]>
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> | |
9d82ec47 | 22 | #include <sys/socket.h> |
064af421 BP |
23 | #include <net/if.h> |
24 | #include <netinet/in.h> | |
25 | #include <stdbool.h> | |
26 | #include <stdlib.h> | |
27 | #include "classifier.h" | |
28 | #include "coverage.h" | |
29 | #include "discovery.h" | |
30 | #include "dpif.h" | |
4f2cad2c | 31 | #include "dynamic-string.h" |
064af421 BP |
32 | #include "fail-open.h" |
33 | #include "in-band.h" | |
34 | #include "mac-learning.h" | |
35 | #include "netdev.h" | |
36 | #include "netflow.h" | |
37 | #include "odp-util.h" | |
38 | #include "ofp-print.h" | |
72b06300 | 39 | #include "ofproto-sflow.h" |
064af421 BP |
40 | #include "ofpbuf.h" |
41 | #include "openflow/nicira-ext.h" | |
42 | #include "openflow/openflow.h" | |
064af421 BP |
43 | #include "openvswitch/datapath-protocol.h" |
44 | #include "packets.h" | |
45 | #include "pinsched.h" | |
46 | #include "pktbuf.h" | |
47 | #include "poll-loop.h" | |
48 | #include "port-array.h" | |
49 | #include "rconn.h" | |
50 | #include "shash.h" | |
51 | #include "status.h" | |
52 | #include "stp.h" | |
fe55ad15 | 53 | #include "stream-ssl.h" |
064af421 BP |
54 | #include "svec.h" |
55 | #include "tag.h" | |
56 | #include "timeval.h" | |
4f2cad2c | 57 | #include "unixctl.h" |
064af421 | 58 | #include "vconn.h" |
064af421 BP |
59 | #include "xtoxll.h" |
60 | ||
61 | #define THIS_MODULE VLM_ofproto | |
62 | #include "vlog.h" | |
63 | ||
72b06300 | 64 | #include "sflow_api.h" |
064af421 BP |
65 | |
66 | enum { | |
67 | TABLEID_HASH = 0, | |
68 | TABLEID_CLASSIFIER = 1 | |
69 | }; | |
70 | ||
71 | struct ofport { | |
72 | struct netdev *netdev; | |
73 | struct ofp_phy_port opp; /* In host byte order. */ | |
74 | }; | |
75 | ||
76 | static void ofport_free(struct ofport *); | |
77 | static void hton_ofp_phy_port(struct ofp_phy_port *); | |
78 | ||
79 | static int xlate_actions(const union ofp_action *in, size_t n_in, | |
80 | const flow_t *flow, struct ofproto *ofproto, | |
81 | const struct ofpbuf *packet, | |
82 | struct odp_actions *out, tag_type *tags, | |
6a07af36 | 83 | bool *may_set_up_flow, uint16_t *nf_output_iface); |
064af421 BP |
84 | |
85 | struct rule { | |
86 | struct cls_rule cr; | |
87 | ||
39997502 JP |
88 | uint64_t flow_cookie; /* Controller-issued identifier. |
89 | (Kept in network-byte order.) */ | |
064af421 BP |
90 | uint16_t idle_timeout; /* In seconds from time of last use. */ |
91 | uint16_t hard_timeout; /* In seconds from time of creation. */ | |
ca069229 | 92 | bool send_flow_removed; /* Send a flow removed message? */ |
064af421 BP |
93 | long long int used; /* Last-used time (0 if never used). */ |
94 | long long int created; /* Creation time. */ | |
95 | uint64_t packet_count; /* Number of packets received. */ | |
96 | uint64_t byte_count; /* Number of bytes received. */ | |
97 | uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */ | |
064af421 | 98 | tag_type tags; /* Tags (set only by hooks). */ |
0193b2af | 99 | struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ |
064af421 BP |
100 | |
101 | /* If 'super' is non-NULL, this rule is a subrule, that is, it is an | |
102 | * exact-match rule (having cr.wc.wildcards of 0) generated from the | |
103 | * wildcard rule 'super'. In this case, 'list' is an element of the | |
104 | * super-rule's list. | |
105 | * | |
106 | * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of | |
107 | * a list of subrules. A super-rule with no wildcards (where | |
108 | * cr.wc.wildcards is 0) will never have any subrules. */ | |
109 | struct rule *super; | |
110 | struct list list; | |
111 | ||
112 | /* OpenFlow actions. | |
79eee1eb BP |
113 | * |
114 | * 'n_actions' is the number of elements in the 'actions' array. A single | |
115 | * action may take up more more than one element's worth of space. | |
064af421 BP |
116 | * |
117 | * A subrule has no actions (it uses the super-rule's actions). */ | |
118 | int n_actions; | |
119 | union ofp_action *actions; | |
120 | ||
121 | /* Datapath actions. | |
122 | * | |
123 | * A super-rule with wildcard fields never has ODP actions (since the | |
124 | * datapath only supports exact-match flows). */ | |
125 | bool installed; /* Installed in datapath? */ | |
126 | bool may_install; /* True ordinarily; false if actions must | |
127 | * be reassessed for every packet. */ | |
128 | int n_odp_actions; | |
129 | union odp_action *odp_actions; | |
130 | }; | |
131 | ||
132 | static inline bool | |
133 | rule_is_hidden(const struct rule *rule) | |
134 | { | |
135 | /* Subrules are merely an implementation detail, so hide them from the | |
136 | * controller. */ | |
137 | if (rule->super != NULL) { | |
138 | return true; | |
139 | } | |
140 | ||
8cd4882f | 141 | /* Rules with priority higher than UINT16_MAX are set up by ofproto itself |
064af421 BP |
142 | * (e.g. by in-band control) and are intentionally hidden from the |
143 | * controller. */ | |
144 | if (rule->cr.priority > UINT16_MAX) { | |
145 | return true; | |
146 | } | |
147 | ||
148 | return false; | |
149 | } | |
150 | ||
0193b2af JG |
151 | static struct rule *rule_create(struct ofproto *, struct rule *super, |
152 | const union ofp_action *, size_t n_actions, | |
ca069229 | 153 | uint16_t idle_timeout, uint16_t hard_timeout, |
39997502 | 154 | uint64_t flow_cookie, bool send_flow_removed); |
064af421 BP |
155 | static void rule_free(struct rule *); |
156 | static void rule_destroy(struct ofproto *, struct rule *); | |
157 | static struct rule *rule_from_cls_rule(const struct cls_rule *); | |
158 | static void rule_insert(struct ofproto *, struct rule *, | |
159 | struct ofpbuf *packet, uint16_t in_port); | |
160 | static void rule_remove(struct ofproto *, struct rule *); | |
161 | static bool rule_make_actions(struct ofproto *, struct rule *, | |
162 | const struct ofpbuf *packet); | |
163 | static void rule_install(struct ofproto *, struct rule *, | |
164 | struct rule *displaced_rule); | |
165 | static void rule_uninstall(struct ofproto *, struct rule *); | |
166 | static void rule_post_uninstall(struct ofproto *, struct rule *); | |
ca069229 JP |
167 | static void send_flow_removed(struct ofproto *p, struct rule *rule, |
168 | long long int now, uint8_t reason); | |
064af421 | 169 | |
76ce9432 BP |
170 | /* ofproto supports two kinds of OpenFlow connections: |
171 | * | |
172 | * - "Controller connections": Connections to ordinary OpenFlow controllers. | |
173 | * ofproto maintains persistent connections to these controllers and by | |
174 | * default sends them asynchronous messages such as packet-ins. | |
175 | * | |
176 | * - "Transient connections", e.g. from ovs-ofctl. When these connections | |
177 | * drop, it is the other side's responsibility to reconnect them if | |
178 | * necessary. ofproto does not send them asynchronous messages by default. | |
179 | */ | |
180 | enum ofconn_type { | |
181 | OFCONN_CONTROLLER, /* An OpenFlow controller. */ | |
182 | OFCONN_TRANSIENT /* A transient connection. */ | |
183 | }; | |
064af421 | 184 | |
76ce9432 BP |
185 | /* An OpenFlow connection. */ |
186 | struct ofconn { | |
187 | struct ofproto *ofproto; /* The ofproto that owns this connection. */ | |
188 | struct list node; /* In struct ofproto's "all_conns" list. */ | |
189 | struct rconn *rconn; /* OpenFlow connection. */ | |
190 | enum ofconn_type type; /* Type. */ | |
191 | ||
192 | /* OFPT_PACKET_IN related data. */ | |
193 | struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */ | |
194 | struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */ | |
195 | struct pktbuf *pktbuf; /* OpenFlow packet buffers. */ | |
196 | int miss_send_len; /* Bytes to send of buffered packets. */ | |
197 | ||
198 | /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow | |
199 | * requests, and the maximum number before we stop reading OpenFlow | |
200 | * requests. */ | |
064af421 BP |
201 | #define OFCONN_REPLY_MAX 100 |
202 | struct rconn_packet_counter *reply_counter; | |
76ce9432 BP |
203 | |
204 | /* type == OFCONN_CONTROLLER only. */ | |
9deba63b | 205 | enum nx_role role; /* Role. */ |
76ce9432 BP |
206 | struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */ |
207 | struct discovery *discovery; /* Controller discovery object, if enabled. */ | |
208 | struct status_category *ss; /* Switch status category. */ | |
d2ede7bc | 209 | enum ofproto_band band; /* In-band or out-of-band? */ |
064af421 BP |
210 | }; |
211 | ||
76ce9432 BP |
212 | /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's |
213 | * "schedulers" array. Their values are 0 and 1, and their meanings and values | |
214 | * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In | |
215 | * case anything ever changes, check their values here. */ | |
216 | #define N_SCHEDULERS 2 | |
217 | BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0); | |
218 | BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR); | |
219 | BUILD_ASSERT_DECL(OFPR_ACTION == 1); | |
220 | BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR); | |
221 | ||
222 | static struct ofconn *ofconn_create(struct ofproto *, struct rconn *, | |
223 | enum ofconn_type); | |
c475ae67 | 224 | static void ofconn_destroy(struct ofconn *); |
064af421 BP |
225 | static void ofconn_run(struct ofconn *, struct ofproto *); |
226 | static void ofconn_wait(struct ofconn *); | |
c91248b3 BP |
227 | static bool ofconn_receives_async_msgs(const struct ofconn *); |
228 | ||
064af421 BP |
229 | static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn, |
230 | struct rconn_packet_counter *counter); | |
231 | ||
76ce9432 BP |
232 | static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg); |
233 | static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn); | |
234 | ||
064af421 BP |
235 | struct ofproto { |
236 | /* Settings. */ | |
237 | uint64_t datapath_id; /* Datapath ID. */ | |
238 | uint64_t fallback_dpid; /* Datapath ID if no better choice found. */ | |
5a719c38 JP |
239 | char *mfr_desc; /* Manufacturer. */ |
240 | char *hw_desc; /* Hardware. */ | |
241 | char *sw_desc; /* Software version. */ | |
242 | char *serial_desc; /* Serial number. */ | |
8abc4ed7 | 243 | char *dp_desc; /* Datapath description. */ |
064af421 BP |
244 | |
245 | /* Datapath. */ | |
c228a364 | 246 | struct dpif *dpif; |
e9e28be3 | 247 | struct netdev_monitor *netdev_monitor; |
064af421 BP |
248 | struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is |
249 | * OFP port nr. */ | |
250 | struct shash port_by_name; | |
251 | uint32_t max_ports; | |
252 | ||
253 | /* Configuration. */ | |
254 | struct switch_status *switch_status; | |
064af421 | 255 | struct fail_open *fail_open; |
064af421 | 256 | struct netflow *netflow; |
72b06300 | 257 | struct ofproto_sflow *sflow; |
064af421 | 258 | |
d2ede7bc BP |
259 | /* In-band control. */ |
260 | struct in_band *in_band; | |
261 | long long int next_in_band_update; | |
917e50e1 BP |
262 | struct sockaddr_in *extra_in_band_remotes; |
263 | size_t n_extra_remotes; | |
264 | ||
064af421 BP |
265 | /* Flow table. */ |
266 | struct classifier cls; | |
267 | bool need_revalidate; | |
268 | long long int next_expiration; | |
269 | struct tag_set revalidate_set; | |
659586ef | 270 | bool tun_id_from_cookie; |
064af421 BP |
271 | |
272 | /* OpenFlow connections. */ | |
76ce9432 BP |
273 | struct hmap controllers; /* Controller "struct ofconn"s. */ |
274 | struct list all_conns; /* Contains "struct ofconn"s. */ | |
064af421 BP |
275 | struct pvconn **listeners; |
276 | size_t n_listeners; | |
277 | struct pvconn **snoops; | |
278 | size_t n_snoops; | |
279 | ||
280 | /* Hooks for ovs-vswitchd. */ | |
281 | const struct ofhooks *ofhooks; | |
282 | void *aux; | |
283 | ||
284 | /* Used by default ofhooks. */ | |
285 | struct mac_learning *ml; | |
286 | }; | |
287 | ||
288 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
289 | ||
290 | static const struct ofhooks default_ofhooks; | |
291 | ||
fa60c019 | 292 | static uint64_t pick_datapath_id(const struct ofproto *); |
064af421 | 293 | static uint64_t pick_fallback_dpid(void); |
76ce9432 | 294 | |
064af421 | 295 | static void update_used(struct ofproto *); |
0193b2af JG |
296 | static void update_stats(struct ofproto *, struct rule *, |
297 | const struct odp_flow_stats *); | |
064af421 | 298 | static void expire_rule(struct cls_rule *, void *ofproto); |
0193b2af | 299 | static void active_timeout(struct ofproto *ofproto, struct rule *rule); |
064af421 BP |
300 | static bool revalidate_rule(struct ofproto *p, struct rule *rule); |
301 | static void revalidate_cb(struct cls_rule *rule_, void *p_); | |
302 | ||
303 | static void handle_odp_msg(struct ofproto *, struct ofpbuf *); | |
304 | ||
305 | static void handle_openflow(struct ofconn *, struct ofproto *, | |
306 | struct ofpbuf *); | |
307 | ||
72b06300 BP |
308 | static void refresh_port_groups(struct ofproto *); |
309 | ||
064af421 BP |
310 | static void update_port(struct ofproto *, const char *devname); |
311 | static int init_ports(struct ofproto *); | |
312 | static void reinit_ports(struct ofproto *); | |
313 | ||
314 | int | |
1a6f1e2a JG |
315 | ofproto_create(const char *datapath, const char *datapath_type, |
316 | const struct ofhooks *ofhooks, void *aux, | |
064af421 BP |
317 | struct ofproto **ofprotop) |
318 | { | |
064af421 BP |
319 | struct odp_stats stats; |
320 | struct ofproto *p; | |
c228a364 | 321 | struct dpif *dpif; |
064af421 BP |
322 | int error; |
323 | ||
324 | *ofprotop = NULL; | |
325 | ||
326 | /* Connect to datapath and start listening for messages. */ | |
1a6f1e2a | 327 | error = dpif_open(datapath, datapath_type, &dpif); |
064af421 BP |
328 | if (error) { |
329 | VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error)); | |
330 | return error; | |
331 | } | |
c228a364 | 332 | error = dpif_get_dp_stats(dpif, &stats); |
064af421 BP |
333 | if (error) { |
334 | VLOG_ERR("failed to obtain stats for datapath %s: %s", | |
335 | datapath, strerror(error)); | |
c228a364 | 336 | dpif_close(dpif); |
064af421 BP |
337 | return error; |
338 | } | |
72b06300 | 339 | error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW); |
064af421 BP |
340 | if (error) { |
341 | VLOG_ERR("failed to listen on datapath %s: %s", | |
342 | datapath, strerror(error)); | |
c228a364 | 343 | dpif_close(dpif); |
064af421 BP |
344 | return error; |
345 | } | |
c228a364 | 346 | dpif_flow_flush(dpif); |
8f24562a | 347 | dpif_recv_purge(dpif); |
064af421 BP |
348 | |
349 | /* Initialize settings. */ | |
ec6fde61 | 350 | p = xzalloc(sizeof *p); |
064af421 | 351 | p->fallback_dpid = pick_fallback_dpid(); |
fa60c019 | 352 | p->datapath_id = p->fallback_dpid; |
5a719c38 JP |
353 | p->mfr_desc = xstrdup(DEFAULT_MFR_DESC); |
354 | p->hw_desc = xstrdup(DEFAULT_HW_DESC); | |
355 | p->sw_desc = xstrdup(DEFAULT_SW_DESC); | |
356 | p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC); | |
23ff2821 | 357 | p->dp_desc = xstrdup(DEFAULT_DP_DESC); |
064af421 BP |
358 | |
359 | /* Initialize datapath. */ | |
360 | p->dpif = dpif; | |
8b61709d | 361 | p->netdev_monitor = netdev_monitor_create(); |
064af421 BP |
362 | port_array_init(&p->ports); |
363 | shash_init(&p->port_by_name); | |
364 | p->max_ports = stats.max_ports; | |
365 | ||
366 | /* Initialize submodules. */ | |
367 | p->switch_status = switch_status_create(p); | |
368 | p->in_band = NULL; | |
064af421 | 369 | p->fail_open = NULL; |
064af421 | 370 | p->netflow = NULL; |
72b06300 | 371 | p->sflow = NULL; |
064af421 BP |
372 | |
373 | /* Initialize flow table. */ | |
374 | classifier_init(&p->cls); | |
375 | p->need_revalidate = false; | |
376 | p->next_expiration = time_msec() + 1000; | |
377 | tag_set_init(&p->revalidate_set); | |
378 | ||
379 | /* Initialize OpenFlow connections. */ | |
380 | list_init(&p->all_conns); | |
76ce9432 | 381 | hmap_init(&p->controllers); |
064af421 BP |
382 | p->listeners = NULL; |
383 | p->n_listeners = 0; | |
384 | p->snoops = NULL; | |
385 | p->n_snoops = 0; | |
386 | ||
387 | /* Initialize hooks. */ | |
388 | if (ofhooks) { | |
389 | p->ofhooks = ofhooks; | |
390 | p->aux = aux; | |
391 | p->ml = NULL; | |
392 | } else { | |
393 | p->ofhooks = &default_ofhooks; | |
394 | p->aux = p; | |
395 | p->ml = mac_learning_create(); | |
396 | } | |
397 | ||
fa60c019 BP |
398 | /* Pick final datapath ID. */ |
399 | p->datapath_id = pick_datapath_id(p); | |
b123cc3c | 400 | VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id); |
fa60c019 | 401 | |
064af421 BP |
402 | *ofprotop = p; |
403 | return 0; | |
404 | } | |
405 | ||
406 | void | |
407 | ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id) | |
408 | { | |
409 | uint64_t old_dpid = p->datapath_id; | |
fa60c019 | 410 | p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p); |
064af421 | 411 | if (p->datapath_id != old_dpid) { |
76ce9432 BP |
412 | struct ofconn *ofconn; |
413 | ||
b123cc3c | 414 | VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id); |
76ce9432 BP |
415 | |
416 | /* Force all active connections to reconnect, since there is no way to | |
417 | * notify a controller that the datapath ID has changed. */ | |
418 | LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) { | |
419 | rconn_reconnect(ofconn->rconn); | |
420 | } | |
064af421 BP |
421 | } |
422 | } | |
423 | ||
76ce9432 BP |
424 | static bool |
425 | is_discovery_controller(const struct ofproto_controller *c) | |
426 | { | |
427 | return !strcmp(c->target, "discover"); | |
428 | } | |
429 | ||
430 | static bool | |
431 | is_in_band_controller(const struct ofproto_controller *c) | |
432 | { | |
433 | return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND; | |
434 | } | |
435 | ||
436 | /* Creates a new controller in 'ofproto'. Some of the settings are initially | |
437 | * drawn from 'c', but update_controller() needs to be called later to finish | |
438 | * the new ofconn's configuration. */ | |
439 | static void | |
440 | add_controller(struct ofproto *ofproto, const struct ofproto_controller *c) | |
441 | { | |
442 | struct discovery *discovery; | |
443 | struct ofconn *ofconn; | |
444 | ||
445 | if (is_discovery_controller(c)) { | |
446 | int error = discovery_create(c->accept_re, c->update_resolv_conf, | |
447 | ofproto->dpif, ofproto->switch_status, | |
448 | &discovery); | |
449 | if (error) { | |
450 | return; | |
451 | } | |
452 | } else { | |
453 | discovery = NULL; | |
454 | } | |
455 | ||
456 | ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_CONTROLLER); | |
457 | ofconn->pktbuf = pktbuf_create(); | |
458 | ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN; | |
459 | if (discovery) { | |
460 | ofconn->discovery = discovery; | |
461 | } else { | |
462 | rconn_connect(ofconn->rconn, c->target); | |
463 | } | |
464 | hmap_insert(&ofproto->controllers, &ofconn->hmap_node, | |
465 | hash_string(c->target, 0)); | |
466 | } | |
467 | ||
468 | /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's | |
469 | * target or turn discovery on or off (these are done by creating new ofconns | |
470 | * and deleting old ones), but it can update the rest of an ofconn's | |
471 | * settings. */ | |
472 | static void | |
473 | update_controller(struct ofconn *ofconn, const struct ofproto_controller *c) | |
064af421 | 474 | { |
76ce9432 BP |
475 | struct ofproto *ofproto = ofconn->ofproto; |
476 | int probe_interval; | |
477 | int i; | |
79c9f2ee | 478 | |
d2ede7bc BP |
479 | ofconn->band = (is_in_band_controller(c) |
480 | ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND); | |
481 | ||
76ce9432 | 482 | rconn_set_max_backoff(ofconn->rconn, c->max_backoff); |
79c9f2ee | 483 | |
76ce9432 BP |
484 | probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0; |
485 | rconn_set_probe_interval(ofconn->rconn, probe_interval); | |
79c9f2ee | 486 | |
76ce9432 BP |
487 | if (ofconn->discovery) { |
488 | discovery_set_update_resolv_conf(ofconn->discovery, | |
489 | c->update_resolv_conf); | |
490 | discovery_set_accept_controller_re(ofconn->discovery, c->accept_re); | |
491 | } | |
79c9f2ee | 492 | |
76ce9432 BP |
493 | for (i = 0; i < N_SCHEDULERS; i++) { |
494 | struct pinsched **s = &ofconn->schedulers[i]; | |
79c9f2ee | 495 | |
76ce9432 BP |
496 | if (c->rate_limit > 0) { |
497 | if (!*s) { | |
498 | *s = pinsched_create(c->rate_limit, c->burst_limit, | |
