]>
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" | |
fa37b408 | 39 | #include "ofp-util.h" |
72b06300 | 40 | #include "ofproto-sflow.h" |
064af421 BP |
41 | #include "ofpbuf.h" |
42 | #include "openflow/nicira-ext.h" | |
43 | #include "openflow/openflow.h" | |
064af421 BP |
44 | #include "openvswitch/datapath-protocol.h" |
45 | #include "packets.h" | |
46 | #include "pinsched.h" | |
47 | #include "pktbuf.h" | |
48 | #include "poll-loop.h" | |
49 | #include "port-array.h" | |
50 | #include "rconn.h" | |
51 | #include "shash.h" | |
52 | #include "status.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" |
5136ce49 | 59 | #include "vlog.h" |
064af421 BP |
60 | #include "xtoxll.h" |
61 | ||
5136ce49 | 62 | VLOG_DEFINE_THIS_MODULE(ofproto) |
064af421 | 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 | * | |
5899143f BP |
172 | * - "Primary" connections to ordinary OpenFlow controllers. ofproto |
173 | * maintains persistent connections to these controllers and by default | |
174 | * sends them asynchronous messages such as packet-ins. | |
76ce9432 | 175 | * |
5899143f | 176 | * - "Service" connections, e.g. from ovs-ofctl. When these connections |
76ce9432 BP |
177 | * drop, it is the other side's responsibility to reconnect them if |
178 | * necessary. ofproto does not send them asynchronous messages by default. | |
7d674866 BP |
179 | * |
180 | * Currently, active (tcp, ssl, unix) connections are always "primary" | |
181 | * connections and passive (ptcp, pssl, punix) connections are always "service" | |
182 | * connections. There is no inherent reason for this, but it reflects the | |
183 | * common case. | |
76ce9432 BP |
184 | */ |
185 | enum ofconn_type { | |
5899143f BP |
186 | OFCONN_PRIMARY, /* An ordinary OpenFlow controller. */ |
187 | OFCONN_SERVICE /* A service connection, e.g. "ovs-ofctl". */ | |
76ce9432 | 188 | }; |
064af421 | 189 | |
7d674866 BP |
190 | /* A listener for incoming OpenFlow "service" connections. */ |
191 | struct ofservice { | |
192 | struct hmap_node node; /* In struct ofproto's "services" hmap. */ | |
193 | struct pvconn *pvconn; /* OpenFlow connection listener. */ | |
194 | ||
195 | /* These are not used by ofservice directly. They are settings for | |
196 | * accepted "struct ofconn"s from the pvconn. */ | |
197 | int probe_interval; /* Max idle time before probing, in seconds. */ | |
198 | int rate_limit; /* Max packet-in rate in packets per second. */ | |
199 | int burst_limit; /* Limit on accumulating packet credits. */ | |
200 | }; | |
201 | ||
202 | static struct ofservice *ofservice_lookup(struct ofproto *, | |
203 | const char *target); | |
204 | static int ofservice_create(struct ofproto *, | |
205 | const struct ofproto_controller *); | |
206 | static void ofservice_reconfigure(struct ofservice *, | |
207 | const struct ofproto_controller *); | |
208 | static void ofservice_destroy(struct ofproto *, struct ofservice *); | |
209 | ||
76ce9432 BP |
210 | /* An OpenFlow connection. */ |
211 | struct ofconn { | |
212 | struct ofproto *ofproto; /* The ofproto that owns this connection. */ | |
213 | struct list node; /* In struct ofproto's "all_conns" list. */ | |
214 | struct rconn *rconn; /* OpenFlow connection. */ | |
215 | enum ofconn_type type; /* Type. */ | |
216 | ||
217 | /* OFPT_PACKET_IN related data. */ | |
218 | struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */ | |
219 | struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */ | |
220 | struct pktbuf *pktbuf; /* OpenFlow packet buffers. */ | |
221 | int miss_send_len; /* Bytes to send of buffered packets. */ | |
222 | ||
223 | /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow | |
224 | * requests, and the maximum number before we stop reading OpenFlow | |
225 | * requests. */ | |
064af421 BP |
226 | #define OFCONN_REPLY_MAX 100 |
227 | struct rconn_packet_counter *reply_counter; | |
76ce9432 | 228 | |
5899143f | 229 | /* type == OFCONN_PRIMARY only. */ |
9deba63b | 230 | enum nx_role role; /* Role. */ |
76ce9432 BP |
231 | struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */ |
232 | struct discovery *discovery; /* Controller discovery object, if enabled. */ | |
233 | struct status_category *ss; /* Switch status category. */ | |
d2ede7bc | 234 | enum ofproto_band band; /* In-band or out-of-band? */ |
064af421 BP |
235 | }; |
236 | ||
76ce9432 BP |
237 | /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's |
238 | * "schedulers" array. Their values are 0 and 1, and their meanings and values | |
239 | * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In | |
240 | * case anything ever changes, check their values here. */ | |
241 | #define N_SCHEDULERS 2 | |
242 | BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0); | |
243 | BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR); | |
244 | BUILD_ASSERT_DECL(OFPR_ACTION == 1); | |
245 | BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR); | |
246 | ||
247 | static struct ofconn *ofconn_create(struct ofproto *, struct rconn *, | |
248 | enum ofconn_type); | |
c475ae67 | 249 | static void ofconn_destroy(struct ofconn *); |
064af421 BP |
250 | static void ofconn_run(struct ofconn *, struct ofproto *); |
251 | static void ofconn_wait(struct ofconn *); | |
c91248b3 | 252 | static bool ofconn_receives_async_msgs(const struct ofconn *); |
eb15cdbb | 253 | static char *ofconn_make_name(const struct ofproto *, const char *target); |
7d674866 | 254 | static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst); |
c91248b3 | 255 | |
064af421 BP |
256 | static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn, |
257 | struct rconn_packet_counter *counter); | |
258 | ||
76ce9432 BP |
259 | static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg); |
260 | static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn); | |
261 | ||
064af421 BP |
262 | struct ofproto { |
263 | /* Settings. */ | |
264 | uint64_t datapath_id; /* Datapath ID. */ | |
265 | uint64_t fallback_dpid; /* Datapath ID if no better choice found. */ | |
5a719c38 JP |
266 | char *mfr_desc; /* Manufacturer. */ |
267 | char *hw_desc; /* Hardware. */ | |
268 | char *sw_desc; /* Software version. */ | |
269 | char *serial_desc; /* Serial number. */ | |
8abc4ed7 | 270 | char *dp_desc; /* Datapath description. */ |
064af421 BP |
271 | |
272 | /* Datapath. */ | |
c228a364 | 273 | struct dpif *dpif; |
e9e28be3 | 274 | struct netdev_monitor *netdev_monitor; |
064af421 BP |
275 | struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is |
276 | * OFP port nr. */ | |
277 | struct shash port_by_name; | |
278 | uint32_t max_ports; | |
279 | ||
280 | /* Configuration. */ | |
281 | struct switch_status *switch_status; | |
064af421 | 282 | struct fail_open *fail_open; |
064af421 | 283 | struct netflow *netflow; |
72b06300 | 284 | struct ofproto_sflow *sflow; |
064af421 | 285 | |
d2ede7bc BP |
286 | /* In-band control. */ |
287 | struct in_band *in_band; | |
288 | long long int next_in_band_update; | |
917e50e1 BP |
289 | struct sockaddr_in *extra_in_band_remotes; |
290 | size_t n_extra_remotes; | |
291 | ||
064af421 BP |
292 | /* Flow table. */ |
293 | struct classifier cls; | |
294 | bool need_revalidate; | |
295 | long long int next_expiration; | |
296 | struct tag_set revalidate_set; | |
659586ef | 297 | bool tun_id_from_cookie; |
064af421 BP |
298 | |
299 | /* OpenFlow connections. */ | |
76ce9432 BP |
300 | struct hmap controllers; /* Controller "struct ofconn"s. */ |
301 | struct list all_conns; /* Contains "struct ofconn"s. */ | |
31681a5d | 302 | enum ofproto_fail_mode fail_mode; |
7d674866 BP |
303 | |
304 | /* OpenFlow listeners. */ | |
305 | struct hmap services; /* Contains "struct ofservice"s. */ | |
064af421 BP |
306 | struct pvconn **snoops; |
307 | size_t n_snoops; | |
308 | ||
309 | /* Hooks for ovs-vswitchd. */ | |
310 | const struct ofhooks *ofhooks; | |
311 | void *aux; | |
312 | ||
313 | /* Used by default ofhooks. */ | |
314 | struct mac_learning *ml; | |
315 | }; | |
316 | ||
317 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
318 | ||
319 | static const struct ofhooks default_ofhooks; | |
320 | ||
fa60c019 | 321 | static uint64_t pick_datapath_id(const struct ofproto *); |
064af421 | 322 | static uint64_t pick_fallback_dpid(void); |
76ce9432 | 323 | |
064af421 | 324 | static void update_used(struct ofproto *); |
0193b2af JG |
325 | static void update_stats(struct ofproto *, struct rule *, |
326 | const struct odp_flow_stats *); | |
064af421 | 327 | static void expire_rule(struct cls_rule *, void *ofproto); |
0193b2af | 328 | static void active_timeout(struct ofproto *ofproto, struct rule *rule); |
064af421 BP |
329 | static bool revalidate_rule(struct ofproto *p, struct rule *rule); |
330 | static void revalidate_cb(struct cls_rule *rule_, void *p_); | |
331 | ||
332 | static void handle_odp_msg(struct ofproto *, struct ofpbuf *); | |
333 | ||
334 | static void handle_openflow(struct ofconn *, struct ofproto *, | |
335 | struct ofpbuf *); | |
336 | ||
72b06300 BP |
337 | static void refresh_port_groups(struct ofproto *); |
338 | ||
064af421 BP |
339 | static void update_port(struct ofproto *, const char *devname); |
340 | static int init_ports(struct ofproto *); | |
341 | static void reinit_ports(struct ofproto *); | |
342 | ||
343 | int | |
1a6f1e2a JG |
344 | ofproto_create(const char *datapath, const char *datapath_type, |
345 | const struct ofhooks *ofhooks, void *aux, | |
064af421 BP |
346 | struct ofproto **ofprotop) |
347 | { | |
064af421 BP |
348 | struct odp_stats stats; |
349 | struct ofproto *p; | |
c228a364 | 350 | struct dpif *dpif; |
064af421 BP |
351 | int error; |
352 | ||
353 | *ofprotop = NULL; | |
354 | ||
355 | /* Connect to datapath and start listening for messages. */ | |
1a6f1e2a | 356 | error = dpif_open(datapath, datapath_type, &dpif); |
064af421 BP |
357 | if (error) { |
358 | VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error)); | |
359 | return error; | |
360 | } | |
c228a364 | 361 | error = dpif_get_dp_stats(dpif, &stats); |
064af421 BP |
362 | if (error) { |
363 | VLOG_ERR("failed to obtain stats for datapath %s: %s", | |
364 | datapath, strerror(error)); | |
c228a364 | 365 | dpif_close(dpif); |
064af421 BP |
366 | return error; |
367 | } | |
72b06300 | 368 | error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW); |
064af421 BP |
369 | if (error) { |
370 | VLOG_ERR("failed to listen on datapath %s: %s", | |
371 | datapath, strerror(error)); | |
c228a364 | 372 | dpif_close(dpif); |
064af421 BP |
373 | return error; |
374 | } | |
c228a364 | 375 | dpif_flow_flush(dpif); |
8f24562a | 376 | dpif_recv_purge(dpif); |
064af421 BP |
377 | |
378 | /* Initialize settings. */ | |
ec6fde61 | 379 | p = xzalloc(sizeof *p); |
064af421 | 380 | p->fallback_dpid = pick_fallback_dpid(); |
fa60c019 | 381 | p->datapath_id = p->fallback_dpid; |
5a719c38 JP |
382 | p->mfr_desc = xstrdup(DEFAULT_MFR_DESC); |
383 | p->hw_desc = xstrdup(DEFAULT_HW_DESC); | |
384 | p->sw_desc = xstrdup(DEFAULT_SW_DESC); | |
385 | p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC); | |
23ff2821 | 386 | p->dp_desc = xstrdup(DEFAULT_DP_DESC); |
064af421 BP |
387 | |
388 | /* Initialize datapath. */ | |
389 | p->dpif = dpif; | |
8b61709d | 390 | p->netdev_monitor = netdev_monitor_create(); |
064af421 BP |
391 | port_array_init(&p->ports); |
392 | shash_init(&p->port_by_name); | |
393 | p->max_ports = stats.max_ports; | |
394 | ||
395 | /* Initialize submodules. */ | |
396 | p->switch_status = switch_status_create(p); | |
397 | p->in_band = NULL; | |
064af421 | 398 | p->fail_open = NULL; |
064af421 | 399 | p->netflow = NULL; |
72b06300 | 400 | p->sflow = NULL; |
064af421 BP |
401 | |
402 | /* Initialize flow table. */ | |
403 | classifier_init(&p->cls); | |
404 | p->need_revalidate = false; | |
405 | p->next_expiration = time_msec() + 1000; | |
406 | tag_set_init(&p->revalidate_set); | |
407 | ||
408 | /* Initialize OpenFlow connections. */ | |
409 | list_init(&p->all_conns); | |
76ce9432 | 410 | hmap_init(&p->controllers); |
7d674866 | 411 | hmap_init(&p->services); |
064af421 BP |
412 | p->snoops = NULL; |
413 | p->n_snoops = 0; | |
414 | ||
415 | /* Initialize hooks. */ | |
416 | if (ofhooks) { | |
417 | p->ofhooks = ofhooks; | |
418 | p->aux = aux; | |
419 | p->ml = NULL; | |
420 | } else { | |
421 | p->ofhooks = &default_ofhooks; | |
422 | p->aux = p; | |
423 | p->ml = mac_learning_create(); | |
424 | } | |
425 | ||
fa60c019 BP |
426 | /* Pick final datapath ID. */ |
427 | p->datapath_id = pick_datapath_id(p); | |
b123cc3c | 428 | VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id); |
fa60c019 | 429 | |
064af421 BP |
430 | *ofprotop = p; |
431 | return 0; | |
432 | } | |
433 | ||
434 | void | |
435 | ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id) | |
436 | { | |
437 | uint64_t old_dpid = p->datapath_id; | |
fa60c019 | 438 | p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p); |
064af421 | 439 | if (p->datapath_id != old_dpid) { |
b123cc3c | 440 | VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id); |
76ce9432 BP |
441 | |
442 | /* Force all active connections to reconnect, since there is no way to | |
443 | * notify a controller that the datapath ID has changed. */ | |
fa05809b | 444 | ofproto_reconnect_controllers(p); |
064af421 BP |
445 | } |
446 | } | |
447 | ||
76ce9432 BP |
448 | static bool |
449 | is_discovery_controller(const struct ofproto_controller *c) | |
450 | { | |
451 | return !strcmp(c->target, "discover"); | |
452 | } | |
453 | ||
454 | static bool | |
455 | is_in_band_controller(const struct ofproto_controller *c) | |
456 | { | |
457 | return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND; | |
458 | } | |
459 | ||
460 | /* Creates a new controller in 'ofproto'. Some of the settings are initially | |
461 | * drawn from 'c', but update_controller() needs to be called later to finish | |
462 | * the new ofconn's configuration. */ | |
463 | static void | |
464 | add_controller(struct ofproto *ofproto, const struct ofproto_controller *c) | |
465 | { | |
466 | struct discovery *discovery; | |
467 | struct ofconn *ofconn; | |
468 | ||
469 | if (is_discovery_controller(c)) { | |
470 | int error = discovery_create(c->accept_re, c->update_resolv_conf, | |
471 | ofproto->dpif, ofproto->switch_status, | |
472 | &discovery); | |
473 | if (error) { | |
474 | return; | |
475 | } | |
476 | } else { | |
477 | discovery = NULL; | |
478 | } | |
479 | ||
5899143f | 480 | ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY); |
76ce9432 BP |
481 | ofconn->pktbuf = pktbuf_create(); |
482 | ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN; | |
483 | if (discovery) { | |
484 | ofconn->discovery = discovery; | |
485 | } else { | |
eb15cdbb BP |
486 | char *name = ofconn_make_name(ofproto, c->target); |
487 | rconn_connect(ofconn->rconn, c->target, name); | |
488 | free(name); | |
76ce9432 BP |
489 | } |
490 | hmap_insert(&ofproto->controllers, &ofconn->hmap_node, | |
491 | hash_string(c->target, 0)); | |
492 | } | |
493 | ||
494 | /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's | |
495 | * target or turn discovery on or off (these are done by creating new ofconns | |
496 | * and deleting old ones), but it can update the rest of an ofconn's | |
497 | * settings. */ | |
498 | static void | |
499 | update_controller(struct ofconn *ofconn, const struct ofproto_controller *c) | |
064af421 | 500 | { |
76ce9432 | 501 | int probe_interval; |
79c9f2ee | 502 | |
d2ede7bc BP |
503 | ofconn->band = (is_in_band_controller(c) |
504 | ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND); | |
505 | ||
76ce9432 | 506 | rconn_set_max_backoff(ofconn->rconn, c->max_backoff); |
79c9f2ee | 507 | |
76ce9432 BP |
508 | probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0; |
509 | rconn_set_probe_interval(ofconn->rconn, probe_interval); | |
79c9f2ee | 510 | |
76ce9432 BP |
511 | if (ofconn->discovery) { |
512 | discovery_set_update_resolv_conf(ofconn->discovery, | |
513 | c->update_resolv_conf); | |
514 | discovery_set_accept_controller_re(ofconn->discovery, c->accept_re); | |
515 | } | |
79c9f2ee | 516 | |
7d674866 | 517 | ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit); |
76ce9432 | 518 | } |
79c9f2ee | 519 | |
76ce9432 BP |
520 | static const char * |
521 | ofconn_get_target(const struct ofconn *ofconn) | |
522 | { | |
eb15cdbb | 523 | return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn); |
76ce9432 BP |
524 | } |
525 | ||
526 | static struct ofconn * | |
527 | find_controller_by_target(struct ofproto *ofproto, const char *target) | |
528 | { | |
529 | struct ofconn *ofconn; | |
530 | ||
531 | HMAP_FOR_EACH_WITH_HASH (ofconn, struct ofconn, hmap_node, | |
532 | hash_string(target, 0), &ofproto->controllers) { | |
533 | if (!strcmp(ofconn_get_target(ofconn), target)) { | |
534 | return ofconn; | |
79c9f2ee | 535 | } |
064af421 | 536 | } |
76ce9432 BP |
537 | return NULL; |
538 | } | |
064af421 | 539 | |
d2ede7bc BP |
540 | static void |
541 | update_in_band_remotes(struct ofproto *ofproto) | |
542 | { | |
543 | const struct ofconn *ofconn; | |
544 | struct sockaddr_in *addrs; | |
917e50e1 | 545 | size_t max_addrs, n_addrs; |
d2ede7bc | 546 | bool discovery; |
917e50e1 | 547 | size_t i; |
d2ede7bc | 548 | |
917e50e1 BP |
549 | /* Allocate enough memory for as many remotes as we could possibly have. */ |
550 | max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers); | |
551 | addrs = xmalloc(max_addrs * sizeof *addrs); | |
d2ede7bc BP |
552 | n_addrs = 0; |
553 | ||
554 | /* Add all the remotes. */ | |
555 | discovery = false; | |
556 | HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &ofproto->controllers) { | |
557 | struct sockaddr_in *sin = &addrs[n_addrs]; | |
558 | ||
487ec65f BP |
559 | if (ofconn->band == OFPROTO_OUT_OF_BAND) { |
560 | continue; | |
561 | } | |
562 | ||
d2ede7bc BP |
563 | sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn); |
564 | if (sin->sin_addr.s_addr) { | |
565 | sin->sin_port = rconn_get_remote_port(ofconn->rconn); | |
566 | n_addrs++; | |
567 | } | |
568 | if (ofconn->discovery) { | |
569 | discovery = true; | |
570 | } | |
571 | } | |
917e50e1 BP |
572 | for (i = 0; i < ofproto->n_extra_remotes; i++) { |
573 | addrs[n_addrs++] = ofproto->extra_in_band_remotes[i]; | |
574 | } | |
d2ede7bc BP |
575 | |
576 | /* Create or update or destroy in-band. | |
577 | * | |
578 | * Ordinarily we only enable in-band if there's at least one remote | |
579 | * address, but discovery needs the in-band rules for DHCP to be installed | |
580 | * even before we know any remote addresses. */ | |
581 | if (n_addrs || discovery) { | |
582 | if (!ofproto->in_band) { | |
583 | in_band_create(ofproto, ofproto->dpif, ofproto->switch_status, | |
584 | &ofproto->in_band); | |
585 | } | |
40cae670 BP |
586 | if (ofproto->in_band) { |
587 | in_band_set_remotes(ofproto->in_band, addrs, n_addrs); | |
588 | } | |
d2ede7bc BP |
589 | ofproto->next_in_band_update = time_msec() + 1000; |
590 | } else { | |
591 | in_band_destroy(ofproto->in_band); | |
592 | ofproto->in_band = NULL; | |
593 | } | |
594 | ||
595 | /* Clean up. */ | |
596 | free(addrs); | |
597 | } | |
598 | ||
31681a5d JP |
599 | static void |
600 | update_fail_open(struct ofproto *p) | |
601 | { | |
602 | struct ofconn *ofconn; | |
603 | ||
604 | if (!hmap_is_empty(&p->controllers) | |
605 | && p->fail_mode == OFPROTO_FAIL_STANDALONE) { | |
606 | struct rconn **rconns; | |
607 | size_t n; | |
608 | ||
609 | if (!p->fail_open) { | |
610 | p->fail_open = fail_open_create(p, p->switch_status); | |
611 | } | |
612 | ||
613 | n = 0; | |
614 | rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns); | |
615 | HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &p->controllers) { | |
616 | rconns[n++] = ofconn->rconn; | |
617 | } | |
