } | |
3177 | ||
0193b2af | 3178 | rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions, |
064af421 | 3179 | n_actions, ntohs(ofm->idle_timeout), |
39997502 | 3180 | ntohs(ofm->hard_timeout), ofm->cookie, |
ca069229 | 3181 | ofm->flags & htons(OFPFF_SEND_FLOW_REM)); |
659586ef JG |
3182 | cls_rule_from_match(&ofm->match, ntohs(ofm->priority), |
3183 | p->tun_id_from_cookie, ofm->cookie, &rule->cr); | |
064af421 | 3184 | |
064af421 BP |
3185 | error = 0; |
3186 | if (ofm->buffer_id != htonl(UINT32_MAX)) { | |
3187 | error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id), | |
3188 | &packet, &in_port); | |
212fe71c BP |
3189 | } else { |
3190 | packet = NULL; | |
165cd8a3 | 3191 | in_port = UINT16_MAX; |
064af421 BP |
3192 | } |
3193 | ||
3194 | rule_insert(p, rule, packet, in_port); | |
3195 | ofpbuf_delete(packet); | |
3196 | return error; | |
3197 | } | |
3198 | ||
79eee1eb BP |
3199 | static struct rule * |
3200 | find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm) | |
064af421 | 3201 | { |
064af421 BP |
3202 | uint32_t wildcards; |
3203 | flow_t flow; | |
3204 | ||
659586ef JG |
3205 | flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie, |
3206 | &flow, &wildcards); | |
79eee1eb | 3207 | return rule_from_cls_rule(classifier_find_rule_exactly( |
064af421 BP |
3208 | &p->cls, &flow, wildcards, |
3209 | ntohs(ofm->priority))); | |
79eee1eb | 3210 | } |
064af421 | 3211 | |
79eee1eb BP |
3212 | static int |
3213 | send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn, | |
3214 | struct rule *rule, const struct ofp_flow_mod *ofm) | |
3215 | { | |
3216 | struct ofpbuf *packet; | |
3217 | uint16_t in_port; | |
3218 | flow_t flow; | |
3219 | int error; | |
064af421 | 3220 | |
79eee1eb BP |
3221 | if (ofm->buffer_id == htonl(UINT32_MAX)) { |
3222 | return 0; | |
064af421 | 3223 | } |
79eee1eb BP |
3224 | |
3225 | error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id), | |
3226 | &packet, &in_port); | |
3227 | if (error) { | |
3228 | return error; | |
3229 | } | |
3230 | ||
659586ef | 3231 | flow_extract(packet, 0, in_port, &flow); |
79eee1eb BP |
3232 | rule_execute(ofproto, rule, packet, &flow); |
3233 | ofpbuf_delete(packet); | |
3234 | ||
064af421 BP |
3235 | return 0; |
3236 | } | |
79eee1eb BP |
3237 | \f |
3238 | /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */ | |
064af421 BP |
3239 | |
3240 | struct modify_flows_cbdata { | |
3241 | struct ofproto *ofproto; | |
3242 | const struct ofp_flow_mod *ofm; | |
064af421 | 3243 | size_t n_actions; |
79eee1eb | 3244 | struct rule *match; |
064af421 BP |
3245 | }; |
3246 | ||
79eee1eb BP |
3247 | static int modify_flow(struct ofproto *, const struct ofp_flow_mod *, |
3248 | size_t n_actions, struct rule *); | |
3249 | static void modify_flows_cb(struct cls_rule *, void *cbdata_); | |
3250 | ||
3251 | /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as | |
3252 | * encoded by ofp_mkerr() on failure. | |
3253 | * | |
3254 | * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, | |
3255 | * if any. */ | |
3256 | static int | |
3257 | modify_flows_loose(struct ofproto *p, struct ofconn *ofconn, | |
3258 | const struct ofp_flow_mod *ofm, size_t n_actions) | |
3259 | { | |
3260 | struct modify_flows_cbdata cbdata; | |
3261 | struct cls_rule target; | |
3262 | ||
3263 | cbdata.ofproto = p; | |
3264 | cbdata.ofm = ofm; | |
3265 | cbdata.n_actions = n_actions; | |
3266 | cbdata.match = NULL; | |
3267 | ||
659586ef JG |
3268 | cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie, |
3269 | &target); | |
79eee1eb BP |
3270 | |
3271 | classifier_for_each_match(&p->cls, &target, CLS_INC_ALL, | |
3272 | modify_flows_cb, &cbdata); | |
3273 | if (cbdata.match) { | |
3274 | /* This credits the packet to whichever flow happened to happened to | |
3275 | * match last. That's weird. Maybe we should do a lookup for the | |
3276 | * flow that actually matches the packet? Who knows. */ | |
3277 | send_buffered_packet(p, ofconn, cbdata.match, ofm); | |
3278 | return 0; | |
3279 | } else { | |
3280 | return add_flow(p, ofconn, ofm, n_actions); | |
3281 | } | |
3282 | } | |
3283 | ||
3284 | /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error | |
3285 | * code as encoded by ofp_mkerr() on failure. | |
3286 | * | |
3287 | * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, | |
3288 | * if any. */ | |
3289 | static int | |
3290 | modify_flow_strict(struct ofproto *p, struct ofconn *ofconn, | |
3291 | struct ofp_flow_mod *ofm, size_t n_actions) | |
3292 | { | |
3293 | struct rule *rule = find_flow_strict(p, ofm); | |
3294 | if (rule && !rule_is_hidden(rule)) { | |
3295 | modify_flow(p, ofm, n_actions, rule); | |
3296 | return send_buffered_packet(p, ofconn, rule, ofm); | |
3297 | } else { | |
3298 | return add_flow(p, ofconn, ofm, n_actions); | |
3299 | } | |
3300 | } | |
3301 | ||
3302 | /* Callback for modify_flows_loose(). */ | |
064af421 BP |
3303 | static void |