499 | ofproto->switch_status); | |
79c9f2ee | 500 | } else { |
76ce9432 | 501 | pinsched_set_limits(*s, c->rate_limit, c->burst_limit); |
79c9f2ee | 502 | } |
76ce9432 BP |
503 | } else { |
504 | pinsched_destroy(*s); | |
505 | *s = NULL; | |
79c9f2ee | 506 | } |
76ce9432 BP |
507 | } |
508 | } | |
79c9f2ee | 509 | |
76ce9432 BP |
510 | static const char * |
511 | ofconn_get_target(const struct ofconn *ofconn) | |
512 | { | |
513 | return ofconn->discovery ? "discover" : rconn_get_name(ofconn->rconn); | |
514 | } | |
515 | ||
516 | static struct ofconn * | |
517 | find_controller_by_target(struct ofproto *ofproto, const char *target) | |
518 | { | |
519 | struct ofconn *ofconn; | |
520 | ||
521 | HMAP_FOR_EACH_WITH_HASH (ofconn, struct ofconn, hmap_node, | |
522 | hash_string(target, 0), &ofproto->controllers) { | |
523 | if (!strcmp(ofconn_get_target(ofconn), target)) { | |
524 | return ofconn; | |
79c9f2ee | 525 | } |
064af421 | 526 | } |
76ce9432 BP |
527 | return NULL; |
528 | } | |
064af421 | 529 | |
d2ede7bc BP |
530 | static void |
531 | update_in_band_remotes(struct ofproto *ofproto) | |
532 | { | |
533 | const struct ofconn *ofconn; | |
534 | struct sockaddr_in *addrs; | |
917e50e1 | 535 | size_t max_addrs, n_addrs; |
d2ede7bc | 536 | bool discovery; |
917e50e1 | 537 | size_t i; |
d2ede7bc | 538 | |
917e50e1 BP |
539 | /* Allocate enough memory for as many remotes as we could possibly have. */ |
540 | max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers); | |
541 | addrs = xmalloc(max_addrs * sizeof *addrs); | |
d2ede7bc BP |
542 | n_addrs = 0; |
543 | ||
544 | /* Add all the remotes. */ | |
545 | discovery = false; | |
546 | HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &ofproto->controllers) { | |
547 | struct sockaddr_in *sin = &addrs[n_addrs]; | |
548 | ||
487ec65f BP |
549 | if (ofconn->band == OFPROTO_OUT_OF_BAND) { |
550 | continue; | |
551 | } | |
552 | ||
d2ede7bc BP |
553 | sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn); |
554 | if (sin->sin_addr.s_addr) { | |
555 | sin->sin_port = rconn_get_remote_port(ofconn->rconn); | |
556 | n_addrs++; | |
557 | } | |
558 | if (ofconn->discovery) { | |
559 | discovery = true; | |
560 | } | |
561 | } | |
917e50e1 BP |
562 | for (i = 0; i < ofproto->n_extra_remotes; i++) { |
563 | addrs[n_addrs++] = ofproto->extra_in_band_remotes[i]; | |
564 | } | |
d2ede7bc BP |
565 | |
566 | /* Create or update or destroy in-band. | |
567 | * | |
568 | * Ordinarily we only enable in-band if there's at least one remote | |
569 | * address, but discovery needs the in-band rules for DHCP to be installed | |
570 | * even before we know any remote addresses. */ | |
571 | if (n_addrs || discovery) { | |
572 | if (!ofproto->in_band) { | |
573 | in_band_create(ofproto, ofproto->dpif, ofproto->switch_status, | |
574 | &ofproto->in_band); | |
575 | } | |
40cae670 BP |
576 | if (ofproto->in_band) { |
577 | in_band_set_remotes(ofproto->in_band, addrs, n_addrs); | |
578 | } | |
d2ede7bc BP |
579 | ofproto->next_in_band_update = time_msec() + 1000; |
580 | } else { | |
581 | in_band_destroy(ofproto->in_band); | |
582 | ofproto->in_band = NULL; | |
583 | } | |
584 | ||
585 | /* Clean up. */ | |
586 | free(addrs); | |
587 | } | |
588 | ||
76ce9432 BP |
589 | void |
590 | ofproto_set_controllers(struct ofproto *p, | |
591 | const struct ofproto_controller *controllers, | |
592 | size_t n_controllers) | |
593 | { | |
594 | struct shash new_controllers; | |
76ce9432 BP |
595 | enum ofproto_fail_mode fail_mode; |
596 | struct ofconn *ofconn, *next; | |
597 | bool ss_exists; | |
76ce9432 | 598 | size_t i; |
79c9f2ee | 599 | |
76ce9432 BP |
600 | shash_init(&new_controllers); |
601 | for (i = 0; i < n_controllers; i++) { | |
602 | const struct ofproto_controller *c = &controllers[i]; | |
603 | ||
604 | shash_add_once(&new_controllers, c->target, &controllers[i]); | |
605 | if (!find_controller_by_target(p, c->target)) { | |
606 | add_controller(p, c); | |
607 | } | |
608 | } | |
609 | ||
76ce9432 BP |
610 | fail_mode = OFPROTO_FAIL_STANDALONE; |
611 | ss_exists = false; | |
612 | HMAP_FOR_EACH_SAFE (ofconn, next, struct ofconn, hmap_node, | |
613 | &p->controllers) { | |
614 | struct ofproto_controller *c; | |
615 | ||
616 | c = shash_find_data(&new_controllers, ofconn_get_target(ofconn)); | |
617 | if (!c) { | |
618 | ofconn_destroy(ofconn); | |
79c9f2ee | 619 | } else { |
76ce9432 | 620 | update_controller(ofconn, c); |
76ce9432 BP |
621 | if (ofconn->ss) { |
622 | ss_exists = true; | |
623 | } | |
76ce9432 BP |
624 | if (c->fail == OFPROTO_FAIL_SECURE) { |
625 | fail_mode = OFPROTO_FAIL_SECURE; | |
626 | } | |
627 | } | |
628 | } | |
629 | shash_destroy(&new_controllers); | |
630 | ||
d2ede7bc | 631 | update_in_band_remotes(p); |
76ce9432 BP |
632 | |
633 | if (!hmap_is_empty(&p->controllers) | |
634 | && fail_mode == OFPROTO_FAIL_STANDALONE) { | |
635 | struct rconn **rconns; | |
636 | size_t n; | |
79c9f2ee | 637 | |
79c9f2ee | 638 | if (!p->fail_open) { |
76ce9432 BP |
639 | p->fail_open = fail_open_create(p, p->switch_status); |
640 | } | |
641 | ||
642 | n = 0; | |
643 | rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns); | |
644 | HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &p->controllers) { | |
645 | rconns[n++] = ofconn->rconn; | |
79c9f2ee | 646 | } |
76ce9432 BP |
647 | |
648 | fail_open_set_controllers(p->fail_open, rconns, n); | |
649 | /* p->fail_open takes ownership of 'rconns'. */ | |
79c9f2ee BP |
650 | } else { |
651 | fail_open_destroy(p->fail_open); | |
652 | p->fail_open = NULL; | |
653 | } | |
654 | ||
76ce9432 BP |
655 | if (!hmap_is_empty(&p->controllers) && !ss_exists) { |
656 | ofconn = CONTAINER_OF(hmap_first(&p->controllers), | |
657 | struct ofconn, hmap_node); | |
658 | ofconn->ss = switch_status_register(p->switch_status, "remote", | |
659 | rconn_status_cb, ofconn->rconn); | |
79c9f2ee | 660 | } |
064af421 BP |
661 | } |
662 | ||
917e50e1 BP |
663 | static bool |
664 | any_extras_changed(const struct ofproto *ofproto, | |
665 | const struct sockaddr_in *extras, size_t n) | |
666 | { | |
667 | size_t i; | |
668 | ||
669 | if (n != ofproto->n_extra_remotes) { | |
670 | return true; | |
671 | } | |
672 | ||
673 | for (i = 0; i < n; i++) { | |
674 | const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i]; | |
675 | const struct sockaddr_in *new = &extras[i]; | |
676 | ||
677 | if (old->sin_addr.s_addr != new->sin_addr.s_addr || | |
678 | old->sin_port != new->sin_port) { | |
679 | return true; | |
680 | } | |
681 | } | |
682 | ||
683 | return false; | |
684 | } | |
685 | ||
686 | /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s | |
687 | * in-band control should guarantee access, in the same way that in-band | |
688 | * control guarantees access to OpenFlow controllers. */ | |
689 | void | |
690 | ofproto_set_extra_in_band_remotes(struct ofproto *ofproto, | |
691 | const struct sockaddr_in *extras, size_t n) | |
692 | { | |
693 | if (!any_extras_changed(ofproto, extras, n)) { | |
694 | return; | |
695 | } | |
696 | ||
697 | free(ofproto->extra_in_band_remotes); | |
698 | ofproto->n_extra_remotes = n; | |
699 | ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras); | |
700 | ||
701 | update_in_band_remotes(ofproto); | |
702 | } | |
703 | ||
064af421 BP |
704 | void |
705 | ofproto_set_desc(struct ofproto *p, | |
5a719c38 JP |
706 | const char *mfr_desc, const char *hw_desc, |
707 | const char *sw_desc, const char *serial_desc, | |
8abc4ed7 | 708 | const char *dp_desc) |
064af421 | 709 | { |
5a719c38 JP |
710 | struct ofp_desc_stats *ods; |
711 | ||
712 | if (mfr_desc) { | |
713 | if (strlen(mfr_desc) >= sizeof ods->mfr_desc) { | |
714 | VLOG_WARN("truncating mfr_desc, must be less than %zu characters", | |
715 | sizeof ods->mfr_desc); | |
716 | } | |
717 | free(p->mfr_desc); | |
718 | p->mfr_desc = xstrdup(mfr_desc); | |
064af421 | 719 | } |
5a719c38 JP |
720 | if (hw_desc) { |
721 | if (strlen(hw_desc) >= sizeof ods->hw_desc) { | |
722 | VLOG_WARN("truncating hw_desc, must be less than %zu characters", | |
723 | sizeof ods->hw_desc); | |
724 | } | |
725 | free(p->hw_desc); | |
726 | p->hw_desc = xstrdup(hw_desc); | |
064af421 | 727 | } |
5a719c38 JP |
728 | if (sw_desc) { |
729 | if (strlen(sw_desc) >= sizeof ods->sw_desc) { | |
730 | VLOG_WARN("truncating sw_desc, must be less than %zu characters", | |
731 | sizeof ods->sw_desc); | |
732 | } | |
733 | free(p->sw_desc); | |
734 | p->sw_desc = xstrdup(sw_desc); | |
735 | } | |
736 | if (serial_desc) { | |
737 | if (strlen(serial_desc) >= sizeof ods->serial_num) { | |
738 | VLOG_WARN("truncating serial_desc, must be less than %zu " | |
739 | "characters", | |
740 | sizeof ods->serial_num); | |
741 | } | |
742 | free(p->serial_desc); | |
743 | p->serial_desc = xstrdup(serial_desc); | |
064af421 | 744 | } |
8abc4ed7 | 745 | if (dp_desc) { |
5a719c38 JP |
746 | if (strlen(dp_desc) >= sizeof ods->dp_desc) { |
747 | VLOG_WARN("truncating dp_desc, must be less than %zu characters", | |
748 | sizeof ods->dp_desc); | |
749 | } | |
8abc4ed7 JP |
750 | free(p->dp_desc); |
751 | p->dp_desc = xstrdup(dp_desc); | |
752 | } | |
064af421 BP |
753 | } |
754 | ||
064af421 BP |
755 | static int |
756 | set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp, | |
757 | const struct svec *svec) | |
758 | { | |
759 | struct pvconn **pvconns = *pvconnsp; | |
760 | size_t n_pvconns = *n_pvconnsp; | |
761 | int retval = 0; | |
762 | size_t i; | |
763 | ||
764 | for (i = 0; i < n_pvconns; i++) { | |
765 | pvconn_close(pvconns[i]); | |
766 | } | |
767 | free(pvconns); | |
768 | ||
769 | pvconns = xmalloc(svec->n * sizeof *pvconns); | |
770 | n_pvconns = 0; | |
771 | for (i = 0; i < svec->n; i++) { | |
772 | const char *name = svec->names[i]; | |
773 | struct pvconn *pvconn; | |
774 | int error; | |
775 | ||
776 | error = pvconn_open(name, &pvconn); | |
777 | if (!error) { | |
778 | pvconns[n_pvconns++] = pvconn; | |
779 | } else { | |
780 | VLOG_ERR("failed to listen on %s: %s", name, strerror(error)); | |
781 | if (!retval) { | |
782 | retval = error; | |
783 | } | |
784 | } | |
785 | } | |
786 | ||
787 | *pvconnsp = pvconns; | |
788 | *n_pvconnsp = n_pvconns; | |
789 | ||
790 | return retval; | |
791 | } | |
792 | ||
793 | int | |
794 | ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners) | |
795 | { | |
796 | return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners); | |
797 | } | |
798 | ||
799 | int | |
800 | ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops) | |
801 | { | |
802 | return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops); | |
803 | } | |
804 | ||
805 | int | |
0193b2af JG |
806 | ofproto_set_netflow(struct ofproto *ofproto, |
807 | const struct netflow_options *nf_options) | |
064af421 | 808 | { |
76343538 | 809 | if (nf_options && nf_options->collectors.n) { |
064af421 BP |
810 | if (!ofproto->netflow) { |
811 | ofproto->netflow = netflow_create(); | |
812 | } | |
0193b2af | 813 | return netflow_set_options(ofproto->netflow, nf_options); |
064af421 BP |
814 | } else { |
815 | netflow_destroy(ofproto->netflow); | |
816 | ofproto->netflow = NULL; | |
817 | return 0; | |
818 | } | |
819 | } | |
820 | ||
72b06300 BP |
821 | void |
822 | ofproto_set_sflow(struct ofproto *ofproto, | |
823 | const struct ofproto_sflow_options *oso) | |
824 | { | |
825 | struct ofproto_sflow *os = ofproto->sflow; | |
826 | if (oso) { | |
827 | if (!os) { | |
828 | struct ofport *ofport; | |
829 | unsigned int odp_port; | |
830 | ||
831 | os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif); | |
832 | refresh_port_groups(ofproto); | |
833 | PORT_ARRAY_FOR_EACH (ofport, &ofproto->ports, odp_port) { | |
834 | ofproto_sflow_add_port(os, odp_port, | |
835 | netdev_get_name(ofport->netdev)); | |
836 | } | |
837 | } | |
838 | ofproto_sflow_set_options(os, oso); | |
839 | } else { | |
840 | ofproto_sflow_destroy(os); | |
841 | ofproto->sflow = NULL; | |
842 | } | |
843 | } | |
844 | ||
064af421 | 845 | int |
67a4917b | 846 | ofproto_set_stp(struct ofproto *ofproto OVS_UNUSED, bool enable_stp) |
064af421 BP |
847 | { |
848 | /* XXX */ | |
849 | if (enable_stp) { | |
850 | VLOG_WARN("STP is not yet implemented"); | |
851 | return EINVAL; | |
852 | } else { | |
853 | return 0; | |
854 | } | |
855 | } | |
856 | ||
064af421 BP |
857 | uint64_t |
858 | ofproto_get_datapath_id(const struct ofproto *ofproto) | |
859 | { | |
860 | return ofproto->datapath_id; | |
861 | } | |
862 | ||
76ce9432 BP |
863 | bool |
864 | ofproto_has_controller(const struct ofproto *ofproto) | |
064af421 | 865 | { |
76ce9432 | 866 | return !hmap_is_empty(&ofproto->controllers); |
064af421 BP |
867 | } |
868 | ||
869 | void | |
870 | ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners) | |
871 | { | |
872 | size_t i; | |
873 | ||
874 | for (i = 0; i < ofproto->n_listeners; i++) { | |
875 | svec_add(listeners, pvconn_get_name(ofproto->listeners[i])); | |
876 | } | |
877 | } | |
878 | ||
879 | void | |
880 | ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops) | |
881 | { | |
882 | size_t i; | |
883 | ||
884 | for (i = 0; i < ofproto->n_snoops; i++) { | |
885 | svec_add(snoops, pvconn_get_name(ofproto->snoops[i])); | |
886 | } | |
887 | } | |
888 | ||
889 | void | |
890 | ofproto_destroy(struct ofproto *p) | |
891 | { | |
892 | struct ofconn *ofconn, *next_ofconn; | |
893 | struct ofport *ofport; | |
894 | unsigned int port_no; | |
895 | size_t i; | |
896 | ||
897 | if (!p) { | |
898 | return; | |
899 | } | |
900 | ||
f7de2cdf | 901 | /* Destroy fail-open and in-band early, since they touch the classifier. */ |
79c9f2ee BP |
902 | fail_open_destroy(p->fail_open); |
903 | p->fail_open = NULL; | |
904 | ||
905 | in_band_destroy(p->in_band); | |
906 | p->in_band = NULL; | |
917e50e1 | 907 | free(p->extra_in_band_remotes); |
2f6d3445 | 908 | |
064af421 BP |
909 | ofproto_flush_flows(p); |
910 | classifier_destroy(&p->cls); | |
911 | ||
912 | LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node, | |
913 | &p->all_conns) { | |
c475ae67 | 914 | ofconn_destroy(ofconn); |
064af421 | 915 | } |
76ce9432 | 916 | hmap_destroy(&p->controllers); |
064af421 | 917 | |
c228a364 | 918 | dpif_close(p->dpif); |
e9e28be3 | 919 | netdev_monitor_destroy(p->netdev_monitor); |
064af421 BP |
920 | PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) { |
921 | ofport_free(ofport); | |
922 | } | |
923 | shash_destroy(&p->port_by_name); | |
924 | ||
925 | switch_status_destroy(p->switch_status); | |
064af421 | 926 | netflow_destroy(p->netflow); |
72b06300 | 927 | ofproto_sflow_destroy(p->sflow); |
064af421 | 928 | |
064af421 BP |
929 | for (i = 0; i < p->n_listeners; i++) { |
930 | pvconn_close(p->listeners[i]); | |
931 | } | |
932 | free(p->listeners); | |
933 | ||
934 | for (i = 0; i < p->n_snoops; i++) { | |
935 | pvconn_close(p->snoops[i]); | |
936 | } | |
937 | free(p->snoops); | |
938 | ||
939 | mac_learning_destroy(p->ml); | |
940 | ||
5a719c38 JP |
941 | free(p->mfr_desc); |
942 | free(p->hw_desc); | |
943 | free(p->sw_desc); | |
944 | free(p->serial_desc); | |
cb871ae0 JP |
945 | free(p->dp_desc); |
946 | ||
3b917492 BP |
947 | port_array_destroy(&p->ports); |
948 | ||
064af421 BP |
949 | free(p); |
950 | } | |
951 | ||
952 | int | |
953 | ofproto_run(struct ofproto *p) | |
954 | { | |
955 | int error = ofproto_run1(p); | |
956 | if (!error) { | |
957 | error = ofproto_run2(p, false); | |
958 | } | |
959 | return error; | |
960 | } | |
961 | ||
e9e28be3 BP |
962 | static void |
963 | process_port_change(struct ofproto *ofproto, int error, char *devname) | |
964 | { | |
965 | if (error == ENOBUFS) { | |
966 | reinit_ports(ofproto); | |
967 | } else if (!error) { | |
968 | update_port(ofproto, devname); | |
969 | free(devname); | |
970 | } | |
971 | } | |
972 | ||
e2bfacb6 BP |
973 | /* Returns a "preference level" for snooping 'ofconn'. A higher return value |
974 | * means that 'ofconn' is more interesting for monitoring than a lower return | |
975 | * value. */ | |
976 | static int | |
977 | snoop_preference(const struct ofconn *ofconn) | |
978 | { | |
979 | switch (ofconn->role) { | |
980 | case NX_ROLE_MASTER: | |
981 | return 3; | |
982 | case NX_ROLE_OTHER: | |
983 | return 2; | |
984 | case NX_ROLE_SLAVE: | |
985 | return 1; | |
986 | default: | |
987 | /* Shouldn't happen. */ | |
988 | return 0; | |
989 | } | |
990 | } | |
991 | ||
76ce9432 BP |
992 | /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'. |
993 | * Connects this vconn to a controller. */ | |
994 | static void | |
995 | add_snooper(struct ofproto *ofproto, struct vconn *vconn) | |
996 | { | |
e2bfacb6 | 997 | struct ofconn *ofconn, *best; |
76ce9432 | 998 | |
e2bfacb6 BP |
999 | /* Pick a controller for monitoring. */ |
1000 | best = NULL; | |
76ce9432 | 1001 | LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) { |
e2bfacb6 BP |
1002 | if (ofconn->type == OFCONN_CONTROLLER |
1003 | && (!best || snoop_preference(ofconn) > snoop_preference(best))) { | |
1004 | best = ofconn; | |
76ce9432 | 1005 | } |
e2bfacb6 | 1006 | } |
76ce9432 | 1007 | |
e2bfacb6 BP |
1008 | if (best) { |
1009 | rconn_add_monitor(best->rconn, vconn); | |
1010 | } else { | |
1011 | VLOG_INFO_RL(&rl, "no controller connection to snoop"); | |
1012 | vconn_close(vconn); | |
76ce9432 | 1013 | } |
76ce9432 BP |
1014 | } |
1015 | ||
064af421 BP |
1016 | int |
1017 | ofproto_run1(struct ofproto *p) | |
1018 | { | |
1019 | struct ofconn *ofconn, *next_ofconn; | |
1020 | char *devname; | |
1021 | int error; | |
1022 | int i; | |
1023 | ||
149f577a JG |
1024 | if (shash_is_empty(&p->port_by_name)) { |
1025 | init_ports(p); | |
1026 | } | |
1027 | ||
064af421 BP |
1028 | for (i = 0; i < 50; i++) { |
1029 | struct ofpbuf *buf; | |
1030 | int error; | |
1031 | ||
c228a364 | 1032 | error = dpif_recv(p->dpif, &buf); |
064af421 BP |
1033 | if (error) { |
1034 | if (error == ENODEV) { | |
1035 | /* Someone destroyed the datapath behind our back. The caller | |
1036 | * better destroy us and give up, because we're just going to | |
1037 | * spin from here on out. */ | |
39a559f2 BP |
1038 | static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5); |
1039 | VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally", | |
c228a364 | 1040 | dpif_name(p->dpif)); |
064af421 BP |
1041 | return ENODEV; |
1042 | } | |
1043 | break; | |
1044 | } | |
1045 | ||
1046 | handle_odp_msg(p, buf); | |
1047 | } | |
1048 | ||
e9e28be3 BP |
1049 | while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) { |
1050 | process_port_change(p, error, devname); | |
1051 | } | |
1052 | while ((error = netdev_monitor_poll(p->netdev_monitor, | |
1053 | &devname)) != EAGAIN) { | |
1054 | process_port_change(p, error, devname); | |
064af421 BP |
1055 | } |
1056 | ||
1057 | if (p->in_band) { | |
d2ede7bc BP |
1058 | if (time_msec() >= p->next_in_band_update) { |
1059 | update_in_band_remotes(p); | |
1060 | } | |
064af421 BP |
1061 | in_band_run(p->in_band); |
1062 | } | |
064af421 BP |
1063 | |
1064 | LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node, | |
1065 | &p->all_conns) { | |
1066 | ofconn_run(ofconn, p); | |
1067 | } | |
1068 | ||
7778bd15 BP |
1069 | /* Fail-open maintenance. Do this after processing the ofconns since |
1070 | * fail-open checks the status of the controller rconn. */ | |
1071 | if (p->fail_open) { | |
1072 | fail_open_run(p->fail_open); | |
1073 | } | |
1074 | ||
064af421 BP |
1075 | for (i = 0; i < p->n_listeners; i++) { |
1076 | struct vconn *vconn; | |
1077 | int retval; | |
1078 | ||
1079 | retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn); | |
1080 | if (!retval) { | |
76ce9432 BP |
1081 | ofconn_create(p, rconn_new_from_vconn("passive", vconn), |
1082 | OFCONN_TRANSIENT); | |
064af421 BP |
1083 | } else if (retval != EAGAIN) { |
1084 | VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval)); | |
1085 | } | |
1086 | } | |
1087 | ||
1088 | for (i = 0; i < p->n_snoops; i++) { | |
1089 | struct vconn *vconn; | |
1090 | int retval; | |
1091 | ||
1092 | retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn); | |
1093 | if (!retval) { | |
76ce9432 | 1094 | add_snooper(p, vconn); |
064af421 BP |
1095 | } else if (retval != EAGAIN) { |
1096 | VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval)); | |
1097 | } | |
1098 | } | |
1099 | ||
1100 | if (time_msec() >= p->next_expiration) { | |