618 | ||
619 | fail_open_set_controllers(p->fail_open, rconns, n); | |
620 | /* p->fail_open takes ownership of 'rconns'. */ | |
621 | } else { | |
622 | fail_open_destroy(p->fail_open); | |
623 | p->fail_open = NULL; | |
624 | } | |
625 | } | |
626 | ||
76ce9432 BP |
627 | void |
628 | ofproto_set_controllers(struct ofproto *p, | |
629 | const struct ofproto_controller *controllers, | |
630 | size_t n_controllers) | |
631 | { | |
632 | struct shash new_controllers; | |
7d674866 BP |
633 | struct ofconn *ofconn, *next_ofconn; |
634 | struct ofservice *ofservice, *next_ofservice; | |
76ce9432 | 635 | bool ss_exists; |
76ce9432 | 636 | size_t i; |
79c9f2ee | 637 | |
7d674866 BP |
638 | /* Create newly configured controllers and services. |
639 | * Create a name to ofproto_controller mapping in 'new_controllers'. */ | |
76ce9432 BP |
640 | shash_init(&new_controllers); |
641 | for (i = 0; i < n_controllers; i++) { | |
642 | const struct ofproto_controller *c = &controllers[i]; | |
643 | ||
7d674866 BP |
644 | if (!vconn_verify_name(c->target) || !strcmp(c->target, "discover")) { |
645 | if (!find_controller_by_target(p, c->target)) { | |
646 | add_controller(p, c); | |
647 | } | |
648 | } else if (!pvconn_verify_name(c->target)) { | |
649 | if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) { | |
650 | continue; | |
651 | } | |
652 | } else { | |
653 | VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"", | |
654 | dpif_name(p->dpif), c->target); | |
655 | continue; | |
76ce9432 | 656 | } |
7d674866 BP |
657 | |
658 | shash_add_once(&new_controllers, c->target, &controllers[i]); | |
76ce9432 BP |
659 | } |
660 | ||
7d674866 BP |
661 | /* Delete controllers that are no longer configured. |
662 | * Update configuration of all now-existing controllers. */ | |
76ce9432 | 663 | ss_exists = false; |
7d674866 | 664 | HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, hmap_node, |
76ce9432 BP |
665 | &p->controllers) { |
666 | struct ofproto_controller *c; | |
667 | ||
668 | c = shash_find_data(&new_controllers, ofconn_get_target(ofconn)); | |
669 | if (!c) { | |
670 | ofconn_destroy(ofconn); | |
79c9f2ee | 671 | } else { |
76ce9432 | 672 | update_controller(ofconn, c); |
76ce9432 BP |
673 | if (ofconn->ss) { |
674 | ss_exists = true; | |
675 | } | |
76ce9432 BP |
676 | } |
677 | } | |
7d674866 BP |
678 | |
679 | /* Delete services that are no longer configured. | |
680 | * Update configuration of all now-existing services. */ | |
681 | HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, struct ofservice, node, | |
682 | &p->services) { | |
683 | struct ofproto_controller *c; | |
684 | ||
685 | c = shash_find_data(&new_controllers, | |
686 | pvconn_get_name(ofservice->pvconn)); | |
687 | if (!c) { | |
688 | ofservice_destroy(p, ofservice); | |
689 | } else { | |
690 | ofservice_reconfigure(ofservice, c); | |
691 | } | |
692 | } | |
693 | ||
76ce9432 BP |
694 | shash_destroy(&new_controllers); |
695 | ||
d2ede7bc | 696 | update_in_band_remotes(p); |
31681a5d | 697 | update_fail_open(p); |
79c9f2ee | 698 | |
76ce9432 BP |
699 | if (!hmap_is_empty(&p->controllers) && !ss_exists) { |
700 | ofconn = CONTAINER_OF(hmap_first(&p->controllers), | |
701 | struct ofconn, hmap_node); | |
702 | ofconn->ss = switch_status_register(p->switch_status, "remote", | |
703 | rconn_status_cb, ofconn->rconn); | |
79c9f2ee | 704 | } |
064af421 BP |
705 | } |
706 | ||
31681a5d JP |
707 | void |
708 | ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode) | |
709 | { | |
710 | p->fail_mode = fail_mode; | |
711 | update_fail_open(p); | |
712 | } | |
713 | ||
fa05809b BP |
714 | /* Drops the connections between 'ofproto' and all of its controllers, forcing |
715 | * them to reconnect. */ | |
716 | void | |
717 | ofproto_reconnect_controllers(struct ofproto *ofproto) | |
718 | { | |
719 | struct ofconn *ofconn; | |
720 | ||
721 | LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) { | |
722 | rconn_reconnect(ofconn->rconn); | |
723 | } | |
724 | } | |
725 | ||
917e50e1 BP |
726 | static bool |
727 | any_extras_changed(const struct ofproto *ofproto, | |
728 | const struct sockaddr_in *extras, size_t n) | |
729 | { | |
730 | size_t i; | |
731 | ||
732 | if (n != ofproto->n_extra_remotes) { | |
733 | return true; | |
734 | } | |
735 | ||
736 | for (i = 0; i < n; i++) { | |
737 | const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i]; | |
738 | const struct sockaddr_in *new = &extras[i]; | |
739 | ||
740 | if (old->sin_addr.s_addr != new->sin_addr.s_addr || | |
741 | old->sin_port != new->sin_port) { | |
742 | return true; | |
743 | } | |
744 | } | |
745 | ||
746 | return false; | |
747 | } | |
748 | ||
749 | /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s | |
750 | * in-band control should guarantee access, in the same way that in-band | |
751 | * control guarantees access to OpenFlow controllers. */ | |
752 | void | |
753 | ofproto_set_extra_in_band_remotes(struct ofproto *ofproto, | |
754 | const struct sockaddr_in *extras, size_t n) | |
755 | { | |
756 | if (!any_extras_changed(ofproto, extras, n)) { | |
757 | return; | |
758 | } | |
759 | ||
760 | free(ofproto->extra_in_band_remotes); | |
761 | ofproto->n_extra_remotes = n; | |
762 | ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras); | |
763 | ||
764 | update_in_band_remotes(ofproto); | |
765 | } | |
766 | ||
064af421 BP |
767 | void |
768 | ofproto_set_desc(struct ofproto *p, | |
5a719c38 JP |
769 | const char *mfr_desc, const char *hw_desc, |
770 | const char *sw_desc, const char *serial_desc, | |
8abc4ed7 | 771 | const char *dp_desc) |
064af421 | 772 | { |
5a719c38 JP |
773 | struct ofp_desc_stats *ods; |
774 | ||
775 | if (mfr_desc) { | |
776 | if (strlen(mfr_desc) >= sizeof ods->mfr_desc) { | |
777 | VLOG_WARN("truncating mfr_desc, must be less than %zu characters", | |
778 | sizeof ods->mfr_desc); | |
779 | } | |
780 | free(p->mfr_desc); | |
781 | p->mfr_desc = xstrdup(mfr_desc); | |
064af421 | 782 | } |
5a719c38 JP |
783 | if (hw_desc) { |
784 | if (strlen(hw_desc) >= sizeof ods->hw_desc) { | |
785 | VLOG_WARN("truncating hw_desc, must be less than %zu characters", | |
786 | sizeof ods->hw_desc); | |
787 | } | |
788 | free(p->hw_desc); | |
789 | p->hw_desc = xstrdup(hw_desc); | |
064af421 | 790 | } |
5a719c38 JP |
791 | if (sw_desc) { |
792 | if (strlen(sw_desc) >= sizeof ods->sw_desc) { | |
793 | VLOG_WARN("truncating sw_desc, must be less than %zu characters", | |
794 | sizeof ods->sw_desc); | |
795 | } | |
796 | free(p->sw_desc); | |
797 | p->sw_desc = xstrdup(sw_desc); | |
798 | } | |
799 | if (serial_desc) { | |
800 | if (strlen(serial_desc) >= sizeof ods->serial_num) { | |
801 | VLOG_WARN("truncating serial_desc, must be less than %zu " | |
802 | "characters", | |
803 | sizeof ods->serial_num); | |
804 | } | |
805 | free(p->serial_desc); | |
806 | p->serial_desc = xstrdup(serial_desc); | |
064af421 | 807 | } |
8abc4ed7 | 808 | if (dp_desc) { |
5a719c38 JP |
809 | if (strlen(dp_desc) >= sizeof ods->dp_desc) { |
810 | VLOG_WARN("truncating dp_desc, must be less than %zu characters", | |
811 | sizeof ods->dp_desc); | |
812 | } | |
8abc4ed7 JP |
813 | free(p->dp_desc); |
814 | p->dp_desc = xstrdup(dp_desc); | |
815 | } | |
064af421 BP |
816 | } |
817 | ||
064af421 BP |
818 | static int |
819 | set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp, | |
820 | const struct svec *svec) | |
821 | { | |
822 | struct pvconn **pvconns = *pvconnsp; | |
823 | size_t n_pvconns = *n_pvconnsp; | |
824 | int retval = 0; | |
825 | size_t i; | |
826 | ||
827 | for (i = 0; i < n_pvconns; i++) { | |
828 | pvconn_close(pvconns[i]); | |
829 | } | |
830 | free(pvconns); | |
831 | ||
832 | pvconns = xmalloc(svec->n * sizeof *pvconns); | |
833 | n_pvconns = 0; | |
834 | for (i = 0; i < svec->n; i++) { | |
835 | const char *name = svec->names[i]; | |
836 | struct pvconn *pvconn; | |
837 | int error; | |
838 | ||
839 | error = pvconn_open(name, &pvconn); | |
840 | if (!error) { | |
841 | pvconns[n_pvconns++] = pvconn; | |
842 | } else { | |
843 | VLOG_ERR("failed to listen on %s: %s", name, strerror(error)); | |
844 | if (!retval) { | |
845 | retval = error; | |
846 | } | |
847 | } | |
848 | } | |
849 | ||
850 | *pvconnsp = pvconns; | |
851 | *n_pvconnsp = n_pvconns; | |
852 | ||
853 | return retval; | |
854 | } | |
855 | ||
064af421 BP |
856 | int |
857 | ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops) | |
858 | { | |
859 | return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops); | |
860 | } | |
861 | ||
862 | int | |
0193b2af JG |
863 | ofproto_set_netflow(struct ofproto *ofproto, |
864 | const struct netflow_options *nf_options) | |
064af421 | 865 | { |
76343538 | 866 | if (nf_options && nf_options->collectors.n) { |
064af421 BP |
867 | if (!ofproto->netflow) { |
868 | ofproto->netflow = netflow_create(); | |
869 | } | |
0193b2af | 870 | return netflow_set_options(ofproto->netflow, nf_options); |
064af421 BP |
871 | } else { |
872 | netflow_destroy(ofproto->netflow); | |
873 | ofproto->netflow = NULL; | |
874 | return 0; | |
875 | } | |
876 | } | |
877 | ||
72b06300 BP |
878 | void |
879 | ofproto_set_sflow(struct ofproto *ofproto, | |
880 | const struct ofproto_sflow_options *oso) | |
881 | { | |
882 | struct ofproto_sflow *os = ofproto->sflow; | |
883 | if (oso) { | |
884 | if (!os) { | |
885 | struct ofport *ofport; | |
886 | unsigned int odp_port; | |
887 | ||
888 | os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif); | |
889 | refresh_port_groups(ofproto); | |
890 | PORT_ARRAY_FOR_EACH (ofport, &ofproto->ports, odp_port) { | |
891 | ofproto_sflow_add_port(os, odp_port, | |
892 | netdev_get_name(ofport->netdev)); | |
893 | } | |
894 | } | |
895 | ofproto_sflow_set_options(os, oso); | |
896 | } else { | |
897 | ofproto_sflow_destroy(os); | |
898 | ofproto->sflow = NULL; | |
899 | } | |
900 | } | |
901 | ||
064af421 BP |
902 | uint64_t |
903 | ofproto_get_datapath_id(const struct ofproto *ofproto) | |
904 | { | |
905 | return ofproto->datapath_id; | |
906 | } | |
907 | ||
76ce9432 | 908 | bool |
7d674866 | 909 | ofproto_has_primary_controller(const struct ofproto *ofproto) |
064af421 | 910 | { |
76ce9432 | 911 | return !hmap_is_empty(&ofproto->controllers); |
064af421 BP |
912 | } |
913 | ||
abdfe474 JP |
914 | enum ofproto_fail_mode |
915 | ofproto_get_fail_mode(const struct ofproto *p) | |
916 | { | |
917 | return p->fail_mode; | |
918 | } | |
919 | ||
064af421 BP |
920 | void |
921 | ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops) | |
922 | { | |
923 | size_t i; | |
924 | ||
925 | for (i = 0; i < ofproto->n_snoops; i++) { | |
926 | svec_add(snoops, pvconn_get_name(ofproto->snoops[i])); | |
927 | } | |
928 | } | |
929 | ||
930 | void | |
931 | ofproto_destroy(struct ofproto *p) | |
932 | { | |
7d674866 | 933 | struct ofservice *ofservice, *next_ofservice; |
064af421 BP |
934 | struct ofconn *ofconn, *next_ofconn; |
935 | struct ofport *ofport; | |
936 | unsigned int port_no; | |
937 | size_t i; | |
938 | ||
939 | if (!p) { | |
940 | return; | |
941 | } | |
942 | ||
f7de2cdf | 943 | /* Destroy fail-open and in-band early, since they touch the classifier. */ |
79c9f2ee BP |
944 | fail_open_destroy(p->fail_open); |
945 | p->fail_open = NULL; | |
946 | ||
947 | in_band_destroy(p->in_band); | |
948 | p->in_band = NULL; | |
917e50e1 | 949 | free(p->extra_in_band_remotes); |
2f6d3445 | 950 | |
064af421 BP |
951 | ofproto_flush_flows(p); |
952 | classifier_destroy(&p->cls); | |
953 | ||
954 | LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node, | |
955 | &p->all_conns) { | |
c475ae67 | 956 | ofconn_destroy(ofconn); |
064af421 | 957 | } |
76ce9432 | 958 | hmap_destroy(&p->controllers); |
064af421 | 959 | |
c228a364 | 960 | dpif_close(p->dpif); |
e9e28be3 | 961 | netdev_monitor_destroy(p->netdev_monitor); |
064af421 BP |
962 | PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) { |
963 | ofport_free(ofport); | |
964 | } | |
965 | shash_destroy(&p->port_by_name); | |
966 | ||
967 | switch_status_destroy(p->switch_status); | |
064af421 | 968 | netflow_destroy(p->netflow); |
72b06300 | 969 | ofproto_sflow_destroy(p->sflow); |
064af421 | 970 | |
7d674866 BP |
971 | HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, struct ofservice, node, |
972 | &p->services) { | |
973 | ofservice_destroy(p, ofservice); | |
064af421 | 974 | } |
7d674866 | 975 | hmap_destroy(&p->services); |
064af421 BP |
976 | |
977 | for (i = 0; i < p->n_snoops; i++) { | |
978 | pvconn_close(p->snoops[i]); | |
979 | } | |
980 | free(p->snoops); | |
981 | ||
982 | mac_learning_destroy(p->ml); | |
983 | ||
5a719c38 JP |
984 | free(p->mfr_desc); |
985 | free(p->hw_desc); | |
986 | free(p->sw_desc); | |
987 | free(p->serial_desc); | |
cb871ae0 JP |
988 | free(p->dp_desc); |
989 | ||
3b917492 BP |
990 | port_array_destroy(&p->ports); |
991 | ||
064af421 BP |
992 | free(p); |
993 | } | |
994 | ||
995 | int | |
996 | ofproto_run(struct ofproto *p) | |
997 | { | |
998 | int error = ofproto_run1(p); | |
999 | if (!error) { | |
1000 | error = ofproto_run2(p, false); | |
1001 | } | |
1002 | return error; | |
1003 | } | |
1004 | ||
e9e28be3 BP |
1005 | static void |
1006 | process_port_change(struct ofproto *ofproto, int error, char *devname) | |
1007 | { | |
1008 | if (error == ENOBUFS) { | |
1009 | reinit_ports(ofproto); | |
1010 | } else if (!error) { | |
1011 | update_port(ofproto, devname); | |
1012 | free(devname); | |
1013 | } | |
1014 | } | |
1015 | ||
e2bfacb6 BP |
1016 | /* Returns a "preference level" for snooping 'ofconn'. A higher return value |
1017 | * means that 'ofconn' is more interesting for monitoring than a lower return | |
1018 | * value. */ | |
1019 | static int | |
1020 | snoop_preference(const struct ofconn *ofconn) | |
1021 | { | |
1022 | switch (ofconn->role) { | |
1023 | case NX_ROLE_MASTER: | |
1024 | return 3; | |
1025 | case NX_ROLE_OTHER: | |
1026 | return 2; | |
1027 | case NX_ROLE_SLAVE: | |
1028 | return 1; | |
1029 | default: | |
1030 | /* Shouldn't happen. */ | |
1031 | return 0; | |
1032 | } | |
1033 | } | |
1034 | ||
76ce9432 BP |
1035 | /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'. |
1036 | * Connects this vconn to a controller. */ | |
1037 | static void | |
1038 | add_snooper(struct ofproto *ofproto, struct vconn *vconn) | |
1039 | { | |
e2bfacb6 | 1040 | struct ofconn *ofconn, *best; |
76ce9432 | 1041 | |
e2bfacb6 BP |
1042 | /* Pick a controller for monitoring. */ |
1043 | best = NULL; | |
76ce9432 | 1044 | LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) { |
5899143f | 1045 | if (ofconn->type == OFCONN_PRIMARY |
e2bfacb6 BP |
1046 | && (!best || snoop_preference(ofconn) > snoop_preference(best))) { |
1047 | best = ofconn; | |
76ce9432 | 1048 | } |
e2bfacb6 | 1049 | } |
76ce9432 | 1050 | |
e2bfacb6 BP |
1051 | if (best) { |
1052 | rconn_add_monitor(best->rconn, vconn); | |
1053 | } else { | |
1054 | VLOG_INFO_RL(&rl, "no controller connection to snoop"); | |
1055 | vconn_close(vconn); | |
76ce9432 | 1056 | } |
76ce9432 BP |
1057 | } |
1058 | ||
064af421 BP |
1059 | int |
1060 | ofproto_run1(struct ofproto *p) | |
1061 | { | |
1062 | struct ofconn *ofconn, *next_ofconn; | |
7d674866 | 1063 | struct ofservice *ofservice; |
064af421 BP |
1064 | char *devname; |
1065 | int error; | |
1066 | int i; | |
1067 | ||
149f577a JG |
1068 | if (shash_is_empty(&p->port_by_name)) { |
1069 | init_ports(p); | |
1070 | } | |
1071 | ||
064af421 BP |
1072 | for (i = 0; i < 50; i++) { |
1073 | struct ofpbuf *buf; | |
1074 | int error; | |
1075 | ||
c228a364 | 1076 | error = dpif_recv(p->dpif, &buf); |
064af421 BP |
1077 | if (error) { |
1078 | if (error == ENODEV) { | |
1079 | /* Someone destroyed the datapath behind our back. The caller | |
1080 | * better destroy us and give up, because we're just going to | |
1081 | * spin from here on out. */ | |
39a559f2 BP |
1082 | static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5); |
1083 | VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally", | |
c228a364 | 1084 | dpif_name(p->dpif)); |
064af421 BP |
1085 | return ENODEV; |
1086 | } | |
1087 | break; | |
1088 | } | |
1089 | ||
1090 | handle_odp_msg(p, buf); | |
1091 | } | |
1092 | ||
e9e28be3 BP |
1093 | while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) { |
1094 | process_port_change(p, error, devname); | |
1095 | } | |
1096 | while ((error = netdev_monitor_poll(p->netdev_monitor, | |
1097 | &devname)) != EAGAIN) { | |
1098 | process_port_change(p, error, devname); | |
064af421 BP |
1099 | } |
1100 | ||
1101 | if (p->in_band) { | |
d2ede7bc BP |
1102 | if (time_msec() >= p->next_in_band_update) { |
1103 | update_in_band_remotes(p); | |
1104 | } | |
064af421 BP |
1105 | in_band_run(p->in_band); |
1106 | } | |
064af421 BP |
1107 | |
1108 | LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node, | |
1109 | &p->all_conns) { | |
1110 | ofconn_run(ofconn, p); | |
1111 | } | |
1112 | ||
7778bd15 BP |
1113 | /* Fail-open maintenance. Do this after processing the ofconns since |
1114 | * fail-open checks the status of the controller rconn. */ | |
1115 | if (p->fail_open) { | |
1116 | fail_open_run(p->fail_open); | |
1117 | } | |
1118 | ||
7d674866 | 1119 | HMAP_FOR_EACH (ofservice, struct ofservice, node, &p->services) { |
064af421 BP |
1120 | struct vconn *vconn; |
1121 | int retval; | |
1122 | ||
7d674866 | 1123 | retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn); |
064af421 | 1124 | if (!retval) { |
7d674866 | 1125 | struct ofconn *ofconn; |
9794e806 | 1126 | struct rconn *rconn; |
eb15cdbb | 1127 | char *name; |
9794e806 | 1128 | |
7d674866 | 1129 | rconn = rconn_create(ofservice->probe_interval, 0); |
eb15cdbb BP |
1130 | name = ofconn_make_name(p, vconn_get_name(vconn)); |
1131 | rconn_connect_unreliably(rconn, vconn, name); | |
1132 | free(name); | |
1133 | ||
7d674866 BP |
1134 | ofconn = ofconn_create(p, rconn, OFCONN_SERVICE); |
1135 | ofconn_set_rate_limit(ofconn, ofservice->rate_limit, | |
1136 | ofservice->burst_limit); | |
064af421 BP |
1137 | } else if (retval != EAGAIN) { |
1138 | VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval)); | |
1139 | } | |
1140 | } | |
1141 | ||
1142 | for (i = 0; i < p->n_snoops; i++) { | |
1143 | struct vconn *vconn; | |
1144 | int retval; | |
1145 | ||
1146 | retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn); | |
1147 | if (!retval) { | |
76ce9432 | 1148 | add_snooper(p, vconn); |
064af421 BP |
1149 | } else if (retval != EAGAIN) { |
1150 | VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval)); | |
1151 | } | |
1152 | } | |
1153 | ||
1154 | if (time_msec() >= p->next_expiration) { | |
1155 | COVERAGE_INC(ofproto_expiration); | |
1156 | p->next_expiration = time_msec() + 1000; | |
1157 | update_used(p); | |
1158 | ||
1159 | classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p); | |
1160 | ||
1161 | /* Let the hook know that we're at a stable point: all outstanding data | |
1162 | * in existing flows has been accounted to the account_cb. Thus, the | |
1163 | * hook can now reasonably do operations that depend on having accurate | |
1164 | * flow volume accounting (currently, that's just bond rebalancing). */ | |
1165 | if (p->ofhooks->account_checkpoint_cb) { | |
1166 | p->ofhooks->account_checkpoint_cb(p->aux); | |
1167 | } | |
1168 | } | |
1169 | ||
1170 | if (p->netflow) { | |
1171 | netflow_run(p->netflow); | |
1172 | } | |
72b06300 BP |
1173 | if (p->sflow) { |
1174 | ofproto_sflow_run(p->sflow); | |
1175 | } | |
064af421 BP |
1176 | |
1177 | return 0; | |
1178 | } | |
1179 | ||
1180 | struct revalidate_cbdata { | |
1181 | struct ofproto *ofproto; | |
1182 | bool revalidate_all; /* Revalidate all exact-match rules? */ | |
1183 | bool revalidate_subrules; /* Revalidate all exact-match subrules? */ | |
1184 | struct tag_set revalidate_set; /* Set of tags to revalidate. */ | |