3304 | modify_flows_cb(struct cls_rule *rule_, void *cbdata_) | |
3305 | { | |
3306 | struct rule *rule = rule_from_cls_rule(rule_); | |
3307 | struct modify_flows_cbdata *cbdata = cbdata_; | |
3308 | ||
79eee1eb BP |
3309 | if (!rule_is_hidden(rule)) { |
3310 | cbdata->match = rule; | |
3311 | modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule); | |
064af421 | 3312 | } |
064af421 BP |
3313 | } |
3314 | ||
79eee1eb BP |
3315 | /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has |
3316 | * been identified as a flow in 'p''s flow table to be modified, by changing | |
3317 | * the rule's actions to match those in 'ofm' (which is followed by 'n_actions' | |
3318 | * ofp_action[] structures). */ | |
064af421 | 3319 | static int |
79eee1eb BP |
3320 | modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm, |
3321 | size_t n_actions, struct rule *rule) | |
064af421 | 3322 | { |
79eee1eb BP |
3323 | size_t actions_len = n_actions * sizeof *rule->actions; |
3324 | ||
3325 | rule->flow_cookie = ofm->cookie; | |
3326 | ||
3327 | /* If the actions are the same, do nothing. */ | |
3328 | if (n_actions == rule->n_actions | |
3329 | && !memcmp(ofm->actions, rule->actions, actions_len)) | |
3330 | { | |
3331 | return 0; | |
3332 | } | |
3333 | ||
3334 | /* Replace actions. */ | |
3335 | free(rule->actions); | |
3336 | rule->actions = xmemdup(ofm->actions, actions_len); | |
3337 | rule->n_actions = n_actions; | |
3338 | ||
3339 | /* Make sure that the datapath gets updated properly. */ | |
3340 | if (rule->cr.wc.wildcards) { | |
3341 | COVERAGE_INC(ofproto_mod_wc_flow); | |
3342 | p->need_revalidate = true; | |
3343 | } else { | |
3344 | rule_update_actions(p, rule); | |
3345 | } | |
3346 | ||
3347 | return 0; | |
3348 | } | |
3349 | \f | |
3350 | /* OFPFC_DELETE implementation. */ | |
3351 | ||
3352 | struct delete_flows_cbdata { | |
3353 | struct ofproto *ofproto; | |
3354 | uint16_t out_port; | |
3355 | }; | |
3356 | ||
3357 | static void delete_flows_cb(struct cls_rule *, void *cbdata_); | |
3358 | static void delete_flow(struct ofproto *, struct rule *, uint16_t out_port); | |
3359 | ||
3360 | /* Implements OFPFC_DELETE. */ | |
3361 | static void | |
3362 | delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm) | |
3363 | { | |
3364 | struct delete_flows_cbdata cbdata; | |
064af421 BP |
3365 | struct cls_rule target; |
3366 | ||
3367 | cbdata.ofproto = p; | |
79eee1eb | 3368 | cbdata.out_port = ofm->out_port; |
064af421 | 3369 | |
659586ef JG |
3370 | cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie, |
3371 | &target); | |
064af421 BP |
3372 | |
3373 | classifier_for_each_match(&p->cls, &target, CLS_INC_ALL, | |
79eee1eb | 3374 | delete_flows_cb, &cbdata); |
064af421 BP |
3375 | } |
3376 | ||
79eee1eb BP |
3377 | /* Implements OFPFC_DELETE_STRICT. */ |
3378 | static void | |
3379 | delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm) | |
3380 | { | |
3381 | struct rule *rule = find_flow_strict(p, ofm); | |
3382 | if (rule) { | |
3383 | delete_flow(p, rule, ofm->out_port); | |
3384 | } | |
3385 | } | |
3386 | ||
3387 | /* Callback for delete_flows_loose(). */ | |
3388 | static void | |
3389 | delete_flows_cb(struct cls_rule *rule_, void *cbdata_) | |
3390 | { | |
3391 | struct rule *rule = rule_from_cls_rule(rule_); | |
3392 | struct delete_flows_cbdata *cbdata = cbdata_; | |
3393 | ||
3394 | delete_flow(cbdata->ofproto, rule, cbdata->out_port); | |
3395 | } | |
3396 | ||
3397 | /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has | |
3398 | * been identified as a flow to delete from 'p''s flow table, by deleting the | |
3399 | * flow and sending out a OFPT_FLOW_REMOVED message to any interested | |
3400 | * controller. | |
3401 | * | |
3402 | * Will not delete 'rule' if it is hidden. Will delete 'rule' only if | |
3403 | * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the | |
3404 | * specified 'out_port'. */ | |
3405 | static void | |
3406 | delete_flow(struct ofproto *p, struct rule *rule, uint16_t out_port) | |
3407 | { | |
3408 | if (rule_is_hidden(rule)) { | |
3409 | return; | |
3410 | } | |
3411 | ||
3412 | if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) { | |
3413 | return; | |
3414 | } | |
3415 | ||
3416 | send_flow_removed(p, rule, time_msec(), OFPRR_DELETE); | |
3417 | rule_remove(p, rule); | |
3418 | } | |
3419 | \f | |
064af421 BP |
3420 | static int |
3421 | handle_flow_mod(struct ofproto *p, struct ofconn *ofconn, | |
3422 | struct ofp_flow_mod *ofm) | |
3423 | { | |
3424 | size_t n_actions; | |
3425 | int error; | |
3426 | ||
9deba63b BP |
3427 | error = reject_slave_controller(ofconn, &ofm->header); |
3428 | if (error) { | |
3429 | return error; | |
3430 | } | |
064af421 BP |
3431 | error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm, |
3432 | sizeof *ofm->actions, &n_actions); | |
3433 | if (error) { | |
3434 | return error; | |
3435 | } | |
3436 | ||
49bdc010 JP |