1101 | COVERAGE_INC(ofproto_expiration); | |
1102 | p->next_expiration = time_msec() + 1000; | |
1103 | update_used(p); | |
1104 | ||
1105 | classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p); | |
1106 | ||
1107 | /* Let the hook know that we're at a stable point: all outstanding data | |
1108 | * in existing flows has been accounted to the account_cb. Thus, the | |
1109 | * hook can now reasonably do operations that depend on having accurate | |
1110 | * flow volume accounting (currently, that's just bond rebalancing). */ | |
1111 | if (p->ofhooks->account_checkpoint_cb) { | |
1112 | p->ofhooks->account_checkpoint_cb(p->aux); | |
1113 | } | |
1114 | } | |
1115 | ||
1116 | if (p->netflow) { | |
1117 | netflow_run(p->netflow); | |
1118 | } | |
72b06300 BP |
1119 | if (p->sflow) { |
1120 | ofproto_sflow_run(p->sflow); | |
1121 | } | |
064af421 BP |
1122 | |
1123 | return 0; | |
1124 | } | |
1125 | ||
1126 | struct revalidate_cbdata { | |
1127 | struct ofproto *ofproto; | |
1128 | bool revalidate_all; /* Revalidate all exact-match rules? */ | |
1129 | bool revalidate_subrules; /* Revalidate all exact-match subrules? */ | |
1130 | struct tag_set revalidate_set; /* Set of tags to revalidate. */ | |
1131 | }; | |
1132 | ||
1133 | int | |
1134 | ofproto_run2(struct ofproto *p, bool revalidate_all) | |
1135 | { | |
1136 | if (p->need_revalidate || revalidate_all | |
1137 | || !tag_set_is_empty(&p->revalidate_set)) { | |
1138 | struct revalidate_cbdata cbdata; | |
1139 | cbdata.ofproto = p; | |
1140 | cbdata.revalidate_all = revalidate_all; | |
1141 | cbdata.revalidate_subrules = p->need_revalidate; | |
1142 | cbdata.revalidate_set = p->revalidate_set; | |
1143 | tag_set_init(&p->revalidate_set); | |
1144 | COVERAGE_INC(ofproto_revalidate); | |
1145 | classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata); | |
1146 | p->need_revalidate = false; | |
1147 | } | |
1148 | ||
1149 | return 0; | |
1150 | } | |
1151 | ||
1152 | void | |
1153 | ofproto_wait(struct ofproto *p) | |
1154 | { | |
1155 | struct ofconn *ofconn; | |
1156 | size_t i; | |
1157 | ||
c228a364 | 1158 | dpif_recv_wait(p->dpif); |
e9e28be3 BP |
1159 | dpif_port_poll_wait(p->dpif); |
1160 | netdev_monitor_poll_wait(p->netdev_monitor); | |
064af421 BP |
1161 | LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) { |
1162 | ofconn_wait(ofconn); | |
1163 | } | |
1164 | if (p->in_band) { | |
7cf8b266 | 1165 | poll_timer_wait_until(p->next_in_band_update); |
064af421 BP |
1166 | in_band_wait(p->in_band); |
1167 | } | |
064af421 BP |
1168 | if (p->fail_open) { |
1169 | fail_open_wait(p->fail_open); | |
1170 | } | |
72b06300 BP |
1171 | if (p->sflow) { |
1172 | ofproto_sflow_wait(p->sflow); | |
1173 | } | |
064af421 BP |
1174 | if (!tag_set_is_empty(&p->revalidate_set)) { |
1175 | poll_immediate_wake(); | |
1176 | } | |
1177 | if (p->need_revalidate) { | |
1178 | /* Shouldn't happen, but if it does just go around again. */ | |
1179 | VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); | |
1180 | poll_immediate_wake(); | |
1181 | } else if (p->next_expiration != LLONG_MAX) { | |
7cf8b266 | 1182 | poll_timer_wait_until(p->next_expiration); |
064af421 BP |
1183 | } |
1184 | for (i = 0; i < p->n_listeners; i++) { | |
1185 | pvconn_wait(p->listeners[i]); | |
1186 | } | |
1187 | for (i = 0; i < p->n_snoops; i++) { | |
1188 | pvconn_wait(p->snoops[i]); | |
1189 | } | |
1190 | } | |
1191 | ||
1192 | void | |
1193 | ofproto_revalidate(struct ofproto *ofproto, tag_type tag) | |
1194 | { | |
1195 | tag_set_add(&ofproto->revalidate_set, tag); | |
1196 | } | |
1197 | ||
1198 | struct tag_set * | |
1199 | ofproto_get_revalidate_set(struct ofproto *ofproto) | |
1200 | { | |
1201 | return &ofproto->revalidate_set; | |
1202 | } | |
1203 | ||
1204 | bool | |
1205 | ofproto_is_alive(const struct ofproto *p) | |
1206 | { | |
76ce9432 | 1207 | return !hmap_is_empty(&p->controllers); |
064af421 BP |
1208 | } |
1209 | ||
1210 | int | |
1211 | ofproto_send_packet(struct ofproto *p, const flow_t *flow, | |
1212 | const union ofp_action *actions, size_t n_actions, | |
1213 | const struct ofpbuf *packet) | |
1214 | { | |
1215 | struct odp_actions odp_actions; | |
1216 | int error; | |
1217 | ||
1218 | error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions, | |
6a07af36 | 1219 | NULL, NULL, NULL); |
064af421 BP |
1220 | if (error) { |
1221 | return error; | |
1222 | } | |
1223 | ||
1224 | /* XXX Should we translate the dpif_execute() errno value into an OpenFlow | |
1225 | * error code? */ | |
c228a364 | 1226 | dpif_execute(p->dpif, flow->in_port, odp_actions.actions, |
064af421 BP |
1227 | odp_actions.n_actions, packet); |
1228 | return 0; | |
1229 | } | |
1230 | ||
1231 | void | |
1232 | ofproto_add_flow(struct ofproto *p, | |
1233 | const flow_t *flow, uint32_t wildcards, unsigned int priority, | |
1234 | const union ofp_action *actions, size_t n_actions, | |
1235 | int idle_timeout) | |
1236 | { | |
1237 | struct rule *rule; | |
0193b2af | 1238 | rule = rule_create(p, NULL, actions, n_actions, |
ca069229 | 1239 | idle_timeout >= 0 ? idle_timeout : 5 /* XXX */, |
39997502 | 1240 | 0, 0, false); |
659586ef | 1241 | cls_rule_from_flow(flow, wildcards, priority, &rule->cr); |
064af421 BP |
1242 | rule_insert(p, rule, NULL, 0); |
1243 | } | |
1244 | ||
1245 | void | |
1246 | ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow, | |
1247 | uint32_t wildcards, unsigned int priority) | |
1248 | { | |
1249 | struct rule *rule; | |
1250 | ||
1251 | rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls, | |
1252 | flow, wildcards, | |
1253 | priority)); | |
1254 | if (rule) { | |
1255 | rule_remove(ofproto, rule); | |
1256 | } | |
1257 | } | |
1258 | ||
1259 | static void | |
1260 | destroy_rule(struct cls_rule *rule_, void *ofproto_) | |
1261 | { | |
1262 | struct rule *rule = rule_from_cls_rule(rule_); | |
1263 | struct ofproto *ofproto = ofproto_; | |
1264 | ||
1265 | /* Mark the flow as not installed, even though it might really be | |
1266 | * installed, so that rule_remove() doesn't bother trying to uninstall it. | |
1267 | * There is no point in uninstalling it individually since we are about to | |
1268 | * blow away all the flows with dpif_flow_flush(). */ | |
1269 | rule->installed = false; | |
1270 | ||
1271 | rule_remove(ofproto, rule); | |
1272 | } | |
1273 | ||
1274 | void | |
1275 | ofproto_flush_flows(struct ofproto *ofproto) | |
1276 | { | |
1277 | COVERAGE_INC(ofproto_flush); | |
1278 | classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto); | |
c228a364 | 1279 | dpif_flow_flush(ofproto->dpif); |
064af421 BP |
1280 | if (ofproto->in_band) { |
1281 | in_band_flushed(ofproto->in_band); | |
1282 | } | |
1283 | if (ofproto->fail_open) { | |
1284 | fail_open_flushed(ofproto->fail_open); | |
1285 | } | |
1286 | } | |
1287 | \f | |
1288 | static void | |
1289 | reinit_ports(struct ofproto *p) | |
1290 | { | |
1291 | struct svec devnames; | |
1292 | struct ofport *ofport; | |
1293 | unsigned int port_no; | |
1294 | struct odp_port *odp_ports; | |
1295 | size_t n_odp_ports; | |
1296 | size_t i; | |
1297 | ||
1298 | svec_init(&devnames); | |
1299 | PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) { | |
1300 | svec_add (&devnames, (char *) ofport->opp.name); | |
1301 | } | |
c228a364 | 1302 | dpif_port_list(p->dpif, &odp_ports, &n_odp_ports); |
064af421 BP |
1303 | for (i = 0; i < n_odp_ports; i++) { |
1304 | svec_add (&devnames, odp_ports[i].devname); | |
1305 | } | |
1306 | free(odp_ports); | |
1307 | ||
1308 | svec_sort_unique(&devnames); | |
1309 | for (i = 0; i < devnames.n; i++) { | |
1310 | update_port(p, devnames.names[i]); | |
1311 | } | |
1312 | svec_destroy(&devnames); | |
1313 | } | |
1314 | ||
72b06300 | 1315 | static size_t |
064af421 BP |
1316 | refresh_port_group(struct ofproto *p, unsigned int group) |
1317 | { | |
1318 | uint16_t *ports; | |
1319 | size_t n_ports; | |
1320 | struct ofport *port; | |
1321 | unsigned int port_no; | |
1322 | ||
1323 | assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD); | |
1324 | ||
1325 | ports = xmalloc(port_array_count(&p->ports) * sizeof *ports); | |
1326 | n_ports = 0; | |
1327 | PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) { | |
1328 | if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) { | |
1329 | ports[n_ports++] = port_no; | |
1330 | } | |
1331 | } | |
c228a364 | 1332 | dpif_port_group_set(p->dpif, group, ports, n_ports); |
064af421 | 1333 | free(ports); |
72b06300 BP |
1334 | |
1335 | return n_ports; | |
064af421 BP |
1336 | } |
1337 | ||
1338 | static void | |
1339 | refresh_port_groups(struct ofproto *p) | |
1340 | { | |
72b06300 BP |
1341 | size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD); |
1342 | size_t n_all = refresh_port_group(p, DP_GROUP_ALL); | |
1343 | if (p->sflow) { | |
1344 | ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all); | |
1345 | } | |
064af421 BP |
1346 | } |
1347 | ||
1348 | static struct ofport * | |
1349 | make_ofport(const struct odp_port *odp_port) | |
1350 | { | |
149f577a | 1351 | struct netdev_options netdev_options; |
064af421 BP |
1352 | enum netdev_flags flags; |
1353 | struct ofport *ofport; | |
1354 | struct netdev *netdev; | |
1355 | bool carrier; | |
1356 | int error; | |
1357 | ||
149f577a JG |
1358 | memset(&netdev_options, 0, sizeof netdev_options); |
1359 | netdev_options.name = odp_port->devname; | |
1360 | netdev_options.ethertype = NETDEV_ETH_TYPE_NONE; | |
149f577a JG |
1361 | |
1362 | error = netdev_open(&netdev_options, &netdev); | |
064af421 BP |
1363 | if (error) { |
1364 | VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s " | |
1365 | "cannot be opened (%s)", | |
1366 | odp_port->devname, odp_port->port, | |
1367 | odp_port->devname, strerror(error)); | |
1368 | return NULL; | |
1369 | } | |
1370 | ||
1371 | ofport = xmalloc(sizeof *ofport); | |
1372 | ofport->netdev = netdev; | |
1373 | ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port); | |
80992a35 | 1374 | netdev_get_etheraddr(netdev, ofport->opp.hw_addr); |
064af421 BP |
1375 | memcpy(ofport->opp.name, odp_port->devname, |
1376 | MIN(sizeof ofport->opp.name, sizeof odp_port->devname)); | |
1377 | ofport->opp.name[sizeof ofport->opp.name - 1] = '\0'; | |
1378 | ||
1379 | netdev_get_flags(netdev, &flags); | |
1380 | ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN; | |
1381 | ||
1382 | netdev_get_carrier(netdev, &carrier); | |
1383 | ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN; | |
1384 | ||
1385 | netdev_get_features(netdev, | |
1386 | &ofport->opp.curr, &ofport->opp.advertised, | |
1387 | &ofport->opp.supported, &ofport->opp.peer); | |
1388 | return ofport; | |
1389 | } | |
1390 | ||
1391 | static bool | |
1392 | ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port) | |
1393 | { | |
1394 | if (port_array_get(&p->ports, odp_port->port)) { | |
1395 | VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath", | |
1396 | odp_port->port); | |
1397 | return true; | |
1398 | } else if (shash_find(&p->port_by_name, odp_port->devname)) { | |
1399 | VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath", | |
1400 | odp_port->devname); | |
1401 | return true; | |
1402 | } else { | |
1403 | return false; | |
1404 | } | |
1405 | } | |
1406 | ||
1407 | static int | |
1408 | ofport_equal(const struct ofport *a_, const struct ofport *b_) | |
1409 | { | |
1410 | const struct ofp_phy_port *a = &a_->opp; | |
1411 | const struct ofp_phy_port *b = &b_->opp; | |
1412 | ||
1413 | BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */ | |
1414 | return (a->port_no == b->port_no | |
1415 | && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr) | |
1416 | && !strcmp((char *) a->name, (char *) b->name) | |
1417 | && a->state == b->state | |
1418 | && a->config == b->config | |
1419 | && a->curr == b->curr | |
1420 | && a->advertised == b->advertised | |
1421 | && a->supported == b->supported | |
1422 | && a->peer == b->peer); | |
1423 | } | |
1424 | ||
1425 | static void | |
1426 | send_port_status(struct ofproto *p, const struct ofport *ofport, | |
1427 | uint8_t reason) | |
1428 | { | |
1429 | /* XXX Should limit the number of queued port status change messages. */ | |
1430 | struct ofconn *ofconn; | |
1431 | LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) { | |
1432 | struct ofp_port_status *ops; | |
1433 | struct ofpbuf *b; | |
1434 | ||
c91248b3 | 1435 | if (!ofconn_receives_async_msgs(ofconn)) { |
9deba63b BP |
1436 | continue; |
1437 | } | |
1438 | ||
064af421 BP |
1439 | ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b); |
1440 | ops->reason = reason; | |
1441 | ops->desc = ofport->opp; | |
1442 | hton_ofp_phy_port(&ops->desc); | |
1443 | queue_tx(b, ofconn, NULL); | |
1444 | } | |
1445 | if (p->ofhooks->port_changed_cb) { | |
1446 | p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux); | |
1447 | } | |
1448 | } | |
1449 | ||
1450 | static void | |
1451 | ofport_install(struct ofproto *p, struct ofport *ofport) | |
1452 | { | |
72b06300 BP |
1453 | uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no); |
1454 | const char *netdev_name = (const char *) ofport->opp.name; | |
1455 | ||
e9e28be3 | 1456 | netdev_monitor_add(p->netdev_monitor, ofport->netdev); |
72b06300 BP |
1457 | port_array_set(&p->ports, odp_port, ofport); |
1458 | shash_add(&p->port_by_name, netdev_name, ofport); | |
1459 | if (p->sflow) { | |
1460 | ofproto_sflow_add_port(p->sflow, odp_port, netdev_name); | |
1461 | } | |
064af421 BP |
1462 | } |
1463 | ||
1464 | static void | |
1465 | ofport_remove(struct ofproto *p, struct ofport *ofport) | |
1466 | { | |
72b06300 BP |
1467 | uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no); |
1468 | ||
e9e28be3 | 1469 | netdev_monitor_remove(p->netdev_monitor, ofport->netdev); |
72b06300 | 1470 | port_array_set(&p->ports, odp_port, NULL); |
064af421 BP |
1471 | shash_delete(&p->port_by_name, |
1472 | shash_find(&p->port_by_name, (char *) ofport->opp.name)); | |
72b06300 BP |
1473 | if (p->sflow) { |
1474 | ofproto_sflow_del_port(p->sflow, odp_port); | |
1475 | } | |
064af421 BP |
1476 | } |
1477 | ||
1478 | static void | |
1479 | ofport_free(struct ofport *ofport) | |
1480 | { | |
1481 | if (ofport) { | |
1482 | netdev_close(ofport->netdev); | |
1483 | free(ofport); | |
1484 | } | |
1485 | } | |
1486 | ||
1487 | static void | |
1488 | update_port(struct ofproto *p, const char *devname) | |
1489 | { | |
1490 | struct odp_port odp_port; | |
c874dc6d BP |
1491 | struct ofport *old_ofport; |
1492 | struct ofport *new_ofport; | |
064af421 BP |
1493 | int error; |
1494 | ||
1495 | COVERAGE_INC(ofproto_update_port); | |
c874dc6d BP |
1496 | |
1497 | /* Query the datapath for port information. */ | |
c228a364 | 1498 | error = dpif_port_query_by_name(p->dpif, devname, &odp_port); |
064af421 | 1499 | |
c874dc6d BP |
1500 | /* Find the old ofport. */ |
1501 | old_ofport = shash_find_data(&p->port_by_name, devname); | |
1502 | if (!error) { | |
1503 | if (!old_ofport) { | |
1504 | /* There's no port named 'devname' but there might be a port with | |
1505 | * the same port number. This could happen if a port is deleted | |
1506 | * and then a new one added in its place very quickly, or if a port | |
1507 | * is renamed. In the former case we want to send an OFPPR_DELETE | |
1508 | * and an OFPPR_ADD, and in the latter case we want to send a | |
1509 | * single OFPPR_MODIFY. We can distinguish the cases by comparing | |
1510 | * the old port's ifindex against the new port, or perhaps less | |
1511 | * reliably but more portably by comparing the old port's MAC | |
1512 | * against the new port's MAC. However, this code isn't that smart | |
1513 | * and always sends an OFPPR_MODIFY (XXX). */ | |
1514 | old_ofport = port_array_get(&p->ports, odp_port.port); | |
064af421 | 1515 | } |
c874dc6d | 1516 | } else if (error != ENOENT && error != ENODEV) { |
064af421 BP |
1517 | VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error " |
1518 | "%s", strerror(error)); | |
1519 | return; | |
1520 | } | |
c874dc6d BP |
1521 | |
1522 | /* Create a new ofport. */ | |
1523 | new_ofport = !error ? make_ofport(&odp_port) : NULL; | |
1524 | ||
1525 | /* Eliminate a few pathological cases. */ | |
1526 | if (!old_ofport && !new_ofport) { | |
1527 | return; | |
1528 | } else if (old_ofport && new_ofport) { | |
1529 | /* Most of the 'config' bits are OpenFlow soft state, but | |
1530 | * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow | |
1531 | * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and | |
1532 | * leaves the other bits 0.) */ | |
1533 | new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN; | |
1534 | ||
1535 | if (ofport_equal(old_ofport, new_ofport)) { | |
1536 | /* False alarm--no change. */ | |
1537 | ofport_free(new_ofport); | |
1538 | return; | |
1539 | } | |
1540 | } | |
1541 | ||
1542 | /* Now deal with the normal cases. */ | |
1543 | if (old_ofport) { | |
1544 | ofport_remove(p, old_ofport); | |
1545 | } | |
1546 | if (new_ofport) { | |
1547 | ofport_install(p, new_ofport); | |
1548 | } | |
1549 | send_port_status(p, new_ofport ? new_ofport : old_ofport, | |
1550 | (!old_ofport ? OFPPR_ADD | |
1551 | : !new_ofport ? OFPPR_DELETE | |
1552 | : OFPPR_MODIFY)); | |
1553 | ofport_free(old_ofport); | |
1554 | ||
1555 | /* Update port groups. */ | |
064af421 BP |
1556 | refresh_port_groups(p); |
1557 | } | |
1558 | ||
1559 | static int | |
1560 | init_ports(struct ofproto *p) | |
1561 | { | |
1562 | struct odp_port *ports; | |
1563 | size_t n_ports; | |
1564 | size_t i; | |
1565 | int error; | |
1566 | ||
c228a364 | 1567 | error = dpif_port_list(p->dpif, &ports, &n_ports); |
064af421 BP |
1568 | if (error) { |
1569 | return error; | |
1570 | } | |
1571 | ||
1572 | for (i = 0; i < n_ports; i++) { | |
1573 | const struct odp_port *odp_port = &ports[i]; | |
1574 | if (!ofport_conflicts(p, odp_port)) { | |
1575 | struct ofport *ofport = make_ofport(odp_port); | |
1576 | if (ofport) { | |
1577 | ofport_install(p, ofport); | |
1578 | } | |
1579 | } | |
1580 | } | |
1581 | free(ports); | |
1582 | refresh_port_groups(p); | |
1583 | return 0; | |
1584 | } | |
1585 | \f | |
1586 | static struct ofconn * | |
76ce9432 | 1587 | ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type) |
064af421 | 1588 | { |
76ce9432 BP |
1589 | struct ofconn *ofconn = xzalloc(sizeof *ofconn); |
1590 | ofconn->ofproto = p; | |
064af421 BP |
1591 | list_push_back(&p->all_conns, &ofconn->node); |
1592 | ofconn->rconn = rconn; | |
76ce9432 | 1593 | ofconn->type = type; |
9deba63b | 1594 | ofconn->role = NX_ROLE_OTHER; |
76ce9432 | 1595 | ofconn->packet_in_counter = rconn_packet_counter_create (); |
064af421 | 1596 | ofconn->pktbuf = NULL; |
064af421 | 1597 | ofconn->miss_send_len = 0; |
064af421 BP |
1598 | ofconn->reply_counter = rconn_packet_counter_create (); |
1599 | return ofconn; | |
1600 | } | |
1601 | ||
1602 | static void | |
c475ae67 | 1603 | ofconn_destroy(struct ofconn *ofconn) |
064af421 | 1604 | { |
76ce9432 BP |
1605 | if (ofconn->type == OFCONN_CONTROLLER) { |
1606 | hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node); | |
1607 | } | |
1608 | discovery_destroy(ofconn->discovery); | |
1609 | ||
064af421 | 1610 | list_remove(&ofconn->node); |
76ce9432 | 1611 | switch_status_unregister(ofconn->ss); |
064af421 BP |
1612 | rconn_destroy(ofconn->rconn); |
1613 | rconn_packet_counter_destroy(ofconn->packet_in_counter); | |
1614 | rconn_packet_counter_destroy(ofconn->reply_counter); | |
1615 | pktbuf_destroy(ofconn->pktbuf); | |
1616 | free(ofconn); | |
1617 | } | |
1618 | ||
1619 | static void | |
1620 | ofconn_run(struct ofconn *ofconn, struct ofproto *p) | |
1621 | { | |
1622 | int iteration; | |
76ce9432 BP |
1623 | size_t i; |
1624 | ||
1625 | if (ofconn->discovery) { | |
1626 | char *controller_name; | |
1627 | if (rconn_is_connectivity_questionable(ofconn->rconn)) { | |
1628 | discovery_question_connectivity(ofconn->discovery); | |
1629 | } | |
1630 | if (discovery_run(ofconn->discovery, &controller_name)) { | |
1631 | if (controller_name) { | |
1632 | rconn_connect(ofconn->rconn, controller_name); | |
1633 | } else { | |
1634 | rconn_disconnect(ofconn->rconn); | |
1635 | } | |
1636 | } | |
1637 | } | |
1638 | ||
1639 | for (i = 0; i < N_SCHEDULERS; i++) { | |
1640 | pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn); | |