1185 | }; | |
1186 | ||
1187 | int | |
1188 | ofproto_run2(struct ofproto *p, bool revalidate_all) | |
1189 | { | |
1190 | if (p->need_revalidate || revalidate_all | |
1191 | || !tag_set_is_empty(&p->revalidate_set)) { | |
1192 | struct revalidate_cbdata cbdata; | |
1193 | cbdata.ofproto = p; | |
1194 | cbdata.revalidate_all = revalidate_all; | |
1195 | cbdata.revalidate_subrules = p->need_revalidate; | |
1196 | cbdata.revalidate_set = p->revalidate_set; | |
1197 | tag_set_init(&p->revalidate_set); | |
1198 | COVERAGE_INC(ofproto_revalidate); | |
1199 | classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata); | |
1200 | p->need_revalidate = false; | |
1201 | } | |
1202 | ||
1203 | return 0; | |
1204 | } | |
1205 | ||
1206 | void | |
1207 | ofproto_wait(struct ofproto *p) | |
1208 | { | |
7d674866 | 1209 | struct ofservice *ofservice; |
064af421 BP |
1210 | struct ofconn *ofconn; |
1211 | size_t i; | |
1212 | ||
c228a364 | 1213 | dpif_recv_wait(p->dpif); |
e9e28be3 BP |
1214 | dpif_port_poll_wait(p->dpif); |
1215 | netdev_monitor_poll_wait(p->netdev_monitor); | |
064af421 BP |
1216 | LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) { |
1217 | ofconn_wait(ofconn); | |
1218 | } | |
1219 | if (p->in_band) { | |
7cf8b266 | 1220 | poll_timer_wait_until(p->next_in_band_update); |
064af421 BP |
1221 | in_band_wait(p->in_band); |
1222 | } | |
064af421 BP |
1223 | if (p->fail_open) { |
1224 | fail_open_wait(p->fail_open); | |
1225 | } | |
72b06300 BP |
1226 | if (p->sflow) { |
1227 | ofproto_sflow_wait(p->sflow); | |
1228 | } | |
064af421 BP |
1229 | if (!tag_set_is_empty(&p->revalidate_set)) { |
1230 | poll_immediate_wake(); | |
1231 | } | |
1232 | if (p->need_revalidate) { | |
1233 | /* Shouldn't happen, but if it does just go around again. */ | |
1234 | VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); | |
1235 | poll_immediate_wake(); | |
1236 | } else if (p->next_expiration != LLONG_MAX) { | |
7cf8b266 | 1237 | poll_timer_wait_until(p->next_expiration); |
064af421 | 1238 | } |
7d674866 BP |
1239 | HMAP_FOR_EACH (ofservice, struct ofservice, node, &p->services) { |
1240 | pvconn_wait(ofservice->pvconn); | |
064af421 BP |
1241 | } |
1242 | for (i = 0; i < p->n_snoops; i++) { | |
1243 | pvconn_wait(p->snoops[i]); | |
1244 | } | |
1245 | } | |
1246 | ||
1247 | void | |
1248 | ofproto_revalidate(struct ofproto *ofproto, tag_type tag) | |
1249 | { | |
1250 | tag_set_add(&ofproto->revalidate_set, tag); | |
1251 | } | |
1252 | ||
1253 | struct tag_set * | |
1254 | ofproto_get_revalidate_set(struct ofproto *ofproto) | |
1255 | { | |
1256 | return &ofproto->revalidate_set; | |
1257 | } | |
1258 | ||
1259 | bool | |
1260 | ofproto_is_alive(const struct ofproto *p) | |
1261 | { | |
76ce9432 | 1262 | return !hmap_is_empty(&p->controllers); |
064af421 BP |
1263 | } |
1264 | ||
1265 | int | |
1266 | ofproto_send_packet(struct ofproto *p, const flow_t *flow, | |
1267 | const union ofp_action *actions, size_t n_actions, | |
1268 | const struct ofpbuf *packet) | |
1269 | { | |
1270 | struct odp_actions odp_actions; | |
1271 | int error; | |
1272 | ||
1273 | error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions, | |
6a07af36 | 1274 | NULL, NULL, NULL); |
064af421 BP |
1275 | if (error) { |
1276 | return error; | |
1277 | } | |
1278 | ||
1279 | /* XXX Should we translate the dpif_execute() errno value into an OpenFlow | |
1280 | * error code? */ | |
c228a364 | 1281 | dpif_execute(p->dpif, flow->in_port, odp_actions.actions, |
064af421 BP |
1282 | odp_actions.n_actions, packet); |
1283 | return 0; | |
1284 | } | |
1285 | ||
1286 | void | |
1287 | ofproto_add_flow(struct ofproto *p, | |
1288 | const flow_t *flow, uint32_t wildcards, unsigned int priority, | |
1289 | const union ofp_action *actions, size_t n_actions, | |
1290 | int idle_timeout) | |
1291 | { | |
1292 | struct rule *rule; | |
0193b2af | 1293 | rule = rule_create(p, NULL, actions, n_actions, |
ca069229 | 1294 | idle_timeout >= 0 ? idle_timeout : 5 /* XXX */, |
39997502 | 1295 | 0, 0, false); |
659586ef | 1296 | cls_rule_from_flow(flow, wildcards, priority, &rule->cr); |
064af421 BP |
1297 | rule_insert(p, rule, NULL, 0); |
1298 | } | |
1299 | ||
1300 | void | |
1301 | ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow, | |
1302 | uint32_t wildcards, unsigned int priority) | |
1303 | { | |
1304 | struct rule *rule; | |
1305 | ||
1306 | rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls, | |
1307 | flow, wildcards, | |
1308 | priority)); | |
1309 | if (rule) { | |
1310 | rule_remove(ofproto, rule); | |
1311 | } | |
1312 | } | |
1313 | ||
1314 | static void | |
1315 | destroy_rule(struct cls_rule *rule_, void *ofproto_) | |
1316 | { | |
1317 | struct rule *rule = rule_from_cls_rule(rule_); | |
1318 | struct ofproto *ofproto = ofproto_; | |
1319 | ||
1320 | /* Mark the flow as not installed, even though it might really be | |
1321 | * installed, so that rule_remove() doesn't bother trying to uninstall it. | |
1322 | * There is no point in uninstalling it individually since we are about to | |
1323 | * blow away all the flows with dpif_flow_flush(). */ | |
1324 | rule->installed = false; | |
1325 | ||
1326 | rule_remove(ofproto, rule); | |
1327 | } | |
1328 | ||
1329 | void | |
1330 | ofproto_flush_flows(struct ofproto *ofproto) | |
1331 | { | |
1332 | COVERAGE_INC(ofproto_flush); | |
1333 | classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto); | |
c228a364 | 1334 | dpif_flow_flush(ofproto->dpif); |
064af421 BP |
1335 | if (ofproto->in_band) { |
1336 | in_band_flushed(ofproto->in_band); | |
1337 | } | |
1338 | if (ofproto->fail_open) { | |
1339 | fail_open_flushed(ofproto->fail_open); | |
1340 | } | |
1341 | } | |
1342 | \f | |
1343 | static void | |
1344 | reinit_ports(struct ofproto *p) | |
1345 | { | |
1346 | struct svec devnames; | |
1347 | struct ofport *ofport; | |
1348 | unsigned int port_no; | |
1349 | struct odp_port *odp_ports; | |
1350 | size_t n_odp_ports; | |
1351 | size_t i; | |
1352 | ||
1353 | svec_init(&devnames); | |
1354 | PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) { | |
1355 | svec_add (&devnames, (char *) ofport->opp.name); | |
1356 | } | |
c228a364 | 1357 | dpif_port_list(p->dpif, &odp_ports, &n_odp_ports); |
064af421 BP |
1358 | for (i = 0; i < n_odp_ports; i++) { |
1359 | svec_add (&devnames, odp_ports[i].devname); | |
1360 | } | |
1361 | free(odp_ports); | |
1362 | ||
1363 | svec_sort_unique(&devnames); | |
1364 | for (i = 0; i < devnames.n; i++) { | |
1365 | update_port(p, devnames.names[i]); | |
1366 | } | |
1367 | svec_destroy(&devnames); | |
1368 | } | |
1369 | ||
72b06300 | 1370 | static size_t |
064af421 BP |
1371 | refresh_port_group(struct ofproto *p, unsigned int group) |
1372 | { | |
1373 | uint16_t *ports; | |
1374 | size_t n_ports; | |
1375 | struct ofport *port; | |
1376 | unsigned int port_no; | |
1377 | ||
1378 | assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD); | |
1379 | ||
1380 | ports = xmalloc(port_array_count(&p->ports) * sizeof *ports); | |
1381 | n_ports = 0; | |
1382 | PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) { | |
1383 | if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) { | |
1384 | ports[n_ports++] = port_no; | |
1385 | } | |
1386 | } | |
c228a364 | 1387 | dpif_port_group_set(p->dpif, group, ports, n_ports); |
064af421 | 1388 | free(ports); |
72b06300 BP |
1389 | |
1390 | return n_ports; | |
064af421 BP |
1391 | } |
1392 | ||
1393 | static void | |
1394 | refresh_port_groups(struct ofproto *p) | |
1395 | { | |
72b06300 BP |
1396 | size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD); |
1397 | size_t n_all = refresh_port_group(p, DP_GROUP_ALL); | |
1398 | if (p->sflow) { | |
1399 | ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all); | |
1400 | } | |
064af421 BP |
1401 | } |
1402 | ||
1403 | static struct ofport * | |
1404 | make_ofport(const struct odp_port *odp_port) | |
1405 | { | |
149f577a | 1406 | struct netdev_options netdev_options; |
064af421 BP |
1407 | enum netdev_flags flags; |
1408 | struct ofport *ofport; | |
1409 | struct netdev *netdev; | |
1410 | bool carrier; | |
1411 | int error; | |
1412 | ||
149f577a JG |
1413 | memset(&netdev_options, 0, sizeof netdev_options); |
1414 | netdev_options.name = odp_port->devname; | |
1415 | netdev_options.ethertype = NETDEV_ETH_TYPE_NONE; | |
149f577a JG |
1416 | |
1417 | error = netdev_open(&netdev_options, &netdev); | |
064af421 BP |
1418 | if (error) { |
1419 | VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s " | |
1420 | "cannot be opened (%s)", | |
1421 | odp_port->devname, odp_port->port, | |
1422 | odp_port->devname, strerror(error)); | |
1423 | return NULL; | |
1424 | } | |
1425 | ||
1426 | ofport = xmalloc(sizeof *ofport); | |
1427 | ofport->netdev = netdev; | |
1428 | ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port); | |
80992a35 | 1429 | netdev_get_etheraddr(netdev, ofport->opp.hw_addr); |
064af421 BP |
1430 | memcpy(ofport->opp.name, odp_port->devname, |
1431 | MIN(sizeof ofport->opp.name, sizeof odp_port->devname)); | |
1432 | ofport->opp.name[sizeof ofport->opp.name - 1] = '\0'; | |
1433 | ||
1434 | netdev_get_flags(netdev, &flags); | |
1435 | ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN; | |
1436 | ||
1437 | netdev_get_carrier(netdev, &carrier); | |
1438 | ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN; | |
1439 | ||
1440 | netdev_get_features(netdev, | |
1441 | &ofport->opp.curr, &ofport->opp.advertised, | |
1442 | &ofport->opp.supported, &ofport->opp.peer); | |
1443 | return ofport; | |
1444 | } | |
1445 | ||
1446 | static bool | |
1447 | ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port) | |
1448 | { | |
1449 | if (port_array_get(&p->ports, odp_port->port)) { | |
1450 | VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath", | |
1451 | odp_port->port); | |
1452 | return true; | |
1453 | } else if (shash_find(&p->port_by_name, odp_port->devname)) { | |
1454 | VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath", | |
1455 | odp_port->devname); | |
1456 | return true; | |
1457 | } else { | |
1458 | return false; | |
1459 | } | |
1460 | } | |
1461 | ||
1462 | static int | |
1463 | ofport_equal(const struct ofport *a_, const struct ofport *b_) | |
1464 | { | |
1465 | const struct ofp_phy_port *a = &a_->opp; | |
1466 | const struct ofp_phy_port *b = &b_->opp; | |
1467 | ||
1468 | BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */ | |
1469 | return (a->port_no == b->port_no | |
1470 | && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr) | |
1471 | && !strcmp((char *) a->name, (char *) b->name) | |
1472 | && a->state == b->state | |
1473 | && a->config == b->config | |
1474 | && a->curr == b->curr | |
1475 | && a->advertised == b->advertised | |
1476 | && a->supported == b->supported | |
1477 | && a->peer == b->peer); | |
1478 | } | |
1479 | ||
1480 | static void | |
1481 | send_port_status(struct ofproto *p, const struct ofport *ofport, | |
1482 | uint8_t reason) | |
1483 | { | |
1484 | /* XXX Should limit the number of queued port status change messages. */ | |
1485 | struct ofconn *ofconn; | |
1486 | LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) { | |
1487 | struct ofp_port_status *ops; | |
1488 | struct ofpbuf *b; | |
1489 | ||
c91248b3 | 1490 | if (!ofconn_receives_async_msgs(ofconn)) { |
9deba63b BP |
1491 | continue; |
1492 | } | |
1493 | ||
064af421 BP |
1494 | ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b); |
1495 | ops->reason = reason; | |
1496 | ops->desc = ofport->opp; | |
1497 | hton_ofp_phy_port(&ops->desc); | |
1498 | queue_tx(b, ofconn, NULL); | |
1499 | } | |
1500 | if (p->ofhooks->port_changed_cb) { | |
1501 | p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux); | |
1502 | } | |
1503 | } | |
1504 | ||
1505 | static void | |
1506 | ofport_install(struct ofproto *p, struct ofport *ofport) | |
1507 | { | |
72b06300 BP |
1508 | uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no); |
1509 | const char *netdev_name = (const char *) ofport->opp.name; | |
1510 | ||
e9e28be3 | 1511 | netdev_monitor_add(p->netdev_monitor, ofport->netdev); |
72b06300 BP |
1512 | port_array_set(&p->ports, odp_port, ofport); |
1513 | shash_add(&p->port_by_name, netdev_name, ofport); | |
1514 | if (p->sflow) { | |
1515 | ofproto_sflow_add_port(p->sflow, odp_port, netdev_name); | |
1516 | } | |
064af421 BP |
1517 | } |
1518 | ||
1519 | static void | |
1520 | ofport_remove(struct ofproto *p, struct ofport *ofport) | |
1521 | { | |
72b06300 BP |
1522 | uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no); |
1523 | ||
e9e28be3 | 1524 | netdev_monitor_remove(p->netdev_monitor, ofport->netdev); |
e3648418 | 1525 | port_array_delete(&p->ports, odp_port); |
064af421 BP |
1526 | shash_delete(&p->port_by_name, |
1527 | shash_find(&p->port_by_name, (char *) ofport->opp.name)); | |
72b06300 BP |
1528 | if (p->sflow) { |
1529 | ofproto_sflow_del_port(p->sflow, odp_port); | |
1530 | } | |
064af421 BP |
1531 | } |
1532 | ||
1533 | static void | |
1534 | ofport_free(struct ofport *ofport) | |
1535 | { | |
1536 | if (ofport) { | |
1537 | netdev_close(ofport->netdev); | |
1538 | free(ofport); | |
1539 | } | |
1540 | } | |
1541 | ||
1542 | static void | |
1543 | update_port(struct ofproto *p, const char *devname) | |
1544 | { | |
1545 | struct odp_port odp_port; | |
c874dc6d BP |
1546 | struct ofport *old_ofport; |
1547 | struct ofport *new_ofport; | |
064af421 BP |
1548 | int error; |
1549 | ||
1550 | COVERAGE_INC(ofproto_update_port); | |
c874dc6d BP |
1551 | |
1552 | /* Query the datapath for port information. */ | |
c228a364 | 1553 | error = dpif_port_query_by_name(p->dpif, devname, &odp_port); |
064af421 | 1554 | |
c874dc6d BP |
1555 | /* Find the old ofport. */ |
1556 | old_ofport = shash_find_data(&p->port_by_name, devname); | |
1557 | if (!error) { | |
1558 | if (!old_ofport) { | |
1559 | /* There's no port named 'devname' but there might be a port with | |
1560 | * the same port number. This could happen if a port is deleted | |
1561 | * and then a new one added in its place very quickly, or if a port | |
1562 | * is renamed. In the former case we want to send an OFPPR_DELETE | |
1563 | * and an OFPPR_ADD, and in the latter case we want to send a | |
1564 | * single OFPPR_MODIFY. We can distinguish the cases by comparing | |
1565 | * the old port's ifindex against the new port, or perhaps less | |
1566 | * reliably but more portably by comparing the old port's MAC | |
1567 | * against the new port's MAC. However, this code isn't that smart | |
1568 | * and always sends an OFPPR_MODIFY (XXX). */ | |
1569 | old_ofport = port_array_get(&p->ports, odp_port.port); | |
064af421 | 1570 | } |
c874dc6d | 1571 | } else if (error != ENOENT && error != ENODEV) { |
064af421 BP |
1572 | VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error " |
1573 | "%s", strerror(error)); | |
1574 | return; | |
1575 | } | |
c874dc6d BP |
1576 | |
1577 | /* Create a new ofport. */ | |
1578 | new_ofport = !error ? make_ofport(&odp_port) : NULL; | |
1579 | ||
1580 | /* Eliminate a few pathological cases. */ | |
1581 | if (!old_ofport && !new_ofport) { | |
1582 | return; | |
1583 | } else if (old_ofport && new_ofport) { | |
1584 | /* Most of the 'config' bits are OpenFlow soft state, but | |
1585 | * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow | |
1586 | * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and | |
1587 | * leaves the other bits 0.) */ | |
1588 | new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN; | |
1589 | ||
1590 | if (ofport_equal(old_ofport, new_ofport)) { | |
1591 | /* False alarm--no change. */ | |
1592 | ofport_free(new_ofport); | |
1593 | return; | |
1594 | } | |
1595 | } | |
1596 | ||
1597 | /* Now deal with the normal cases. */ | |
1598 | if (old_ofport) { | |
1599 | ofport_remove(p, old_ofport); | |
1600 | } | |
1601 | if (new_ofport) { | |
1602 | ofport_install(p, new_ofport); | |
1603 | } | |
1604 | send_port_status(p, new_ofport ? new_ofport : old_ofport, | |
1605 | (!old_ofport ? OFPPR_ADD | |
1606 | : !new_ofport ? OFPPR_DELETE | |
1607 | : OFPPR_MODIFY)); | |
1608 | ofport_free(old_ofport); | |
1609 | ||
1610 | /* Update port groups. */ | |
064af421 BP |
1611 | refresh_port_groups(p); |
1612 | } | |
1613 | ||
1614 | static int | |
1615 | init_ports(struct ofproto *p) | |
1616 | { | |
1617 | struct odp_port *ports; | |
1618 | size_t n_ports; | |
1619 | size_t i; | |
1620 | int error; | |
1621 | ||
c228a364 | 1622 | error = dpif_port_list(p->dpif, &ports, &n_ports); |
064af421 BP |
1623 | if (error) { |
1624 | return error; | |
1625 | } | |
1626 | ||
1627 | for (i = 0; i < n_ports; i++) { | |
1628 | const struct odp_port *odp_port = &ports[i]; | |
1629 | if (!ofport_conflicts(p, odp_port)) { | |
1630 | struct ofport *ofport = make_ofport(odp_port); | |
1631 | if (ofport) { | |
1632 | ofport_install(p, ofport); | |
1633 | } | |
1634 | } | |
1635 | } | |
1636 | free(ports); | |
1637 | refresh_port_groups(p); | |
1638 | return 0; | |
1639 | } | |
1640 | \f | |
1641 | static struct ofconn * | |
76ce9432 | 1642 | ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type) |
064af421 | 1643 | { |
76ce9432 BP |
1644 | struct ofconn *ofconn = xzalloc(sizeof *ofconn); |
1645 | ofconn->ofproto = p; | |
064af421 BP |
1646 | list_push_back(&p->all_conns, &ofconn->node); |
1647 | ofconn->rconn = rconn; | |
76ce9432 | 1648 | ofconn->type = type; |
9deba63b | 1649 | ofconn->role = NX_ROLE_OTHER; |
76ce9432 | 1650 | ofconn->packet_in_counter = rconn_packet_counter_create (); |
064af421 | 1651 | ofconn->pktbuf = NULL; |
064af421 | 1652 | ofconn->miss_send_len = 0; |
064af421 BP |
1653 | ofconn->reply_counter = rconn_packet_counter_create (); |
1654 | return ofconn; | |
1655 | } | |
1656 | ||
1657 | static void | |
c475ae67 | 1658 | ofconn_destroy(struct ofconn *ofconn) |
064af421 | 1659 | { |
5899143f | 1660 | if (ofconn->type == OFCONN_PRIMARY) { |
76ce9432 BP |
1661 | hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node); |
1662 | } | |
1663 | discovery_destroy(ofconn->discovery); | |
1664 | ||
064af421 | 1665 | list_remove(&ofconn->node); |
76ce9432 | 1666 | switch_status_unregister(ofconn->ss); |
064af421 BP |
1667 | rconn_destroy(ofconn->rconn); |
1668 | rconn_packet_counter_destroy(ofconn->packet_in_counter); | |
1669 | rconn_packet_counter_destroy(ofconn->reply_counter); | |
1670 | pktbuf_destroy(ofconn->pktbuf); | |
1671 | free(ofconn); | |
1672 | } | |
1673 | ||
1674 | static void | |
1675 | ofconn_run(struct ofconn *ofconn, struct ofproto *p) | |
1676 | { | |
1677 | int iteration; | |
76ce9432 BP |
1678 | size_t i; |
1679 | ||
1680 | if (ofconn->discovery) { | |
1681 | char *controller_name; | |
1682 | if (rconn_is_connectivity_questionable(ofconn->rconn)) { | |
1683 | discovery_question_connectivity(ofconn->discovery); | |
1684 | } | |
1685 | if (discovery_run(ofconn->discovery, &controller_name)) { | |
1686 | if (controller_name) { | |
eb15cdbb BP |
1687 | char *ofconn_name = ofconn_make_name(p, controller_name); |
1688 | rconn_connect(ofconn->rconn, controller_name, ofconn_name); | |
1689 | free(ofconn_name); | |
76ce9432 BP |
1690 | } else { |
1691 | rconn_disconnect(ofconn->rconn); | |
1692 | } | |
1693 | } | |
1694 | } | |
1695 | ||
1696 | for (i = 0; i < N_SCHEDULERS; i++) { | |
1697 | pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn); | |
1698 | } | |
064af421 BP |
1699 | |
1700 | rconn_run(ofconn->rconn); | |
1701 | ||
1702 | if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) { | |
1703 | /* Limit the number of iterations to prevent other tasks from | |
1704 | * starving. */ | |
1705 | for (iteration = 0; iteration < 50; iteration++) { | |
1706 | struct ofpbuf *of_msg = rconn_recv(ofconn->rconn); | |
1707 | if (!of_msg) { | |
1708 | break; | |
1709 | } | |
7778bd15 BP |