3437 | /* We do not support the emergency flow cache. It will hopefully |
3438 | * get dropped from OpenFlow in the near future. */ | |
3439 | if (ofm->flags & htons(OFPFF_EMERG)) { | |
3440 | /* There isn't a good fit for an error code, so just state that the | |
3441 | * flow table is full. */ | |
3442 | return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL); | |
3443 | } | |
3444 | ||
064af421 BP |
3445 | normalize_match(&ofm->match); |
3446 | if (!ofm->match.wildcards) { | |
3447 | ofm->priority = htons(UINT16_MAX); | |
3448 | } | |
3449 | ||
3450 | error = validate_actions((const union ofp_action *) ofm->actions, | |
3451 | n_actions, p->max_ports); | |
3452 | if (error) { | |
3453 | return error; | |
3454 | } | |
3455 | ||
3456 | switch (ntohs(ofm->command)) { | |
3457 | case OFPFC_ADD: | |
3458 | return add_flow(p, ofconn, ofm, n_actions); | |
3459 | ||
3460 | case OFPFC_MODIFY: | |
79eee1eb | 3461 | return modify_flows_loose(p, ofconn, ofm, n_actions); |
064af421 BP |
3462 | |
3463 | case OFPFC_MODIFY_STRICT: | |
79eee1eb | 3464 | return modify_flow_strict(p, ofconn, ofm, n_actions); |
064af421 BP |
3465 | |
3466 | case OFPFC_DELETE: | |
79eee1eb BP |
3467 | delete_flows_loose(p, ofm); |
3468 | return 0; | |
064af421 BP |
3469 | |
3470 | case OFPFC_DELETE_STRICT: | |
79eee1eb BP |
3471 | delete_flow_strict(p, ofm); |
3472 | return 0; | |
064af421 BP |
3473 | |
3474 | default: | |
3475 | return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND); | |
3476 | } | |
3477 | } | |
3478 | ||
659586ef JG |
3479 | static int |
3480 | handle_tun_id_from_cookie(struct ofproto *p, struct nxt_tun_id_cookie *msg) | |
3481 | { | |
3482 | int error; | |
3483 | ||
3484 | error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg); | |
3485 | if (error) { | |
3486 | return error; | |
3487 | } | |
3488 | ||
3489 | p->tun_id_from_cookie = !!msg->set; | |
3490 | return 0; | |
3491 | } | |
3492 | ||
9deba63b BP |
3493 | static int |
3494 | handle_role_request(struct ofproto *ofproto, | |
3495 | struct ofconn *ofconn, struct nicira_header *msg) | |
3496 | { | |
3497 | struct nx_role_request *nrr; | |
3498 | struct nx_role_request *reply; | |
3499 | struct ofpbuf *buf; | |
3500 | uint32_t role; | |
3501 | ||
3502 | if (ntohs(msg->header.length) != sizeof *nrr) { | |
3503 | VLOG_WARN_RL(&rl, "received role request of length %zu (expected %zu)", | |
3504 | ntohs(msg->header.length), sizeof *nrr); | |
3505 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
3506 | } | |
3507 | nrr = (struct nx_role_request *) msg; | |
3508 | ||
3509 | if (ofconn->type != OFCONN_CONTROLLER) { | |
3510 | VLOG_WARN_RL(&rl, "ignoring role request on non-controller " | |
3511 | "connection"); | |
3512 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM); | |
3513 | } | |
3514 | ||
3515 | role = ntohl(nrr->role); | |
3516 | if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER | |
3517 | && role != NX_ROLE_SLAVE) { | |
3518 | VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role); | |
3519 | ||
3520 | /* There's no good error code for this. */ | |
3521 | return ofp_mkerr(OFPET_BAD_REQUEST, -1); | |
3522 | } | |
3523 | ||
3524 | if (role == NX_ROLE_MASTER) { | |
3525 | struct ofconn *other; | |
3526 | ||
3527 | HMAP_FOR_EACH (other, struct ofconn, hmap_node, | |
3528 | &ofproto->controllers) { | |
3529 | if (other->role == NX_ROLE_MASTER) { | |
3530 | other->role = NX_ROLE_SLAVE; | |
3531 | } | |
3532 | } | |
3533 | } | |
3534 | ofconn->role = role; | |
3535 | ||
3536 | reply = make_openflow_xid(sizeof *reply, OFPT_VENDOR, msg->header.xid, | |
3537 | &buf); | |
3538 | reply->nxh.vendor = htonl(NX_VENDOR_ID); | |
3539 | reply->nxh.subtype = htonl(NXT_ROLE_REPLY); | |
3540 | reply->role = htonl(role); | |
3541 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
3542 | ||
3543 | return 0; | |
3544 | } | |
3545 | ||
064af421 BP |
3546 | static int |
3547 | handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg) | |
3548 | { | |
3549 | struct ofp_vendor_header *ovh = msg; | |
3550 | struct nicira_header *nh; | |
3551 | ||
3552 | if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) { | |
659586ef JG |
3553 | VLOG_WARN_RL(&rl, "received vendor message of length %zu " |
3554 | "(expected at least %zu)", | |
3555 | ntohs(ovh->header.length), sizeof(struct ofp_vendor_header)); | |
49bdc010 | 3556 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); |
064af421 BP |
3557 | } |
3558 | if (ovh->vendor != htonl(NX_VENDOR_ID)) { | |
3559 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR); | |
3560 | } | |
3561 | if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) { | |
659586ef JG |
3562 | VLOG_WARN_RL(&rl, "received Nicira vendor message of length %zu " |
3563 | "(expected at least %zu)", | |
3564 | ntohs(ovh->header.length), sizeof(struct nicira_header)); | |
49bdc010 | 3565 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); |
064af421 BP |
3566 | } |
3567 | ||