1641 | } | |
064af421 BP |
1642 | |
1643 | rconn_run(ofconn->rconn); | |
1644 | ||
1645 | if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) { | |
1646 | /* Limit the number of iterations to prevent other tasks from | |
1647 | * starving. */ | |
1648 | for (iteration = 0; iteration < 50; iteration++) { | |
1649 | struct ofpbuf *of_msg = rconn_recv(ofconn->rconn); | |
1650 | if (!of_msg) { | |
1651 | break; | |
1652 | } | |
7778bd15 BP |
1653 | if (p->fail_open) { |
1654 | fail_open_maybe_recover(p->fail_open); | |
1655 | } | |
064af421 BP |
1656 | handle_openflow(ofconn, p, of_msg); |
1657 | ofpbuf_delete(of_msg); | |
1658 | } | |
1659 | } | |
1660 | ||
76ce9432 | 1661 | if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) { |
c475ae67 | 1662 | ofconn_destroy(ofconn); |
064af421 BP |
1663 | } |
1664 | } | |
1665 | ||
1666 | static void | |
1667 | ofconn_wait(struct ofconn *ofconn) | |
1668 | { | |
76ce9432 BP |
1669 | int i; |
1670 | ||
1671 | if (ofconn->discovery) { | |
1672 | discovery_wait(ofconn->discovery); | |
1673 | } | |
1674 | for (i = 0; i < N_SCHEDULERS; i++) { | |
1675 | pinsched_wait(ofconn->schedulers[i]); | |
1676 | } | |
064af421 BP |
1677 | rconn_run_wait(ofconn->rconn); |
1678 | if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) { | |
1679 | rconn_recv_wait(ofconn->rconn); | |
1680 | } else { | |
1681 | COVERAGE_INC(ofproto_ofconn_stuck); | |
1682 | } | |
1683 | } | |
c91248b3 BP |
1684 | |
1685 | /* Returns true if 'ofconn' should receive asynchronous messages. */ | |
1686 | static bool | |
1687 | ofconn_receives_async_msgs(const struct ofconn *ofconn) | |
1688 | { | |
1689 | if (ofconn->type == OFCONN_CONTROLLER) { | |
1690 | /* Ordinary controllers always get asynchronous messages unless they | |
1691 | * have configured themselves as "slaves". */ | |
1692 | return ofconn->role != NX_ROLE_SLAVE; | |
1693 | } else { | |
1694 | /* Transient connections don't get asynchronous messages unless they | |
1695 | * have explicitly asked for them by setting a nonzero miss send | |
1696 | * length. */ | |
1697 | return ofconn->miss_send_len > 0; | |
1698 | } | |
1699 | } | |
064af421 BP |
1700 | \f |
1701 | /* Caller is responsible for initializing the 'cr' member of the returned | |
1702 | * rule. */ | |
1703 | static struct rule * | |
0193b2af | 1704 | rule_create(struct ofproto *ofproto, struct rule *super, |
064af421 | 1705 | const union ofp_action *actions, size_t n_actions, |
ca069229 | 1706 | uint16_t idle_timeout, uint16_t hard_timeout, |
39997502 | 1707 | uint64_t flow_cookie, bool send_flow_removed) |
064af421 | 1708 | { |
ec6fde61 | 1709 | struct rule *rule = xzalloc(sizeof *rule); |
064af421 BP |
1710 | rule->idle_timeout = idle_timeout; |
1711 | rule->hard_timeout = hard_timeout; | |
39997502 | 1712 | rule->flow_cookie = flow_cookie; |
064af421 | 1713 | rule->used = rule->created = time_msec(); |
ca069229 | 1714 | rule->send_flow_removed = send_flow_removed; |
064af421 BP |
1715 | rule->super = super; |
1716 | if (super) { | |
1717 | list_push_back(&super->list, &rule->list); | |
1718 | } else { | |
1719 | list_init(&rule->list); | |
1720 | } | |
1721 | rule->n_actions = n_actions; | |
1722 | rule->actions = xmemdup(actions, n_actions * sizeof *actions); | |
0193b2af JG |
1723 | netflow_flow_clear(&rule->nf_flow); |
1724 | netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created); | |
1725 | ||
064af421 BP |
1726 | return rule; |
1727 | } | |
1728 | ||
1729 | static struct rule * | |
1730 | rule_from_cls_rule(const struct cls_rule *cls_rule) | |
1731 | { | |
1732 | return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL; | |
1733 | } | |
1734 | ||
1735 | static void | |
1736 | rule_free(struct rule *rule) | |
1737 | { | |
1738 | free(rule->actions); | |
1739 | free(rule->odp_actions); | |
1740 | free(rule); | |
1741 | } | |
1742 | ||
1743 | /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its | |
1744 | * super-rule's list of subrules. If 'rule' is a super-rule, also iterates | |
1745 | * through all of its subrules and revalidates them, destroying any that no | |
1746 | * longer has a super-rule (which is probably all of them). | |
1747 | * | |
1748 | * Before calling this function, the caller must make have removed 'rule' from | |
1749 | * the classifier. If 'rule' is an exact-match rule, the caller is also | |
1750 | * responsible for ensuring that it has been uninstalled from the datapath. */ | |
1751 | static void | |
1752 | rule_destroy(struct ofproto *ofproto, struct rule *rule) | |
1753 | { | |
1754 | if (!rule->super) { | |
1755 | struct rule *subrule, *next; | |
1756 | LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) { | |
1757 | revalidate_rule(ofproto, subrule); | |
1758 | } | |
1759 | } else { | |
1760 | list_remove(&rule->list); | |
1761 | } | |
1762 | rule_free(rule); | |
1763 | } | |
1764 | ||
1765 | static bool | |
1766 | rule_has_out_port(const struct rule *rule, uint16_t out_port) | |
1767 | { | |
1768 | const union ofp_action *oa; | |
1769 | struct actions_iterator i; | |
1770 | ||
1771 | if (out_port == htons(OFPP_NONE)) { | |
1772 | return true; | |
1773 | } | |
1774 | for (oa = actions_first(&i, rule->actions, rule->n_actions); oa; | |
1775 | oa = actions_next(&i)) { | |
1776 | if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) { | |
1777 | return true; | |
1778 | } | |
1779 | } | |
1780 | return false; | |
1781 | } | |
1782 | ||
1783 | /* Executes the actions indicated by 'rule' on 'packet', which is in flow | |
1784 | * 'flow' and is considered to have arrived on ODP port 'in_port'. | |
1785 | * | |
1786 | * The flow that 'packet' actually contains does not need to actually match | |
1787 | * 'rule'; the actions in 'rule' will be applied to it either way. Likewise, | |
1788 | * the packet and byte counters for 'rule' will be credited for the packet sent | |
1789 | * out whether or not the packet actually matches 'rule'. | |
1790 | * | |
1791 | * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow, | |
1792 | * the caller must already have accurately composed ODP actions for it given | |
1793 | * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if | |
1794 | * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this | |
1795 | * function will compose a set of ODP actions based on 'rule''s OpenFlow | |
1796 | * actions and apply them to 'packet'. */ | |
1797 | static void | |
1798 | rule_execute(struct ofproto *ofproto, struct rule *rule, | |
1799 | struct ofpbuf *packet, const flow_t *flow) | |
1800 | { | |
1801 | const union odp_action *actions; | |
1802 | size_t n_actions; | |
1803 | struct odp_actions a; | |
1804 | ||
1805 | /* Grab or compose the ODP actions. | |
1806 | * | |
1807 | * The special case for an exact-match 'rule' where 'flow' is not the | |
1808 | * rule's flow is important to avoid, e.g., sending a packet out its input | |
1809 | * port simply because the ODP actions were composed for the wrong | |
1810 | * scenario. */ | |
1811 | if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) { | |
1812 | struct rule *super = rule->super ? rule->super : rule; | |
1813 | if (xlate_actions(super->actions, super->n_actions, flow, ofproto, | |
6a07af36 | 1814 | packet, &a, NULL, 0, NULL)) { |
064af421 BP |
1815 | return; |
1816 | } | |
1817 | actions = a.actions; | |
1818 | n_actions = a.n_actions; | |
1819 | } else { | |
1820 | actions = rule->odp_actions; | |
1821 | n_actions = rule->n_odp_actions; | |
1822 | } | |
1823 | ||
1824 | /* Execute the ODP actions. */ | |
c228a364 | 1825 | if (!dpif_execute(ofproto->dpif, flow->in_port, |
064af421 BP |
1826 | actions, n_actions, packet)) { |
1827 | struct odp_flow_stats stats; | |
1828 | flow_extract_stats(flow, packet, &stats); | |
0193b2af | 1829 | update_stats(ofproto, rule, &stats); |
064af421 | 1830 | rule->used = time_msec(); |
0193b2af | 1831 | netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used); |
064af421 BP |
1832 | } |
1833 | } | |
1834 | ||
1835 | static void | |
1836 | rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet, | |
1837 | uint16_t in_port) | |
1838 | { | |
1839 | struct rule *displaced_rule; | |
1840 | ||
1841 | /* Insert the rule in the classifier. */ | |
1842 | displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr)); | |
1843 | if (!rule->cr.wc.wildcards) { | |
1844 | rule_make_actions(p, rule, packet); | |
1845 | } | |
1846 | ||
1847 | /* Send the packet and credit it to the rule. */ | |
1848 | if (packet) { | |
1849 | flow_t flow; | |
659586ef | 1850 | flow_extract(packet, 0, in_port, &flow); |
064af421 BP |
1851 | rule_execute(p, rule, packet, &flow); |
1852 | } | |
1853 | ||
1854 | /* Install the rule in the datapath only after sending the packet, to | |
1855 | * avoid packet reordering. */ | |
1856 | if (rule->cr.wc.wildcards) { | |
1857 | COVERAGE_INC(ofproto_add_wc_flow); | |
1858 | p->need_revalidate = true; | |
1859 | } else { | |
1860 | rule_install(p, rule, displaced_rule); | |
1861 | } | |
1862 | ||
1863 | /* Free the rule that was displaced, if any. */ | |
1864 | if (displaced_rule) { | |
1865 | rule_destroy(p, displaced_rule); | |
1866 | } | |
1867 | } | |
1868 | ||
1869 | static struct rule * | |
1870 | rule_create_subrule(struct ofproto *ofproto, struct rule *rule, | |
1871 | const flow_t *flow) | |
1872 | { | |
0193b2af | 1873 | struct rule *subrule = rule_create(ofproto, rule, NULL, 0, |
ca069229 | 1874 | rule->idle_timeout, rule->hard_timeout, |
39997502 | 1875 | 0, false); |
064af421 | 1876 | COVERAGE_INC(ofproto_subrule_create); |
659586ef JG |
1877 | cls_rule_from_flow(flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX |
1878 | : rule->cr.priority), &subrule->cr); | |
064af421 BP |
1879 | classifier_insert_exact(&ofproto->cls, &subrule->cr); |
1880 | ||
1881 | return subrule; | |
1882 | } | |
1883 | ||
1884 | static void | |
1885 | rule_remove(struct ofproto *ofproto, struct rule *rule) | |
1886 | { | |
1887 | if (rule->cr.wc.wildcards) { | |
1888 | COVERAGE_INC(ofproto_del_wc_flow); | |
1889 | ofproto->need_revalidate = true; | |
1890 | } else { | |
1891 | rule_uninstall(ofproto, rule); | |
1892 | } | |
1893 | classifier_remove(&ofproto->cls, &rule->cr); | |
1894 | rule_destroy(ofproto, rule); | |
1895 | } | |
1896 | ||
1897 | /* Returns true if the actions changed, false otherwise. */ | |
1898 | static bool | |
1899 | rule_make_actions(struct ofproto *p, struct rule *rule, | |
1900 | const struct ofpbuf *packet) | |
1901 | { | |
1902 | const struct rule *super; | |
1903 | struct odp_actions a; | |
1904 | size_t actions_len; | |
1905 | ||
1906 | assert(!rule->cr.wc.wildcards); | |
1907 | ||
1908 | super = rule->super ? rule->super : rule; | |
1909 | rule->tags = 0; | |
1910 | xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p, | |
6a07af36 | 1911 | packet, &a, &rule->tags, &rule->may_install, |
0193b2af | 1912 | &rule->nf_flow.output_iface); |
064af421 BP |
1913 | |
1914 | actions_len = a.n_actions * sizeof *a.actions; | |
1915 | if (rule->n_odp_actions != a.n_actions | |
1916 | || memcmp(rule->odp_actions, a.actions, actions_len)) { | |
1917 | COVERAGE_INC(ofproto_odp_unchanged); | |
1918 | free(rule->odp_actions); | |
1919 | rule->n_odp_actions = a.n_actions; | |
1920 | rule->odp_actions = xmemdup(a.actions, actions_len); | |
1921 | return true; | |
1922 | } else { | |
1923 | return false; | |
1924 | } | |
1925 | } | |
1926 | ||
1927 | static int | |
1928 | do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags, | |
1929 | struct odp_flow_put *put) | |
1930 | { | |
1931 | memset(&put->flow.stats, 0, sizeof put->flow.stats); | |
1932 | put->flow.key = rule->cr.flow; | |
1933 | put->flow.actions = rule->odp_actions; | |
1934 | put->flow.n_actions = rule->n_odp_actions; | |
ab48643b | 1935 | put->flow.flags = 0; |
064af421 | 1936 | put->flags = flags; |
c228a364 | 1937 | return dpif_flow_put(ofproto->dpif, put); |
064af421 BP |
1938 | } |
1939 | ||
1940 | static void | |
1941 | rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule) | |
1942 | { | |
1943 | assert(!rule->cr.wc.wildcards); | |
1944 | ||
1945 | if (rule->may_install) { | |
1946 | struct odp_flow_put put; | |
1947 | if (!do_put_flow(p, rule, | |
1948 | ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS, | |
1949 | &put)) { | |
1950 | rule->installed = true; | |
1951 | if (displaced_rule) { | |
14986b31 | 1952 | update_stats(p, displaced_rule, &put.flow.stats); |
064af421 BP |
1953 | rule_post_uninstall(p, displaced_rule); |
1954 | } | |
1955 | } | |
1956 | } else if (displaced_rule) { | |
1957 | rule_uninstall(p, displaced_rule); | |
1958 | } | |
1959 | } | |
1960 | ||
1961 | static void | |
1962 | rule_reinstall(struct ofproto *ofproto, struct rule *rule) | |
1963 | { | |
1964 | if (rule->installed) { | |
1965 | struct odp_flow_put put; | |
1966 | COVERAGE_INC(ofproto_dp_missed); | |
1967 | do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put); | |
1968 | } else { | |
1969 | rule_install(ofproto, rule, NULL); | |
1970 | } | |
1971 | } | |
1972 | ||
1973 | static void | |
1974 | rule_update_actions(struct ofproto *ofproto, struct rule *rule) | |
1975 | { | |
42c3641c JG |
1976 | bool actions_changed; |
1977 | uint16_t new_out_iface, old_out_iface; | |
1978 | ||
1979 | old_out_iface = rule->nf_flow.output_iface; | |
1980 | actions_changed = rule_make_actions(ofproto, rule, NULL); | |
1981 | ||
064af421 BP |
1982 | if (rule->may_install) { |
1983 | if (rule->installed) { | |
1984 | if (actions_changed) { | |
064af421 | 1985 | struct odp_flow_put put; |
42c3641c JG |
1986 | do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY |
1987 | | ODPPF_ZERO_STATS, &put); | |
1988 | update_stats(ofproto, rule, &put.flow.stats); | |
1989 | ||
1990 | /* Temporarily set the old output iface so that NetFlow | |
1991 | * messages have the correct output interface for the old | |
1992 | * stats. */ | |
1993 | new_out_iface = rule->nf_flow.output_iface; | |
1994 | rule->nf_flow.output_iface = old_out_iface; | |
1995 | rule_post_uninstall(ofproto, rule); | |
1996 | rule->nf_flow.output_iface = new_out_iface; | |
064af421 BP |
1997 | } |
1998 | } else { | |
1999 | rule_install(ofproto, rule, NULL); | |
2000 | } | |
2001 | } else { | |
2002 | rule_uninstall(ofproto, rule); | |
2003 | } | |
2004 | } | |
2005 | ||
2006 | static void | |
2007 | rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes) | |
2008 | { | |
2009 | uint64_t total_bytes = rule->byte_count + extra_bytes; | |
2010 | ||
2011 | if (ofproto->ofhooks->account_flow_cb | |
2012 | && total_bytes > rule->accounted_bytes) | |
2013 | { | |
2014 | ofproto->ofhooks->account_flow_cb( | |
2015 | &rule->cr.flow, rule->odp_actions, rule->n_odp_actions, | |
2016 | total_bytes - rule->accounted_bytes, ofproto->aux); | |
2017 | rule->accounted_bytes = total_bytes; | |
2018 | } | |
2019 | } | |
2020 | ||
2021 | static void | |
2022 | rule_uninstall(struct ofproto *p, struct rule *rule) | |
2023 | { | |
2024 | assert(!rule->cr.wc.wildcards); | |
2025 | if (rule->installed) { | |
2026 | struct odp_flow odp_flow; | |
2027 | ||
2028 | odp_flow.key = rule->cr.flow; | |
2029 | odp_flow.actions = NULL; | |
2030 | odp_flow.n_actions = 0; | |
ab48643b | 2031 | odp_flow.flags = 0; |
c228a364 | 2032 | if (!dpif_flow_del(p->dpif, &odp_flow)) { |
0193b2af | 2033 | update_stats(p, rule, &odp_flow.stats); |
064af421 BP |
2034 | } |
2035 | rule->installed = false; | |
2036 | ||
2037 | rule_post_uninstall(p, rule); | |
2038 | } | |
2039 | } | |
2040 | ||
0193b2af JG |
2041 | static bool |
2042 | is_controller_rule(struct rule *rule) | |
2043 | { | |
2044 | /* If the only action is send to the controller then don't report | |
2045 | * NetFlow expiration messages since it is just part of the control | |
2046 | * logic for the network and not real traffic. */ | |
2047 | ||
2048 | if (rule && rule->super) { | |
2049 | struct rule *super = rule->super; | |
2050 | ||
2051 | return super->n_actions == 1 && | |
2052 | super->actions[0].type == htons(OFPAT_OUTPUT) && | |
2053 | super->actions[0].output.port == htons(OFPP_CONTROLLER); | |
2054 | } | |
2055 | ||
2056 | return false; | |
2057 | } | |
2058 | ||
064af421 BP |
2059 | static void |
2060 | rule_post_uninstall(struct ofproto *ofproto, struct rule *rule) | |
2061 | { | |
2062 | struct rule *super = rule->super; | |
2063 | ||
2064 | rule_account(ofproto, rule, 0); | |
6a07af36 | 2065 | |
0193b2af | 2066 | if (ofproto->netflow && !is_controller_rule(rule)) { |
064af421 BP |
2067 | struct ofexpired expired; |
2068 | expired.flow = rule->cr.flow; | |
2069 | expired.packet_count = rule->packet_count; | |
2070 | expired.byte_count = rule->byte_count; | |
2071 | expired.used = rule->used; | |
0193b2af | 2072 | netflow_expire(ofproto->netflow, &rule->nf_flow, &expired); |
064af421 BP |
2073 | } |
2074 | if (super) { | |
2075 | super->packet_count += rule->packet_count; | |
2076 | super->byte_count += rule->byte_count; | |
064af421 | 2077 | |
0c0afbec JG |
2078 | /* Reset counters to prevent double counting if the rule ever gets |
2079 | * reinstalled. */ | |
2080 | rule->packet_count = 0; | |
2081 | rule->byte_count = 0; | |
2082 | rule->accounted_bytes = 0; | |
0193b2af JG |
2083 | |
2084 | netflow_flow_clear(&rule->nf_flow); | |
0c0afbec | 2085 | } |
064af421 BP |
2086 | } |
2087 | \f | |
2088 | static void | |
2089 | queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn, | |
2090 | struct rconn_packet_counter *counter) | |
2091 | { | |
2092 | update_openflow_length(msg); | |
2093 | if (rconn_send(ofconn->rconn, msg, counter)) { | |
2094 | ofpbuf_delete(msg); | |
2095 | } | |
2096 | } | |
2097 | ||
2098 | static void | |
2099 | send_error(const struct ofconn *ofconn, const struct ofp_header *oh, | |
2100 | int error, const void *data, size_t len) | |
2101 | { | |
2102 | struct ofpbuf *buf; | |
2103 | struct ofp_error_msg *oem; | |
2104 | ||
2105 | if (!(error >> 16)) { | |
2106 | VLOG_WARN_RL(&rl, "not sending bad error code %d to controller", | |
2107 | error); | |
2108 | return; | |
2109 | } | |
2110 | ||
2111 | COVERAGE_INC(ofproto_error); | |
2112 | oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR, | |
2113 | oh ? oh->xid : 0, &buf); | |
2114 | oem->type = htons((unsigned int) error >> 16); | |
2115 | oem->code = htons(error & 0xffff); | |
2116 | memcpy(oem->data, data, len); | |
2117 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
2118 | } | |
2119 | ||
2120 | static void | |
2121 | send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh, | |
2122 | int error) | |
2123 | { | |
2124 | size_t oh_length = ntohs(oh->length); | |
2125 | send_error(ofconn, oh, error, oh, MIN(oh_length, 64)); | |
2126 | } | |
2127 | ||
2128 | static void | |
2129 | hton_ofp_phy_port(struct ofp_phy_port *opp) | |
2130 | { | |
2131 | opp->port_no = htons(opp->port_no); | |
2132 | opp->config = htonl(opp->config); | |
2133 | opp->state = htonl(opp->state); | |
2134 | opp->curr = htonl(opp->curr); | |
2135 | opp->advertised = htonl(opp->advertised); | |
2136 | opp->supported = htonl(opp->supported); | |
2137 | opp->peer = htonl(opp->peer); | |
2138 | } | |
2139 | ||
2140 | static int | |
2141 | handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh) | |
2142 | { | |
2143 | struct ofp_header *rq = oh; | |
2144 | queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter); | |
2145 | return 0; | |
2146 | } | |
2147 | ||
2148 | static int | |
2149 | handle_features_request(struct ofproto *p, struct ofconn *ofconn, | |
2150 | struct ofp_header *oh) | |
2151 | { | |
2152 | struct ofp_switch_features *osf; | |
2153 | struct ofpbuf *buf; | |
2154 | unsigned int port_no; | |
2155 | struct ofport *port; | |
2156 | ||
2157 | osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf); | |
2158 | osf->datapath_id = htonll(p->datapath_id); | |
2159 | osf->n_buffers = htonl(pktbuf_capacity()); | |
2160 | osf->n_tables = 2; | |
2161 | osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS | | |
0254ae23 | 2162 | OFPC_PORT_STATS | OFPC_ARP_MATCH_IP); |
064af421 BP |
2163 | osf->actions = htonl((1u << OFPAT_OUTPUT) | |
2164 | (1u << OFPAT_SET_VLAN_VID) | | |
2165 | (1u << OFPAT_SET_VLAN_PCP) | | |
2166 | (1u << OFPAT_STRIP_VLAN) | | |
2167 | (1u << OFPAT_SET_DL_SRC) | | |
2168 | (1u << OFPAT_SET_DL_DST) | | |
2169 | (1u << OFPAT_SET_NW_SRC) | | |