1710 | if (p->fail_open) { |
1711 | fail_open_maybe_recover(p->fail_open); | |
1712 | } | |
064af421 BP |
1713 | handle_openflow(ofconn, p, of_msg); |
1714 | ofpbuf_delete(of_msg); | |
1715 | } | |
1716 | } | |
1717 | ||
76ce9432 | 1718 | if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) { |
c475ae67 | 1719 | ofconn_destroy(ofconn); |
064af421 BP |
1720 | } |
1721 | } | |
1722 | ||
1723 | static void | |
1724 | ofconn_wait(struct ofconn *ofconn) | |
1725 | { | |
76ce9432 BP |
1726 | int i; |
1727 | ||
1728 | if (ofconn->discovery) { | |
1729 | discovery_wait(ofconn->discovery); | |
1730 | } | |
1731 | for (i = 0; i < N_SCHEDULERS; i++) { | |
1732 | pinsched_wait(ofconn->schedulers[i]); | |
1733 | } | |
064af421 BP |
1734 | rconn_run_wait(ofconn->rconn); |
1735 | if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) { | |
1736 | rconn_recv_wait(ofconn->rconn); | |
1737 | } else { | |
1738 | COVERAGE_INC(ofproto_ofconn_stuck); | |
1739 | } | |
1740 | } | |
c91248b3 BP |
1741 | |
1742 | /* Returns true if 'ofconn' should receive asynchronous messages. */ | |
1743 | static bool | |
1744 | ofconn_receives_async_msgs(const struct ofconn *ofconn) | |
1745 | { | |
5899143f BP |
1746 | if (ofconn->type == OFCONN_PRIMARY) { |
1747 | /* Primary controllers always get asynchronous messages unless they | |
c91248b3 BP |
1748 | * have configured themselves as "slaves". */ |
1749 | return ofconn->role != NX_ROLE_SLAVE; | |
1750 | } else { | |
5899143f BP |
1751 | /* Service connections don't get asynchronous messages unless they have |
1752 | * explicitly asked for them by setting a nonzero miss send length. */ | |
c91248b3 BP |
1753 | return ofconn->miss_send_len > 0; |
1754 | } | |
1755 | } | |
eb15cdbb BP |
1756 | |
1757 | /* Returns a human-readable name for an OpenFlow connection between 'ofproto' | |
1758 | * and 'target', suitable for use in log messages for identifying the | |
1759 | * connection. | |
1760 | * | |
1761 | * The name is dynamically allocated. The caller should free it (with free()) | |
1762 | * when it is no longer needed. */ | |
1763 | static char * | |
1764 | ofconn_make_name(const struct ofproto *ofproto, const char *target) | |
1765 | { | |
1766 | return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target); | |
1767 | } | |
7d674866 BP |
1768 | |
1769 | static void | |
1770 | ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst) | |
1771 | { | |
1772 | int i; | |
1773 | ||
1774 | for (i = 0; i < N_SCHEDULERS; i++) { | |
1775 | struct pinsched **s = &ofconn->schedulers[i]; | |
1776 | ||
1777 | if (rate > 0) { | |
1778 | if (!*s) { | |
1779 | *s = pinsched_create(rate, burst, | |
1780 | ofconn->ofproto->switch_status); | |
1781 | } else { | |
1782 | pinsched_set_limits(*s, rate, burst); | |
1783 | } | |
1784 | } else { | |
1785 | pinsched_destroy(*s); | |
1786 | *s = NULL; | |
1787 | } | |
1788 | } | |
1789 | } | |
1790 | \f | |
1791 | static void | |
1792 | ofservice_reconfigure(struct ofservice *ofservice, | |
1793 | const struct ofproto_controller *c) | |
1794 | { | |
1795 | ofservice->probe_interval = c->probe_interval; | |
1796 | ofservice->rate_limit = c->rate_limit; | |
1797 | ofservice->burst_limit = c->burst_limit; | |
1798 | } | |
1799 | ||
1800 | /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a | |
1801 | * positive errno value. */ | |
1802 | static int | |
1803 | ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c) | |
1804 | { | |
1805 | struct ofservice *ofservice; | |
1806 | struct pvconn *pvconn; | |
1807 | int error; | |
1808 | ||
1809 | error = pvconn_open(c->target, &pvconn); | |
1810 | if (error) { | |
1811 | return error; | |
1812 | } | |
1813 | ||
1814 | ofservice = xzalloc(sizeof *ofservice); | |
1815 | hmap_insert(&ofproto->services, &ofservice->node, | |
1816 | hash_string(c->target, 0)); | |
1817 | ofservice->pvconn = pvconn; | |
1818 | ||
1819 | ofservice_reconfigure(ofservice, c); | |
1820 | ||
1821 | return 0; | |
1822 | } | |
1823 | ||
1824 | static void | |
1825 | ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice) | |
1826 | { | |
1827 | hmap_remove(&ofproto->services, &ofservice->node); | |
1828 | pvconn_close(ofservice->pvconn); | |
1829 | free(ofservice); | |
1830 | } | |
1831 | ||
1832 | /* Finds and returns the ofservice within 'ofproto' that has the given | |
1833 | * 'target', or a null pointer if none exists. */ | |
1834 | static struct ofservice * | |
1835 | ofservice_lookup(struct ofproto *ofproto, const char *target) | |
1836 | { | |
1837 | struct ofservice *ofservice; | |
1838 | ||
1839 | HMAP_FOR_EACH_WITH_HASH (ofservice, struct ofservice, node, | |
1840 | hash_string(target, 0), &ofproto->services) { | |
1841 | if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) { | |
1842 | return ofservice; | |
1843 | } | |
1844 | } | |
1845 | return NULL; | |
1846 | } | |
064af421 BP |
1847 | \f |
1848 | /* Caller is responsible for initializing the 'cr' member of the returned | |
1849 | * rule. */ | |
1850 | static struct rule * | |
0193b2af | 1851 | rule_create(struct ofproto *ofproto, struct rule *super, |
064af421 | 1852 | const union ofp_action *actions, size_t n_actions, |
ca069229 | 1853 | uint16_t idle_timeout, uint16_t hard_timeout, |
39997502 | 1854 | uint64_t flow_cookie, bool send_flow_removed) |
064af421 | 1855 | { |
ec6fde61 | 1856 | struct rule *rule = xzalloc(sizeof *rule); |
064af421 BP |
1857 | rule->idle_timeout = idle_timeout; |
1858 | rule->hard_timeout = hard_timeout; | |
39997502 | 1859 | rule->flow_cookie = flow_cookie; |
064af421 | 1860 | rule->used = rule->created = time_msec(); |
ca069229 | 1861 | rule->send_flow_removed = send_flow_removed; |
064af421 BP |
1862 | rule->super = super; |
1863 | if (super) { | |
1864 | list_push_back(&super->list, &rule->list); | |
1865 | } else { | |
1866 | list_init(&rule->list); | |
1867 | } | |
1868 | rule->n_actions = n_actions; | |
1869 | rule->actions = xmemdup(actions, n_actions * sizeof *actions); | |
0193b2af JG |
1870 | netflow_flow_clear(&rule->nf_flow); |
1871 | netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created); | |
1872 | ||
064af421 BP |
1873 | return rule; |
1874 | } | |
1875 | ||
1876 | static struct rule * | |
1877 | rule_from_cls_rule(const struct cls_rule *cls_rule) | |
1878 | { | |
1879 | return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL; | |
1880 | } | |
1881 | ||
1882 | static void | |
1883 | rule_free(struct rule *rule) | |
1884 | { | |
1885 | free(rule->actions); | |
1886 | free(rule->odp_actions); | |
1887 | free(rule); | |
1888 | } | |
1889 | ||
1890 | /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its | |
1891 | * super-rule's list of subrules. If 'rule' is a super-rule, also iterates | |
1892 | * through all of its subrules and revalidates them, destroying any that no | |
1893 | * longer has a super-rule (which is probably all of them). | |
1894 | * | |
1895 | * Before calling this function, the caller must make have removed 'rule' from | |
1896 | * the classifier. If 'rule' is an exact-match rule, the caller is also | |
1897 | * responsible for ensuring that it has been uninstalled from the datapath. */ | |
1898 | static void | |
1899 | rule_destroy(struct ofproto *ofproto, struct rule *rule) | |
1900 | { | |
1901 | if (!rule->super) { | |
1902 | struct rule *subrule, *next; | |
1903 | LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) { | |
1904 | revalidate_rule(ofproto, subrule); | |
1905 | } | |
1906 | } else { | |
1907 | list_remove(&rule->list); | |
1908 | } | |
1909 | rule_free(rule); | |
1910 | } | |
1911 | ||
1912 | static bool | |
1913 | rule_has_out_port(const struct rule *rule, uint16_t out_port) | |
1914 | { | |
1915 | const union ofp_action *oa; | |
1916 | struct actions_iterator i; | |
1917 | ||
1918 | if (out_port == htons(OFPP_NONE)) { | |
1919 | return true; | |
1920 | } | |
1921 | for (oa = actions_first(&i, rule->actions, rule->n_actions); oa; | |
1922 | oa = actions_next(&i)) { | |
c1c9c9c4 | 1923 | if (action_outputs_to_port(oa, out_port)) { |
064af421 BP |
1924 | return true; |
1925 | } | |
1926 | } | |
1927 | return false; | |
1928 | } | |
1929 | ||
1930 | /* Executes the actions indicated by 'rule' on 'packet', which is in flow | |
1931 | * 'flow' and is considered to have arrived on ODP port 'in_port'. | |
1932 | * | |
1933 | * The flow that 'packet' actually contains does not need to actually match | |
1934 | * 'rule'; the actions in 'rule' will be applied to it either way. Likewise, | |
1935 | * the packet and byte counters for 'rule' will be credited for the packet sent | |
1936 | * out whether or not the packet actually matches 'rule'. | |
1937 | * | |
1938 | * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow, | |
1939 | * the caller must already have accurately composed ODP actions for it given | |
1940 | * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if | |
1941 | * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this | |
1942 | * function will compose a set of ODP actions based on 'rule''s OpenFlow | |
1943 | * actions and apply them to 'packet'. */ | |
1944 | static void | |
1945 | rule_execute(struct ofproto *ofproto, struct rule *rule, | |
1946 | struct ofpbuf *packet, const flow_t *flow) | |
1947 | { | |
1948 | const union odp_action *actions; | |
1949 | size_t n_actions; | |
1950 | struct odp_actions a; | |
1951 | ||
1952 | /* Grab or compose the ODP actions. | |
1953 | * | |
1954 | * The special case for an exact-match 'rule' where 'flow' is not the | |
1955 | * rule's flow is important to avoid, e.g., sending a packet out its input | |
1956 | * port simply because the ODP actions were composed for the wrong | |
1957 | * scenario. */ | |
1958 | if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) { | |
1959 | struct rule *super = rule->super ? rule->super : rule; | |
1960 | if (xlate_actions(super->actions, super->n_actions, flow, ofproto, | |
6a07af36 | 1961 | packet, &a, NULL, 0, NULL)) { |
064af421 BP |
1962 | return; |
1963 | } | |
1964 | actions = a.actions; | |
1965 | n_actions = a.n_actions; | |
1966 | } else { | |
1967 | actions = rule->odp_actions; | |
1968 | n_actions = rule->n_odp_actions; | |
1969 | } | |
1970 | ||
1971 | /* Execute the ODP actions. */ | |
c228a364 | 1972 | if (!dpif_execute(ofproto->dpif, flow->in_port, |
064af421 BP |
1973 | actions, n_actions, packet)) { |
1974 | struct odp_flow_stats stats; | |
1975 | flow_extract_stats(flow, packet, &stats); | |
0193b2af | 1976 | update_stats(ofproto, rule, &stats); |
064af421 | 1977 | rule->used = time_msec(); |
0193b2af | 1978 | netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used); |
064af421 BP |
1979 | } |
1980 | } | |
1981 | ||
1982 | static void | |
1983 | rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet, | |
1984 | uint16_t in_port) | |
1985 | { | |
1986 | struct rule *displaced_rule; | |
1987 | ||
1988 | /* Insert the rule in the classifier. */ | |
1989 | displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr)); | |
1990 | if (!rule->cr.wc.wildcards) { | |
1991 | rule_make_actions(p, rule, packet); | |
1992 | } | |
1993 | ||
1994 | /* Send the packet and credit it to the rule. */ | |
1995 | if (packet) { | |
1996 | flow_t flow; | |
659586ef | 1997 | flow_extract(packet, 0, in_port, &flow); |
064af421 BP |
1998 | rule_execute(p, rule, packet, &flow); |
1999 | } | |
2000 | ||
2001 | /* Install the rule in the datapath only after sending the packet, to | |
2002 | * avoid packet reordering. */ | |
2003 | if (rule->cr.wc.wildcards) { | |
2004 | COVERAGE_INC(ofproto_add_wc_flow); | |
2005 | p->need_revalidate = true; | |
2006 | } else { | |
2007 | rule_install(p, rule, displaced_rule); | |
2008 | } | |
2009 | ||
2010 | /* Free the rule that was displaced, if any. */ | |
2011 | if (displaced_rule) { | |
2012 | rule_destroy(p, displaced_rule); | |
2013 | } | |
2014 | } | |
2015 | ||
2016 | static struct rule * | |
2017 | rule_create_subrule(struct ofproto *ofproto, struct rule *rule, | |
2018 | const flow_t *flow) | |
2019 | { | |
0193b2af | 2020 | struct rule *subrule = rule_create(ofproto, rule, NULL, 0, |
ca069229 | 2021 | rule->idle_timeout, rule->hard_timeout, |
39997502 | 2022 | 0, false); |
064af421 | 2023 | COVERAGE_INC(ofproto_subrule_create); |
659586ef JG |
2024 | cls_rule_from_flow(flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX |
2025 | : rule->cr.priority), &subrule->cr); | |
064af421 BP |
2026 | classifier_insert_exact(&ofproto->cls, &subrule->cr); |
2027 | ||
2028 | return subrule; | |
2029 | } | |
2030 | ||
2031 | static void | |
2032 | rule_remove(struct ofproto *ofproto, struct rule *rule) | |
2033 | { | |
2034 | if (rule->cr.wc.wildcards) { | |
2035 | COVERAGE_INC(ofproto_del_wc_flow); | |
2036 | ofproto->need_revalidate = true; | |
2037 | } else { | |
2038 | rule_uninstall(ofproto, rule); | |
2039 | } | |
2040 | classifier_remove(&ofproto->cls, &rule->cr); | |
2041 | rule_destroy(ofproto, rule); | |
2042 | } | |
2043 | ||
2044 | /* Returns true if the actions changed, false otherwise. */ | |
2045 | static bool | |
2046 | rule_make_actions(struct ofproto *p, struct rule *rule, | |
2047 | const struct ofpbuf *packet) | |
2048 | { | |
2049 | const struct rule *super; | |
2050 | struct odp_actions a; | |
2051 | size_t actions_len; | |
2052 | ||
2053 | assert(!rule->cr.wc.wildcards); | |
2054 | ||
2055 | super = rule->super ? rule->super : rule; | |
2056 | rule->tags = 0; | |
2057 | xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p, | |
6a07af36 | 2058 | packet, &a, &rule->tags, &rule->may_install, |
0193b2af | 2059 | &rule->nf_flow.output_iface); |
064af421 BP |
2060 | |
2061 | actions_len = a.n_actions * sizeof *a.actions; | |
2062 | if (rule->n_odp_actions != a.n_actions | |
2063 | || memcmp(rule->odp_actions, a.actions, actions_len)) { | |
2064 | COVERAGE_INC(ofproto_odp_unchanged); | |
2065 | free(rule->odp_actions); | |
2066 | rule->n_odp_actions = a.n_actions; | |
2067 | rule->odp_actions = xmemdup(a.actions, actions_len); | |
2068 | return true; | |
2069 | } else { | |
2070 | return false; | |
2071 | } | |
2072 | } | |
2073 | ||
2074 | static int | |
2075 | do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags, | |
2076 | struct odp_flow_put *put) | |
2077 | { | |
2078 | memset(&put->flow.stats, 0, sizeof put->flow.stats); | |
2079 | put->flow.key = rule->cr.flow; | |
2080 | put->flow.actions = rule->odp_actions; | |
2081 | put->flow.n_actions = rule->n_odp_actions; | |
ab48643b | 2082 | put->flow.flags = 0; |
064af421 | 2083 | put->flags = flags; |
c228a364 | 2084 | return dpif_flow_put(ofproto->dpif, put); |
064af421 BP |
2085 | } |
2086 | ||
2087 | static void | |
2088 | rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule) | |
2089 | { | |
2090 | assert(!rule->cr.wc.wildcards); | |
2091 | ||
2092 | if (rule->may_install) { | |
2093 | struct odp_flow_put put; | |
2094 | if (!do_put_flow(p, rule, | |
2095 | ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS, | |
2096 | &put)) { | |
2097 | rule->installed = true; | |
2098 | if (displaced_rule) { | |
14986b31 | 2099 | update_stats(p, displaced_rule, &put.flow.stats); |
064af421 BP |
2100 | rule_post_uninstall(p, displaced_rule); |
2101 | } | |
2102 | } | |
2103 | } else if (displaced_rule) { | |
2104 | rule_uninstall(p, displaced_rule); | |
2105 | } | |
2106 | } | |
2107 | ||
2108 | static void | |
2109 | rule_reinstall(struct ofproto *ofproto, struct rule *rule) | |
2110 | { | |
2111 | if (rule->installed) { | |
2112 | struct odp_flow_put put; | |
2113 | COVERAGE_INC(ofproto_dp_missed); | |
2114 | do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put); | |
2115 | } else { | |
2116 | rule_install(ofproto, rule, NULL); | |
2117 | } | |
2118 | } | |
2119 | ||
2120 | static void | |
2121 | rule_update_actions(struct ofproto *ofproto, struct rule *rule) | |
2122 | { | |
42c3641c JG |
2123 | bool actions_changed; |
2124 | uint16_t new_out_iface, old_out_iface; | |
2125 | ||
2126 | old_out_iface = rule->nf_flow.output_iface; | |
2127 | actions_changed = rule_make_actions(ofproto, rule, NULL); | |
2128 | ||
064af421 BP |
2129 | if (rule->may_install) { |
2130 | if (rule->installed) { | |
2131 | if (actions_changed) { | |
064af421 | 2132 | struct odp_flow_put put; |
42c3641c JG |
2133 | do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY |
2134 | | ODPPF_ZERO_STATS, &put); | |
2135 | update_stats(ofproto, rule, &put.flow.stats); | |
2136 | ||
2137 | /* Temporarily set the old output iface so that NetFlow | |
2138 | * messages have the correct output interface for the old | |
2139 | * stats. */ | |
2140 | new_out_iface = rule->nf_flow.output_iface; | |
2141 | rule->nf_flow.output_iface = old_out_iface; | |
2142 | rule_post_uninstall(ofproto, rule); | |
2143 | rule->nf_flow.output_iface = new_out_iface; | |
064af421 BP |
2144 | } |
2145 | } else { | |
2146 | rule_install(ofproto, rule, NULL); | |
2147 | } | |
2148 | } else { | |
2149 | rule_uninstall(ofproto, rule); | |
2150 | } | |
2151 | } | |
2152 | ||
2153 | static void | |
2154 | rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes) | |
2155 | { | |
2156 | uint64_t total_bytes = rule->byte_count + extra_bytes; | |
2157 | ||
2158 | if (ofproto->ofhooks->account_flow_cb | |
2159 | && total_bytes > rule->accounted_bytes) | |
2160 | { | |
2161 | ofproto->ofhooks->account_flow_cb( | |
26d79bf2 | 2162 | &rule->cr.flow, rule->tags, rule->odp_actions, rule->n_odp_actions, |
064af421 BP |
2163 | total_bytes - rule->accounted_bytes, ofproto->aux); |
2164 | rule->accounted_bytes = total_bytes; | |
2165 | } | |
2166 | } | |
2167 | ||
2168 | static void | |
2169 | rule_uninstall(struct ofproto *p, struct rule *rule) | |
2170 | { | |
2171 | assert(!rule->cr.wc.wildcards); | |
2172 | if (rule->installed) { | |
2173 | struct odp_flow odp_flow; | |
2174 | ||
2175 | odp_flow.key = rule->cr.flow; | |
2176 | odp_flow.actions = NULL; | |
2177 | odp_flow.n_actions = 0; | |
ab48643b | 2178 | odp_flow.flags = 0; |
c228a364 | 2179 | if (!dpif_flow_del(p->dpif, &odp_flow)) { |
0193b2af | 2180 | update_stats(p, rule, &odp_flow.stats); |
064af421 BP |
2181 | } |
2182 | rule->installed = false; | |
2183 | ||
2184 | rule_post_uninstall(p, rule); | |
2185 | } | |
2186 | } | |
2187 | ||
0193b2af JG |
2188 | static bool |
2189 | is_controller_rule(struct rule *rule) | |
2190 | { | |
2191 | /* If the only action is send to the controller then don't report | |
2192 | * NetFlow expiration messages since it is just part of the control | |
2193 | * logic for the network and not real traffic. */ | |
2194 | ||
c1c9c9c4 BP |
2195 | return (rule |
2196 | && rule->super | |
2197 | && rule->super->n_actions == 1 | |
2198 | && action_outputs_to_port(&rule->super->actions[0], | |
2199 | htons(OFPP_CONTROLLER))); | |
0193b2af JG |
2200 | } |
2201 | ||
064af421 BP |
2202 | static void |
2203 | rule_post_uninstall(struct ofproto *ofproto, struct rule *rule) | |
2204 | { | |
2205 | struct rule *super = rule->super; | |
2206 | ||
2207 | rule_account(ofproto, rule, 0); | |
6a07af36 | 2208 | |
0193b2af | 2209 | if (ofproto->netflow && !is_controller_rule(rule)) { |
064af421 BP |
2210 | struct ofexpired expired; |
2211 | expired.flow = rule->cr.flow; | |
2212 | expired.packet_count = rule->packet_count; | |
2213 | expired.byte_count = rule->byte_count; | |
2214 | expired.used = rule->used; | |
0193b2af | 2215 | netflow_expire(ofproto->netflow, &rule->nf_flow, &expired); |
064af421 BP |
2216 | } |
2217 | if (super) { | |
2218 | super->packet_count += rule->packet_count; | |
2219 | super->byte_count += rule->byte_count; | |
064af421 | 2220 | |
0c0afbec JG |
2221 | /* Reset counters to prevent double counting if the rule ever gets |
2222 | * reinstalled. */ | |
2223 | rule->packet_count = 0; | |
2224 | rule->byte_count = 0; | |
2225 | rule->accounted_bytes = 0; | |
0193b2af JG |
2226 | |