3568 | nh = msg; | |
3569 | switch (ntohl(nh->subtype)) { | |
3570 | case NXT_STATUS_REQUEST: | |
3571 | return switch_status_handle_request(p->switch_status, ofconn->rconn, | |
3572 | msg); | |
659586ef JG |
3573 | |
3574 | case NXT_TUN_ID_FROM_COOKIE: | |
3575 | return handle_tun_id_from_cookie(p, msg); | |
9deba63b BP |
3576 | |
3577 | case NXT_ROLE_REQUEST: | |
3578 | return handle_role_request(p, ofconn, msg); | |
064af421 BP |
3579 | } |
3580 | ||
3581 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); | |
3582 | } | |
3583 | ||
246e61ea JP |
3584 | static int |
3585 | handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh) | |
3586 | { | |
3587 | struct ofp_header *ob; | |
3588 | struct ofpbuf *buf; | |
3589 | ||
3590 | /* Currently, everything executes synchronously, so we can just | |
3591 | * immediately send the barrier reply. */ | |
3592 | ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf); | |
3593 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
3594 | return 0; | |
3595 | } | |
3596 | ||
064af421 BP |
3597 | static void |
3598 | handle_openflow(struct ofconn *ofconn, struct ofproto *p, | |
3599 | struct ofpbuf *ofp_msg) | |
3600 | { | |
3601 | struct ofp_header *oh = ofp_msg->data; | |
3602 | int error; | |
3603 | ||
3604 | COVERAGE_INC(ofproto_recv_openflow); | |
3605 | switch (oh->type) { | |
3606 | case OFPT_ECHO_REQUEST: | |
3607 | error = handle_echo_request(ofconn, oh); | |
3608 | break; | |
3609 | ||
3610 | case OFPT_ECHO_REPLY: | |
3611 | error = 0; | |
3612 | break; | |
3613 | ||
3614 | case OFPT_FEATURES_REQUEST: | |
3615 | error = handle_features_request(p, ofconn, oh); | |
3616 | break; | |
3617 | ||
3618 | case OFPT_GET_CONFIG_REQUEST: | |
3619 | error = handle_get_config_request(p, ofconn, oh); | |
3620 | break; | |
3621 | ||
3622 | case OFPT_SET_CONFIG: | |
3623 | error = handle_set_config(p, ofconn, ofp_msg->data); | |
3624 | break; | |
3625 | ||
3626 | case OFPT_PACKET_OUT: | |
3627 | error = handle_packet_out(p, ofconn, ofp_msg->data); | |
3628 | break; | |
3629 | ||
3630 | case OFPT_PORT_MOD: | |
9deba63b | 3631 | error = handle_port_mod(p, ofconn, oh); |
064af421 BP |
3632 | break; |
3633 | ||
3634 | case OFPT_FLOW_MOD: | |
3635 | error = handle_flow_mod(p, ofconn, ofp_msg->data); | |
3636 | break; | |
3637 | ||
3638 | case OFPT_STATS_REQUEST: | |
3639 | error = handle_stats_request(p, ofconn, oh); | |
3640 | break; | |
3641 | ||
3642 | case OFPT_VENDOR: | |
3643 | error = handle_vendor(p, ofconn, ofp_msg->data); | |
3644 | break; | |
3645 | ||
246e61ea JP |
3646 | case OFPT_BARRIER_REQUEST: |
3647 | error = handle_barrier_request(ofconn, oh); | |
3648 | break; | |
3649 | ||
064af421 BP |
3650 | default: |
3651 | if (VLOG_IS_WARN_ENABLED()) { | |
3652 | char *s = ofp_to_string(oh, ntohs(oh->length), 2); | |
3653 | VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s); | |
3654 | free(s); | |
3655 | } | |
3656 | error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE); | |
3657 | break; | |
3658 | } | |
3659 | ||
3660 | if (error) { | |
3661 | send_error_oh(ofconn, ofp_msg->data, error); | |
3662 | } | |
3663 | } | |
3664 | \f | |
3665 | static void | |
72b06300 | 3666 | handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet) |
064af421 BP |
3667 | { |
3668 | struct odp_msg *msg = packet->data; | |
064af421 BP |
3669 | struct rule *rule; |
3670 | struct ofpbuf payload; | |
3671 | flow_t flow; | |
3672 | ||
064af421 BP |
3673 | payload.data = msg + 1; |
3674 | payload.size = msg->length - sizeof *msg; | |
659586ef | 3675 | flow_extract(&payload, msg->arg, msg->port, &flow); |
064af421 | 3676 | |
0ad9b732 JP |
3677 | /* Check with in-band control to see if this packet should be sent |
3678 | * to the local port regardless of the flow table. */ | |
3679 | if (in_band_msg_in_hook(p->in_band, &flow, &payload)) { | |
3680 | union odp_action action; | |
3681 | ||
3682 | memset(&action, 0, sizeof(action)); | |
3683 | action.output.type = ODPAT_OUTPUT; | |
3684 | action.output.port = ODPP_LOCAL; | |
f1acd62b | 3685 | dpif_execute(p->dpif, flow.in_port, &action, 1, &payload); |
0ad9b732 JP |
3686 | } |
3687 | ||
064af421 BP |
3688 | rule = lookup_valid_rule(p, &flow); |
3689 | if (!rule) { | |
3690 | /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */ | |
3691 | struct ofport *port = port_array_get(&p->ports, msg->port); | |
3692 | if (port) { | |
3693 | if (port->opp.config & OFPPC_NO_PACKET_IN) { | |
3694 | COVERAGE_INC(ofproto_no_packet_in); | |
3695 | /* XXX install 'drop' flow entry */ | |
3696 | ofpbuf_delete(packet); | |
3697 | return; | |
3698 | } | |
3699 | } else { | |
3700 | VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port); | |
3701 | } | |
3702 | ||
3703 | COVERAGE_INC(ofproto_packet_in); | |
76ce9432 | 3704 | send_packet_in(p, packet); |
064af421 BP |
3705 | return; |
3706 | } | |
3707 | ||
3708 | if (rule->cr.wc.wildcards) { | |
3709 | rule = rule_create_subrule(p, rule, &flow); | |