2170 | (1u << OFPAT_SET_NW_DST) | | |
959a2ecd | 2171 | (1u << OFPAT_SET_NW_TOS) | |
064af421 BP |
2172 | (1u << OFPAT_SET_TP_SRC) | |
2173 | (1u << OFPAT_SET_TP_DST)); | |
2174 | ||
2175 | PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) { | |
2176 | hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp)); | |
2177 | } | |
2178 | ||
2179 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
2180 | return 0; | |
2181 | } | |
2182 | ||
2183 | static int | |
2184 | handle_get_config_request(struct ofproto *p, struct ofconn *ofconn, | |
2185 | struct ofp_header *oh) | |
2186 | { | |
2187 | struct ofpbuf *buf; | |
2188 | struct ofp_switch_config *osc; | |
2189 | uint16_t flags; | |
2190 | bool drop_frags; | |
2191 | ||
2192 | /* Figure out flags. */ | |
c228a364 | 2193 | dpif_get_drop_frags(p->dpif, &drop_frags); |
064af421 | 2194 | flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL; |
064af421 BP |
2195 | |
2196 | /* Send reply. */ | |
2197 | osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf); | |
2198 | osc->flags = htons(flags); | |
2199 | osc->miss_send_len = htons(ofconn->miss_send_len); | |
2200 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
2201 | ||
2202 | return 0; | |
2203 | } | |
2204 | ||
2205 | static int | |
2206 | handle_set_config(struct ofproto *p, struct ofconn *ofconn, | |
2207 | struct ofp_switch_config *osc) | |
2208 | { | |
2209 | uint16_t flags; | |
2210 | int error; | |
2211 | ||
2212 | error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc); | |
2213 | if (error) { | |
2214 | return error; | |
2215 | } | |
2216 | flags = ntohs(osc->flags); | |
2217 | ||
9deba63b | 2218 | if (ofconn->type == OFCONN_CONTROLLER && ofconn->role != NX_ROLE_SLAVE) { |
064af421 BP |
2219 | switch (flags & OFPC_FRAG_MASK) { |
2220 | case OFPC_FRAG_NORMAL: | |
c228a364 | 2221 | dpif_set_drop_frags(p->dpif, false); |
064af421 BP |
2222 | break; |
2223 | case OFPC_FRAG_DROP: | |
c228a364 | 2224 | dpif_set_drop_frags(p->dpif, true); |
064af421 BP |
2225 | break; |
2226 | default: | |
2227 | VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")", | |
2228 | osc->flags); | |
2229 | break; | |
2230 | } | |
2231 | } | |
2232 | ||
064af421 BP |
2233 | ofconn->miss_send_len = ntohs(osc->miss_send_len); |
2234 | ||
2235 | return 0; | |
2236 | } | |
2237 | ||
2238 | static void | |
6a07af36 JG |
2239 | add_output_group_action(struct odp_actions *actions, uint16_t group, |
2240 | uint16_t *nf_output_iface) | |
064af421 BP |
2241 | { |
2242 | odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group; | |
6a07af36 JG |
2243 | |
2244 | if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) { | |
2245 | *nf_output_iface = NF_OUT_FLOOD; | |
2246 | } | |
064af421 BP |
2247 | } |
2248 | ||
2249 | static void | |
2250 | add_controller_action(struct odp_actions *actions, | |
2251 | const struct ofp_action_output *oao) | |
2252 | { | |
2253 | union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER); | |
30ea5d93 | 2254 | a->controller.arg = ntohs(oao->max_len); |
064af421 BP |
2255 | } |
2256 | ||
2257 | struct action_xlate_ctx { | |
2258 | /* Input. */ | |
e18fe8a2 | 2259 | flow_t flow; /* Flow to which these actions correspond. */ |
064af421 BP |
2260 | int recurse; /* Recursion level, via xlate_table_action. */ |
2261 | struct ofproto *ofproto; | |
2262 | const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a | |
2263 | * null pointer if we are revalidating | |
2264 | * without a packet to refer to. */ | |
2265 | ||
2266 | /* Output. */ | |
2267 | struct odp_actions *out; /* Datapath actions. */ | |
2268 | tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */ | |
d6fbec6d | 2269 | bool may_set_up_flow; /* True ordinarily; false if the actions must |
064af421 | 2270 | * be reassessed for every packet. */ |
6a07af36 | 2271 | uint16_t nf_output_iface; /* Output interface index for NetFlow. */ |
064af421 BP |
2272 | }; |
2273 | ||
2274 | static void do_xlate_actions(const union ofp_action *in, size_t n_in, | |
2275 | struct action_xlate_ctx *ctx); | |
2276 | ||
2277 | static void | |
2278 | add_output_action(struct action_xlate_ctx *ctx, uint16_t port) | |
2279 | { | |
2280 | const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port); | |
6cfaf517 BP |
2281 | |
2282 | if (ofport) { | |
2283 | if (ofport->opp.config & OFPPC_NO_FWD) { | |
2284 | /* Forwarding disabled on port. */ | |
2285 | return; | |
2286 | } | |
2287 | } else { | |
2288 | /* | |
2289 | * We don't have an ofport record for this port, but it doesn't hurt to | |
2290 | * allow forwarding to it anyhow. Maybe such a port will appear later | |
2291 | * and we're pre-populating the flow table. | |
2292 | */ | |
064af421 | 2293 | } |
6cfaf517 BP |
2294 | |
2295 | odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port; | |
6a07af36 | 2296 | ctx->nf_output_iface = port; |
064af421 BP |
2297 | } |
2298 | ||
2299 | static struct rule * | |
2300 | lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow) | |
2301 | { | |
2302 | struct rule *rule; | |
2303 | rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow)); | |
2304 | ||
2305 | /* The rule we found might not be valid, since we could be in need of | |
2306 | * revalidation. If it is not valid, don't return it. */ | |
2307 | if (rule | |
2308 | && rule->super | |
2309 | && ofproto->need_revalidate | |
2310 | && !revalidate_rule(ofproto, rule)) { | |
2311 | COVERAGE_INC(ofproto_invalidated); | |
2312 | return NULL; | |
2313 | } | |
2314 | ||
2315 | return rule; | |
2316 | } | |
2317 | ||
2318 | static void | |
2319 | xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port) | |
2320 | { | |
2321 | if (!ctx->recurse) { | |
2c5d1389 | 2322 | uint16_t old_in_port; |
064af421 | 2323 | struct rule *rule; |
064af421 | 2324 | |
2c5d1389 BP |
2325 | /* Look up a flow with 'in_port' as the input port. Then restore the |
2326 | * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will | |
2327 | * have surprising behavior). */ | |
2328 | old_in_port = ctx->flow.in_port; | |
e18fe8a2 BP |
2329 | ctx->flow.in_port = in_port; |
2330 | rule = lookup_valid_rule(ctx->ofproto, &ctx->flow); | |
2c5d1389 BP |
2331 | ctx->flow.in_port = old_in_port; |
2332 | ||
064af421 BP |
2333 | if (rule) { |
2334 | if (rule->super) { | |
2335 | rule = rule->super; | |
2336 | } | |
2337 | ||
2338 | ctx->recurse++; | |
2339 | do_xlate_actions(rule->actions, rule->n_actions, ctx); | |
2340 | ctx->recurse--; | |
2341 | } | |
2342 | } | |
2343 | } | |
2344 | ||
2345 | static void | |
2346 | xlate_output_action(struct action_xlate_ctx *ctx, | |
2347 | const struct ofp_action_output *oao) | |
2348 | { | |
2349 | uint16_t odp_port; | |
6a07af36 JG |
2350 | uint16_t prev_nf_output_iface = ctx->nf_output_iface; |
2351 | ||
2352 | ctx->nf_output_iface = NF_OUT_DROP; | |
064af421 BP |
2353 | |
2354 | switch (ntohs(oao->port)) { | |
2355 | case OFPP_IN_PORT: | |
e18fe8a2 | 2356 | add_output_action(ctx, ctx->flow.in_port); |
064af421 BP |
2357 | break; |
2358 | case OFPP_TABLE: | |
e18fe8a2 | 2359 | xlate_table_action(ctx, ctx->flow.in_port); |
064af421 BP |
2360 | break; |
2361 | case OFPP_NORMAL: | |
e18fe8a2 | 2362 | if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet, |
064af421 | 2363 | ctx->out, ctx->tags, |
6a07af36 | 2364 | &ctx->nf_output_iface, |
064af421 BP |
2365 | ctx->ofproto->aux)) { |
2366 | COVERAGE_INC(ofproto_uninstallable); | |
d6fbec6d | 2367 | ctx->may_set_up_flow = false; |
064af421 BP |
2368 | } |
2369 | break; | |
2370 | case OFPP_FLOOD: | |
6a07af36 JG |
2371 | add_output_group_action(ctx->out, DP_GROUP_FLOOD, |
2372 | &ctx->nf_output_iface); | |
064af421 BP |
2373 | break; |
2374 | case OFPP_ALL: | |
6a07af36 | 2375 | add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface); |
064af421 BP |
2376 | break; |
2377 | case OFPP_CONTROLLER: | |
2378 | add_controller_action(ctx->out, oao); | |
2379 | break; | |
2380 | case OFPP_LOCAL: | |
2381 | add_output_action(ctx, ODPP_LOCAL); | |
2382 | break; | |
2383 | default: | |
2384 | odp_port = ofp_port_to_odp_port(ntohs(oao->port)); | |
e18fe8a2 | 2385 | if (odp_port != ctx->flow.in_port) { |
064af421 BP |
2386 | add_output_action(ctx, odp_port); |
2387 | } | |
2388 | break; | |
2389 | } | |
6a07af36 JG |
2390 | |
2391 | if (prev_nf_output_iface == NF_OUT_FLOOD) { | |
2392 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
2393 | } else if (ctx->nf_output_iface == NF_OUT_DROP) { | |
2394 | ctx->nf_output_iface = prev_nf_output_iface; | |
2395 | } else if (prev_nf_output_iface != NF_OUT_DROP && | |
2396 | ctx->nf_output_iface != NF_OUT_FLOOD) { | |
2397 | ctx->nf_output_iface = NF_OUT_MULTI; | |
2398 | } | |
064af421 BP |
2399 | } |
2400 | ||
2401 | static void | |
2402 | xlate_nicira_action(struct action_xlate_ctx *ctx, | |
2403 | const struct nx_action_header *nah) | |
2404 | { | |
2405 | const struct nx_action_resubmit *nar; | |
659586ef JG |
2406 | const struct nx_action_set_tunnel *nast; |
2407 | union odp_action *oa; | |
064af421 BP |
2408 | int subtype = ntohs(nah->subtype); |
2409 | ||
2410 | assert(nah->vendor == htonl(NX_VENDOR_ID)); | |
2411 | switch (subtype) { | |
2412 | case NXAST_RESUBMIT: | |
2413 | nar = (const struct nx_action_resubmit *) nah; | |
2414 | xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port))); | |
2415 | break; | |
2416 | ||
659586ef JG |
2417 | case NXAST_SET_TUNNEL: |
2418 | nast = (const struct nx_action_set_tunnel *) nah; | |
2419 | oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL); | |
2420 | ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id; | |
2421 | break; | |
2422 | ||
999f0d45 BP |
2423 | /* If you add a new action here that modifies flow data, don't forget to |
2424 | * update the flow key in ctx->flow in the same key. */ | |
2425 | ||
064af421 BP |
2426 | default: |
2427 | VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype); | |
2428 | break; | |
2429 | } | |
2430 | } | |
2431 | ||
2432 | static void | |
2433 | do_xlate_actions(const union ofp_action *in, size_t n_in, | |
2434 | struct action_xlate_ctx *ctx) | |
2435 | { | |
2436 | struct actions_iterator iter; | |
2437 | const union ofp_action *ia; | |
2438 | const struct ofport *port; | |
2439 | ||
e18fe8a2 | 2440 | port = port_array_get(&ctx->ofproto->ports, ctx->flow.in_port); |
064af421 | 2441 | if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) && |
e18fe8a2 | 2442 | port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, stp_eth_addr) |
064af421 BP |
2443 | ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) { |
2444 | /* Drop this flow. */ | |
2445 | return; | |
2446 | } | |
2447 | ||
2448 | for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) { | |
2449 | uint16_t type = ntohs(ia->type); | |
2450 | union odp_action *oa; | |
2451 | ||
2452 | switch (type) { | |
2453 | case OFPAT_OUTPUT: | |
2454 | xlate_output_action(ctx, &ia->output); | |
2455 | break; | |
2456 | ||
2457 | case OFPAT_SET_VLAN_VID: | |
2458 | oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID); | |
999f0d45 | 2459 | ctx->flow.dl_vlan = oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid; |
064af421 BP |
2460 | break; |
2461 | ||
2462 | case OFPAT_SET_VLAN_PCP: | |
2463 | oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP); | |
999f0d45 | 2464 | ctx->flow.dl_vlan_pcp = oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp; |
064af421 BP |
2465 | break; |
2466 | ||
2467 | case OFPAT_STRIP_VLAN: | |
2468 | odp_actions_add(ctx->out, ODPAT_STRIP_VLAN); | |
999f0d45 BP |
2469 | ctx->flow.dl_vlan = OFP_VLAN_NONE; |
2470 | ctx->flow.dl_vlan_pcp = 0; | |
064af421 BP |
2471 | break; |
2472 | ||
2473 | case OFPAT_SET_DL_SRC: | |
2474 | oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC); | |
2475 | memcpy(oa->dl_addr.dl_addr, | |
2476 | ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); | |
999f0d45 BP |
2477 | memcpy(ctx->flow.dl_src, |
2478 | ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); | |
064af421 BP |
2479 | break; |
2480 | ||
2481 | case OFPAT_SET_DL_DST: | |
2482 | oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST); | |
2483 | memcpy(oa->dl_addr.dl_addr, | |
2484 | ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); | |
999f0d45 BP |
2485 | memcpy(ctx->flow.dl_dst, |
2486 | ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); | |
064af421 BP |
2487 | break; |
2488 | ||
2489 | case OFPAT_SET_NW_SRC: | |
2490 | oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC); | |
999f0d45 | 2491 | ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr; |
064af421 BP |
2492 | break; |
2493 | ||
2d70a31a JP |
2494 | case OFPAT_SET_NW_DST: |
2495 | oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST); | |
999f0d45 | 2496 | ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr; |
2d38e234 | 2497 | break; |
959a2ecd JP |
2498 | |
2499 | case OFPAT_SET_NW_TOS: | |
2500 | oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS); | |
999f0d45 | 2501 | ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos; |
2d70a31a JP |
2502 | break; |
2503 | ||
064af421 BP |
2504 | case OFPAT_SET_TP_SRC: |
2505 | oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC); | |
999f0d45 | 2506 | ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port; |
064af421 BP |
2507 | break; |
2508 | ||
2d70a31a JP |
2509 | case OFPAT_SET_TP_DST: |
2510 | oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST); | |
999f0d45 | 2511 | ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port; |
2d70a31a JP |
2512 | break; |
2513 | ||
064af421 BP |
2514 | case OFPAT_VENDOR: |
2515 | xlate_nicira_action(ctx, (const struct nx_action_header *) ia); | |
2516 | break; | |
2517 | ||
2518 | default: | |
2519 | VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type); | |
2520 | break; | |
2521 | } | |
2522 | } | |
2523 | } | |
2524 | ||
2525 | static int | |
2526 | xlate_actions(const union ofp_action *in, size_t n_in, | |
2527 | const flow_t *flow, struct ofproto *ofproto, | |
2528 | const struct ofpbuf *packet, | |
6a07af36 JG |
2529 | struct odp_actions *out, tag_type *tags, bool *may_set_up_flow, |
2530 | uint16_t *nf_output_iface) | |
064af421 BP |
2531 | { |
2532 | tag_type no_tags = 0; | |
2533 | struct action_xlate_ctx ctx; | |
2534 | COVERAGE_INC(ofproto_ofp2odp); | |
2535 | odp_actions_init(out); | |
e18fe8a2 | 2536 | ctx.flow = *flow; |
064af421 BP |
2537 | ctx.recurse = 0; |
2538 | ctx.ofproto = ofproto; | |
2539 | ctx.packet = packet; | |
2540 | ctx.out = out; | |
2541 | ctx.tags = tags ? tags : &no_tags; | |
d6fbec6d | 2542 | ctx.may_set_up_flow = true; |
6a07af36 | 2543 | ctx.nf_output_iface = NF_OUT_DROP; |
064af421 | 2544 | do_xlate_actions(in, n_in, &ctx); |
0ad9b732 | 2545 | |
d6fbec6d | 2546 | /* Check with in-band control to see if we're allowed to set up this |
0ad9b732 JP |
2547 | * flow. */ |
2548 | if (!in_band_rule_check(ofproto->in_band, flow, out)) { | |
d6fbec6d | 2549 | ctx.may_set_up_flow = false; |
0ad9b732 JP |
2550 | } |
2551 | ||
d6fbec6d BP |
2552 | if (may_set_up_flow) { |
2553 | *may_set_up_flow = ctx.may_set_up_flow; | |
064af421 | 2554 | } |
6a07af36 JG |
2555 | if (nf_output_iface) { |
2556 | *nf_output_iface = ctx.nf_output_iface; | |
064af421 BP |
2557 | } |
2558 | if (odp_actions_overflow(out)) { | |
2559 | odp_actions_init(out); | |
2560 | return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY); | |
2561 | } | |
2562 | return 0; | |
2563 | } | |
2564 | ||
9deba63b BP |
2565 | /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow |
2566 | * error message code (composed with ofp_mkerr()) for the caller to propagate | |
2567 | * upward. Otherwise, returns 0. | |
2568 | * | |
2569 | * 'oh' is used to make log messages more informative. */ | |
2570 | static int | |
2571 | reject_slave_controller(struct ofconn *ofconn, const struct ofp_header *oh) | |
2572 | { | |
2573 | if (ofconn->type == OFCONN_CONTROLLER && ofconn->role == NX_ROLE_SLAVE) { | |
2574 | static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
2575 | char *type_name; | |
2576 | ||
2577 | type_name = ofp_message_type_to_string(oh->type); | |
2578 | VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller", | |
2579 | type_name); | |
2580 | free(type_name); | |
2581 | ||
2582 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM); | |
2583 | } else { | |
2584 | return 0; | |
2585 | } | |
2586 | } | |
2587 | ||
064af421 BP |
2588 | static int |
2589 | handle_packet_out(struct ofproto *p, struct ofconn *ofconn, | |
2590 | struct ofp_header *oh) | |
2591 | { | |
2592 | struct ofp_packet_out *opo; | |
2593 | struct ofpbuf payload, *buffer; | |
2594 | struct odp_actions actions; | |
2595 | int n_actions; | |
2596 | uint16_t in_port; | |
2597 | flow_t flow; | |
2598 | int error; | |
2599 | ||
9deba63b BP |
2600 | error = reject_slave_controller(ofconn, oh); |
2601 | if (error) { | |
2602 | return error; | |
2603 | } | |
2604 | ||
064af421 BP |
2605 | error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports); |
2606 | if (error) { | |
2607 | return error; | |
2608 | } | |
2609 | opo = (struct ofp_packet_out *) oh; | |
2610 | ||
2611 | COVERAGE_INC(ofproto_packet_out); | |
2612 | if (opo->buffer_id != htonl(UINT32_MAX)) { | |
2613 | error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id), | |
2614 | &buffer, &in_port); | |
7778bd15 | 2615 | if (error || !buffer) { |
064af421 BP |
2616 | return error; |
2617 | } | |
2618 | payload = *buffer; | |
2619 | } else { | |
2620 | buffer = NULL; | |
2621 | } | |
2622 | ||
659586ef | 2623 | flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow); |
064af421 | 2624 | error = xlate_actions((const union ofp_action *) opo->actions, n_actions, |
6a07af36 | 2625 | &flow, p, &payload, &actions, NULL, NULL, NULL); |
064af421 BP |
2626 | if (error) { |
2627 | return error; | |
2628 | } | |
2629 | ||
c228a364 | 2630 | dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions, |
064af421 BP |
2631 | &payload); |
2632 | ofpbuf_delete(buffer); | |
2633 | ||
2634 | return 0; | |
2635 | } | |
2636 | ||
2637 | static void | |
2638 | update_port_config(struct ofproto *p, struct ofport *port, | |
2639 | uint32_t config, uint32_t mask) | |
2640 | { | |
2641 | mask &= config ^ port->opp.config; | |
2642 | if (mask & OFPPC_PORT_DOWN) { | |
2643 | if (config & OFPPC_PORT_DOWN) { | |
2644 | netdev_turn_flags_off(port->netdev, NETDEV_UP, true); | |
2645 | } else { | |
2646 | netdev_turn_flags_on(port->netdev, NETDEV_UP, true); | |
2647 | } | |
2648 | } | |
2649 | #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD) | |
2650 | if (mask & REVALIDATE_BITS) { | |
2651 | COVERAGE_INC(ofproto_costly_flags); | |
2652 | port->opp.config ^= mask & REVALIDATE_BITS; | |
2653 | p->need_revalidate = true; | |
2654 | } | |
2655 | #undef REVALIDATE_BITS | |
2656 | if (mask & OFPPC_NO_FLOOD) { | |
2657 | port->opp.config ^= OFPPC_NO_FLOOD; | |
72b06300 | 2658 | refresh_port_groups(p); |
064af421 BP |
2659 | } |
2660 | if (mask & OFPPC_NO_PACKET_IN) { | |
2661 | port->opp.config ^= OFPPC_NO_PACKET_IN; | |
2662 | } | |
2663 | } | |
2664 | ||
2665 | static int | |
9deba63b BP |
2666 | handle_port_mod(struct ofproto *p, struct ofconn *ofconn, |
2667 | struct ofp_header *oh) | |
064af421 BP |
2668 | { |
2669 | const struct ofp_port_mod *opm; | |
2670 | struct ofport *port; | |
2671 | int error; | |
2672 | ||
9deba63b BP |
2673 | error = reject_slave_controller(ofconn, oh); |
2674 | if (error) { | |
2675 | return error; | |
2676 | } | |
064af421 BP |
2677 | error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm); |
2678 | if (error) { | |
2679 | return error; | |
2680 | } | |
2681 | opm = (struct ofp_port_mod *) oh; | |
2682 | ||
2683 | port = port_array_get(&p->ports, | |
2684 | ofp_port_to_odp_port(ntohs(opm->port_no))); | |
2685 | if (!port) { | |
2686 | return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT); | |
2687 | } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) { | |
2688 | return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR); | |
2689 | } else { | |
2690 | update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask)); | |
2691 | if (opm->advertise) { | |
2692 | netdev_set_advertisements(port->netdev, ntohl(opm->advertise)); | |
2693 | } | |
2694 | } | |
2695 | return 0; | |
2696 | } | |
2697 | ||
2698 | static struct ofpbuf * | |
2699 | make_stats_reply(uint32_t xid, uint16_t type, size_t body_len) | |