2227 | netflow_flow_clear(&rule->nf_flow); | |
0c0afbec | 2228 | } |
064af421 BP |
2229 | } |
2230 | \f | |
2231 | static void | |
2232 | queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn, | |
2233 | struct rconn_packet_counter *counter) | |
2234 | { | |
2235 | update_openflow_length(msg); | |
2236 | if (rconn_send(ofconn->rconn, msg, counter)) { | |
2237 | ofpbuf_delete(msg); | |
2238 | } | |
2239 | } | |
2240 | ||
2241 | static void | |
2242 | send_error(const struct ofconn *ofconn, const struct ofp_header *oh, | |
2243 | int error, const void *data, size_t len) | |
2244 | { | |
2245 | struct ofpbuf *buf; | |
2246 | struct ofp_error_msg *oem; | |
2247 | ||
2248 | if (!(error >> 16)) { | |
2249 | VLOG_WARN_RL(&rl, "not sending bad error code %d to controller", | |
2250 | error); | |
2251 | return; | |
2252 | } | |
2253 | ||
2254 | COVERAGE_INC(ofproto_error); | |
2255 | oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR, | |
2256 | oh ? oh->xid : 0, &buf); | |
2257 | oem->type = htons((unsigned int) error >> 16); | |
2258 | oem->code = htons(error & 0xffff); | |
2259 | memcpy(oem->data, data, len); | |
2260 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
2261 | } | |
2262 | ||
2263 | static void | |
2264 | send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh, | |
2265 | int error) | |
2266 | { | |
2267 | size_t oh_length = ntohs(oh->length); | |
2268 | send_error(ofconn, oh, error, oh, MIN(oh_length, 64)); | |
2269 | } | |
2270 | ||
2271 | static void | |
2272 | hton_ofp_phy_port(struct ofp_phy_port *opp) | |
2273 | { | |
2274 | opp->port_no = htons(opp->port_no); | |
2275 | opp->config = htonl(opp->config); | |
2276 | opp->state = htonl(opp->state); | |
2277 | opp->curr = htonl(opp->curr); | |
2278 | opp->advertised = htonl(opp->advertised); | |
2279 | opp->supported = htonl(opp->supported); | |
2280 | opp->peer = htonl(opp->peer); | |
2281 | } | |
2282 | ||
2283 | static int | |
2284 | handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh) | |
2285 | { | |
2286 | struct ofp_header *rq = oh; | |
2287 | queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter); | |
2288 | return 0; | |
2289 | } | |
2290 | ||
2291 | static int | |
2292 | handle_features_request(struct ofproto *p, struct ofconn *ofconn, | |
2293 | struct ofp_header *oh) | |
2294 | { | |
2295 | struct ofp_switch_features *osf; | |
2296 | struct ofpbuf *buf; | |
2297 | unsigned int port_no; | |
2298 | struct ofport *port; | |
2299 | ||
2300 | osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf); | |
2301 | osf->datapath_id = htonll(p->datapath_id); | |
2302 | osf->n_buffers = htonl(pktbuf_capacity()); | |
2303 | osf->n_tables = 2; | |
2304 | osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS | | |
0254ae23 | 2305 | OFPC_PORT_STATS | OFPC_ARP_MATCH_IP); |
064af421 BP |
2306 | osf->actions = htonl((1u << OFPAT_OUTPUT) | |
2307 | (1u << OFPAT_SET_VLAN_VID) | | |
2308 | (1u << OFPAT_SET_VLAN_PCP) | | |
2309 | (1u << OFPAT_STRIP_VLAN) | | |
2310 | (1u << OFPAT_SET_DL_SRC) | | |
2311 | (1u << OFPAT_SET_DL_DST) | | |
2312 | (1u << OFPAT_SET_NW_SRC) | | |
2313 | (1u << OFPAT_SET_NW_DST) | | |
959a2ecd | 2314 | (1u << OFPAT_SET_NW_TOS) | |
064af421 | 2315 | (1u << OFPAT_SET_TP_SRC) | |
c1c9c9c4 BP |
2316 | (1u << OFPAT_SET_TP_DST) | |
2317 | (1u << OFPAT_ENQUEUE)); | |
064af421 BP |
2318 | |
2319 | PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) { | |
2320 | hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp)); | |
2321 | } | |
2322 | ||
2323 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
2324 | return 0; | |
2325 | } | |
2326 | ||
2327 | static int | |
2328 | handle_get_config_request(struct ofproto *p, struct ofconn *ofconn, | |
2329 | struct ofp_header *oh) | |
2330 | { | |
2331 | struct ofpbuf *buf; | |
2332 | struct ofp_switch_config *osc; | |
2333 | uint16_t flags; | |
2334 | bool drop_frags; | |
2335 | ||
2336 | /* Figure out flags. */ | |
c228a364 | 2337 | dpif_get_drop_frags(p->dpif, &drop_frags); |
064af421 | 2338 | flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL; |
064af421 BP |
2339 | |
2340 | /* Send reply. */ | |
2341 | osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf); | |
2342 | osc->flags = htons(flags); | |
2343 | osc->miss_send_len = htons(ofconn->miss_send_len); | |
2344 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
2345 | ||
2346 | return 0; | |
2347 | } | |
2348 | ||
2349 | static int | |
2350 | handle_set_config(struct ofproto *p, struct ofconn *ofconn, | |
2351 | struct ofp_switch_config *osc) | |
2352 | { | |
2353 | uint16_t flags; | |
2354 | int error; | |
2355 | ||
2356 | error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc); | |
2357 | if (error) { | |
2358 | return error; | |
2359 | } | |
2360 | flags = ntohs(osc->flags); | |
2361 | ||
5899143f | 2362 | if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) { |
064af421 BP |
2363 | switch (flags & OFPC_FRAG_MASK) { |
2364 | case OFPC_FRAG_NORMAL: | |
c228a364 | 2365 | dpif_set_drop_frags(p->dpif, false); |
064af421 BP |
2366 | break; |
2367 | case OFPC_FRAG_DROP: | |
c228a364 | 2368 | dpif_set_drop_frags(p->dpif, true); |
064af421 BP |
2369 | break; |
2370 | default: | |
2371 | VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")", | |
2372 | osc->flags); | |
2373 | break; | |
2374 | } | |
2375 | } | |
2376 | ||
064af421 BP |
2377 | ofconn->miss_send_len = ntohs(osc->miss_send_len); |
2378 | ||
2379 | return 0; | |
2380 | } | |
2381 | ||
2382 | static void | |
6a07af36 JG |
2383 | add_output_group_action(struct odp_actions *actions, uint16_t group, |
2384 | uint16_t *nf_output_iface) | |
064af421 BP |
2385 | { |
2386 | odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group; | |
6a07af36 JG |
2387 | |
2388 | if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) { | |
2389 | *nf_output_iface = NF_OUT_FLOOD; | |
2390 | } | |
064af421 BP |
2391 | } |
2392 | ||
2393 | static void | |
aae51f53 | 2394 | add_controller_action(struct odp_actions *actions, uint16_t max_len) |
064af421 BP |
2395 | { |
2396 | union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER); | |
aae51f53 | 2397 | a->controller.arg = max_len; |
064af421 BP |
2398 | } |
2399 | ||
2400 | struct action_xlate_ctx { | |
2401 | /* Input. */ | |
e18fe8a2 | 2402 | flow_t flow; /* Flow to which these actions correspond. */ |
064af421 BP |
2403 | int recurse; /* Recursion level, via xlate_table_action. */ |
2404 | struct ofproto *ofproto; | |
2405 | const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a | |
2406 | * null pointer if we are revalidating | |
2407 | * without a packet to refer to. */ | |
2408 | ||
2409 | /* Output. */ | |
2410 | struct odp_actions *out; /* Datapath actions. */ | |
2411 | tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */ | |
d6fbec6d | 2412 | bool may_set_up_flow; /* True ordinarily; false if the actions must |
064af421 | 2413 | * be reassessed for every packet. */ |
6a07af36 | 2414 | uint16_t nf_output_iface; /* Output interface index for NetFlow. */ |
064af421 BP |
2415 | }; |
2416 | ||
2417 | static void do_xlate_actions(const union ofp_action *in, size_t n_in, | |
2418 | struct action_xlate_ctx *ctx); | |
2419 | ||
2420 | static void | |
2421 | add_output_action(struct action_xlate_ctx *ctx, uint16_t port) | |
2422 | { | |
2423 | const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port); | |
6cfaf517 BP |
2424 | |
2425 | if (ofport) { | |
2426 | if (ofport->opp.config & OFPPC_NO_FWD) { | |
2427 | /* Forwarding disabled on port. */ | |
2428 | return; | |
2429 | } | |
2430 | } else { | |
2431 | /* | |
2432 | * We don't have an ofport record for this port, but it doesn't hurt to | |
2433 | * allow forwarding to it anyhow. Maybe such a port will appear later | |
2434 | * and we're pre-populating the flow table. | |
2435 | */ | |
064af421 | 2436 | } |
6cfaf517 BP |
2437 | |
2438 | odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port; | |
6a07af36 | 2439 | ctx->nf_output_iface = port; |
064af421 BP |
2440 | } |
2441 | ||
2442 | static struct rule * | |
2443 | lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow) | |
2444 | { | |
2445 | struct rule *rule; | |
2446 | rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow)); | |
2447 | ||
2448 | /* The rule we found might not be valid, since we could be in need of | |
2449 | * revalidation. If it is not valid, don't return it. */ | |
2450 | if (rule | |
2451 | && rule->super | |
2452 | && ofproto->need_revalidate | |
2453 | && !revalidate_rule(ofproto, rule)) { | |
2454 | COVERAGE_INC(ofproto_invalidated); | |
2455 | return NULL; | |
2456 | } | |
2457 | ||
2458 | return rule; | |
2459 | } | |
2460 | ||
2461 | static void | |
2462 | xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port) | |
2463 | { | |
2464 | if (!ctx->recurse) { | |
2c5d1389 | 2465 | uint16_t old_in_port; |
064af421 | 2466 | struct rule *rule; |
064af421 | 2467 | |
2c5d1389 BP |
2468 | /* Look up a flow with 'in_port' as the input port. Then restore the |
2469 | * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will | |
2470 | * have surprising behavior). */ | |
2471 | old_in_port = ctx->flow.in_port; | |
e18fe8a2 BP |
2472 | ctx->flow.in_port = in_port; |
2473 | rule = lookup_valid_rule(ctx->ofproto, &ctx->flow); | |
2c5d1389 BP |
2474 | ctx->flow.in_port = old_in_port; |
2475 | ||
064af421 BP |
2476 | if (rule) { |
2477 | if (rule->super) { | |
2478 | rule = rule->super; | |
2479 | } | |
2480 | ||
2481 | ctx->recurse++; | |
2482 | do_xlate_actions(rule->actions, rule->n_actions, ctx); | |
2483 | ctx->recurse--; | |
2484 | } | |
2485 | } | |
2486 | } | |
2487 | ||
2488 | static void | |
aae51f53 BP |
2489 | xlate_output_action__(struct action_xlate_ctx *ctx, |
2490 | uint16_t port, uint16_t max_len) | |
064af421 BP |
2491 | { |
2492 | uint16_t odp_port; | |
6a07af36 JG |
2493 | uint16_t prev_nf_output_iface = ctx->nf_output_iface; |
2494 | ||
2495 | ctx->nf_output_iface = NF_OUT_DROP; | |
064af421 | 2496 | |
aae51f53 | 2497 | switch (port) { |
064af421 | 2498 | case OFPP_IN_PORT: |
e18fe8a2 | 2499 | add_output_action(ctx, ctx->flow.in_port); |
064af421 BP |
2500 | break; |
2501 | case OFPP_TABLE: | |
e18fe8a2 | 2502 | xlate_table_action(ctx, ctx->flow.in_port); |
064af421 BP |
2503 | break; |
2504 | case OFPP_NORMAL: | |
e18fe8a2 | 2505 | if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet, |
064af421 | 2506 | ctx->out, ctx->tags, |
6a07af36 | 2507 | &ctx->nf_output_iface, |
064af421 BP |
2508 | ctx->ofproto->aux)) { |
2509 | COVERAGE_INC(ofproto_uninstallable); | |
d6fbec6d | 2510 | ctx->may_set_up_flow = false; |
064af421 BP |
2511 | } |
2512 | break; | |
2513 | case OFPP_FLOOD: | |
6a07af36 JG |
2514 | add_output_group_action(ctx->out, DP_GROUP_FLOOD, |
2515 | &ctx->nf_output_iface); | |
064af421 BP |
2516 | break; |
2517 | case OFPP_ALL: | |
6a07af36 | 2518 | add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface); |
064af421 BP |
2519 | break; |
2520 | case OFPP_CONTROLLER: | |
aae51f53 | 2521 | add_controller_action(ctx->out, max_len); |
064af421 BP |
2522 | break; |
2523 | case OFPP_LOCAL: | |
2524 | add_output_action(ctx, ODPP_LOCAL); | |
2525 | break; | |
2526 | default: | |
aae51f53 | 2527 | odp_port = ofp_port_to_odp_port(port); |
e18fe8a2 | 2528 | if (odp_port != ctx->flow.in_port) { |
064af421 BP |
2529 | add_output_action(ctx, odp_port); |
2530 | } | |
2531 | break; | |
2532 | } | |
6a07af36 JG |
2533 | |
2534 | if (prev_nf_output_iface == NF_OUT_FLOOD) { | |
2535 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
2536 | } else if (ctx->nf_output_iface == NF_OUT_DROP) { | |
2537 | ctx->nf_output_iface = prev_nf_output_iface; | |
2538 | } else if (prev_nf_output_iface != NF_OUT_DROP && | |
2539 | ctx->nf_output_iface != NF_OUT_FLOOD) { | |
2540 | ctx->nf_output_iface = NF_OUT_MULTI; | |
2541 | } | |
064af421 BP |
2542 | } |
2543 | ||
aae51f53 BP |
2544 | static void |
2545 | xlate_output_action(struct action_xlate_ctx *ctx, | |
2546 | const struct ofp_action_output *oao) | |
2547 | { | |
2548 | xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len)); | |
2549 | } | |
2550 | ||
c1c9c9c4 BP |
2551 | /* If the final ODP action in 'ctx' is "pop priority", drop it, as an |
2552 | * optimization, because we're going to add another action that sets the | |
2553 | * priority immediately after, or because there are no actions following the | |
2554 | * pop. */ | |
2555 | static void | |
2556 | remove_pop_action(struct action_xlate_ctx *ctx) | |
2557 | { | |
2558 | size_t n = ctx->out->n_actions; | |
2559 | if (n > 0 && ctx->out->actions[n - 1].type == ODPAT_POP_PRIORITY) { | |
2560 | ctx->out->n_actions--; | |
2561 | } | |
2562 | } | |
2563 | ||
2564 | static void | |
2565 | xlate_enqueue_action(struct action_xlate_ctx *ctx, | |
2566 | const struct ofp_action_enqueue *oae) | |
2567 | { | |
2568 | uint16_t ofp_port, odp_port; | |
aae51f53 BP |
2569 | uint32_t priority; |
2570 | int error; | |
2571 | ||
2572 | error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id), | |
2573 | &priority); | |
2574 | if (error) { | |
2575 | /* Fall back to ordinary output action. */ | |
2576 | xlate_output_action__(ctx, ntohs(oae->port), 0); | |
2577 | return; | |
2578 | } | |
c1c9c9c4 BP |
2579 | |
2580 | /* Figure out ODP output port. */ | |
2581 | ofp_port = ntohs(oae->port); | |
2582 | if (ofp_port != OFPP_IN_PORT) { | |
2583 | odp_port = ofp_port_to_odp_port(ofp_port); | |
2584 | } else { | |
2585 | odp_port = ctx->flow.in_port; | |
2586 | } | |
2587 | ||
2588 | /* Add ODP actions. */ | |
2589 | remove_pop_action(ctx); | |
2590 | odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority | |
aae51f53 | 2591 | = priority; |
c1c9c9c4 BP |
2592 | add_output_action(ctx, odp_port); |
2593 | odp_actions_add(ctx->out, ODPAT_POP_PRIORITY); | |
2594 | ||
2595 | /* Update NetFlow output port. */ | |
2596 | if (ctx->nf_output_iface == NF_OUT_DROP) { | |
2597 | ctx->nf_output_iface = odp_port; | |
2598 | } else if (ctx->nf_output_iface != NF_OUT_FLOOD) { | |
2599 | ctx->nf_output_iface = NF_OUT_MULTI; | |
2600 | } | |
2601 | } | |
2602 | ||
064af421 BP |
2603 | static void |
2604 | xlate_nicira_action(struct action_xlate_ctx *ctx, | |
2605 | const struct nx_action_header *nah) | |
2606 | { | |
2607 | const struct nx_action_resubmit *nar; | |
659586ef JG |
2608 | const struct nx_action_set_tunnel *nast; |
2609 | union odp_action *oa; | |
064af421 BP |
2610 | int subtype = ntohs(nah->subtype); |
2611 | ||
2612 | assert(nah->vendor == htonl(NX_VENDOR_ID)); | |
2613 | switch (subtype) { | |
2614 | case NXAST_RESUBMIT: | |
2615 | nar = (const struct nx_action_resubmit *) nah; | |
2616 | xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port))); | |
2617 | break; | |
2618 | ||
659586ef JG |
2619 | case NXAST_SET_TUNNEL: |
2620 | nast = (const struct nx_action_set_tunnel *) nah; | |
2621 | oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL); | |
2622 | ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id; | |
2623 | break; | |
2624 | ||
999f0d45 | 2625 | /* If you add a new action here that modifies flow data, don't forget to |
c1c9c9c4 | 2626 | * update the flow key in ctx->flow at the same time. */ |
999f0d45 | 2627 | |
064af421 BP |
2628 | default: |
2629 | VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype); | |
2630 | break; | |
2631 | } | |
2632 | } | |
2633 | ||
2634 | static void | |
2635 | do_xlate_actions(const union ofp_action *in, size_t n_in, | |
2636 | struct action_xlate_ctx *ctx) | |
2637 | { | |
2638 | struct actions_iterator iter; | |
2639 | const union ofp_action *ia; | |
2640 | const struct ofport *port; | |
2641 | ||
e18fe8a2 | 2642 | port = port_array_get(&ctx->ofproto->ports, ctx->flow.in_port); |
064af421 | 2643 | if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) && |
ba186119 | 2644 | port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp) |
064af421 BP |
2645 | ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) { |
2646 | /* Drop this flow. */ | |
2647 | return; | |
2648 | } | |
2649 | ||
2650 | for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) { | |
2651 | uint16_t type = ntohs(ia->type); | |
2652 | union odp_action *oa; | |
2653 | ||
2654 | switch (type) { | |
2655 | case OFPAT_OUTPUT: | |
2656 | xlate_output_action(ctx, &ia->output); | |
2657 | break; | |
2658 | ||
2659 | case OFPAT_SET_VLAN_VID: | |
2660 | oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID); | |
999f0d45 | 2661 | ctx->flow.dl_vlan = oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid; |
064af421 BP |
2662 | break; |
2663 | ||
2664 | case OFPAT_SET_VLAN_PCP: | |
2665 | oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP); | |
999f0d45 | 2666 | ctx->flow.dl_vlan_pcp = oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp; |
064af421 BP |
2667 | break; |
2668 | ||
2669 | case OFPAT_STRIP_VLAN: | |
2670 | odp_actions_add(ctx->out, ODPAT_STRIP_VLAN); | |
7ac0f8db | 2671 | ctx->flow.dl_vlan = htons(OFP_VLAN_NONE); |
999f0d45 | 2672 | ctx->flow.dl_vlan_pcp = 0; |
064af421 BP |
2673 | break; |
2674 | ||
2675 | case OFPAT_SET_DL_SRC: | |
2676 | oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC); | |
2677 | memcpy(oa->dl_addr.dl_addr, | |
2678 | ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); | |
999f0d45 BP |
2679 | memcpy(ctx->flow.dl_src, |
2680 | ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); | |
064af421 BP |
2681 | break; |
2682 | ||
2683 | case OFPAT_SET_DL_DST: | |
2684 | oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST); | |
2685 | memcpy(oa->dl_addr.dl_addr, | |
2686 | ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); | |
999f0d45 BP |
2687 | memcpy(ctx->flow.dl_dst, |
2688 | ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); | |
064af421 BP |
2689 | break; |
2690 | ||
2691 | case OFPAT_SET_NW_SRC: | |
2692 | oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC); | |
999f0d45 | 2693 | ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr; |
064af421 BP |
2694 | break; |
2695 | ||
2d70a31a JP |
2696 | case OFPAT_SET_NW_DST: |
2697 | oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST); | |
999f0d45 | 2698 | ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr; |
2d38e234 | 2699 | break; |
959a2ecd JP |
2700 | |
2701 | case OFPAT_SET_NW_TOS: | |
2702 | oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS); | |
999f0d45 | 2703 | ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos; |
2d70a31a JP |
2704 | break; |
2705 | ||
064af421 BP |
2706 | case OFPAT_SET_TP_SRC: |
2707 | oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC); | |
999f0d45 | 2708 | ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port; |
064af421 BP |
2709 | break; |
2710 | ||
2d70a31a JP |
2711 | case OFPAT_SET_TP_DST: |
2712 | oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST); | |
999f0d45 | 2713 | ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port; |
2d70a31a JP |
2714 | break; |
2715 | ||
064af421 BP |
2716 | case OFPAT_VENDOR: |
2717 | xlate_nicira_action(ctx, (const struct nx_action_header *) ia); | |
2718 | break; | |
2719 | ||
c1c9c9c4 BP |
2720 | case OFPAT_ENQUEUE: |
2721 | xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia); | |
2722 | break; | |
2723 | ||
064af421 BP |
2724 | default: |
2725 | VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type); | |
2726 | break; | |
2727 | } | |
2728 | } | |
2729 | } | |
2730 | ||
2731 | static int | |
2732 | xlate_actions(const union ofp_action *in, size_t n_in, | |
2733 | const flow_t *flow, struct ofproto *ofproto, | |
2734 | const struct ofpbuf *packet, | |
6a07af36 JG |
2735 | struct odp_actions *out, tag_type *tags, bool *may_set_up_flow, |
2736 | uint16_t *nf_output_iface) | |
064af421 BP |
2737 | { |
2738 | tag_type no_tags = 0; | |
2739 | struct action_xlate_ctx ctx; | |
2740 | COVERAGE_INC(ofproto_ofp2odp); | |