3710 | rule_make_actions(p, rule, packet); | |
3711 | } else { | |
3712 | if (!rule->may_install) { | |
3713 | /* The rule is not installable, that is, we need to process every | |
3714 | * packet, so process the current packet and set its actions into | |
3715 | * 'subrule'. */ | |
3716 | rule_make_actions(p, rule, packet); | |
3717 | } else { | |
3718 | /* XXX revalidate rule if it needs it */ | |
3719 | } | |
3720 | } | |
3721 | ||
3722 | rule_execute(p, rule, &payload, &flow); | |
3723 | rule_reinstall(p, rule); | |
7778bd15 | 3724 | |
76ce9432 | 3725 | if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY) { |
7778bd15 BP |
3726 | /* |
3727 | * Extra-special case for fail-open mode. | |
3728 | * | |
3729 | * We are in fail-open mode and the packet matched the fail-open rule, | |
3730 | * but we are connected to a controller too. We should send the packet | |
3731 | * up to the controller in the hope that it will try to set up a flow | |
3732 | * and thereby allow us to exit fail-open. | |
3733 | * | |
3734 | * See the top-level comment in fail-open.c for more information. | |
3735 | */ | |
76ce9432 | 3736 | send_packet_in(p, packet); |
7778bd15 BP |
3737 | } else { |
3738 | ofpbuf_delete(packet); | |
3739 | } | |
064af421 | 3740 | } |
72b06300 BP |
3741 | |
3742 | static void | |
3743 | handle_odp_msg(struct ofproto *p, struct ofpbuf *packet) | |
3744 | { | |
3745 | struct odp_msg *msg = packet->data; | |
3746 | ||
3747 | switch (msg->type) { | |
3748 | case _ODPL_ACTION_NR: | |
3749 | COVERAGE_INC(ofproto_ctlr_action); | |
76ce9432 | 3750 | send_packet_in(p, packet); |
72b06300 BP |
3751 | break; |
3752 | ||
3753 | case _ODPL_SFLOW_NR: | |
3754 | if (p->sflow) { | |
3755 | ofproto_sflow_received(p->sflow, msg); | |
3756 | } | |
3757 | ofpbuf_delete(packet); | |
3758 | break; | |
3759 | ||
3760 | case _ODPL_MISS_NR: | |
3761 | handle_odp_miss_msg(p, packet); | |
3762 | break; | |
3763 | ||
3764 | default: | |
3765 | VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32, | |
3766 | msg->type); | |
3767 | break; | |
3768 | } | |
3769 | } | |
064af421 BP |
3770 | \f |
3771 | static void | |
3772 | revalidate_cb(struct cls_rule *sub_, void *cbdata_) | |
3773 | { | |
3774 | struct rule *sub = rule_from_cls_rule(sub_); | |
3775 | struct revalidate_cbdata *cbdata = cbdata_; | |
3776 | ||
3777 | if (cbdata->revalidate_all | |
3778 | || (cbdata->revalidate_subrules && sub->super) | |
3779 | || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) { | |
3780 | revalidate_rule(cbdata->ofproto, sub); | |
3781 | } | |
3782 | } | |
3783 | ||
3784 | static bool | |
3785 | revalidate_rule(struct ofproto *p, struct rule *rule) | |
3786 | { | |
3787 | const flow_t *flow = &rule->cr.flow; | |
3788 | ||
3789 | COVERAGE_INC(ofproto_revalidate_rule); | |
3790 | if (rule->super) { | |
3791 | struct rule *super; | |
3792 | super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow)); | |
3793 | if (!super) { | |
3794 | rule_remove(p, rule); | |
3795 | return false; | |
3796 | } else if (super != rule->super) { | |
3797 | COVERAGE_INC(ofproto_revalidate_moved); | |
3798 | list_remove(&rule->list); | |
3799 | list_push_back(&super->list, &rule->list); | |
3800 | rule->super = super; | |
3801 | rule->hard_timeout = super->hard_timeout; | |
3802 | rule->idle_timeout = super->idle_timeout; | |
3803 | rule->created = super->created; | |
3804 | rule->used = 0; | |
3805 | } | |
3806 | } | |
3807 | ||
3808 | rule_update_actions(p, rule); | |
3809 | return true; | |
3810 | } | |
3811 | ||
3812 | static struct ofpbuf * | |
659586ef JG |
3813 | compose_flow_removed(struct ofproto *p, const struct rule *rule, |
3814 | long long int now, uint8_t reason) | |
064af421 | 3815 | { |
ca069229 | 3816 | struct ofp_flow_removed *ofr; |
064af421 | 3817 | struct ofpbuf *buf; |
9ca76894 | 3818 | long long int tdiff = now - rule->created; |
26c3f94a JP |
3819 | uint32_t sec = tdiff / 1000; |
3820 | uint32_t msec = tdiff - (sec * 1000); | |
064af421 | 3821 | |
ca069229 | 3822 | ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf); |
659586ef JG |
3823 | flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, p->tun_id_from_cookie, |
3824 | &ofr->match); | |
39997502 | 3825 | ofr->cookie = rule->flow_cookie; |
ca069229 JP |
3826 | ofr->priority = htons(rule->cr.priority); |
3827 | ofr->reason = reason; | |
26c3f94a JP |
3828 | ofr->duration_sec = htonl(sec); |
3829 | ofr->duration_nsec = htonl(msec * 1000000); | |
ca069229 JP |
3830 | ofr->idle_timeout = htons(rule->idle_timeout); |
3831 | ofr->packet_count = htonll(rule->packet_count); | |
3832 | ofr->byte_count = htonll(rule->byte_count); | |
064af421 BP |
3833 | |
3834 | return buf; | |
3835 | } | |
3836 | ||
3837 | static void | |
ca069229 JP |
3838 | uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule) |
3839 | { | |
3840 | assert(rule->installed); | |
3841 | assert(!rule->cr.wc.wildcards); | |
3842 | ||