2700 | { | |
2701 | struct ofp_stats_reply *osr; | |
2702 | struct ofpbuf *msg; | |
2703 | ||
2704 | msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX)); | |
2705 | osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg); | |
2706 | osr->type = type; | |
2707 | osr->flags = htons(0); | |
2708 | return msg; | |
2709 | } | |
2710 | ||
2711 | static struct ofpbuf * | |
2712 | start_stats_reply(const struct ofp_stats_request *request, size_t body_len) | |
2713 | { | |
2714 | return make_stats_reply(request->header.xid, request->type, body_len); | |
2715 | } | |
2716 | ||
2717 | static void * | |
2718 | append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp) | |
2719 | { | |
2720 | struct ofpbuf *msg = *msgp; | |
2721 | assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply)); | |
2722 | if (nbytes + msg->size > UINT16_MAX) { | |
2723 | struct ofp_stats_reply *reply = msg->data; | |
2724 | reply->flags = htons(OFPSF_REPLY_MORE); | |
2725 | *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes); | |
2726 | queue_tx(msg, ofconn, ofconn->reply_counter); | |
2727 | } | |
2728 | return ofpbuf_put_uninit(*msgp, nbytes); | |
2729 | } | |
2730 | ||
2731 | static int | |
2732 | handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn, | |
2733 | struct ofp_stats_request *request) | |
2734 | { | |
2735 | struct ofp_desc_stats *ods; | |
2736 | struct ofpbuf *msg; | |
2737 | ||
2738 | msg = start_stats_reply(request, sizeof *ods); | |
2739 | ods = append_stats_reply(sizeof *ods, ofconn, &msg); | |
5a719c38 JP |
2740 | memset(ods, 0, sizeof *ods); |
2741 | ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc); | |
2742 | ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc); | |
2743 | ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc); | |
2744 | ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num); | |
2745 | ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc); | |
064af421 BP |
2746 | queue_tx(msg, ofconn, ofconn->reply_counter); |
2747 | ||
2748 | return 0; | |
2749 | } | |
2750 | ||
2751 | static void | |
2752 | count_subrules(struct cls_rule *cls_rule, void *n_subrules_) | |
2753 | { | |
2754 | struct rule *rule = rule_from_cls_rule(cls_rule); | |
2755 | int *n_subrules = n_subrules_; | |
2756 | ||
2757 | if (rule->super) { | |
2758 | (*n_subrules)++; | |
2759 | } | |
2760 | } | |
2761 | ||
2762 | static int | |
2763 | handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn, | |
2764 | struct ofp_stats_request *request) | |
2765 | { | |
2766 | struct ofp_table_stats *ots; | |
2767 | struct ofpbuf *msg; | |
2768 | struct odp_stats dpstats; | |
2769 | int n_exact, n_subrules, n_wild; | |
2770 | ||
2771 | msg = start_stats_reply(request, sizeof *ots * 2); | |
2772 | ||
2773 | /* Count rules of various kinds. */ | |
2774 | n_subrules = 0; | |
2775 | classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules); | |
2776 | n_exact = classifier_count_exact(&p->cls) - n_subrules; | |
2777 | n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls); | |
2778 | ||
2779 | /* Hash table. */ | |
c228a364 | 2780 | dpif_get_dp_stats(p->dpif, &dpstats); |
064af421 BP |
2781 | ots = append_stats_reply(sizeof *ots, ofconn, &msg); |
2782 | memset(ots, 0, sizeof *ots); | |
2783 | ots->table_id = TABLEID_HASH; | |
2784 | strcpy(ots->name, "hash"); | |
2785 | ots->wildcards = htonl(0); | |
2786 | ots->max_entries = htonl(dpstats.max_capacity); | |
2787 | ots->active_count = htonl(n_exact); | |
2788 | ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit + | |
2789 | dpstats.n_missed); | |
2790 | ots->matched_count = htonll(dpstats.n_hit); /* XXX */ | |
2791 | ||
2792 | /* Classifier table. */ | |
2793 | ots = append_stats_reply(sizeof *ots, ofconn, &msg); | |
2794 | memset(ots, 0, sizeof *ots); | |
2795 | ots->table_id = TABLEID_CLASSIFIER; | |
2796 | strcpy(ots->name, "classifier"); | |
659586ef JG |
2797 | ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL) |
2798 | : htonl(OFPFW_ALL); | |
064af421 BP |
2799 | ots->max_entries = htonl(65536); |
2800 | ots->active_count = htonl(n_wild); | |
2801 | ots->lookup_count = htonll(0); /* XXX */ | |
2802 | ots->matched_count = htonll(0); /* XXX */ | |
2803 | ||
2804 | queue_tx(msg, ofconn, ofconn->reply_counter); | |
2805 | return 0; | |
2806 | } | |
2807 | ||
abaad8cf JP |
2808 | static void |
2809 | append_port_stat(struct ofport *port, uint16_t port_no, struct ofconn *ofconn, | |
a4948b95 | 2810 | struct ofpbuf **msgp) |
abaad8cf JP |
2811 | { |
2812 | struct netdev_stats stats; | |
2813 | struct ofp_port_stats *ops; | |
2814 | ||
2815 | /* Intentionally ignore return value, since errors will set | |
2816 | * 'stats' to all-1s, which is correct for OpenFlow, and | |
2817 | * netdev_get_stats() will log errors. */ | |
2818 | netdev_get_stats(port->netdev, &stats); | |
2819 | ||
a4948b95 | 2820 | ops = append_stats_reply(sizeof *ops, ofconn, msgp); |
abaad8cf JP |
2821 | ops->port_no = htons(odp_port_to_ofp_port(port_no)); |
2822 | memset(ops->pad, 0, sizeof ops->pad); | |
2823 | ops->rx_packets = htonll(stats.rx_packets); | |
2824 | ops->tx_packets = htonll(stats.tx_packets); | |
2825 | ops->rx_bytes = htonll(stats.rx_bytes); | |
2826 | ops->tx_bytes = htonll(stats.tx_bytes); | |
2827 | ops->rx_dropped = htonll(stats.rx_dropped); | |
2828 | ops->tx_dropped = htonll(stats.tx_dropped); | |
2829 | ops->rx_errors = htonll(stats.rx_errors); | |
2830 | ops->tx_errors = htonll(stats.tx_errors); | |
2831 | ops->rx_frame_err = htonll(stats.rx_frame_errors); | |
2832 | ops->rx_over_err = htonll(stats.rx_over_errors); | |
2833 | ops->rx_crc_err = htonll(stats.rx_crc_errors); | |
2834 | ops->collisions = htonll(stats.collisions); | |
2835 | } | |
2836 | ||
064af421 BP |
2837 | static int |
2838 | handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn, | |
abaad8cf JP |
2839 | struct ofp_stats_request *osr, |
2840 | size_t arg_size) | |
064af421 | 2841 | { |
abaad8cf | 2842 | struct ofp_port_stats_request *psr; |
064af421 BP |
2843 | struct ofp_port_stats *ops; |
2844 | struct ofpbuf *msg; | |
2845 | struct ofport *port; | |
2846 | unsigned int port_no; | |
2847 | ||
abaad8cf JP |
2848 | if (arg_size != sizeof *psr) { |
2849 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
2850 | } | |
2851 | psr = (struct ofp_port_stats_request *) osr->body; | |
2852 | ||
2853 | msg = start_stats_reply(osr, sizeof *ops * 16); | |
2854 | if (psr->port_no != htons(OFPP_NONE)) { | |
2855 | port = port_array_get(&p->ports, | |
2856 | ofp_port_to_odp_port(ntohs(psr->port_no))); | |
2857 | if (port) { | |
a4948b95 | 2858 | append_port_stat(port, ntohs(psr->port_no), ofconn, &msg); |
abaad8cf JP |
2859 | } |
2860 | } else { | |
2861 | PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) { | |
a4948b95 | 2862 | append_port_stat(port, port_no, ofconn, &msg); |
abaad8cf | 2863 | } |
064af421 BP |
2864 | } |
2865 | ||
2866 | queue_tx(msg, ofconn, ofconn->reply_counter); | |
2867 | return 0; | |
2868 | } | |
2869 | ||
2870 | struct flow_stats_cbdata { | |
2871 | struct ofproto *ofproto; | |
2872 | struct ofconn *ofconn; | |
2873 | uint16_t out_port; | |
2874 | struct ofpbuf *msg; | |
2875 | }; | |
2876 | ||
01149cfd BP |
2877 | /* Obtains statistic counters for 'rule' within 'p' and stores them into |
2878 | * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the | |
2879 | * returned statistic include statistics for all of 'rule''s subrules. */ | |
064af421 BP |
2880 | static void |
2881 | query_stats(struct ofproto *p, struct rule *rule, | |
2882 | uint64_t *packet_countp, uint64_t *byte_countp) | |
2883 | { | |
2884 | uint64_t packet_count, byte_count; | |
2885 | struct rule *subrule; | |
2886 | struct odp_flow *odp_flows; | |
2887 | size_t n_odp_flows; | |
2888 | ||
01149cfd BP |
2889 | /* Start from historical data for 'rule' itself that are no longer tracked |
2890 | * by the datapath. This counts, for example, subrules that have | |
2891 | * expired. */ | |
b3137fe8 JG |
2892 | packet_count = rule->packet_count; |
2893 | byte_count = rule->byte_count; | |
2894 | ||
01149cfd BP |
2895 | /* Prepare to ask the datapath for statistics on 'rule', or if it is |
2896 | * wildcarded then on all of its subrules. | |
2897 | * | |
2898 | * Also, add any statistics that are not tracked by the datapath for each | |
2899 | * subrule. This includes, for example, statistics for packets that were | |
2900 | * executed "by hand" by ofproto via dpif_execute() but must be accounted | |
2901 | * to a flow. */ | |
064af421 | 2902 | n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1; |
ec6fde61 | 2903 | odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows); |
064af421 BP |
2904 | if (rule->cr.wc.wildcards) { |
2905 | size_t i = 0; | |
2906 | LIST_FOR_EACH (subrule, struct rule, list, &rule->list) { | |
2907 | odp_flows[i++].key = subrule->cr.flow; | |
b3137fe8 JG |
2908 | packet_count += subrule->packet_count; |
2909 | byte_count += subrule->byte_count; | |
064af421 BP |
2910 | } |
2911 | } else { | |
2912 | odp_flows[0].key = rule->cr.flow; | |
2913 | } | |
2914 | ||
28998b22 | 2915 | /* Fetch up-to-date statistics from the datapath and add them in. */ |
c228a364 | 2916 | if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) { |
064af421 BP |
2917 | size_t i; |
2918 | for (i = 0; i < n_odp_flows; i++) { | |
2919 | struct odp_flow *odp_flow = &odp_flows[i]; | |
2920 | packet_count += odp_flow->stats.n_packets; | |
2921 | byte_count += odp_flow->stats.n_bytes; | |
2922 | } | |
2923 | } | |
2924 | free(odp_flows); | |
2925 | ||
01149cfd | 2926 | /* Return the stats to the caller. */ |
064af421 BP |
2927 | *packet_countp = packet_count; |
2928 | *byte_countp = byte_count; | |
2929 | } | |
2930 | ||
2931 | static void | |
2932 | flow_stats_cb(struct cls_rule *rule_, void *cbdata_) | |
2933 | { | |
2934 | struct rule *rule = rule_from_cls_rule(rule_); | |
2935 | struct flow_stats_cbdata *cbdata = cbdata_; | |
2936 | struct ofp_flow_stats *ofs; | |
2937 | uint64_t packet_count, byte_count; | |
2938 | size_t act_len, len; | |
26c3f94a JP |
2939 | long long int tdiff = time_msec() - rule->created; |
2940 | uint32_t sec = tdiff / 1000; | |
2941 | uint32_t msec = tdiff - (sec * 1000); | |
064af421 BP |
2942 | |
2943 | if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) { | |
2944 | return; | |
2945 | } | |
2946 | ||
2947 | act_len = sizeof *rule->actions * rule->n_actions; | |
2948 | len = offsetof(struct ofp_flow_stats, actions) + act_len; | |
2949 | ||
2950 | query_stats(cbdata->ofproto, rule, &packet_count, &byte_count); | |
2951 | ||
2952 | ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg); | |
2953 | ofs->length = htons(len); | |
2954 | ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH; | |
2955 | ofs->pad = 0; | |
659586ef JG |
2956 | flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, |
2957 | cbdata->ofproto->tun_id_from_cookie, &ofs->match); | |
26c3f94a JP |
2958 | ofs->duration_sec = htonl(sec); |
2959 | ofs->duration_nsec = htonl(msec * 1000000); | |
39997502 | 2960 | ofs->cookie = rule->flow_cookie; |
064af421 BP |
2961 | ofs->priority = htons(rule->cr.priority); |
2962 | ofs->idle_timeout = htons(rule->idle_timeout); | |
2963 | ofs->hard_timeout = htons(rule->hard_timeout); | |
39997502 | 2964 | memset(ofs->pad2, 0, sizeof ofs->pad2); |
064af421 BP |
2965 | ofs->packet_count = htonll(packet_count); |
2966 | ofs->byte_count = htonll(byte_count); | |
2967 | memcpy(ofs->actions, rule->actions, act_len); | |
2968 | } | |
2969 | ||
2970 | static int | |
2971 | table_id_to_include(uint8_t table_id) | |
2972 | { | |
2973 | return (table_id == TABLEID_HASH ? CLS_INC_EXACT | |
2974 | : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD | |
2975 | : table_id == 0xff ? CLS_INC_ALL | |
2976 | : 0); | |
2977 | } | |
2978 | ||
2979 | static int | |
2980 | handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn, | |
2981 | const struct ofp_stats_request *osr, | |
2982 | size_t arg_size) | |
2983 | { | |
2984 | struct ofp_flow_stats_request *fsr; | |
2985 | struct flow_stats_cbdata cbdata; | |
2986 | struct cls_rule target; | |
2987 | ||
2988 | if (arg_size != sizeof *fsr) { | |
49bdc010 | 2989 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); |
064af421 BP |
2990 | } |
2991 | fsr = (struct ofp_flow_stats_request *) osr->body; | |
2992 | ||
2993 | COVERAGE_INC(ofproto_flows_req); | |
2994 | cbdata.ofproto = p; | |
2995 | cbdata.ofconn = ofconn; | |
2996 | cbdata.out_port = fsr->out_port; | |
2997 | cbdata.msg = start_stats_reply(osr, 1024); | |
659586ef | 2998 | cls_rule_from_match(&fsr->match, 0, false, 0, &target); |
064af421 BP |
2999 | classifier_for_each_match(&p->cls, &target, |
3000 | table_id_to_include(fsr->table_id), | |
3001 | flow_stats_cb, &cbdata); | |
3002 | queue_tx(cbdata.msg, ofconn, ofconn->reply_counter); | |
3003 | return 0; | |
3004 | } | |
3005 | ||
4f2cad2c JP |
3006 | struct flow_stats_ds_cbdata { |
3007 | struct ofproto *ofproto; | |
3008 | struct ds *results; | |
3009 | }; | |
3010 | ||
3011 | static void | |
3012 | flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_) | |
3013 | { | |
3014 | struct rule *rule = rule_from_cls_rule(rule_); | |
3015 | struct flow_stats_ds_cbdata *cbdata = cbdata_; | |
3016 | struct ds *results = cbdata->results; | |
3017 | struct ofp_match match; | |
3018 | uint64_t packet_count, byte_count; | |
3019 | size_t act_len = sizeof *rule->actions * rule->n_actions; | |
3020 | ||
3021 | /* Don't report on subrules. */ | |
3022 | if (rule->super != NULL) { | |
3023 | return; | |
3024 | } | |
3025 | ||
3026 | query_stats(cbdata->ofproto, rule, &packet_count, &byte_count); | |
659586ef JG |
3027 | flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, |
3028 | cbdata->ofproto->tun_id_from_cookie, &match); | |
4f2cad2c JP |
3029 | |
3030 | ds_put_format(results, "duration=%llds, ", | |
3031 | (time_msec() - rule->created) / 1000); | |
52ae00b3 | 3032 | ds_put_format(results, "priority=%u, ", rule->cr.priority); |
4f2cad2c JP |
3033 | ds_put_format(results, "n_packets=%"PRIu64", ", packet_count); |
3034 | ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count); | |
3035 | ofp_print_match(results, &match, true); | |
3036 | ofp_print_actions(results, &rule->actions->header, act_len); | |
3037 | ds_put_cstr(results, "\n"); | |
3038 | } | |
3039 | ||
3040 | /* Adds a pretty-printed description of all flows to 'results', including | |
3041 | * those marked hidden by secchan (e.g., by in-band control). */ | |
3042 | void | |
3043 | ofproto_get_all_flows(struct ofproto *p, struct ds *results) | |
3044 | { | |
3045 | struct ofp_match match; | |
3046 | struct cls_rule target; | |
3047 | struct flow_stats_ds_cbdata cbdata; | |
3048 | ||
3049 | memset(&match, 0, sizeof match); | |
659586ef | 3050 | match.wildcards = htonl(OVSFW_ALL); |
4f2cad2c JP |
3051 | |
3052 | cbdata.ofproto = p; | |
3053 | cbdata.results = results; | |
3054 | ||
659586ef | 3055 | cls_rule_from_match(&match, 0, false, 0, &target); |
4f2cad2c JP |
3056 | classifier_for_each_match(&p->cls, &target, CLS_INC_ALL, |
3057 | flow_stats_ds_cb, &cbdata); | |
3058 | } | |
3059 | ||
064af421 BP |
3060 | struct aggregate_stats_cbdata { |
3061 | struct ofproto *ofproto; | |
3062 | uint16_t out_port; | |
3063 | uint64_t packet_count; | |
3064 | uint64_t byte_count; | |
3065 | uint32_t n_flows; | |
3066 | }; | |
3067 | ||
3068 | static void | |
3069 | aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_) | |
3070 | { | |
3071 | struct rule *rule = rule_from_cls_rule(rule_); | |
3072 | struct aggregate_stats_cbdata *cbdata = cbdata_; | |
3073 | uint64_t packet_count, byte_count; | |
3074 | ||
3075 | if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) { | |
3076 | return; | |
3077 | } | |
3078 | ||
3079 | query_stats(cbdata->ofproto, rule, &packet_count, &byte_count); | |
3080 | ||
3081 | cbdata->packet_count += packet_count; | |
3082 | cbdata->byte_count += byte_count; | |
3083 | cbdata->n_flows++; | |
3084 | } | |
3085 | ||
3086 | static int | |
3087 | handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn, | |
3088 | const struct ofp_stats_request *osr, | |
3089 | size_t arg_size) | |
3090 | { | |
3091 | struct ofp_aggregate_stats_request *asr; | |
3092 | struct ofp_aggregate_stats_reply *reply; | |
3093 | struct aggregate_stats_cbdata cbdata; | |
3094 | struct cls_rule target; | |
3095 | struct ofpbuf *msg; | |
3096 | ||
3097 | if (arg_size != sizeof *asr) { | |
49bdc010 | 3098 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); |
064af421 BP |
3099 | } |
3100 | asr = (struct ofp_aggregate_stats_request *) osr->body; | |
3101 | ||
3102 | COVERAGE_INC(ofproto_agg_request); | |
3103 | cbdata.ofproto = p; | |
3104 | cbdata.out_port = asr->out_port; | |
3105 | cbdata.packet_count = 0; | |
3106 | cbdata.byte_count = 0; | |
3107 | cbdata.n_flows = 0; | |
659586ef | 3108 | cls_rule_from_match(&asr->match, 0, false, 0, &target); |
064af421 BP |
3109 | classifier_for_each_match(&p->cls, &target, |
3110 | table_id_to_include(asr->table_id), | |
3111 | aggregate_stats_cb, &cbdata); | |
3112 | ||
3113 | msg = start_stats_reply(osr, sizeof *reply); | |
3114 | reply = append_stats_reply(sizeof *reply, ofconn, &msg); | |
3115 | reply->flow_count = htonl(cbdata.n_flows); | |
3116 | reply->packet_count = htonll(cbdata.packet_count); | |
3117 | reply->byte_count = htonll(cbdata.byte_count); | |
3118 | queue_tx(msg, ofconn, ofconn->reply_counter); | |
3119 | return 0; | |
3120 | } | |
3121 | ||
3122 | static int | |
3123 | handle_stats_request(struct ofproto *p, struct ofconn *ofconn, | |
3124 | struct ofp_header *oh) | |
3125 | { | |
3126 | struct ofp_stats_request *osr; | |
3127 | size_t arg_size; | |
3128 | int error; | |
3129 | ||
3130 | error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr, | |
3131 | 1, &arg_size); | |
3132 | if (error) { | |
3133 | return error; | |
3134 | } | |
3135 | osr = (struct ofp_stats_request *) oh; | |
3136 | ||
3137 | switch (ntohs(osr->type)) { | |
3138 | case OFPST_DESC: | |
3139 | return handle_desc_stats_request(p, ofconn, osr); | |
3140 | ||
3141 | case OFPST_FLOW: | |
3142 | return handle_flow_stats_request(p, ofconn, osr, arg_size); | |
3143 | ||
3144 | case OFPST_AGGREGATE: | |
3145 | return handle_aggregate_stats_request(p, ofconn, osr, arg_size); | |
3146 | ||
3147 | case OFPST_TABLE: | |
3148 | return handle_table_stats_request(p, ofconn, osr); | |
3149 | ||
3150 | case OFPST_PORT: | |
abaad8cf | 3151 | return handle_port_stats_request(p, ofconn, osr, arg_size); |
064af421 BP |
3152 | |
3153 | case OFPST_VENDOR: | |
3154 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR); | |
3155 | ||
3156 | default: | |
3157 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT); | |
3158 | } | |
3159 | } | |
3160 | ||
3161 | static long long int | |
3162 | msec_from_nsec(uint64_t sec, uint32_t nsec) | |
3163 | { | |
3164 | return !sec ? 0 : sec * 1000 + nsec / 1000000; | |
3165 | } | |
3166 | ||
3167 | static void | |
0193b2af JG |
3168 | update_time(struct ofproto *ofproto, struct rule *rule, |
3169 | const struct odp_flow_stats *stats) | |
064af421 BP |
3170 | { |
3171 | long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec); | |
3172 | if (used > rule->used) { | |
3173 | rule->used = used; | |
4836f9f2 JP |
3174 | if (rule->super && used > rule->super->used) { |
3175 | rule->super->used = used; | |
3176 | } | |
0193b2af | 3177 | netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used); |
064af421 BP |
3178 | } |
3179 | } | |
3180 | ||
3181 | static void | |
0193b2af JG |
3182 | update_stats(struct ofproto *ofproto, struct rule *rule, |
3183 | const struct odp_flow_stats *stats) | |
064af421 | 3184 | { |
064af421 | 3185 | if (stats->n_packets) { |
0193b2af JG |
3186 | update_time(ofproto, rule, stats); |
3187 | rule->packet_count += stats->n_packets; | |
3188 | rule->byte_count += stats->n_bytes; | |
3189 | netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos, | |
3190 | stats->tcp_flags); | |
064af421 BP |
3191 | } |
3192 | } | |
3193 | ||
79eee1eb BP |
3194 | /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT |
3195 | * in which no matching flow already exists in the flow table. | |
3196 | * | |
3197 | * Adds the flow specified by 'ofm', which is followed by 'n_actions' | |