2741 | odp_actions_init(out); | |
e18fe8a2 | 2742 | ctx.flow = *flow; |
064af421 BP |
2743 | ctx.recurse = 0; |
2744 | ctx.ofproto = ofproto; | |
2745 | ctx.packet = packet; | |
2746 | ctx.out = out; | |
2747 | ctx.tags = tags ? tags : &no_tags; | |
d6fbec6d | 2748 | ctx.may_set_up_flow = true; |
6a07af36 | 2749 | ctx.nf_output_iface = NF_OUT_DROP; |
064af421 | 2750 | do_xlate_actions(in, n_in, &ctx); |
c1c9c9c4 | 2751 | remove_pop_action(&ctx); |
0ad9b732 | 2752 | |
d6fbec6d | 2753 | /* Check with in-band control to see if we're allowed to set up this |
0ad9b732 JP |
2754 | * flow. */ |
2755 | if (!in_band_rule_check(ofproto->in_band, flow, out)) { | |
d6fbec6d | 2756 | ctx.may_set_up_flow = false; |
0ad9b732 JP |
2757 | } |
2758 | ||
d6fbec6d BP |
2759 | if (may_set_up_flow) { |
2760 | *may_set_up_flow = ctx.may_set_up_flow; | |
064af421 | 2761 | } |
6a07af36 JG |
2762 | if (nf_output_iface) { |
2763 | *nf_output_iface = ctx.nf_output_iface; | |
064af421 BP |
2764 | } |
2765 | if (odp_actions_overflow(out)) { | |
2766 | odp_actions_init(out); | |
2767 | return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY); | |
2768 | } | |
2769 | return 0; | |
2770 | } | |
2771 | ||
9deba63b BP |
2772 | /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow |
2773 | * error message code (composed with ofp_mkerr()) for the caller to propagate | |
2774 | * upward. Otherwise, returns 0. | |
2775 | * | |
2776 | * 'oh' is used to make log messages more informative. */ | |
2777 | static int | |
2778 | reject_slave_controller(struct ofconn *ofconn, const struct ofp_header *oh) | |
2779 | { | |
5899143f | 2780 | if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) { |
9deba63b BP |
2781 | static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5); |
2782 | char *type_name; | |
2783 | ||
2784 | type_name = ofp_message_type_to_string(oh->type); | |
2785 | VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller", | |
2786 | type_name); | |
2787 | free(type_name); | |
2788 | ||
2789 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM); | |
2790 | } else { | |
2791 | return 0; | |
2792 | } | |
2793 | } | |
2794 | ||
064af421 BP |
2795 | static int |
2796 | handle_packet_out(struct ofproto *p, struct ofconn *ofconn, | |
2797 | struct ofp_header *oh) | |
2798 | { | |
2799 | struct ofp_packet_out *opo; | |
2800 | struct ofpbuf payload, *buffer; | |
2801 | struct odp_actions actions; | |
2802 | int n_actions; | |
2803 | uint16_t in_port; | |
2804 | flow_t flow; | |
2805 | int error; | |
2806 | ||
9deba63b BP |
2807 | error = reject_slave_controller(ofconn, oh); |
2808 | if (error) { | |
2809 | return error; | |
2810 | } | |
2811 | ||
064af421 BP |
2812 | error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports); |
2813 | if (error) { | |
2814 | return error; | |
2815 | } | |
2816 | opo = (struct ofp_packet_out *) oh; | |
2817 | ||
2818 | COVERAGE_INC(ofproto_packet_out); | |
2819 | if (opo->buffer_id != htonl(UINT32_MAX)) { | |
2820 | error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id), | |
2821 | &buffer, &in_port); | |
7778bd15 | 2822 | if (error || !buffer) { |
064af421 BP |
2823 | return error; |
2824 | } | |
2825 | payload = *buffer; | |
2826 | } else { | |
2827 | buffer = NULL; | |
2828 | } | |
2829 | ||
659586ef | 2830 | flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow); |
064af421 | 2831 | error = xlate_actions((const union ofp_action *) opo->actions, n_actions, |
6a07af36 | 2832 | &flow, p, &payload, &actions, NULL, NULL, NULL); |
064af421 BP |
2833 | if (error) { |
2834 | return error; | |
2835 | } | |
2836 | ||
c228a364 | 2837 | dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions, |
064af421 BP |
2838 | &payload); |
2839 | ofpbuf_delete(buffer); | |
2840 | ||
2841 | return 0; | |
2842 | } | |
2843 | ||
2844 | static void | |
2845 | update_port_config(struct ofproto *p, struct ofport *port, | |
2846 | uint32_t config, uint32_t mask) | |
2847 | { | |
2848 | mask &= config ^ port->opp.config; | |
2849 | if (mask & OFPPC_PORT_DOWN) { | |
2850 | if (config & OFPPC_PORT_DOWN) { | |
2851 | netdev_turn_flags_off(port->netdev, NETDEV_UP, true); | |
2852 | } else { | |
2853 | netdev_turn_flags_on(port->netdev, NETDEV_UP, true); | |
2854 | } | |
2855 | } | |
2856 | #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD) | |
2857 | if (mask & REVALIDATE_BITS) { | |
2858 | COVERAGE_INC(ofproto_costly_flags); | |
2859 | port->opp.config ^= mask & REVALIDATE_BITS; | |
2860 | p->need_revalidate = true; | |
2861 | } | |
2862 | #undef REVALIDATE_BITS | |
2863 | if (mask & OFPPC_NO_FLOOD) { | |
2864 | port->opp.config ^= OFPPC_NO_FLOOD; | |
72b06300 | 2865 | refresh_port_groups(p); |
064af421 BP |
2866 | } |
2867 | if (mask & OFPPC_NO_PACKET_IN) { | |
2868 | port->opp.config ^= OFPPC_NO_PACKET_IN; | |
2869 | } | |
2870 | } | |
2871 | ||
2872 | static int | |
9deba63b BP |
2873 | handle_port_mod(struct ofproto *p, struct ofconn *ofconn, |
2874 | struct ofp_header *oh) | |
064af421 BP |
2875 | { |
2876 | const struct ofp_port_mod *opm; | |
2877 | struct ofport *port; | |
2878 | int error; | |
2879 | ||
9deba63b BP |
2880 | error = reject_slave_controller(ofconn, oh); |
2881 | if (error) { | |
2882 | return error; | |
2883 | } | |
064af421 BP |
2884 | error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm); |
2885 | if (error) { | |
2886 | return error; | |
2887 | } | |
2888 | opm = (struct ofp_port_mod *) oh; | |
2889 | ||
2890 | port = port_array_get(&p->ports, | |
2891 | ofp_port_to_odp_port(ntohs(opm->port_no))); | |
2892 | if (!port) { | |
2893 | return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT); | |
2894 | } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) { | |
2895 | return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR); | |
2896 | } else { | |
2897 | update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask)); | |
2898 | if (opm->advertise) { | |
2899 | netdev_set_advertisements(port->netdev, ntohl(opm->advertise)); | |
2900 | } | |
2901 | } | |
2902 | return 0; | |
2903 | } | |
2904 | ||
2905 | static struct ofpbuf * | |
2906 | make_stats_reply(uint32_t xid, uint16_t type, size_t body_len) | |
2907 | { | |
2908 | struct ofp_stats_reply *osr; | |
2909 | struct ofpbuf *msg; | |
2910 | ||
2911 | msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX)); | |
2912 | osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg); | |
2913 | osr->type = type; | |
2914 | osr->flags = htons(0); | |
2915 | return msg; | |
2916 | } | |
2917 | ||
2918 | static struct ofpbuf * | |
2919 | start_stats_reply(const struct ofp_stats_request *request, size_t body_len) | |
2920 | { | |
2921 | return make_stats_reply(request->header.xid, request->type, body_len); | |
2922 | } | |
2923 | ||
2924 | static void * | |
2925 | append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp) | |
2926 | { | |
2927 | struct ofpbuf *msg = *msgp; | |
2928 | assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply)); | |
2929 | if (nbytes + msg->size > UINT16_MAX) { | |
2930 | struct ofp_stats_reply *reply = msg->data; | |
2931 | reply->flags = htons(OFPSF_REPLY_MORE); | |
2932 | *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes); | |
2933 | queue_tx(msg, ofconn, ofconn->reply_counter); | |
2934 | } | |
2935 | return ofpbuf_put_uninit(*msgp, nbytes); | |
2936 | } | |
2937 | ||
2938 | static int | |
2939 | handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn, | |
2940 | struct ofp_stats_request *request) | |
2941 | { | |
2942 | struct ofp_desc_stats *ods; | |
2943 | struct ofpbuf *msg; | |
2944 | ||
2945 | msg = start_stats_reply(request, sizeof *ods); | |
2946 | ods = append_stats_reply(sizeof *ods, ofconn, &msg); | |
5a719c38 JP |
2947 | memset(ods, 0, sizeof *ods); |
2948 | ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc); | |
2949 | ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc); | |
2950 | ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc); | |
2951 | ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num); | |
2952 | ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc); | |
064af421 BP |
2953 | queue_tx(msg, ofconn, ofconn->reply_counter); |
2954 | ||
2955 | return 0; | |
2956 | } | |
2957 | ||
2958 | static void | |
2959 | count_subrules(struct cls_rule *cls_rule, void *n_subrules_) | |
2960 | { | |
2961 | struct rule *rule = rule_from_cls_rule(cls_rule); | |
2962 | int *n_subrules = n_subrules_; | |
2963 | ||
2964 | if (rule->super) { | |
2965 | (*n_subrules)++; | |
2966 | } | |
2967 | } | |
2968 | ||
2969 | static int | |
2970 | handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn, | |
2971 | struct ofp_stats_request *request) | |
2972 | { | |
2973 | struct ofp_table_stats *ots; | |
2974 | struct ofpbuf *msg; | |
2975 | struct odp_stats dpstats; | |
2976 | int n_exact, n_subrules, n_wild; | |
2977 | ||
2978 | msg = start_stats_reply(request, sizeof *ots * 2); | |
2979 | ||
2980 | /* Count rules of various kinds. */ | |
2981 | n_subrules = 0; | |
2982 | classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules); | |
2983 | n_exact = classifier_count_exact(&p->cls) - n_subrules; | |
2984 | n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls); | |
2985 | ||
2986 | /* Hash table. */ | |
c228a364 | 2987 | dpif_get_dp_stats(p->dpif, &dpstats); |
064af421 BP |
2988 | ots = append_stats_reply(sizeof *ots, ofconn, &msg); |
2989 | memset(ots, 0, sizeof *ots); | |
2990 | ots->table_id = TABLEID_HASH; | |
2991 | strcpy(ots->name, "hash"); | |
2992 | ots->wildcards = htonl(0); | |
2993 | ots->max_entries = htonl(dpstats.max_capacity); | |
2994 | ots->active_count = htonl(n_exact); | |
2995 | ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit + | |
2996 | dpstats.n_missed); | |
2997 | ots->matched_count = htonll(dpstats.n_hit); /* XXX */ | |
2998 | ||
2999 | /* Classifier table. */ | |
3000 | ots = append_stats_reply(sizeof *ots, ofconn, &msg); | |
3001 | memset(ots, 0, sizeof *ots); | |
3002 | ots->table_id = TABLEID_CLASSIFIER; | |
3003 | strcpy(ots->name, "classifier"); | |
659586ef JG |
3004 | ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL) |
3005 | : htonl(OFPFW_ALL); | |
064af421 BP |
3006 | ots->max_entries = htonl(65536); |
3007 | ots->active_count = htonl(n_wild); | |
3008 | ots->lookup_count = htonll(0); /* XXX */ | |
3009 | ots->matched_count = htonll(0); /* XXX */ | |
3010 | ||
3011 | queue_tx(msg, ofconn, ofconn->reply_counter); | |
3012 | return 0; | |
3013 | } | |
3014 | ||
abaad8cf JP |
3015 | static void |
3016 | append_port_stat(struct ofport *port, uint16_t port_no, struct ofconn *ofconn, | |
a4948b95 | 3017 | struct ofpbuf **msgp) |
abaad8cf JP |
3018 | { |
3019 | struct netdev_stats stats; | |
3020 | struct ofp_port_stats *ops; | |
3021 | ||
3022 | /* Intentionally ignore return value, since errors will set | |
3023 | * 'stats' to all-1s, which is correct for OpenFlow, and | |
3024 | * netdev_get_stats() will log errors. */ | |
3025 | netdev_get_stats(port->netdev, &stats); | |
3026 | ||
a4948b95 | 3027 | ops = append_stats_reply(sizeof *ops, ofconn, msgp); |
abaad8cf JP |
3028 | ops->port_no = htons(odp_port_to_ofp_port(port_no)); |
3029 | memset(ops->pad, 0, sizeof ops->pad); | |
3030 | ops->rx_packets = htonll(stats.rx_packets); | |
3031 | ops->tx_packets = htonll(stats.tx_packets); | |
3032 | ops->rx_bytes = htonll(stats.rx_bytes); | |
3033 | ops->tx_bytes = htonll(stats.tx_bytes); | |
3034 | ops->rx_dropped = htonll(stats.rx_dropped); | |
3035 | ops->tx_dropped = htonll(stats.tx_dropped); | |
3036 | ops->rx_errors = htonll(stats.rx_errors); | |
3037 | ops->tx_errors = htonll(stats.tx_errors); | |
3038 | ops->rx_frame_err = htonll(stats.rx_frame_errors); | |
3039 | ops->rx_over_err = htonll(stats.rx_over_errors); | |
3040 | ops->rx_crc_err = htonll(stats.rx_crc_errors); | |
3041 | ops->collisions = htonll(stats.collisions); | |
3042 | } | |
3043 | ||
064af421 BP |
3044 | static int |
3045 | handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn, | |
abaad8cf JP |
3046 | struct ofp_stats_request *osr, |
3047 | size_t arg_size) | |
064af421 | 3048 | { |
abaad8cf | 3049 | struct ofp_port_stats_request *psr; |
064af421 BP |
3050 | struct ofp_port_stats *ops; |
3051 | struct ofpbuf *msg; | |
3052 | struct ofport *port; | |
3053 | unsigned int port_no; | |
3054 | ||
abaad8cf JP |
3055 | if (arg_size != sizeof *psr) { |
3056 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
3057 | } | |
3058 | psr = (struct ofp_port_stats_request *) osr->body; | |
3059 | ||
3060 | msg = start_stats_reply(osr, sizeof *ops * 16); | |
3061 | if (psr->port_no != htons(OFPP_NONE)) { | |
3062 | port = port_array_get(&p->ports, | |
3063 | ofp_port_to_odp_port(ntohs(psr->port_no))); | |
3064 | if (port) { | |
a4948b95 | 3065 | append_port_stat(port, ntohs(psr->port_no), ofconn, &msg); |
abaad8cf JP |
3066 | } |
3067 | } else { | |
3068 | PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) { | |
a4948b95 | 3069 | append_port_stat(port, port_no, ofconn, &msg); |
abaad8cf | 3070 | } |
064af421 BP |
3071 | } |
3072 | ||
3073 | queue_tx(msg, ofconn, ofconn->reply_counter); | |
3074 | return 0; | |
3075 | } | |
3076 | ||
3077 | struct flow_stats_cbdata { | |
3078 | struct ofproto *ofproto; | |
3079 | struct ofconn *ofconn; | |
3080 | uint16_t out_port; | |
3081 | struct ofpbuf *msg; | |
3082 | }; | |
3083 | ||
01149cfd BP |
3084 | /* Obtains statistic counters for 'rule' within 'p' and stores them into |
3085 | * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the | |
3086 | * returned statistic include statistics for all of 'rule''s subrules. */ | |
064af421 BP |
3087 | static void |
3088 | query_stats(struct ofproto *p, struct rule *rule, | |
3089 | uint64_t *packet_countp, uint64_t *byte_countp) | |
3090 | { | |
3091 | uint64_t packet_count, byte_count; | |
3092 | struct rule *subrule; | |
3093 | struct odp_flow *odp_flows; | |
3094 | size_t n_odp_flows; | |
3095 | ||
01149cfd BP |
3096 | /* Start from historical data for 'rule' itself that are no longer tracked |
3097 | * by the datapath. This counts, for example, subrules that have | |
3098 | * expired. */ | |
b3137fe8 JG |
3099 | packet_count = rule->packet_count; |
3100 | byte_count = rule->byte_count; | |
3101 | ||
01149cfd BP |
3102 | /* Prepare to ask the datapath for statistics on 'rule', or if it is |
3103 | * wildcarded then on all of its subrules. | |
3104 | * | |
3105 | * Also, add any statistics that are not tracked by the datapath for each | |
3106 | * subrule. This includes, for example, statistics for packets that were | |
3107 | * executed "by hand" by ofproto via dpif_execute() but must be accounted | |
3108 | * to a flow. */ | |
064af421 | 3109 | n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1; |
ec6fde61 | 3110 | odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows); |
064af421 BP |
3111 | if (rule->cr.wc.wildcards) { |
3112 | size_t i = 0; | |
3113 | LIST_FOR_EACH (subrule, struct rule, list, &rule->list) { | |
3114 | odp_flows[i++].key = subrule->cr.flow; | |
b3137fe8 JG |
3115 | packet_count += subrule->packet_count; |
3116 | byte_count += subrule->byte_count; | |
064af421 BP |
3117 | } |
3118 | } else { | |
3119 | odp_flows[0].key = rule->cr.flow; | |
3120 | } | |
3121 | ||
28998b22 | 3122 | /* Fetch up-to-date statistics from the datapath and add them in. */ |
c228a364 | 3123 | if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) { |
064af421 BP |
3124 | size_t i; |
3125 | for (i = 0; i < n_odp_flows; i++) { | |
3126 | struct odp_flow *odp_flow = &odp_flows[i]; | |
3127 | packet_count += odp_flow->stats.n_packets; | |
3128 | byte_count += odp_flow->stats.n_bytes; | |
3129 | } | |
3130 | } | |
3131 | free(odp_flows); | |
3132 | ||
01149cfd | 3133 | /* Return the stats to the caller. */ |
064af421 BP |
3134 | *packet_countp = packet_count; |
3135 | *byte_countp = byte_count; | |
3136 | } | |
3137 | ||
3138 | static void | |
3139 | flow_stats_cb(struct cls_rule *rule_, void *cbdata_) | |
3140 | { | |
3141 | struct rule *rule = rule_from_cls_rule(rule_); | |
3142 | struct flow_stats_cbdata *cbdata = cbdata_; | |
3143 | struct ofp_flow_stats *ofs; | |
3144 | uint64_t packet_count, byte_count; | |
3145 | size_t act_len, len; | |
26c3f94a JP |
3146 | long long int tdiff = time_msec() - rule->created; |
3147 | uint32_t sec = tdiff / 1000; | |
3148 | uint32_t msec = tdiff - (sec * 1000); | |
064af421 BP |
3149 | |
3150 | if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) { | |
3151 | return; | |
3152 | } | |
3153 | ||
3154 | act_len = sizeof *rule->actions * rule->n_actions; | |
3155 | len = offsetof(struct ofp_flow_stats, actions) + act_len; | |
3156 | ||
3157 | query_stats(cbdata->ofproto, rule, &packet_count, &byte_count); | |
3158 | ||
3159 | ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg); | |
3160 | ofs->length = htons(len); | |
3161 | ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH; | |
3162 | ofs->pad = 0; | |
659586ef JG |
3163 | flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, |
3164 | cbdata->ofproto->tun_id_from_cookie, &ofs->match); | |
26c3f94a JP |
3165 | ofs->duration_sec = htonl(sec); |
3166 | ofs->duration_nsec = htonl(msec * 1000000); | |
39997502 | 3167 | ofs->cookie = rule->flow_cookie; |
064af421 BP |
3168 | ofs->priority = htons(rule->cr.priority); |
3169 | ofs->idle_timeout = htons(rule->idle_timeout); | |
3170 | ofs->hard_timeout = htons(rule->hard_timeout); | |
39997502 | 3171 | memset(ofs->pad2, 0, sizeof ofs->pad2); |
064af421 BP |
3172 | ofs->packet_count = htonll(packet_count); |
3173 | ofs->byte_count = htonll(byte_count); | |
3174 | memcpy(ofs->actions, rule->actions, act_len); | |
3175 | } | |
3176 | ||
3177 | static int | |
3178 | table_id_to_include(uint8_t table_id) | |
3179 | { | |
3180 | return (table_id == TABLEID_HASH ? CLS_INC_EXACT | |
3181 | : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD | |
3182 | : table_id == 0xff ? CLS_INC_ALL | |
3183 | : 0); | |
3184 | } | |
3185 | ||
3186 | static int | |
3187 | handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn, | |
3188 | const struct ofp_stats_request *osr, | |
3189 | size_t arg_size) | |
3190 | { | |
3191 | struct ofp_flow_stats_request *fsr; | |
3192 | struct flow_stats_cbdata cbdata; | |
3193 | struct cls_rule target; | |
3194 | ||
3195 | if (arg_size != sizeof *fsr) { | |
49bdc010 | 3196 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); |
064af421 BP |
3197 | } |
3198 | fsr = (struct ofp_flow_stats_request *) osr->body; | |
3199 | ||
3200 | COVERAGE_INC(ofproto_flows_req); | |
3201 | cbdata.ofproto = p; | |
3202 | cbdata.ofconn = ofconn; | |
3203 | cbdata.out_port = fsr->out_port; | |
3204 | cbdata.msg = start_stats_reply(osr, 1024); | |
659586ef | 3205 | cls_rule_from_match(&fsr->match, 0, false, 0, &target); |
064af421 BP |
3206 | classifier_for_each_match(&p->cls, &target, |
3207 | table_id_to_include(fsr->table_id), | |
3208 | flow_stats_cb, &cbdata); | |
3209 | queue_tx(cbdata.msg, ofconn, ofconn->reply_counter); | |
3210 | return 0; | |
3211 | } | |
3212 | ||
4f2cad2c JP |
3213 | struct flow_stats_ds_cbdata { |
3214 | struct ofproto *ofproto; | |
3215 | struct ds *results; | |
3216 | }; | |
3217 | ||
3218 | static void | |
3219 | flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_) | |
3220 | { | |
3221 | struct rule *rule = rule_from_cls_rule(rule_); | |
3222 | struct flow_stats_ds_cbdata *cbdata = cbdata_; | |
3223 | struct ds *results = cbdata->results; | |
3224 | struct ofp_match match; | |
3225 | uint64_t packet_count, byte_count; | |
3226 | size_t act_len = sizeof *rule->actions * rule->n_actions; | |
3227 | ||
3228 | /* Don't report on subrules. */ | |
3229 | if (rule->super != NULL) { | |
3230 | return; | |
3231 | } | |
3232 | ||
3233 | query_stats(cbdata->ofproto, rule, &packet_count, &byte_count); | |
659586ef JG |