3843 | if (rule->super) { | |
3844 | rule_remove(ofproto, rule); | |
3845 | } else { | |
3846 | rule_uninstall(ofproto, rule); | |
3847 | } | |
3848 | } | |
9deba63b | 3849 | |
ca069229 JP |
3850 | static void |
3851 | send_flow_removed(struct ofproto *p, struct rule *rule, | |
3852 | long long int now, uint8_t reason) | |
064af421 BP |
3853 | { |
3854 | struct ofconn *ofconn; | |
3855 | struct ofconn *prev; | |
b9b0ce61 | 3856 | struct ofpbuf *buf = NULL; |
064af421 BP |
3857 | |
3858 | /* We limit the maximum number of queued flow expirations it by accounting | |
3859 | * them under the counter for replies. That works because preventing | |
3860 | * OpenFlow requests from being processed also prevents new flows from | |
3861 | * being added (and expiring). (It also prevents processing OpenFlow | |
3862 | * requests that would not add new flows, so it is imperfect.) */ | |
3863 | ||
3864 | prev = NULL; | |
3865 | LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) { | |
9deba63b BP |
3866 | if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn) |
3867 | && ofconn->role != NX_ROLE_SLAVE) { | |
064af421 | 3868 | if (prev) { |
431d8ad2 | 3869 | queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter); |
064af421 | 3870 | } else { |
659586ef | 3871 | buf = compose_flow_removed(p, rule, now, reason); |
064af421 BP |
3872 | } |
3873 | prev = ofconn; | |
3874 | } | |
3875 | } | |
3876 | if (prev) { | |
431d8ad2 | 3877 | queue_tx(buf, prev, prev->reply_counter); |
064af421 BP |
3878 | } |
3879 | } | |
3880 | ||
064af421 BP |
3881 | |
3882 | static void | |
3883 | expire_rule(struct cls_rule *cls_rule, void *p_) | |
3884 | { | |
3885 | struct ofproto *p = p_; | |
3886 | struct rule *rule = rule_from_cls_rule(cls_rule); | |
3887 | long long int hard_expire, idle_expire, expire, now; | |
3888 | ||
3889 | hard_expire = (rule->hard_timeout | |
3890 | ? rule->created + rule->hard_timeout * 1000 | |
3891 | : LLONG_MAX); | |
3892 | idle_expire = (rule->idle_timeout | |
3893 | && (rule->super || list_is_empty(&rule->list)) | |
3894 | ? rule->used + rule->idle_timeout * 1000 | |
3895 | : LLONG_MAX); | |
3896 | expire = MIN(hard_expire, idle_expire); | |
064af421 BP |
3897 | |
3898 | now = time_msec(); | |
3899 | if (now < expire) { | |
3900 | if (rule->installed && now >= rule->used + 5000) { | |
3901 | uninstall_idle_flow(p, rule); | |
0193b2af JG |
3902 | } else if (!rule->cr.wc.wildcards) { |
3903 | active_timeout(p, rule); | |
064af421 | 3904 | } |
0193b2af | 3905 | |
064af421 BP |
3906 | return; |
3907 | } | |
3908 | ||
3909 | COVERAGE_INC(ofproto_expired); | |
46d6f36f JG |
3910 | |
3911 | /* Update stats. This code will be a no-op if the rule expired | |
3912 | * due to an idle timeout. */ | |
064af421 | 3913 | if (rule->cr.wc.wildcards) { |
064af421 BP |
3914 | struct rule *subrule, *next; |
3915 | LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) { | |
3916 | rule_remove(p, subrule); | |
3917 | } | |
46d6f36f JG |
3918 | } else { |
3919 | rule_uninstall(p, rule); | |
064af421 BP |
3920 | } |
3921 | ||
8fe1a59d | 3922 | if (!rule_is_hidden(rule)) { |
ca069229 JP |
3923 | send_flow_removed(p, rule, now, |
3924 | (now >= hard_expire | |
3925 | ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT)); | |
8fe1a59d | 3926 | } |
064af421 BP |
3927 | rule_remove(p, rule); |
3928 | } | |
3929 | ||
0193b2af JG |
3930 | static void |
3931 | active_timeout(struct ofproto *ofproto, struct rule *rule) | |
3932 | { | |
3933 | if (ofproto->netflow && !is_controller_rule(rule) && | |
3934 | netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) { | |
3935 | struct ofexpired expired; | |
3936 | struct odp_flow odp_flow; | |
3937 | ||
3938 | /* Get updated flow stats. */ | |
3939 | memset(&odp_flow, 0, sizeof odp_flow); | |
094e1514 JG |
3940 | if (rule->installed) { |
3941 | odp_flow.key = rule->cr.flow; | |
3942 | odp_flow.flags = ODPFF_ZERO_TCP_FLAGS; | |
d65349ea | 3943 | dpif_flow_get(ofproto->dpif, &odp_flow); |
094e1514 JG |
3944 | |
3945 | if (odp_flow.stats.n_packets) { | |
3946 | update_time(ofproto, rule, &odp_flow.stats); | |
3947 | netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos, | |
3948 | odp_flow.stats.tcp_flags); | |
3949 | } | |
0193b2af JG |
3950 | } |
3951 | ||
3952 | expired.flow = rule->cr.flow; | |
3953 | expired.packet_count = rule->packet_count + | |
3954 | odp_flow.stats.n_packets; | |
3955 | expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes; | |
3956 | expired.used = rule->used; | |
3957 | ||
3958 | netflow_expire(ofproto->netflow, &rule->nf_flow, &expired); | |
3959 | ||
3960 | /* Schedule us to send the accumulated records once we have | |
3961 | * collected all of them. */ | |
3962 | poll_immediate_wake(); | |
3963 | } | |
3964 | } | |
3965 | ||
064af421 BP |
3966 | static void |
3967 | update_used(struct ofproto *p) | |
3968 | { | |
3969 | struct odp_flow *flows; | |
3970 | size_t n_flows; | |
3971 | size_t i; | |