3198 | * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error | |
3199 | * code as encoded by ofp_mkerr() on failure. | |
3200 | * | |
3201 | * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, | |
3202 | * if any. */ | |
064af421 BP |
3203 | static int |
3204 | add_flow(struct ofproto *p, struct ofconn *ofconn, | |
79eee1eb | 3205 | const struct ofp_flow_mod *ofm, size_t n_actions) |
064af421 BP |
3206 | { |
3207 | struct ofpbuf *packet; | |
3208 | struct rule *rule; | |
3209 | uint16_t in_port; | |
3210 | int error; | |
3211 | ||
49bdc010 JP |
3212 | if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) { |
3213 | flow_t flow; | |
3214 | uint32_t wildcards; | |
3215 | ||
659586ef JG |
3216 | flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie, |
3217 | &flow, &wildcards); | |
49bdc010 JP |
3218 | if (classifier_rule_overlaps(&p->cls, &flow, wildcards, |
3219 | ntohs(ofm->priority))) { | |
3220 | return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP); | |
3221 | } | |
3222 | } | |
3223 | ||
0193b2af | 3224 | rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions, |
064af421 | 3225 | n_actions, ntohs(ofm->idle_timeout), |
39997502 | 3226 | ntohs(ofm->hard_timeout), ofm->cookie, |
ca069229 | 3227 | ofm->flags & htons(OFPFF_SEND_FLOW_REM)); |
659586ef JG |
3228 | cls_rule_from_match(&ofm->match, ntohs(ofm->priority), |
3229 | p->tun_id_from_cookie, ofm->cookie, &rule->cr); | |
064af421 | 3230 | |
064af421 BP |
3231 | error = 0; |
3232 | if (ofm->buffer_id != htonl(UINT32_MAX)) { | |
3233 | error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id), | |
3234 | &packet, &in_port); | |
212fe71c BP |
3235 | } else { |
3236 | packet = NULL; | |
165cd8a3 | 3237 | in_port = UINT16_MAX; |
064af421 BP |
3238 | } |
3239 | ||
3240 | rule_insert(p, rule, packet, in_port); | |
3241 | ofpbuf_delete(packet); | |
3242 | return error; | |
3243 | } | |
3244 | ||
79eee1eb BP |
3245 | static struct rule * |
3246 | find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm) | |
064af421 | 3247 | { |
064af421 BP |
3248 | uint32_t wildcards; |
3249 | flow_t flow; | |
3250 | ||
659586ef JG |
3251 | flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie, |
3252 | &flow, &wildcards); | |
79eee1eb | 3253 | return rule_from_cls_rule(classifier_find_rule_exactly( |
064af421 BP |
3254 | &p->cls, &flow, wildcards, |
3255 | ntohs(ofm->priority))); | |
79eee1eb | 3256 | } |
064af421 | 3257 | |
79eee1eb BP |
3258 | static int |
3259 | send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn, | |
3260 | struct rule *rule, const struct ofp_flow_mod *ofm) | |
3261 | { | |
3262 | struct ofpbuf *packet; | |
3263 | uint16_t in_port; | |
3264 | flow_t flow; | |
3265 | int error; | |
064af421 | 3266 | |
79eee1eb BP |
3267 | if (ofm->buffer_id == htonl(UINT32_MAX)) { |
3268 | return 0; | |
064af421 | 3269 | } |
79eee1eb BP |
3270 | |
3271 | error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id), | |
3272 | &packet, &in_port); | |
3273 | if (error) { | |
3274 | return error; | |
3275 | } | |
3276 | ||
659586ef | 3277 | flow_extract(packet, 0, in_port, &flow); |
79eee1eb BP |
3278 | rule_execute(ofproto, rule, packet, &flow); |
3279 | ofpbuf_delete(packet); | |
3280 | ||
064af421 BP |
3281 | return 0; |
3282 | } | |
79eee1eb BP |
3283 | \f |
3284 | /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */ | |
064af421 BP |
3285 | |
3286 | struct modify_flows_cbdata { | |
3287 | struct ofproto *ofproto; | |
3288 | const struct ofp_flow_mod *ofm; | |
064af421 | 3289 | size_t n_actions; |
79eee1eb | 3290 | struct rule *match; |
064af421 BP |
3291 | }; |
3292 | ||
79eee1eb BP |
3293 | static int modify_flow(struct ofproto *, const struct ofp_flow_mod *, |
3294 | size_t n_actions, struct rule *); | |
3295 | static void modify_flows_cb(struct cls_rule *, void *cbdata_); | |
3296 | ||
3297 | /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as | |
3298 | * encoded by ofp_mkerr() on failure. | |
3299 | * | |
3300 | * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, | |
3301 | * if any. */ | |
3302 | static int | |
3303 | modify_flows_loose(struct ofproto *p, struct ofconn *ofconn, | |
3304 | const struct ofp_flow_mod *ofm, size_t n_actions) | |
3305 | { | |
3306 | struct modify_flows_cbdata cbdata; | |
3307 | struct cls_rule target; | |
3308 | ||
3309 | cbdata.ofproto = p; | |
3310 | cbdata.ofm = ofm; | |
3311 | cbdata.n_actions = n_actions; | |
3312 | cbdata.match = NULL; | |
3313 | ||
659586ef JG |
3314 | cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie, |
3315 | &target); | |
79eee1eb BP |
3316 | |
3317 | classifier_for_each_match(&p->cls, &target, CLS_INC_ALL, | |
3318 | modify_flows_cb, &cbdata); | |
3319 | if (cbdata.match) { | |
3320 | /* This credits the packet to whichever flow happened to happened to | |
3321 | * match last. That's weird. Maybe we should do a lookup for the | |
3322 | * flow that actually matches the packet? Who knows. */ | |
3323 | send_buffered_packet(p, ofconn, cbdata.match, ofm); | |
3324 | return 0; | |
3325 | } else { | |
3326 | return add_flow(p, ofconn, ofm, n_actions); | |
3327 | } | |
3328 | } | |
3329 | ||
3330 | /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error | |
3331 | * code as encoded by ofp_mkerr() on failure. | |
3332 | * | |
3333 | * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, | |
3334 | * if any. */ | |
3335 | static int | |
3336 | modify_flow_strict(struct ofproto *p, struct ofconn *ofconn, | |
3337 | struct ofp_flow_mod *ofm, size_t n_actions) | |
3338 | { | |
3339 | struct rule *rule = find_flow_strict(p, ofm); | |
3340 | if (rule && !rule_is_hidden(rule)) { | |
3341 | modify_flow(p, ofm, n_actions, rule); | |
3342 | return send_buffered_packet(p, ofconn, rule, ofm); | |
3343 | } else { | |
3344 | return add_flow(p, ofconn, ofm, n_actions); | |
3345 | } | |
3346 | } | |
3347 | ||
3348 | /* Callback for modify_flows_loose(). */ | |
064af421 BP |
3349 | static void |
3350 | modify_flows_cb(struct cls_rule *rule_, void *cbdata_) | |
3351 | { | |
3352 | struct rule *rule = rule_from_cls_rule(rule_); | |
3353 | struct modify_flows_cbdata *cbdata = cbdata_; | |
3354 | ||
79eee1eb BP |
3355 | if (!rule_is_hidden(rule)) { |
3356 | cbdata->match = rule; | |
3357 | modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule); | |
064af421 | 3358 | } |
064af421 BP |
3359 | } |
3360 | ||
79eee1eb BP |
3361 | /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has |
3362 | * been identified as a flow in 'p''s flow table to be modified, by changing | |
3363 | * the rule's actions to match those in 'ofm' (which is followed by 'n_actions' | |
3364 | * ofp_action[] structures). */ | |
064af421 | 3365 | static int |
79eee1eb BP |
3366 | modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm, |
3367 | size_t n_actions, struct rule *rule) | |
064af421 | 3368 | { |
79eee1eb BP |
3369 | size_t actions_len = n_actions * sizeof *rule->actions; |
3370 | ||
3371 | rule->flow_cookie = ofm->cookie; | |
3372 | ||
3373 | /* If the actions are the same, do nothing. */ | |
3374 | if (n_actions == rule->n_actions | |
3375 | && !memcmp(ofm->actions, rule->actions, actions_len)) | |
3376 | { | |
3377 | return 0; | |
3378 | } | |
3379 | ||
3380 | /* Replace actions. */ | |
3381 | free(rule->actions); | |
3382 | rule->actions = xmemdup(ofm->actions, actions_len); | |
3383 | rule->n_actions = n_actions; | |
3384 | ||
3385 | /* Make sure that the datapath gets updated properly. */ | |
3386 | if (rule->cr.wc.wildcards) { | |
3387 | COVERAGE_INC(ofproto_mod_wc_flow); | |
3388 | p->need_revalidate = true; | |
3389 | } else { | |
3390 | rule_update_actions(p, rule); | |
3391 | } | |
3392 | ||
3393 | return 0; | |
3394 | } | |
3395 | \f | |
3396 | /* OFPFC_DELETE implementation. */ | |
3397 | ||
3398 | struct delete_flows_cbdata { | |
3399 | struct ofproto *ofproto; | |
3400 | uint16_t out_port; | |
3401 | }; | |
3402 | ||
3403 | static void delete_flows_cb(struct cls_rule *, void *cbdata_); | |
3404 | static void delete_flow(struct ofproto *, struct rule *, uint16_t out_port); | |
3405 | ||
3406 | /* Implements OFPFC_DELETE. */ | |
3407 | static void | |
3408 | delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm) | |
3409 | { | |
3410 | struct delete_flows_cbdata cbdata; | |
064af421 BP |
3411 | struct cls_rule target; |
3412 | ||
3413 | cbdata.ofproto = p; | |
79eee1eb | 3414 | cbdata.out_port = ofm->out_port; |
064af421 | 3415 | |
659586ef JG |
3416 | cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie, |
3417 | &target); | |
064af421 BP |
3418 | |
3419 | classifier_for_each_match(&p->cls, &target, CLS_INC_ALL, | |
79eee1eb | 3420 | delete_flows_cb, &cbdata); |
064af421 BP |
3421 | } |
3422 | ||
79eee1eb BP |
3423 | /* Implements OFPFC_DELETE_STRICT. */ |
3424 | static void | |
3425 | delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm) | |
3426 | { | |
3427 | struct rule *rule = find_flow_strict(p, ofm); | |
3428 | if (rule) { | |
3429 | delete_flow(p, rule, ofm->out_port); | |
3430 | } | |
3431 | } | |
3432 | ||
3433 | /* Callback for delete_flows_loose(). */ | |
3434 | static void | |
3435 | delete_flows_cb(struct cls_rule *rule_, void *cbdata_) | |
3436 | { | |
3437 | struct rule *rule = rule_from_cls_rule(rule_); | |
3438 | struct delete_flows_cbdata *cbdata = cbdata_; | |
3439 | ||
3440 | delete_flow(cbdata->ofproto, rule, cbdata->out_port); | |
3441 | } | |
3442 | ||
3443 | /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has | |
3444 | * been identified as a flow to delete from 'p''s flow table, by deleting the | |
3445 | * flow and sending out a OFPT_FLOW_REMOVED message to any interested | |
3446 | * controller. | |
3447 | * | |
3448 | * Will not delete 'rule' if it is hidden. Will delete 'rule' only if | |
3449 | * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the | |
3450 | * specified 'out_port'. */ | |
3451 | static void | |
3452 | delete_flow(struct ofproto *p, struct rule *rule, uint16_t out_port) | |
3453 | { | |
3454 | if (rule_is_hidden(rule)) { | |
3455 | return; | |
3456 | } | |
3457 | ||
3458 | if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) { | |
3459 | return; | |
3460 | } | |
3461 | ||
3462 | send_flow_removed(p, rule, time_msec(), OFPRR_DELETE); | |
3463 | rule_remove(p, rule); | |
3464 | } | |
3465 | \f | |
064af421 BP |
3466 | static int |
3467 | handle_flow_mod(struct ofproto *p, struct ofconn *ofconn, | |
3468 | struct ofp_flow_mod *ofm) | |
3469 | { | |
3470 | size_t n_actions; | |
3471 | int error; | |
3472 | ||
9deba63b BP |
3473 | error = reject_slave_controller(ofconn, &ofm->header); |
3474 | if (error) { | |
3475 | return error; | |
3476 | } | |
064af421 BP |
3477 | error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm, |
3478 | sizeof *ofm->actions, &n_actions); | |
3479 | if (error) { | |
3480 | return error; | |
3481 | } | |
3482 | ||
49bdc010 JP |
3483 | /* We do not support the emergency flow cache. It will hopefully |
3484 | * get dropped from OpenFlow in the near future. */ | |
3485 | if (ofm->flags & htons(OFPFF_EMERG)) { | |
3486 | /* There isn't a good fit for an error code, so just state that the | |
3487 | * flow table is full. */ | |
3488 | return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL); | |
3489 | } | |
3490 | ||
064af421 BP |
3491 | normalize_match(&ofm->match); |
3492 | if (!ofm->match.wildcards) { | |
3493 | ofm->priority = htons(UINT16_MAX); | |
3494 | } | |
3495 | ||
3496 | error = validate_actions((const union ofp_action *) ofm->actions, | |
3497 | n_actions, p->max_ports); | |
3498 | if (error) { | |
3499 | return error; | |
3500 | } | |
3501 | ||
3502 | switch (ntohs(ofm->command)) { | |
3503 | case OFPFC_ADD: | |
3504 | return add_flow(p, ofconn, ofm, n_actions); | |
3505 | ||
3506 | case OFPFC_MODIFY: | |
79eee1eb | 3507 | return modify_flows_loose(p, ofconn, ofm, n_actions); |
064af421 BP |
3508 | |
3509 | case OFPFC_MODIFY_STRICT: | |
79eee1eb | 3510 | return modify_flow_strict(p, ofconn, ofm, n_actions); |
064af421 BP |
3511 | |
3512 | case OFPFC_DELETE: | |
79eee1eb BP |
3513 | delete_flows_loose(p, ofm); |
3514 | return 0; | |
064af421 BP |
3515 | |
3516 | case OFPFC_DELETE_STRICT: | |
79eee1eb BP |
3517 | delete_flow_strict(p, ofm); |
3518 | return 0; | |
064af421 BP |
3519 | |
3520 | default: | |
3521 | return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND); | |
3522 | } | |
3523 | } | |
3524 | ||
659586ef JG |
3525 | static int |
3526 | handle_tun_id_from_cookie(struct ofproto *p, struct nxt_tun_id_cookie *msg) | |
3527 | { | |
3528 | int error; | |
3529 | ||
3530 | error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg); | |
3531 | if (error) { | |
3532 | return error; | |
3533 | } | |
3534 | ||
3535 | p->tun_id_from_cookie = !!msg->set; | |
3536 | return 0; | |
3537 | } | |
3538 | ||
9deba63b BP |
3539 | static int |
3540 | handle_role_request(struct ofproto *ofproto, | |
3541 | struct ofconn *ofconn, struct nicira_header *msg) | |
3542 | { | |
3543 | struct nx_role_request *nrr; | |
3544 | struct nx_role_request *reply; | |
3545 | struct ofpbuf *buf; | |
3546 | uint32_t role; | |
3547 | ||
3548 | if (ntohs(msg->header.length) != sizeof *nrr) { | |
100e95db | 3549 | VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)", |
9deba63b BP |
3550 | ntohs(msg->header.length), sizeof *nrr); |
3551 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
3552 | } | |
3553 | nrr = (struct nx_role_request *) msg; | |
3554 | ||
3555 | if (ofconn->type != OFCONN_CONTROLLER) { | |
3556 | VLOG_WARN_RL(&rl, "ignoring role request on non-controller " | |
3557 | "connection"); | |
3558 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM); | |
3559 | } | |
3560 | ||
3561 | role = ntohl(nrr->role); | |
3562 | if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER | |
3563 | && role != NX_ROLE_SLAVE) { | |
3564 | VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role); | |
3565 | ||
3566 | /* There's no good error code for this. */ | |
3567 | return ofp_mkerr(OFPET_BAD_REQUEST, -1); | |
3568 | } | |
3569 | ||
3570 | if (role == NX_ROLE_MASTER) { | |
3571 | struct ofconn *other; | |
3572 | ||
3573 | HMAP_FOR_EACH (other, struct ofconn, hmap_node, | |
3574 | &ofproto->controllers) { | |
3575 | if (other->role == NX_ROLE_MASTER) { | |
3576 | other->role = NX_ROLE_SLAVE; | |
3577 | } | |
3578 | } | |
3579 | } | |
3580 | ofconn->role = role; | |
3581 | ||
3582 | reply = make_openflow_xid(sizeof *reply, OFPT_VENDOR, msg->header.xid, | |
3583 | &buf); | |
3584 | reply->nxh.vendor = htonl(NX_VENDOR_ID); | |
3585 | reply->nxh.subtype = htonl(NXT_ROLE_REPLY); | |
3586 | reply->role = htonl(role); | |
3587 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
3588 | ||
3589 | return 0; | |
3590 | } | |
3591 | ||
064af421 BP |
3592 | static int |
3593 | handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg) | |
3594 | { | |
3595 | struct ofp_vendor_header *ovh = msg; | |
3596 | struct nicira_header *nh; | |
3597 | ||
3598 | if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) { | |
100e95db | 3599 | VLOG_WARN_RL(&rl, "received vendor message of length %u " |
659586ef JG |
3600 | "(expected at least %zu)", |
3601 | ntohs(ovh->header.length), sizeof(struct ofp_vendor_header)); | |
49bdc010 | 3602 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); |
064af421 BP |
3603 | } |
3604 | if (ovh->vendor != htonl(NX_VENDOR_ID)) { | |
3605 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR); | |
3606 | } | |
3607 | if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) { | |
100e95db | 3608 | VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u " |
659586ef JG |
3609 | "(expected at least %zu)", |
3610 | ntohs(ovh->header.length), sizeof(struct nicira_header)); | |
49bdc010 | 3611 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); |
064af421 BP |
3612 | } |
3613 | ||
3614 | nh = msg; | |
3615 | switch (ntohl(nh->subtype)) { | |
3616 | case NXT_STATUS_REQUEST: | |
3617 | return switch_status_handle_request(p->switch_status, ofconn->rconn, | |
3618 | msg); | |
659586ef JG |
3619 | |
3620 | case NXT_TUN_ID_FROM_COOKIE: | |
3621 | return handle_tun_id_from_cookie(p, msg); | |
9deba63b BP |
3622 | |
3623 | case NXT_ROLE_REQUEST: | |
3624 | return handle_role_request(p, ofconn, msg); | |
064af421 BP |
3625 | } |
3626 | ||
3627 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); | |
3628 | } | |
3629 | ||
246e61ea JP |
3630 | static int |
3631 | handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh) | |
3632 | { | |
3633 | struct ofp_header *ob; | |
3634 | struct ofpbuf *buf; | |
3635 | ||
3636 | /* Currently, everything executes synchronously, so we can just | |
3637 | * immediately send the barrier reply. */ | |
3638 | ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf); | |
3639 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
3640 | return 0; | |
3641 | } | |
3642 | ||
064af421 BP |
3643 | static void |
3644 | handle_openflow(struct ofconn *ofconn, struct ofproto *p, | |
3645 | struct ofpbuf *ofp_msg) | |
3646 | { | |
3647 | struct ofp_header *oh = ofp_msg->data; | |
3648 | int error; | |
3649 | ||
3650 | COVERAGE_INC(ofproto_recv_openflow); | |
3651 | switch (oh->type) { | |
3652 | case OFPT_ECHO_REQUEST: | |
3653 | error = handle_echo_request(ofconn, oh); | |
3654 | break; | |
3655 | ||
3656 | case OFPT_ECHO_REPLY: | |
3657 | error = 0; | |
3658 | break; | |
3659 | ||
3660 | case OFPT_FEATURES_REQUEST: | |
3661 | error = handle_features_request(p, ofconn, oh); | |
3662 | break; | |
3663 | ||
3664 | case OFPT_GET_CONFIG_REQUEST: | |
3665 | error = handle_get_config_request(p, ofconn, oh); | |
3666 | break; | |
3667 | ||
3668 | case OFPT_SET_CONFIG: | |
3669 | error = handle_set_config(p, ofconn, ofp_msg->data); | |
3670 | break; | |
3671 | ||
3672 | case OFPT_PACKET_OUT: | |
3673 | error = handle_packet_out(p, ofconn, ofp_msg->data); | |
3674 | break; | |
3675 | ||
3676 | case OFPT_PORT_MOD: | |
9deba63b | 3677 | error = handle_port_mod(p, ofconn, oh); |
064af421 BP |
3678 | break; |
3679 | ||
3680 | case OFPT_FLOW_MOD: | |
3681 | error = handle_flow_mod(p, ofconn, ofp_msg->data); | |
3682 | break; | |
3683 | ||
3684 | case OFPT_STATS_REQUEST: | |
3685 | error = handle_stats_request(p, ofconn, oh); | |
3686 | break; | |
3687 | ||
3688 | case OFPT_VENDOR: | |
3689 | error = handle_vendor(p, ofconn, ofp_msg->data); | |
3690 | break; | |
3691 | ||
246e61ea JP |
3692 | case OFPT_BARRIER_REQUEST: |
3693 | error = handle_barrier_request(ofconn, oh); | |
3694 | break; | |
3695 | ||
064af421 BP |
3696 | default: |
3697 | if (VLOG_IS_WARN_ENABLED()) { | |
3698 | char *s = ofp_to_string(oh, ntohs(oh->length), 2); | |
3699 | VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s); | |
3700 | free(s); | |
3701 | } | |
3702 | error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE); | |
3703 | break; | |
3704 | } | |
3705 | ||
3706 | if (error) { | |
3707 | send_error_oh(ofconn, ofp_msg->data, error); | |
3708 | } | |
3709 | } | |
3710 | \f | |
3711 | static void | |
72b06300 | 3712 | handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet) |
064af421 BP |
3713 | { |
3714 | struct odp_msg *msg = packet->data; | |
064af421 BP |
3715 | struct rule *rule; |
3716 | struct ofpbuf payload; | |
3717 | flow_t flow; | |
3718 | ||
064af421 BP |
3719 | payload.data = msg + 1; |
3720 | payload.size = msg->length - sizeof *msg; | |
659586ef | 3721 | flow_extract(&payload, msg->arg, msg->port, &flow); |
064af421 | 3722 | |
0ad9b732 JP |
3723 | /* Check with in-band control to see if this packet should be sent |
3724 | * to the local port regardless of the flow table. */ | |
3725 | if (in_band_msg_in_hook(p->in_band, &flow, &payload)) { | |
3726 | union odp_action action; | |
3727 | ||
3728 | memset(&action, 0, sizeof(action)); | |
3729 | action.output.type = ODPAT_OUTPUT; | |
3730 | action.output.port = ODPP_LOCAL; | |
f1acd62b | 3731 | dpif_execute(p->dpif, flow.in_port, &action, 1, &payload); |
0ad9b732 JP |
3732 | } |
3733 | ||
064af421 BP |
3734 | rule = lookup_valid_rule(p, &flow); |
3735 | if (!rule) { | |
3736 | /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */ | |
3737 | struct ofport *port = port_array_get(&p->ports, msg->port); | |
3738 | if (port) { | |
3739 | if (port->opp.config & OFPPC_NO_PACKET_IN) { | |
3740 | COVERAGE_INC(ofproto_no_packet_in); | |