3234 | flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, |
3235 | cbdata->ofproto->tun_id_from_cookie, &match); | |
4f2cad2c JP |
3236 | |
3237 | ds_put_format(results, "duration=%llds, ", | |
3238 | (time_msec() - rule->created) / 1000); | |
52ae00b3 | 3239 | ds_put_format(results, "priority=%u, ", rule->cr.priority); |
4f2cad2c JP |
3240 | ds_put_format(results, "n_packets=%"PRIu64", ", packet_count); |
3241 | ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count); | |
3242 | ofp_print_match(results, &match, true); | |
3243 | ofp_print_actions(results, &rule->actions->header, act_len); | |
3244 | ds_put_cstr(results, "\n"); | |
3245 | } | |
3246 | ||
3247 | /* Adds a pretty-printed description of all flows to 'results', including | |
3248 | * those marked hidden by secchan (e.g., by in-band control). */ | |
3249 | void | |
3250 | ofproto_get_all_flows(struct ofproto *p, struct ds *results) | |
3251 | { | |
3252 | struct ofp_match match; | |
3253 | struct cls_rule target; | |
3254 | struct flow_stats_ds_cbdata cbdata; | |
3255 | ||
3256 | memset(&match, 0, sizeof match); | |
659586ef | 3257 | match.wildcards = htonl(OVSFW_ALL); |
4f2cad2c JP |
3258 | |
3259 | cbdata.ofproto = p; | |
3260 | cbdata.results = results; | |
3261 | ||
659586ef | 3262 | cls_rule_from_match(&match, 0, false, 0, &target); |
4f2cad2c JP |
3263 | classifier_for_each_match(&p->cls, &target, CLS_INC_ALL, |
3264 | flow_stats_ds_cb, &cbdata); | |
3265 | } | |
3266 | ||
064af421 BP |
3267 | struct aggregate_stats_cbdata { |
3268 | struct ofproto *ofproto; | |
3269 | uint16_t out_port; | |
3270 | uint64_t packet_count; | |
3271 | uint64_t byte_count; | |
3272 | uint32_t n_flows; | |
3273 | }; | |
3274 | ||
3275 | static void | |
3276 | aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_) | |
3277 | { | |
3278 | struct rule *rule = rule_from_cls_rule(rule_); | |
3279 | struct aggregate_stats_cbdata *cbdata = cbdata_; | |
3280 | uint64_t packet_count, byte_count; | |
3281 | ||
3282 | if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) { | |
3283 | return; | |
3284 | } | |
3285 | ||
3286 | query_stats(cbdata->ofproto, rule, &packet_count, &byte_count); | |
3287 | ||
3288 | cbdata->packet_count += packet_count; | |
3289 | cbdata->byte_count += byte_count; | |
3290 | cbdata->n_flows++; | |
3291 | } | |
3292 | ||
3293 | static int | |
3294 | handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn, | |
3295 | const struct ofp_stats_request *osr, | |
3296 | size_t arg_size) | |
3297 | { | |
3298 | struct ofp_aggregate_stats_request *asr; | |
3299 | struct ofp_aggregate_stats_reply *reply; | |
3300 | struct aggregate_stats_cbdata cbdata; | |
3301 | struct cls_rule target; | |
3302 | struct ofpbuf *msg; | |
3303 | ||
3304 | if (arg_size != sizeof *asr) { | |
49bdc010 | 3305 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); |
064af421 BP |
3306 | } |
3307 | asr = (struct ofp_aggregate_stats_request *) osr->body; | |
3308 | ||
3309 | COVERAGE_INC(ofproto_agg_request); | |
3310 | cbdata.ofproto = p; | |
3311 | cbdata.out_port = asr->out_port; | |
3312 | cbdata.packet_count = 0; | |
3313 | cbdata.byte_count = 0; | |
3314 | cbdata.n_flows = 0; | |
659586ef | 3315 | cls_rule_from_match(&asr->match, 0, false, 0, &target); |
064af421 BP |
3316 | classifier_for_each_match(&p->cls, &target, |
3317 | table_id_to_include(asr->table_id), | |
3318 | aggregate_stats_cb, &cbdata); | |
3319 | ||
3320 | msg = start_stats_reply(osr, sizeof *reply); | |
3321 | reply = append_stats_reply(sizeof *reply, ofconn, &msg); | |
3322 | reply->flow_count = htonl(cbdata.n_flows); | |
3323 | reply->packet_count = htonll(cbdata.packet_count); | |
3324 | reply->byte_count = htonll(cbdata.byte_count); | |
3325 | queue_tx(msg, ofconn, ofconn->reply_counter); | |
3326 | return 0; | |
3327 | } | |
3328 | ||
c1c9c9c4 BP |
3329 | struct queue_stats_cbdata { |
3330 | struct ofconn *ofconn; | |
3331 | struct ofpbuf *msg; | |
3332 | uint16_t port_no; | |
3333 | }; | |
3334 | ||
3335 | static void | |
db9220c3 | 3336 | put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id, |
c1c9c9c4 BP |
3337 | const struct netdev_queue_stats *stats) |
3338 | { | |
3339 | struct ofp_queue_stats *reply; | |
3340 | ||
3341 | reply = append_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg); | |
3342 | reply->port_no = htons(cbdata->port_no); | |
3343 | memset(reply->pad, 0, sizeof reply->pad); | |
3344 | reply->queue_id = htonl(queue_id); | |
3345 | reply->tx_bytes = htonll(stats->tx_bytes); | |
3346 | reply->tx_packets = htonll(stats->tx_packets); | |
3347 | reply->tx_errors = htonll(stats->tx_errors); | |
3348 | } | |
3349 | ||
3350 | static void | |
db9220c3 | 3351 | handle_queue_stats_dump_cb(uint32_t queue_id, |
c1c9c9c4 BP |
3352 | struct netdev_queue_stats *stats, |
3353 | void *cbdata_) | |
3354 | { | |
3355 | struct queue_stats_cbdata *cbdata = cbdata_; | |
3356 | ||
3357 | put_queue_stats(cbdata, queue_id, stats); | |
3358 | } | |
3359 | ||
3360 | static void | |
3361 | handle_queue_stats_for_port(struct ofport *port, uint16_t port_no, | |
db9220c3 | 3362 | uint32_t queue_id, |
c1c9c9c4 BP |
3363 | struct queue_stats_cbdata *cbdata) |
3364 | { | |
3365 | cbdata->port_no = port_no; | |
3366 | if (queue_id == OFPQ_ALL) { | |
3367 | netdev_dump_queue_stats(port->netdev, | |
3368 | handle_queue_stats_dump_cb, cbdata); | |
3369 | } else { | |
3370 | struct netdev_queue_stats stats; | |
3371 | ||
3372 | netdev_get_queue_stats(port->netdev, queue_id, &stats); | |
3373 | put_queue_stats(cbdata, queue_id, &stats); | |
3374 | } | |
3375 | } | |
3376 | ||
3377 | static int | |
3378 | handle_queue_stats_request(struct ofproto *ofproto, struct ofconn *ofconn, | |
3379 | const struct ofp_stats_request *osr, | |
3380 | size_t arg_size) | |
3381 | { | |
3382 | struct ofp_queue_stats_request *qsr; | |
3383 | struct queue_stats_cbdata cbdata; | |
3384 | struct ofport *port; | |
3385 | unsigned int port_no; | |
3386 | uint32_t queue_id; | |
3387 | ||
3388 | if (arg_size != sizeof *qsr) { | |
3389 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
3390 | } | |
3391 | qsr = (struct ofp_queue_stats_request *) osr->body; | |
3392 | ||
3393 | COVERAGE_INC(ofproto_queue_req); | |
3394 | ||
3395 | cbdata.ofconn = ofconn; | |
3396 | cbdata.msg = start_stats_reply(osr, 128); | |
3397 | ||
3398 | port_no = ntohs(qsr->port_no); | |
3399 | queue_id = ntohl(qsr->queue_id); | |
3400 | if (port_no == OFPP_ALL) { | |
3401 | PORT_ARRAY_FOR_EACH (port, &ofproto->ports, port_no) { | |
3402 | handle_queue_stats_for_port(port, port_no, queue_id, &cbdata); | |
3403 | } | |
3404 | } else if (port_no < ofproto->max_ports) { | |
3405 | port = port_array_get(&ofproto->ports, port_no); | |
3406 | if (port) { | |
3407 | handle_queue_stats_for_port(port, port_no, queue_id, &cbdata); | |
3408 | } | |
3409 | } else { | |
3410 | ofpbuf_delete(cbdata.msg); | |
3411 | return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT); | |
3412 | } | |
3413 | queue_tx(cbdata.msg, ofconn, ofconn->reply_counter); | |
3414 | ||
3415 | return 0; | |
3416 | } | |
3417 | ||
064af421 BP |
3418 | static int |
3419 | handle_stats_request(struct ofproto *p, struct ofconn *ofconn, | |
3420 | struct ofp_header *oh) | |
3421 | { | |
3422 | struct ofp_stats_request *osr; | |
3423 | size_t arg_size; | |
3424 | int error; | |
3425 | ||
3426 | error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr, | |
3427 | 1, &arg_size); | |
3428 | if (error) { | |
3429 | return error; | |
3430 | } | |
3431 | osr = (struct ofp_stats_request *) oh; | |
3432 | ||
3433 | switch (ntohs(osr->type)) { | |
3434 | case OFPST_DESC: | |
3435 | return handle_desc_stats_request(p, ofconn, osr); | |
3436 | ||
3437 | case OFPST_FLOW: | |
3438 | return handle_flow_stats_request(p, ofconn, osr, arg_size); | |
3439 | ||
3440 | case OFPST_AGGREGATE: | |
3441 | return handle_aggregate_stats_request(p, ofconn, osr, arg_size); | |
3442 | ||
3443 | case OFPST_TABLE: | |
3444 | return handle_table_stats_request(p, ofconn, osr); | |
3445 | ||
3446 | case OFPST_PORT: | |
abaad8cf | 3447 | return handle_port_stats_request(p, ofconn, osr, arg_size); |
064af421 | 3448 | |
c1c9c9c4 BP |
3449 | case OFPST_QUEUE: |
3450 | return handle_queue_stats_request(p, ofconn, osr, arg_size); | |
3451 | ||
064af421 BP |
3452 | case OFPST_VENDOR: |
3453 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR); | |
3454 | ||
3455 | default: | |
3456 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT); | |
3457 | } | |
3458 | } | |
3459 | ||
3460 | static long long int | |
3461 | msec_from_nsec(uint64_t sec, uint32_t nsec) | |
3462 | { | |
3463 | return !sec ? 0 : sec * 1000 + nsec / 1000000; | |
3464 | } | |
3465 | ||
3466 | static void | |
0193b2af JG |
3467 | update_time(struct ofproto *ofproto, struct rule *rule, |
3468 | const struct odp_flow_stats *stats) | |
064af421 BP |
3469 | { |
3470 | long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec); | |
3471 | if (used > rule->used) { | |
3472 | rule->used = used; | |
4836f9f2 JP |
3473 | if (rule->super && used > rule->super->used) { |
3474 | rule->super->used = used; | |
3475 | } | |
0193b2af | 3476 | netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used); |
064af421 BP |
3477 | } |
3478 | } | |
3479 | ||
3480 | static void | |
0193b2af JG |
3481 | update_stats(struct ofproto *ofproto, struct rule *rule, |
3482 | const struct odp_flow_stats *stats) | |
064af421 | 3483 | { |
064af421 | 3484 | if (stats->n_packets) { |
0193b2af JG |
3485 | update_time(ofproto, rule, stats); |
3486 | rule->packet_count += stats->n_packets; | |
3487 | rule->byte_count += stats->n_bytes; | |
abfec865 | 3488 | netflow_flow_update_flags(&rule->nf_flow, stats->tcp_flags); |
064af421 BP |
3489 | } |
3490 | } | |
3491 | ||
79eee1eb BP |
3492 | /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT |
3493 | * in which no matching flow already exists in the flow table. | |
3494 | * | |
3495 | * Adds the flow specified by 'ofm', which is followed by 'n_actions' | |
3496 | * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error | |
3497 | * code as encoded by ofp_mkerr() on failure. | |
3498 | * | |
3499 | * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, | |
3500 | * if any. */ | |
064af421 BP |
3501 | static int |
3502 | add_flow(struct ofproto *p, struct ofconn *ofconn, | |
79eee1eb | 3503 | const struct ofp_flow_mod *ofm, size_t n_actions) |
064af421 BP |
3504 | { |
3505 | struct ofpbuf *packet; | |
3506 | struct rule *rule; | |
3507 | uint16_t in_port; | |
3508 | int error; | |
3509 | ||
49bdc010 JP |
3510 | if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) { |
3511 | flow_t flow; | |
3512 | uint32_t wildcards; | |
3513 | ||
659586ef JG |
3514 | flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie, |
3515 | &flow, &wildcards); | |
49bdc010 JP |
3516 | if (classifier_rule_overlaps(&p->cls, &flow, wildcards, |
3517 | ntohs(ofm->priority))) { | |
3518 | return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP); | |
3519 | } | |
3520 | } | |
3521 | ||
0193b2af | 3522 | rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions, |
064af421 | 3523 | n_actions, ntohs(ofm->idle_timeout), |
39997502 | 3524 | ntohs(ofm->hard_timeout), ofm->cookie, |
ca069229 | 3525 | ofm->flags & htons(OFPFF_SEND_FLOW_REM)); |
659586ef JG |
3526 | cls_rule_from_match(&ofm->match, ntohs(ofm->priority), |
3527 | p->tun_id_from_cookie, ofm->cookie, &rule->cr); | |
064af421 | 3528 | |
064af421 BP |
3529 | error = 0; |
3530 | if (ofm->buffer_id != htonl(UINT32_MAX)) { | |
3531 | error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id), | |
3532 | &packet, &in_port); | |
212fe71c BP |
3533 | } else { |
3534 | packet = NULL; | |
165cd8a3 | 3535 | in_port = UINT16_MAX; |
064af421 BP |
3536 | } |
3537 | ||
3538 | rule_insert(p, rule, packet, in_port); | |
3539 | ofpbuf_delete(packet); | |
3540 | return error; | |
3541 | } | |
3542 | ||
79eee1eb BP |
3543 | static struct rule * |
3544 | find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm) | |
064af421 | 3545 | { |
064af421 BP |
3546 | uint32_t wildcards; |
3547 | flow_t flow; | |
3548 | ||
659586ef JG |
3549 | flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie, |
3550 | &flow, &wildcards); | |
79eee1eb | 3551 | return rule_from_cls_rule(classifier_find_rule_exactly( |
064af421 BP |
3552 | &p->cls, &flow, wildcards, |
3553 | ntohs(ofm->priority))); | |
79eee1eb | 3554 | } |
064af421 | 3555 | |
79eee1eb BP |
3556 | static int |
3557 | send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn, | |
3558 | struct rule *rule, const struct ofp_flow_mod *ofm) | |
3559 | { | |
3560 | struct ofpbuf *packet; | |
3561 | uint16_t in_port; | |
3562 | flow_t flow; | |
3563 | int error; | |
064af421 | 3564 | |
79eee1eb BP |
3565 | if (ofm->buffer_id == htonl(UINT32_MAX)) { |
3566 | return 0; | |
064af421 | 3567 | } |
79eee1eb BP |
3568 | |
3569 | error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id), | |
3570 | &packet, &in_port); | |
3571 | if (error) { | |
3572 | return error; | |
3573 | } | |
3574 | ||
659586ef | 3575 | flow_extract(packet, 0, in_port, &flow); |
79eee1eb BP |
3576 | rule_execute(ofproto, rule, packet, &flow); |
3577 | ofpbuf_delete(packet); | |
3578 | ||
064af421 BP |
3579 | return 0; |
3580 | } | |
79eee1eb BP |
3581 | \f |
3582 | /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */ | |
064af421 BP |
3583 | |
3584 | struct modify_flows_cbdata { | |
3585 | struct ofproto *ofproto; | |
3586 | const struct ofp_flow_mod *ofm; | |
064af421 | 3587 | size_t n_actions; |
79eee1eb | 3588 | struct rule *match; |
064af421 BP |
3589 | }; |
3590 | ||
79eee1eb BP |
3591 | static int modify_flow(struct ofproto *, const struct ofp_flow_mod *, |
3592 | size_t n_actions, struct rule *); | |
3593 | static void modify_flows_cb(struct cls_rule *, void *cbdata_); | |
3594 | ||
3595 | /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as | |
3596 | * encoded by ofp_mkerr() on failure. | |
3597 | * | |
3598 | * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, | |
3599 | * if any. */ | |
3600 | static int | |
3601 | modify_flows_loose(struct ofproto *p, struct ofconn *ofconn, | |
3602 | const struct ofp_flow_mod *ofm, size_t n_actions) | |
3603 | { | |
3604 | struct modify_flows_cbdata cbdata; | |
3605 | struct cls_rule target; | |
3606 | ||
3607 | cbdata.ofproto = p; | |
3608 | cbdata.ofm = ofm; | |
3609 | cbdata.n_actions = n_actions; | |
3610 | cbdata.match = NULL; | |
3611 | ||
659586ef JG |
3612 | cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie, |
3613 | &target); | |
79eee1eb BP |
3614 | |
3615 | classifier_for_each_match(&p->cls, &target, CLS_INC_ALL, | |
3616 | modify_flows_cb, &cbdata); | |
3617 | if (cbdata.match) { | |
3618 | /* This credits the packet to whichever flow happened to happened to | |
3619 | * match last. That's weird. Maybe we should do a lookup for the | |
3620 | * flow that actually matches the packet? Who knows. */ | |
3621 | send_buffered_packet(p, ofconn, cbdata.match, ofm); | |
3622 | return 0; | |
3623 | } else { | |
3624 | return add_flow(p, ofconn, ofm, n_actions); | |
3625 | } | |
3626 | } | |
3627 | ||
3628 | /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error | |
3629 | * code as encoded by ofp_mkerr() on failure. | |
3630 | * | |
3631 | * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, | |
3632 | * if any. */ | |
3633 | static int | |
3634 | modify_flow_strict(struct ofproto *p, struct ofconn *ofconn, | |
3635 | struct ofp_flow_mod *ofm, size_t n_actions) | |
3636 | { | |
3637 | struct rule *rule = find_flow_strict(p, ofm); | |
3638 | if (rule && !rule_is_hidden(rule)) { | |
3639 | modify_flow(p, ofm, n_actions, rule); | |
3640 | return send_buffered_packet(p, ofconn, rule, ofm); | |
3641 | } else { | |
3642 | return add_flow(p, ofconn, ofm, n_actions); | |
3643 | } | |
3644 | } | |
3645 | ||
3646 | /* Callback for modify_flows_loose(). */ | |
064af421 BP |
3647 | static void |
3648 | modify_flows_cb(struct cls_rule *rule_, void *cbdata_) | |
3649 | { | |
3650 | struct rule *rule = rule_from_cls_rule(rule_); | |
3651 | struct modify_flows_cbdata *cbdata = cbdata_; | |
3652 | ||
79eee1eb BP |
3653 | if (!rule_is_hidden(rule)) { |
3654 | cbdata->match = rule; | |
3655 | modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule); | |
064af421 | 3656 | } |
064af421 BP |
3657 | } |
3658 | ||
79eee1eb BP |
3659 | /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has |
3660 | * been identified as a flow in 'p''s flow table to be modified, by changing | |
3661 | * the rule's actions to match those in 'ofm' (which is followed by 'n_actions' | |
3662 | * ofp_action[] structures). */ | |
064af421 | 3663 | static int |
79eee1eb BP |
3664 | modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm, |
3665 | size_t n_actions, struct rule *rule) | |
064af421 | 3666 | { |
79eee1eb BP |
3667 | size_t actions_len = n_actions * sizeof *rule->actions; |
3668 | ||
3669 | rule->flow_cookie = ofm->cookie; | |
3670 | ||
3671 | /* If the actions are the same, do nothing. */ | |
3672 | if (n_actions == rule->n_actions | |
3673 | && !memcmp(ofm->actions, rule->actions, actions_len)) | |
3674 | { | |
3675 | return 0; | |
3676 | } | |
3677 | ||
3678 | /* Replace actions. */ | |
3679 | free(rule->actions); | |
3680 | rule->actions = xmemdup(ofm->actions, actions_len); | |
3681 | rule->n_actions = n_actions; | |
3682 | ||
3683 | /* Make sure that the datapath gets updated properly. */ | |
3684 | if (rule->cr.wc.wildcards) { | |
3685 | COVERAGE_INC(ofproto_mod_wc_flow); | |
3686 | p->need_revalidate = true; | |
3687 | } else { | |
3688 | rule_update_actions(p, rule); | |
3689 | } | |
3690 | ||
3691 | return 0; | |
3692 | } | |
3693 | \f | |
3694 | /* OFPFC_DELETE implementation. */ | |
3695 | ||
3696 | struct delete_flows_cbdata { | |
3697 | struct ofproto *ofproto; | |
3698 | uint16_t out_port; | |
3699 | }; | |
3700 | ||
3701 | static void delete_flows_cb(struct cls_rule *, void *cbdata_); | |
3702 | static void delete_flow(struct ofproto *, struct rule *, uint16_t out_port); | |
3703 | ||
3704 | /* Implements OFPFC_DELETE. */ | |
3705 | static void | |
3706 | delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm) | |
3707 | { | |
3708 | struct delete_flows_cbdata cbdata; | |
064af421 BP |
3709 | struct cls_rule target; |
3710 | ||
3711 | cbdata.ofproto = p; | |
79eee1eb | 3712 | cbdata.out_port = ofm->out_port; |
064af421 | 3713 | |
659586ef JG |
3714 | cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie, |
3715 | &target); | |
064af421 BP |
3716 | |
3717 | classifier_for_each_match(&p->cls, &target, CLS_INC_ALL, | |
79eee1eb | 3718 | delete_flows_cb, &cbdata); |
064af421 BP |
3719 | } |
3720 | ||
79eee1eb BP |
3721 | /* Implements OFPFC_DELETE_STRICT. */ |
3722 | static void | |
3723 | delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm) | |
3724 | { | |
3725 | struct rule *rule = find_flow_strict(p, ofm); | |
3726 | if (rule) { | |
3727 | delete_flow(p, rule, ofm->out_port); | |
3728 | } | |
3729 | } | |
3730 | ||
3731 | /* Callback for delete_flows_loose(). */ | |
3732 | static void | |
3733 | delete_flows_cb(struct cls_rule *rule_, void *cbdata_) | |
3734 | { | |
3735 | struct rule *rule = rule_from_cls_rule(rule_); | |
3736 | struct delete_flows_cbdata *cbdata = cbdata_; | |
3737 | ||
3738 | delete_flow(cbdata->ofproto, rule, cbdata->out_port); | |
3739 | } | |
3740 | ||
3741 | /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has | |
3742 | * been identified as a flow to delete from 'p''s flow table, by deleting the | |
3743 | * flow and sending out a OFPT_FLOW_REMOVED message to any interested | |
3744 | * controller. | |
3745 | * | |
3746 | * Will not delete 'rule' if it is hidden. Will delete 'rule' only if | |
3747 | * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the | |