3972 | int error; | |
3973 | ||
c228a364 | 3974 | error = dpif_flow_list_all(p->dpif, &flows, &n_flows); |
064af421 BP |
3975 | if (error) { |
3976 | return; | |
3977 | } | |
3978 | ||
3979 | for (i = 0; i < n_flows; i++) { | |
3980 | struct odp_flow *f = &flows[i]; | |
3981 | struct rule *rule; | |
3982 | ||
3983 | rule = rule_from_cls_rule( | |
3984 | classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX)); | |
3985 | if (!rule || !rule->installed) { | |
3986 | COVERAGE_INC(ofproto_unexpected_rule); | |
c228a364 | 3987 | dpif_flow_del(p->dpif, f); |
064af421 BP |
3988 | continue; |
3989 | } | |
3990 | ||
0193b2af | 3991 | update_time(p, rule, &f->stats); |
064af421 BP |
3992 | rule_account(p, rule, f->stats.n_bytes); |
3993 | } | |
3994 | free(flows); | |
3995 | } | |
3996 | ||
43253595 | 3997 | /* pinsched callback for sending 'packet' on 'ofconn'. */ |
064af421 | 3998 | static void |
76ce9432 | 3999 | do_send_packet_in(struct ofpbuf *packet, void *ofconn_) |
064af421 | 4000 | { |
76ce9432 | 4001 | struct ofconn *ofconn = ofconn_; |
43253595 BP |
4002 | |
4003 | rconn_send_with_limit(ofconn->rconn, packet, | |
4004 | ofconn->packet_in_counter, 100); | |
4005 | } | |
4006 | ||
4007 | /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and | |
4008 | * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s | |
4009 | * packet scheduler for sending. | |
4010 | * | |
30ea5d93 BP |
4011 | * 'max_len' specifies the maximum number of bytes of the packet to send on |
4012 | * 'ofconn' (INT_MAX specifies no limit). | |
4013 | * | |
43253595 BP |
4014 | * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise, |
4015 | * ownership is transferred to this function. */ | |
4016 | static void | |
30ea5d93 BP |
4017 | schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len, |
4018 | bool clone) | |
43253595 | 4019 | { |
76ce9432 | 4020 | struct ofproto *ofproto = ofconn->ofproto; |
43253595 BP |
4021 | struct ofp_packet_in *opi = packet->data; |
4022 | uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port)); | |
4023 | int send_len, trim_size; | |
76ce9432 | 4024 | uint32_t buffer_id; |
064af421 | 4025 | |
43253595 BP |
4026 | /* Get buffer. */ |
4027 | if (opi->reason == OFPR_ACTION) { | |
76ce9432 | 4028 | buffer_id = UINT32_MAX; |
43253595 BP |
4029 | } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) { |
4030 | buffer_id = pktbuf_get_null(); | |
76ce9432 | 4031 | } else { |
43253595 BP |
4032 | struct ofpbuf payload; |
4033 | payload.data = opi->data; | |
4034 | payload.size = packet->size - offsetof(struct ofp_packet_in, data); | |
4035 | buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port); | |
76ce9432 | 4036 | } |
372179d4 | 4037 | |
43253595 BP |
4038 | /* Figure out how much of the packet to send. */ |
4039 | send_len = ntohs(opi->total_len); | |
4040 | if (buffer_id != UINT32_MAX) { | |
4041 | send_len = MIN(send_len, ofconn->miss_send_len); | |
4042 | } | |
30ea5d93 | 4043 | send_len = MIN(send_len, max_len); |
064af421 | 4044 | |
43253595 BP |
4045 | /* Adjust packet length and clone if necessary. */ |
4046 | trim_size = offsetof(struct ofp_packet_in, data) + send_len; | |
4047 | if (clone) { | |
4048 | packet = ofpbuf_clone_data(packet->data, trim_size); | |
4049 | opi = packet->data; | |
4050 | } else { | |
4051 | packet->size = trim_size; | |
4052 | } | |
4053 | ||
4054 | /* Update packet headers. */ | |
4055 | opi->buffer_id = htonl(buffer_id); | |
4056 | update_openflow_length(packet); | |
4057 | ||
4058 | /* Hand over to packet scheduler. It might immediately call into | |
4059 | * do_send_packet_in() or it might buffer it for a while (until a later | |
4060 | * call to pinsched_run()). */ | |
4061 | pinsched_send(ofconn->schedulers[opi->reason], in_port, | |
4062 | packet, do_send_packet_in, ofconn); | |
064af421 BP |
4063 | } |
4064 | ||
43253595 BP |
4065 | /* Replace struct odp_msg header in 'packet' by equivalent struct |
4066 | * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as | |
4067 | * returned by dpif_recv()). | |
4068 | * | |
4069 | * The conversion is not complete: the caller still needs to trim any unneeded | |
4070 | * payload off the end of the buffer, set the length in the OpenFlow header, | |
4071 | * and set buffer_id. Those require us to know the controller settings and so | |
30ea5d93 BP |
4072 | * must be done on a per-controller basis. |
4073 | * | |
4074 | * Returns the maximum number of bytes of the packet that should be sent to | |
4075 | * the controller (INT_MAX if no limit). */ | |
4076 | static int | |
43253595 | 4077 | do_convert_to_packet_in(struct ofpbuf *packet) |
064af421 | 4078 | { |
76ce9432 | 4079 | struct odp_msg *msg = packet->data; |
43253595 BP |
4080 | struct ofp_packet_in *opi; |
4081 | uint8_t reason; | |
4082 | uint16_t total_len; | |
4083 | uint16_t in_port; | |
30ea5d93 | 4084 | int max_len; |
43253595 BP |
4085 | |
4086 | /* Extract relevant header fields */ | |