3741 | /* XXX install 'drop' flow entry */ | |
3742 | ofpbuf_delete(packet); | |
3743 | return; | |
3744 | } | |
3745 | } else { | |
3746 | VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port); | |
3747 | } | |
3748 | ||
3749 | COVERAGE_INC(ofproto_packet_in); | |
76ce9432 | 3750 | send_packet_in(p, packet); |
064af421 BP |
3751 | return; |
3752 | } | |
3753 | ||
3754 | if (rule->cr.wc.wildcards) { | |
3755 | rule = rule_create_subrule(p, rule, &flow); | |
3756 | rule_make_actions(p, rule, packet); | |
3757 | } else { | |
3758 | if (!rule->may_install) { | |
3759 | /* The rule is not installable, that is, we need to process every | |
3760 | * packet, so process the current packet and set its actions into | |
3761 | * 'subrule'. */ | |
3762 | rule_make_actions(p, rule, packet); | |
3763 | } else { | |
3764 | /* XXX revalidate rule if it needs it */ | |
3765 | } | |
3766 | } | |
3767 | ||
3768 | rule_execute(p, rule, &payload, &flow); | |
3769 | rule_reinstall(p, rule); | |
7778bd15 | 3770 | |
76ce9432 | 3771 | if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY) { |
7778bd15 BP |
3772 | /* |
3773 | * Extra-special case for fail-open mode. | |
3774 | * | |
3775 | * We are in fail-open mode and the packet matched the fail-open rule, | |
3776 | * but we are connected to a controller too. We should send the packet | |
3777 | * up to the controller in the hope that it will try to set up a flow | |
3778 | * and thereby allow us to exit fail-open. | |
3779 | * | |
3780 | * See the top-level comment in fail-open.c for more information. | |
3781 | */ | |
76ce9432 | 3782 | send_packet_in(p, packet); |
7778bd15 BP |
3783 | } else { |
3784 | ofpbuf_delete(packet); | |
3785 | } | |
064af421 | 3786 | } |
72b06300 BP |
3787 | |
3788 | static void | |
3789 | handle_odp_msg(struct ofproto *p, struct ofpbuf *packet) | |
3790 | { | |
3791 | struct odp_msg *msg = packet->data; | |
3792 | ||
3793 | switch (msg->type) { | |
3794 | case _ODPL_ACTION_NR: | |
3795 | COVERAGE_INC(ofproto_ctlr_action); | |
76ce9432 | 3796 | send_packet_in(p, packet); |
72b06300 BP |
3797 | break; |
3798 | ||
3799 | case _ODPL_SFLOW_NR: | |
3800 | if (p->sflow) { | |
3801 | ofproto_sflow_received(p->sflow, msg); | |
3802 | } | |
3803 | ofpbuf_delete(packet); | |
3804 | break; | |
3805 | ||
3806 | case _ODPL_MISS_NR: | |
3807 | handle_odp_miss_msg(p, packet); | |
3808 | break; | |
3809 | ||
3810 | default: | |
3811 | VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32, | |
3812 | msg->type); | |
3813 | break; | |
3814 | } | |
3815 | } | |
064af421 BP |
3816 | \f |
3817 | static void | |
3818 | revalidate_cb(struct cls_rule *sub_, void *cbdata_) | |
3819 | { | |
3820 | struct rule *sub = rule_from_cls_rule(sub_); | |
3821 | struct revalidate_cbdata *cbdata = cbdata_; | |
3822 | ||
3823 | if (cbdata->revalidate_all | |
3824 | || (cbdata->revalidate_subrules && sub->super) | |
3825 | || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) { | |
3826 | revalidate_rule(cbdata->ofproto, sub); | |
3827 | } | |
3828 | } | |
3829 | ||
3830 | static bool | |
3831 | revalidate_rule(struct ofproto *p, struct rule *rule) | |
3832 | { | |
3833 | const flow_t *flow = &rule->cr.flow; | |
3834 | ||
3835 | COVERAGE_INC(ofproto_revalidate_rule); | |
3836 | if (rule->super) { | |
3837 | struct rule *super; | |
3838 | super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow)); | |
3839 | if (!super) { | |
3840 | rule_remove(p, rule); | |
3841 | return false; | |
3842 | } else if (super != rule->super) { | |
3843 | COVERAGE_INC(ofproto_revalidate_moved); | |
3844 | list_remove(&rule->list); | |
3845 | list_push_back(&super->list, &rule->list); | |
3846 | rule->super = super; | |
3847 | rule->hard_timeout = super->hard_timeout; | |
3848 | rule->idle_timeout = super->idle_timeout; | |
3849 | rule->created = super->created; | |
3850 | rule->used = 0; | |
3851 | } | |
3852 | } | |
3853 | ||
3854 | rule_update_actions(p, rule); | |
3855 | return true; | |
3856 | } | |
3857 | ||
3858 | static struct ofpbuf * | |
659586ef JG |
3859 | compose_flow_removed(struct ofproto *p, const struct rule *rule, |
3860 | long long int now, uint8_t reason) | |
064af421 | 3861 | { |
ca069229 | 3862 | struct ofp_flow_removed *ofr; |
064af421 | 3863 | struct ofpbuf *buf; |
9ca76894 | 3864 | long long int tdiff = now - rule->created; |
26c3f94a JP |
3865 | uint32_t sec = tdiff / 1000; |
3866 | uint32_t msec = tdiff - (sec * 1000); | |
064af421 | 3867 | |
ca069229 | 3868 | ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf); |
659586ef JG |
3869 | flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, p->tun_id_from_cookie, |
3870 | &ofr->match); | |
39997502 | 3871 | ofr->cookie = rule->flow_cookie; |
ca069229 JP |
3872 | ofr->priority = htons(rule->cr.priority); |
3873 | ofr->reason = reason; | |
26c3f94a JP |
3874 | ofr->duration_sec = htonl(sec); |
3875 | ofr->duration_nsec = htonl(msec * 1000000); | |
ca069229 JP |
3876 | ofr->idle_timeout = htons(rule->idle_timeout); |
3877 | ofr->packet_count = htonll(rule->packet_count); | |
3878 | ofr->byte_count = htonll(rule->byte_count); | |
064af421 BP |
3879 | |
3880 | return buf; | |
3881 | } | |
3882 | ||
3883 | static void | |
ca069229 JP |
3884 | uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule) |
3885 | { | |
3886 | assert(rule->installed); | |
3887 | assert(!rule->cr.wc.wildcards); | |
3888 | ||
3889 | if (rule->super) { | |
3890 | rule_remove(ofproto, rule); | |
3891 | } else { | |
3892 | rule_uninstall(ofproto, rule); | |
3893 | } | |
3894 | } | |
9deba63b | 3895 | |
ca069229 JP |
3896 | static void |
3897 | send_flow_removed(struct ofproto *p, struct rule *rule, | |
3898 | long long int now, uint8_t reason) | |
064af421 BP |
3899 | { |
3900 | struct ofconn *ofconn; | |
3901 | struct ofconn *prev; | |
b9b0ce61 | 3902 | struct ofpbuf *buf = NULL; |
064af421 BP |
3903 | |
3904 | /* We limit the maximum number of queued flow expirations it by accounting | |
3905 | * them under the counter for replies. That works because preventing | |
3906 | * OpenFlow requests from being processed also prevents new flows from | |
3907 | * being added (and expiring). (It also prevents processing OpenFlow | |
3908 | * requests that would not add new flows, so it is imperfect.) */ | |
3909 | ||
3910 | prev = NULL; | |
3911 | LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) { | |
9deba63b | 3912 | if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn) |
c91248b3 | 3913 | && ofconn_receives_async_msgs(ofconn)) { |
064af421 | 3914 | if (prev) { |
431d8ad2 | 3915 | queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter); |
064af421 | 3916 | } else { |
659586ef | 3917 | buf = compose_flow_removed(p, rule, now, reason); |
064af421 BP |
3918 | } |
3919 | prev = ofconn; | |
3920 | } | |
3921 | } | |
3922 | if (prev) { | |
431d8ad2 | 3923 | queue_tx(buf, prev, prev->reply_counter); |
064af421 BP |
3924 | } |
3925 | } | |
3926 | ||
064af421 BP |
3927 | |
3928 | static void | |
3929 | expire_rule(struct cls_rule *cls_rule, void *p_) | |
3930 | { | |
3931 | struct ofproto *p = p_; | |
3932 | struct rule *rule = rule_from_cls_rule(cls_rule); | |
3933 | long long int hard_expire, idle_expire, expire, now; | |
3934 | ||
3935 | hard_expire = (rule->hard_timeout | |
3936 | ? rule->created + rule->hard_timeout * 1000 | |
3937 | : LLONG_MAX); | |
3938 | idle_expire = (rule->idle_timeout | |
3939 | && (rule->super || list_is_empty(&rule->list)) | |
3940 | ? rule->used + rule->idle_timeout * 1000 | |
3941 | : LLONG_MAX); | |
3942 | expire = MIN(hard_expire, idle_expire); | |
064af421 BP |
3943 | |
3944 | now = time_msec(); | |
3945 | if (now < expire) { | |
3946 | if (rule->installed && now >= rule->used + 5000) { | |
3947 | uninstall_idle_flow(p, rule); | |
0193b2af JG |
3948 | } else if (!rule->cr.wc.wildcards) { |
3949 | active_timeout(p, rule); | |
064af421 | 3950 | } |
0193b2af | 3951 | |
064af421 BP |
3952 | return; |
3953 | } | |
3954 | ||
3955 | COVERAGE_INC(ofproto_expired); | |
46d6f36f JG |
3956 | |
3957 | /* Update stats. This code will be a no-op if the rule expired | |
3958 | * due to an idle timeout. */ | |
064af421 | 3959 | if (rule->cr.wc.wildcards) { |
064af421 BP |
3960 | struct rule *subrule, *next; |
3961 | LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) { | |
3962 | rule_remove(p, subrule); | |
3963 | } | |
46d6f36f JG |
3964 | } else { |
3965 | rule_uninstall(p, rule); | |
064af421 BP |
3966 | } |
3967 | ||
8fe1a59d | 3968 | if (!rule_is_hidden(rule)) { |
ca069229 JP |
3969 | send_flow_removed(p, rule, now, |
3970 | (now >= hard_expire | |
3971 | ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT)); | |
8fe1a59d | 3972 | } |
064af421 BP |
3973 | rule_remove(p, rule); |
3974 | } | |
3975 | ||
0193b2af JG |
3976 | static void |
3977 | active_timeout(struct ofproto *ofproto, struct rule *rule) | |
3978 | { | |
3979 | if (ofproto->netflow && !is_controller_rule(rule) && | |
3980 | netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) { | |
3981 | struct ofexpired expired; | |
3982 | struct odp_flow odp_flow; | |
3983 | ||
3984 | /* Get updated flow stats. */ | |
3985 | memset(&odp_flow, 0, sizeof odp_flow); | |
094e1514 JG |
3986 | if (rule->installed) { |
3987 | odp_flow.key = rule->cr.flow; | |
3988 | odp_flow.flags = ODPFF_ZERO_TCP_FLAGS; | |
d65349ea | 3989 | dpif_flow_get(ofproto->dpif, &odp_flow); |
094e1514 JG |
3990 | |
3991 | if (odp_flow.stats.n_packets) { | |
3992 | update_time(ofproto, rule, &odp_flow.stats); | |
3993 | netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos, | |
3994 | odp_flow.stats.tcp_flags); | |
3995 | } | |
0193b2af JG |
3996 | } |
3997 | ||
3998 | expired.flow = rule->cr.flow; | |
3999 | expired.packet_count = rule->packet_count + | |
4000 | odp_flow.stats.n_packets; | |
4001 | expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes; | |
4002 | expired.used = rule->used; | |
4003 | ||
4004 | netflow_expire(ofproto->netflow, &rule->nf_flow, &expired); | |
4005 | ||
4006 | /* Schedule us to send the accumulated records once we have | |
4007 | * collected all of them. */ | |
4008 | poll_immediate_wake(); | |
4009 | } | |
4010 | } | |
4011 | ||
064af421 BP |
4012 | static void |
4013 | update_used(struct ofproto *p) | |
4014 | { | |
4015 | struct odp_flow *flows; | |
4016 | size_t n_flows; | |
4017 | size_t i; | |
4018 | int error; | |
4019 | ||
c228a364 | 4020 | error = dpif_flow_list_all(p->dpif, &flows, &n_flows); |
064af421 BP |
4021 | if (error) { |
4022 | return; | |
4023 | } | |
4024 | ||
4025 | for (i = 0; i < n_flows; i++) { | |
4026 | struct odp_flow *f = &flows[i]; | |
4027 | struct rule *rule; | |
4028 | ||
4029 | rule = rule_from_cls_rule( | |
4030 | classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX)); | |
4031 | if (!rule || !rule->installed) { | |
4032 | COVERAGE_INC(ofproto_unexpected_rule); | |
c228a364 | 4033 | dpif_flow_del(p->dpif, f); |
064af421 BP |
4034 | continue; |
4035 | } | |
4036 | ||
0193b2af | 4037 | update_time(p, rule, &f->stats); |
064af421 BP |
4038 | rule_account(p, rule, f->stats.n_bytes); |
4039 | } | |
4040 | free(flows); | |
4041 | } | |
4042 | ||
43253595 | 4043 | /* pinsched callback for sending 'packet' on 'ofconn'. */ |
064af421 | 4044 | static void |
76ce9432 | 4045 | do_send_packet_in(struct ofpbuf *packet, void *ofconn_) |
064af421 | 4046 | { |
76ce9432 | 4047 | struct ofconn *ofconn = ofconn_; |
43253595 BP |
4048 | |
4049 | rconn_send_with_limit(ofconn->rconn, packet, | |
4050 | ofconn->packet_in_counter, 100); | |
4051 | } | |
4052 | ||
4053 | /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and | |
4054 | * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s | |
4055 | * packet scheduler for sending. | |
4056 | * | |
30ea5d93 BP |
4057 | * 'max_len' specifies the maximum number of bytes of the packet to send on |
4058 | * 'ofconn' (INT_MAX specifies no limit). | |
4059 | * | |
43253595 BP |
4060 | * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise, |
4061 | * ownership is transferred to this function. */ | |
4062 | static void | |
30ea5d93 BP |
4063 | schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len, |
4064 | bool clone) | |
43253595 | 4065 | { |
76ce9432 | 4066 | struct ofproto *ofproto = ofconn->ofproto; |
43253595 BP |
4067 | struct ofp_packet_in *opi = packet->data; |
4068 | uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port)); | |
4069 | int send_len, trim_size; | |
76ce9432 | 4070 | uint32_t buffer_id; |
064af421 | 4071 | |
43253595 BP |
4072 | /* Get buffer. */ |
4073 | if (opi->reason == OFPR_ACTION) { | |
76ce9432 | 4074 | buffer_id = UINT32_MAX; |
43253595 BP |
4075 | } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) { |
4076 | buffer_id = pktbuf_get_null(); | |
89b9612d BP |
4077 | } else if (!ofconn->pktbuf) { |
4078 | buffer_id = UINT32_MAX; | |
76ce9432 | 4079 | } else { |
43253595 BP |
4080 | struct ofpbuf payload; |
4081 | payload.data = opi->data; | |
4082 | payload.size = packet->size - offsetof(struct ofp_packet_in, data); | |
4083 | buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port); | |
76ce9432 | 4084 | } |
372179d4 | 4085 | |
43253595 BP |
4086 | /* Figure out how much of the packet to send. */ |
4087 | send_len = ntohs(opi->total_len); | |
4088 | if (buffer_id != UINT32_MAX) { | |
4089 | send_len = MIN(send_len, ofconn->miss_send_len); | |
4090 | } | |
30ea5d93 | 4091 | send_len = MIN(send_len, max_len); |
064af421 | 4092 | |
43253595 BP |
4093 | /* Adjust packet length and clone if necessary. */ |
4094 | trim_size = offsetof(struct ofp_packet_in, data) + send_len; | |
4095 | if (clone) { | |
4096 | packet = ofpbuf_clone_data(packet->data, trim_size); | |
4097 | opi = packet->data; | |
4098 | } else { | |
4099 | packet->size = trim_size; | |
4100 | } | |
4101 | ||
4102 | /* Update packet headers. */ | |
4103 | opi->buffer_id = htonl(buffer_id); | |
4104 | update_openflow_length(packet); | |
4105 | ||
4106 | /* Hand over to packet scheduler. It might immediately call into | |
4107 | * do_send_packet_in() or it might buffer it for a while (until a later | |
4108 | * call to pinsched_run()). */ | |
4109 | pinsched_send(ofconn->schedulers[opi->reason], in_port, | |
4110 | packet, do_send_packet_in, ofconn); | |
064af421 BP |
4111 | } |
4112 | ||
43253595 BP |
4113 | /* Replace struct odp_msg header in 'packet' by equivalent struct |
4114 | * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as | |
4115 | * returned by dpif_recv()). | |
4116 | * | |
4117 | * The conversion is not complete: the caller still needs to trim any unneeded | |
4118 | * payload off the end of the buffer, set the length in the OpenFlow header, | |
4119 | * and set buffer_id. Those require us to know the controller settings and so | |
30ea5d93 BP |
4120 | * must be done on a per-controller basis. |
4121 | * | |
4122 | * Returns the maximum number of bytes of the packet that should be sent to | |
4123 | * the controller (INT_MAX if no limit). */ | |
4124 | static int | |
43253595 | 4125 | do_convert_to_packet_in(struct ofpbuf *packet) |
064af421 | 4126 | { |
76ce9432 | 4127 | struct odp_msg *msg = packet->data; |
43253595 BP |
4128 | struct ofp_packet_in *opi; |
4129 | uint8_t reason; | |
4130 | uint16_t total_len; | |
4131 | uint16_t in_port; | |
30ea5d93 | 4132 | int max_len; |
43253595 BP |
4133 | |
4134 | /* Extract relevant header fields */ | |
30ea5d93 BP |
4135 | if (msg->type == _ODPL_ACTION_NR) { |
4136 | reason = OFPR_ACTION; | |
4137 | max_len = msg->arg; | |
4138 | } else { | |
4139 | reason = OFPR_NO_MATCH; | |
4140 | max_len = INT_MAX; | |
4141 | } | |
43253595 BP |
4142 | total_len = msg->length - sizeof *msg; |
4143 | in_port = odp_port_to_ofp_port(msg->port); | |
4144 | ||
4145 | /* Repurpose packet buffer by overwriting header. */ | |
4146 | ofpbuf_pull(packet, sizeof(struct odp_msg)); | |
4147 | opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data)); | |
4148 | opi->header.version = OFP_VERSION; | |
4149 | opi->header.type = OFPT_PACKET_IN; | |
4150 | opi->total_len = htons(total_len); | |
4151 | opi->in_port = htons(in_port); | |
4152 | opi->reason = reason; | |
30ea5d93 BP |
4153 | |
4154 | return max_len; | |
43253595 BP |
4155 | } |
4156 | ||
4157 | /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or | |
4158 | * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller | |
4159 | * as necessary according to their individual configurations. | |
4160 | * | |
4161 | * 'packet' must have sufficient headroom to convert it into a struct | |
4162 | * ofp_packet_in (e.g. as returned by dpif_recv()). | |
4163 | * | |
4164 | * Takes ownership of 'packet'. */ | |
4165 | static void | |
4166 | send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet) | |
4167 | { | |
76ce9432 | 4168 | struct ofconn *ofconn, *prev; |
30ea5d93 | 4169 | int max_len; |
064af421 | 4170 | |
30ea5d93 | 4171 | max_len = do_convert_to_packet_in(packet); |
76ce9432 BP |
4172 | |
4173 | prev = NULL; | |
4174 | LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) { | |
c91248b3 | 4175 | if (ofconn_receives_async_msgs(ofconn)) { |
9deba63b | 4176 | if (prev) { |
30ea5d93 | 4177 | schedule_packet_in(prev, packet, max_len, true); |
9deba63b BP |
4178 | } |
4179 | prev = ofconn; | |
064af421 | 4180 | } |
76ce9432 BP |
4181 | } |
4182 | if (prev) { | |
30ea5d93 | 4183 | schedule_packet_in(prev, packet, max_len, false); |
76ce9432 BP |
4184 | } else { |
4185 | ofpbuf_delete(packet); | |
064af421 | 4186 | } |
064af421 BP |
4187 | } |
4188 | ||
4189 | static uint64_t | |
fa60c019 | 4190 | pick_datapath_id(const struct ofproto *ofproto) |
064af421 | 4191 | { |
fa60c019 | 4192 | const struct ofport *port; |
064af421 | 4193 | |
fa60c019 BP |
4194 | port = port_array_get(&ofproto->ports, ODPP_LOCAL); |
4195 | if (port) { | |
4196 | uint8_t ea[ETH_ADDR_LEN]; | |
4197 | int error; | |
4198 | ||
4199 | error = netdev_get_etheraddr(port->netdev, ea); | |
064af421 BP |
4200 | if (!error) { |
4201 | return eth_addr_to_uint64(ea); | |
4202 | } | |
4203 | VLOG_WARN("could not get MAC address for %s (%s)", | |
fa60c019 | 4204 | netdev_get_name(port->netdev), strerror(error)); |
064af421 | 4205 | } |
fa60c019 | 4206 | return ofproto->fallback_dpid; |
064af421 BP |
4207 | } |
4208 | ||
4209 | static uint64_t | |
4210 | pick_fallback_dpid(void) | |
4211 | { | |
4212 | uint8_t ea[ETH_ADDR_LEN]; | |
70150daf | 4213 | eth_addr_nicira_random(ea); |
064af421 BP |
4214 | return eth_addr_to_uint64(ea); |
4215 | } | |
4216 | \f | |
4217 | static bool | |
4218 | default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet, | |
4219 | struct odp_actions *actions, tag_type *tags, | |
6a07af36 | 4220 | uint16_t *nf_output_iface, void *ofproto_) |
064af421 BP |
4221 | { |
4222 | struct ofproto *ofproto = ofproto_; | |
4223 | int out_port; | |
4224 | ||
4225 | /* Drop frames for reserved multicast addresses. */ | |
4226 | if (eth_addr_is_reserved(flow->dl_dst)) { | |
4227 | return true; | |
4228 | } | |
4229 | ||
4230 | /* Learn source MAC (but don't try to learn from revalidation). */ | |
4231 | if (packet != NULL) { | |
4232 | tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src, | |
4233 | 0, flow->in_port); | |
4234 | if (rev_tag) { | |
4235 | /* The log messages here could actually be useful in debugging, | |
4236 | * so keep the rate limit relatively high. */ | |
4237 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); | |
4238 | VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16, | |
4239 | ETH_ADDR_ARGS(flow->dl_src), flow->in_port); | |
4240 | ofproto_revalidate(ofproto, rev_tag); | |
4241 | } | |
4242 | } | |
4243 | ||
4244 | /* Determine output port. */ | |
4245 | out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags); | |
4246 | if (out_port < 0) { | |
6a07af36 | 4247 | add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface); |
064af421 BP |
4248 | } else if (out_port != flow->in_port) { |
4249 | odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port; | |
6a07af36 | 4250 | *nf_output_iface = out_port; |
064af421 BP |
4251 | } else { |
4252 | /* Drop. */ | |
4253 | } | |
4254 | ||
4255 | return true; | |
4256 | } | |
4257 | ||
4258 | static const struct ofhooks default_ofhooks = { | |
4259 | NULL, | |
4260 | default_normal_ofhook_cb, | |
4261 | NULL, | |
4262 | NULL | |
4263 | }; |