30ea5d93 BP |
4087 | if (msg->type == _ODPL_ACTION_NR) { |
4088 | reason = OFPR_ACTION; | |
4089 | max_len = msg->arg; | |
4090 | } else { | |
4091 | reason = OFPR_NO_MATCH; | |
4092 | max_len = INT_MAX; | |
4093 | } | |
43253595 BP |
4094 | total_len = msg->length - sizeof *msg; |
4095 | in_port = odp_port_to_ofp_port(msg->port); | |
4096 | ||
4097 | /* Repurpose packet buffer by overwriting header. */ | |
4098 | ofpbuf_pull(packet, sizeof(struct odp_msg)); | |
4099 | opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data)); | |
4100 | opi->header.version = OFP_VERSION; | |
4101 | opi->header.type = OFPT_PACKET_IN; | |
4102 | opi->total_len = htons(total_len); | |
4103 | opi->in_port = htons(in_port); | |
4104 | opi->reason = reason; | |
30ea5d93 BP |
4105 | |
4106 | return max_len; | |
43253595 BP |
4107 | } |
4108 | ||
4109 | /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or | |
4110 | * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller | |
4111 | * as necessary according to their individual configurations. | |
4112 | * | |
4113 | * 'packet' must have sufficient headroom to convert it into a struct | |
4114 | * ofp_packet_in (e.g. as returned by dpif_recv()). | |
4115 | * | |
4116 | * Takes ownership of 'packet'. */ | |
4117 | static void | |
4118 | send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet) | |
4119 | { | |
76ce9432 | 4120 | struct ofconn *ofconn, *prev; |
30ea5d93 | 4121 | int max_len; |
064af421 | 4122 | |
30ea5d93 | 4123 | max_len = do_convert_to_packet_in(packet); |
76ce9432 BP |
4124 | |
4125 | prev = NULL; | |
4126 | LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) { | |
9deba63b BP |
4127 | if (ofconn->role != NX_ROLE_SLAVE) { |
4128 | if (prev) { | |
30ea5d93 | 4129 | schedule_packet_in(prev, packet, max_len, true); |
9deba63b BP |
4130 | } |
4131 | prev = ofconn; | |
064af421 | 4132 | } |
76ce9432 BP |
4133 | } |
4134 | if (prev) { | |
30ea5d93 | 4135 | schedule_packet_in(prev, packet, max_len, false); |
76ce9432 BP |
4136 | } else { |
4137 | ofpbuf_delete(packet); | |
064af421 | 4138 | } |
064af421 BP |
4139 | } |
4140 | ||
4141 | static uint64_t | |
fa60c019 | 4142 | pick_datapath_id(const struct ofproto *ofproto) |
064af421 | 4143 | { |
fa60c019 | 4144 | const struct ofport *port; |
064af421 | 4145 | |
fa60c019 BP |
4146 | port = port_array_get(&ofproto->ports, ODPP_LOCAL); |
4147 | if (port) { | |
4148 | uint8_t ea[ETH_ADDR_LEN]; | |
4149 | int error; | |
4150 | ||
4151 | error = netdev_get_etheraddr(port->netdev, ea); | |
064af421 BP |
4152 | if (!error) { |
4153 | return eth_addr_to_uint64(ea); | |
4154 | } | |
4155 | VLOG_WARN("could not get MAC address for %s (%s)", | |
fa60c019 | 4156 | netdev_get_name(port->netdev), strerror(error)); |
064af421 | 4157 | } |
fa60c019 | 4158 | return ofproto->fallback_dpid; |
064af421 BP |
4159 | } |
4160 | ||
4161 | static uint64_t | |
4162 | pick_fallback_dpid(void) | |
4163 | { | |
4164 | uint8_t ea[ETH_ADDR_LEN]; | |
70150daf | 4165 | eth_addr_nicira_random(ea); |
064af421 BP |
4166 | return eth_addr_to_uint64(ea); |
4167 | } | |
4168 | \f | |
4169 | static bool | |
4170 | default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet, | |
4171 | struct odp_actions *actions, tag_type *tags, | |
6a07af36 | 4172 | uint16_t *nf_output_iface, void *ofproto_) |
064af421 BP |
4173 | { |
4174 | struct ofproto *ofproto = ofproto_; | |
4175 | int out_port; | |
4176 | ||
4177 | /* Drop frames for reserved multicast addresses. */ | |
4178 | if (eth_addr_is_reserved(flow->dl_dst)) { | |
4179 | return true; | |
4180 | } | |
4181 | ||
4182 | /* Learn source MAC (but don't try to learn from revalidation). */ | |
4183 | if (packet != NULL) { | |
4184 | tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src, | |
4185 | 0, flow->in_port); | |
4186 | if (rev_tag) { | |
4187 | /* The log messages here could actually be useful in debugging, | |
4188 | * so keep the rate limit relatively high. */ | |
4189 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); | |
4190 | VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16, | |
4191 | ETH_ADDR_ARGS(flow->dl_src), flow->in_port); | |
4192 | ofproto_revalidate(ofproto, rev_tag); | |
4193 | } | |
4194 | } | |
4195 | ||
4196 | /* Determine output port. */ | |
4197 | out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags); | |
4198 | if (out_port < 0) { | |
6a07af36 | 4199 | add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface); |
064af421 BP |
4200 | } else if (out_port != flow->in_port) { |
4201 | odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port; | |
6a07af36 | 4202 | *nf_output_iface = out_port; |
064af421 BP |
4203 | } else { |
4204 | /* Drop. */ | |
4205 | } | |
4206 | ||
4207 | return true; | |
4208 | } | |
4209 | ||
4210 | static const struct ofhooks default_ofhooks = { | |
4211 | NULL, | |
4212 | default_normal_ofhook_cb, | |
4213 | NULL, | |
4214 | NULL | |
4215 | }; |