3748 | * specified 'out_port'. */ | |
3749 | static void | |
3750 | delete_flow(struct ofproto *p, struct rule *rule, uint16_t out_port) | |
3751 | { | |
3752 | if (rule_is_hidden(rule)) { | |
3753 | return; | |
3754 | } | |
3755 | ||
3756 | if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) { | |
3757 | return; | |
3758 | } | |
3759 | ||
3760 | send_flow_removed(p, rule, time_msec(), OFPRR_DELETE); | |
3761 | rule_remove(p, rule); | |
3762 | } | |
3763 | \f | |
064af421 BP |
3764 | static int |
3765 | handle_flow_mod(struct ofproto *p, struct ofconn *ofconn, | |
3766 | struct ofp_flow_mod *ofm) | |
3767 | { | |
3f09c339 | 3768 | struct ofp_match orig_match; |
064af421 BP |
3769 | size_t n_actions; |
3770 | int error; | |
3771 | ||
9deba63b BP |
3772 | error = reject_slave_controller(ofconn, &ofm->header); |
3773 | if (error) { | |
3774 | return error; | |
3775 | } | |
064af421 BP |
3776 | error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm, |
3777 | sizeof *ofm->actions, &n_actions); | |
3778 | if (error) { | |
3779 | return error; | |
3780 | } | |
3781 | ||
49bdc010 JP |
3782 | /* We do not support the emergency flow cache. It will hopefully |
3783 | * get dropped from OpenFlow in the near future. */ | |
3784 | if (ofm->flags & htons(OFPFF_EMERG)) { | |
3785 | /* There isn't a good fit for an error code, so just state that the | |
3786 | * flow table is full. */ | |
3787 | return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL); | |
3788 | } | |
3789 | ||
3f09c339 BP |
3790 | /* Normalize ofp->match. If normalization actually changes anything, then |
3791 | * log the differences. */ | |
3792 | ofm->match.pad1[0] = ofm->match.pad2[0] = 0; | |
3793 | orig_match = ofm->match; | |
064af421 | 3794 | normalize_match(&ofm->match); |
3f09c339 BP |
3795 | if (memcmp(&ofm->match, &orig_match, sizeof orig_match)) { |
3796 | static struct vlog_rate_limit normal_rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
3797 | if (!VLOG_DROP_INFO(&normal_rl)) { | |
3798 | char *old = ofp_match_to_literal_string(&orig_match); | |
3799 | char *new = ofp_match_to_literal_string(&ofm->match); | |
3800 | VLOG_INFO("%s: normalization changed ofp_match, details:", | |
3801 | rconn_get_name(ofconn->rconn)); | |
3802 | VLOG_INFO(" pre: %s", old); | |
3803 | VLOG_INFO("post: %s", new); | |
3804 | free(old); | |
3805 | free(new); | |
3806 | } | |
3807 | } | |
3808 | ||
064af421 BP |
3809 | if (!ofm->match.wildcards) { |
3810 | ofm->priority = htons(UINT16_MAX); | |
3811 | } | |
3812 | ||
3813 | error = validate_actions((const union ofp_action *) ofm->actions, | |
3814 | n_actions, p->max_ports); | |
3815 | if (error) { | |
3816 | return error; | |
3817 | } | |
3818 | ||
3819 | switch (ntohs(ofm->command)) { | |
3820 | case OFPFC_ADD: | |
3821 | return add_flow(p, ofconn, ofm, n_actions); | |
3822 | ||
3823 | case OFPFC_MODIFY: | |
79eee1eb | 3824 | return modify_flows_loose(p, ofconn, ofm, n_actions); |
064af421 BP |
3825 | |
3826 | case OFPFC_MODIFY_STRICT: | |
79eee1eb | 3827 | return modify_flow_strict(p, ofconn, ofm, n_actions); |
064af421 BP |
3828 | |
3829 | case OFPFC_DELETE: | |
79eee1eb BP |
3830 | delete_flows_loose(p, ofm); |
3831 | return 0; | |
064af421 BP |
3832 | |
3833 | case OFPFC_DELETE_STRICT: | |
79eee1eb BP |
3834 | delete_flow_strict(p, ofm); |
3835 | return 0; | |
064af421 BP |
3836 | |
3837 | default: | |
3838 | return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND); | |
3839 | } | |
3840 | } | |
3841 | ||
659586ef JG |
3842 | static int |
3843 | handle_tun_id_from_cookie(struct ofproto *p, struct nxt_tun_id_cookie *msg) | |
3844 | { | |
3845 | int error; | |
3846 | ||
3847 | error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg); | |
3848 | if (error) { | |
3849 | return error; | |
3850 | } | |
3851 | ||
3852 | p->tun_id_from_cookie = !!msg->set; | |
3853 | return 0; | |
3854 | } | |
3855 | ||
9deba63b BP |
3856 | static int |
3857 | handle_role_request(struct ofproto *ofproto, | |
3858 | struct ofconn *ofconn, struct nicira_header *msg) | |
3859 | { | |
3860 | struct nx_role_request *nrr; | |
3861 | struct nx_role_request *reply; | |
3862 | struct ofpbuf *buf; | |
3863 | uint32_t role; | |
3864 | ||
3865 | if (ntohs(msg->header.length) != sizeof *nrr) { | |
100e95db | 3866 | VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)", |
9deba63b BP |
3867 | ntohs(msg->header.length), sizeof *nrr); |
3868 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
3869 | } | |
3870 | nrr = (struct nx_role_request *) msg; | |
3871 | ||
5899143f | 3872 | if (ofconn->type != OFCONN_PRIMARY) { |
9deba63b BP |
3873 | VLOG_WARN_RL(&rl, "ignoring role request on non-controller " |
3874 | "connection"); | |
3875 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM); | |
3876 | } | |
3877 | ||
3878 | role = ntohl(nrr->role); | |
3879 | if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER | |
3880 | && role != NX_ROLE_SLAVE) { | |
3881 | VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role); | |
3882 | ||
3883 | /* There's no good error code for this. */ | |
3884 | return ofp_mkerr(OFPET_BAD_REQUEST, -1); | |
3885 | } | |
3886 | ||
3887 | if (role == NX_ROLE_MASTER) { | |
3888 | struct ofconn *other; | |
3889 | ||
3890 | HMAP_FOR_EACH (other, struct ofconn, hmap_node, | |
3891 | &ofproto->controllers) { | |
3892 | if (other->role == NX_ROLE_MASTER) { | |
3893 | other->role = NX_ROLE_SLAVE; | |
3894 | } | |
3895 | } | |
3896 | } | |
3897 | ofconn->role = role; | |
3898 | ||
3899 | reply = make_openflow_xid(sizeof *reply, OFPT_VENDOR, msg->header.xid, | |
3900 | &buf); | |
3901 | reply->nxh.vendor = htonl(NX_VENDOR_ID); | |
3902 | reply->nxh.subtype = htonl(NXT_ROLE_REPLY); | |
3903 | reply->role = htonl(role); | |
3904 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
3905 | ||
3906 | return 0; | |
3907 | } | |
3908 | ||
064af421 BP |
3909 | static int |
3910 | handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg) | |
3911 | { | |
3912 | struct ofp_vendor_header *ovh = msg; | |
3913 | struct nicira_header *nh; | |
3914 | ||
3915 | if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) { | |
100e95db | 3916 | VLOG_WARN_RL(&rl, "received vendor message of length %u " |
659586ef JG |
3917 | "(expected at least %zu)", |
3918 | ntohs(ovh->header.length), sizeof(struct ofp_vendor_header)); | |
49bdc010 | 3919 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); |
064af421 BP |
3920 | } |
3921 | if (ovh->vendor != htonl(NX_VENDOR_ID)) { | |
3922 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR); | |
3923 | } | |
3924 | if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) { | |
100e95db | 3925 | VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u " |
659586ef JG |
3926 | "(expected at least %zu)", |
3927 | ntohs(ovh->header.length), sizeof(struct nicira_header)); | |
49bdc010 | 3928 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); |
064af421 BP |
3929 | } |
3930 | ||
3931 | nh = msg; | |
3932 | switch (ntohl(nh->subtype)) { | |
3933 | case NXT_STATUS_REQUEST: | |
3934 | return switch_status_handle_request(p->switch_status, ofconn->rconn, | |
3935 | msg); | |
659586ef JG |
3936 | |
3937 | case NXT_TUN_ID_FROM_COOKIE: | |
3938 | return handle_tun_id_from_cookie(p, msg); | |
9deba63b BP |
3939 | |
3940 | case NXT_ROLE_REQUEST: | |
3941 | return handle_role_request(p, ofconn, msg); | |
064af421 BP |
3942 | } |
3943 | ||
3944 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); | |
3945 | } | |
3946 | ||
246e61ea JP |
3947 | static int |
3948 | handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh) | |
3949 | { | |
3950 | struct ofp_header *ob; | |
3951 | struct ofpbuf *buf; | |
3952 | ||
3953 | /* Currently, everything executes synchronously, so we can just | |
3954 | * immediately send the barrier reply. */ | |
3955 | ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf); | |
3956 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
3957 | return 0; | |
3958 | } | |
3959 | ||
064af421 BP |
3960 | static void |
3961 | handle_openflow(struct ofconn *ofconn, struct ofproto *p, | |
3962 | struct ofpbuf *ofp_msg) | |
3963 | { | |
3964 | struct ofp_header *oh = ofp_msg->data; | |
3965 | int error; | |
3966 | ||
3967 | COVERAGE_INC(ofproto_recv_openflow); | |
3968 | switch (oh->type) { | |
3969 | case OFPT_ECHO_REQUEST: | |
3970 | error = handle_echo_request(ofconn, oh); | |
3971 | break; | |
3972 | ||
3973 | case OFPT_ECHO_REPLY: | |
3974 | error = 0; | |
3975 | break; | |
3976 | ||
3977 | case OFPT_FEATURES_REQUEST: | |
3978 | error = handle_features_request(p, ofconn, oh); | |
3979 | break; | |
3980 | ||
3981 | case OFPT_GET_CONFIG_REQUEST: | |
3982 | error = handle_get_config_request(p, ofconn, oh); | |
3983 | break; | |
3984 | ||
3985 | case OFPT_SET_CONFIG: | |
3986 | error = handle_set_config(p, ofconn, ofp_msg->data); | |
3987 | break; | |
3988 | ||
3989 | case OFPT_PACKET_OUT: | |
3990 | error = handle_packet_out(p, ofconn, ofp_msg->data); | |
3991 | break; | |
3992 | ||
3993 | case OFPT_PORT_MOD: | |
9deba63b | 3994 | error = handle_port_mod(p, ofconn, oh); |
064af421 BP |
3995 | break; |
3996 | ||
3997 | case OFPT_FLOW_MOD: | |
3998 | error = handle_flow_mod(p, ofconn, ofp_msg->data); | |
3999 | break; | |
4000 | ||
4001 | case OFPT_STATS_REQUEST: | |
4002 | error = handle_stats_request(p, ofconn, oh); | |
4003 | break; | |
4004 | ||
4005 | case OFPT_VENDOR: | |
4006 | error = handle_vendor(p, ofconn, ofp_msg->data); | |
4007 | break; | |
4008 | ||
246e61ea JP |
4009 | case OFPT_BARRIER_REQUEST: |
4010 | error = handle_barrier_request(ofconn, oh); | |
4011 | break; | |
4012 | ||
064af421 BP |
4013 | default: |
4014 | if (VLOG_IS_WARN_ENABLED()) { | |
4015 | char *s = ofp_to_string(oh, ntohs(oh->length), 2); | |
4016 | VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s); | |
4017 | free(s); | |
4018 | } | |
4019 | error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE); | |
4020 | break; | |
4021 | } | |
4022 | ||
4023 | if (error) { | |
4024 | send_error_oh(ofconn, ofp_msg->data, error); | |
4025 | } | |
4026 | } | |
4027 | \f | |
4028 | static void | |
72b06300 | 4029 | handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet) |
064af421 BP |
4030 | { |
4031 | struct odp_msg *msg = packet->data; | |
064af421 BP |
4032 | struct rule *rule; |
4033 | struct ofpbuf payload; | |
4034 | flow_t flow; | |
4035 | ||
064af421 BP |
4036 | payload.data = msg + 1; |
4037 | payload.size = msg->length - sizeof *msg; | |
659586ef | 4038 | flow_extract(&payload, msg->arg, msg->port, &flow); |
064af421 | 4039 | |
0ad9b732 JP |
4040 | /* Check with in-band control to see if this packet should be sent |
4041 | * to the local port regardless of the flow table. */ | |
4042 | if (in_band_msg_in_hook(p->in_band, &flow, &payload)) { | |
4043 | union odp_action action; | |
4044 | ||
4045 | memset(&action, 0, sizeof(action)); | |
4046 | action.output.type = ODPAT_OUTPUT; | |
4047 | action.output.port = ODPP_LOCAL; | |
f1acd62b | 4048 | dpif_execute(p->dpif, flow.in_port, &action, 1, &payload); |
0ad9b732 JP |
4049 | } |
4050 | ||
064af421 BP |
4051 | rule = lookup_valid_rule(p, &flow); |
4052 | if (!rule) { | |
4053 | /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */ | |
4054 | struct ofport *port = port_array_get(&p->ports, msg->port); | |
4055 | if (port) { | |
4056 | if (port->opp.config & OFPPC_NO_PACKET_IN) { | |
4057 | COVERAGE_INC(ofproto_no_packet_in); | |
4058 | /* XXX install 'drop' flow entry */ | |
4059 | ofpbuf_delete(packet); | |
4060 | return; | |
4061 | } | |
4062 | } else { | |
4063 | VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port); | |
4064 | } | |
4065 | ||
4066 | COVERAGE_INC(ofproto_packet_in); | |
76ce9432 | 4067 | send_packet_in(p, packet); |
064af421 BP |
4068 | return; |
4069 | } | |
4070 | ||
4071 | if (rule->cr.wc.wildcards) { | |
4072 | rule = rule_create_subrule(p, rule, &flow); | |
4073 | rule_make_actions(p, rule, packet); | |
4074 | } else { | |
4075 | if (!rule->may_install) { | |
4076 | /* The rule is not installable, that is, we need to process every | |
4077 | * packet, so process the current packet and set its actions into | |
4078 | * 'subrule'. */ | |
4079 | rule_make_actions(p, rule, packet); | |
4080 | } else { | |
4081 | /* XXX revalidate rule if it needs it */ | |
4082 | } | |
4083 | } | |
4084 | ||
4085 | rule_execute(p, rule, &payload, &flow); | |
4086 | rule_reinstall(p, rule); | |
7778bd15 | 4087 | |
76ce9432 | 4088 | if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY) { |
7778bd15 BP |
4089 | /* |
4090 | * Extra-special case for fail-open mode. | |
4091 | * | |
4092 | * We are in fail-open mode and the packet matched the fail-open rule, | |
4093 | * but we are connected to a controller too. We should send the packet | |
4094 | * up to the controller in the hope that it will try to set up a flow | |
4095 | * and thereby allow us to exit fail-open. | |
4096 | * | |
4097 | * See the top-level comment in fail-open.c for more information. | |
4098 | */ | |
76ce9432 | 4099 | send_packet_in(p, packet); |
7778bd15 BP |
4100 | } else { |
4101 | ofpbuf_delete(packet); | |
4102 | } | |
064af421 | 4103 | } |
72b06300 BP |
4104 | |
4105 | static void | |
4106 | handle_odp_msg(struct ofproto *p, struct ofpbuf *packet) | |
4107 | { | |
4108 | struct odp_msg *msg = packet->data; | |
4109 | ||
4110 | switch (msg->type) { | |
4111 | case _ODPL_ACTION_NR: | |
4112 | COVERAGE_INC(ofproto_ctlr_action); | |
76ce9432 | 4113 | send_packet_in(p, packet); |
72b06300 BP |
4114 | break; |
4115 | ||
4116 | case _ODPL_SFLOW_NR: | |
4117 | if (p->sflow) { | |
4118 | ofproto_sflow_received(p->sflow, msg); | |
4119 | } | |
4120 | ofpbuf_delete(packet); | |
4121 | break; | |
4122 | ||
4123 | case _ODPL_MISS_NR: | |
4124 | handle_odp_miss_msg(p, packet); | |
4125 | break; | |
4126 | ||
4127 | default: | |
4128 | VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32, | |
4129 | msg->type); | |
4130 | break; | |
4131 | } | |
4132 | } | |
064af421 BP |
4133 | \f |
4134 | static void | |
4135 | revalidate_cb(struct cls_rule *sub_, void *cbdata_) | |
4136 | { | |
4137 | struct rule *sub = rule_from_cls_rule(sub_); | |
4138 | struct revalidate_cbdata *cbdata = cbdata_; | |
4139 | ||
4140 | if (cbdata->revalidate_all | |
4141 | || (cbdata->revalidate_subrules && sub->super) | |
4142 | || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) { | |
4143 | revalidate_rule(cbdata->ofproto, sub); | |
4144 | } | |
4145 | } | |
4146 | ||
4147 | static bool | |
4148 | revalidate_rule(struct ofproto *p, struct rule *rule) | |
4149 | { | |
4150 | const flow_t *flow = &rule->cr.flow; | |
4151 | ||
4152 | COVERAGE_INC(ofproto_revalidate_rule); | |
4153 | if (rule->super) { | |
4154 | struct rule *super; | |
4155 | super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow)); | |
4156 | if (!super) { | |
4157 | rule_remove(p, rule); | |
4158 | return false; | |
4159 | } else if (super != rule->super) { | |
4160 | COVERAGE_INC(ofproto_revalidate_moved); | |
4161 | list_remove(&rule->list); | |
4162 | list_push_back(&super->list, &rule->list); | |
4163 | rule->super = super; | |
4164 | rule->hard_timeout = super->hard_timeout; | |
4165 | rule->idle_timeout = super->idle_timeout; | |
4166 | rule->created = super->created; | |
4167 | rule->used = 0; | |
4168 | } | |
4169 | } | |
4170 | ||
4171 | rule_update_actions(p, rule); | |
4172 | return true; | |
4173 | } | |
4174 | ||
4175 | static struct ofpbuf * | |
659586ef JG |
4176 | compose_flow_removed(struct ofproto *p, const struct rule *rule, |
4177 | long long int now, uint8_t reason) | |
064af421 | 4178 | { |
ca069229 | 4179 | struct ofp_flow_removed *ofr; |
064af421 | 4180 | struct ofpbuf *buf; |
9ca76894 | 4181 | long long int tdiff = now - rule->created; |
26c3f94a JP |
4182 | uint32_t sec = tdiff / 1000; |
4183 | uint32_t msec = tdiff - (sec * 1000); | |
064af421 | 4184 | |
ca069229 | 4185 | ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf); |
659586ef JG |
4186 | flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, p->tun_id_from_cookie, |
4187 | &ofr->match); | |
39997502 | 4188 | ofr->cookie = rule->flow_cookie; |
ca069229 JP |
4189 | ofr->priority = htons(rule->cr.priority); |
4190 | ofr->reason = reason; | |
26c3f94a JP |
4191 | ofr->duration_sec = htonl(sec); |
4192 | ofr->duration_nsec = htonl(msec * 1000000); | |
ca069229 JP |
4193 | ofr->idle_timeout = htons(rule->idle_timeout); |
4194 | ofr->packet_count = htonll(rule->packet_count); | |
4195 | ofr->byte_count = htonll(rule->byte_count); | |
064af421 BP |
4196 | |
4197 | return buf; | |
4198 | } | |
4199 | ||
4200 | static void | |
ca069229 JP |
4201 | uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule) |
4202 | { | |
4203 | assert(rule->installed); | |
4204 | assert(!rule->cr.wc.wildcards); | |
4205 | ||
4206 | if (rule->super) { | |
4207 | rule_remove(ofproto, rule); | |
4208 | } else { | |
4209 | rule_uninstall(ofproto, rule); | |
4210 | } | |
4211 | } | |
9deba63b | 4212 | |
ca069229 JP |
4213 | static void |
4214 | send_flow_removed(struct ofproto *p, struct rule *rule, | |
4215 | long long int now, uint8_t reason) | |
064af421 BP |
4216 | { |
4217 | struct ofconn *ofconn; | |
4218 | struct ofconn *prev; | |
b9b0ce61 | 4219 | struct ofpbuf *buf = NULL; |
064af421 BP |
4220 | |
4221 | /* We limit the maximum number of queued flow expirations it by accounting | |
4222 | * them under the counter for replies. That works because preventing | |
4223 | * OpenFlow requests from being processed also prevents new flows from | |
4224 | * being added (and expiring). (It also prevents processing OpenFlow | |
4225 | * requests that would not add new flows, so it is imperfect.) */ | |
4226 | ||
4227 | prev = NULL; | |
4228 | LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) { | |
9deba63b | 4229 | if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn) |
c91248b3 | 4230 | && ofconn_receives_async_msgs(ofconn)) { |
064af421 | 4231 | if (prev) { |
431d8ad2 | 4232 | queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter); |
064af421 | 4233 | } else { |
659586ef | 4234 | buf = compose_flow_removed(p, rule, now, reason); |
064af421 BP |
4235 | } |
4236 | prev = ofconn; | |
4237 | } | |
4238 | } | |
4239 | if (prev) { | |
431d8ad2 | 4240 | queue_tx(buf, prev, prev->reply_counter); |
064af421 BP |
4241 | } |
4242 | } | |
4243 | ||
064af421 BP |
4244 | |
4245 | static void | |
4246 | expire_rule(struct cls_rule *cls_rule, void *p_) | |
4247 | { | |
4248 | struct ofproto *p = p_; | |
4249 | struct rule *rule = rule_from_cls_rule(cls_rule); | |
4250 | long long int hard_expire, idle_expire, expire, now; | |
4251 | ||
4252 | hard_expire = (rule->hard_timeout | |
4253 | ? rule->created + rule->hard_timeout * 1000 | |
4254 | : LLONG_MAX); | |
4255 | idle_expire = (rule->idle_timeout | |
4256 | && (rule->super || list_is_empty(&rule->list)) | |
4257 | ? rule->used + rule->idle_timeout * 1000 | |
4258 | : LLONG_MAX); | |
4259 | expire = MIN(hard_expire, idle_expire); | |
064af421 BP |
4260 | |
4261 | now = time_msec(); | |
4262 | if (now < expire) { | |
4263 | if (rule->installed && now >= rule->used + 5000) { | |
4264 | uninstall_idle_flow(p, rule); | |
0193b2af JG |
4265 | } else if (!rule->cr.wc.wildcards) { |
4266 | active_timeout(p, rule); | |
064af421 | 4267 | } |
0193b2af | 4268 | |
064af421 BP |
4269 | return; |
4270 | } | |
4271 | ||
4272 | COVERAGE_INC(ofproto_expired); | |
46d6f36f JG |
4273 | |
4274 | /* Update stats. This code will be a no-op if the rule expired | |
4275 | * due to an idle timeout. */ | |
064af421 | 4276 | if (rule->cr.wc.wildcards) { |
064af421 BP |
4277 | struct rule *subrule, *next; |
4278 | LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) { | |
4279 | rule_remove(p, subrule); | |
4280 | } | |
46d6f36f JG |
4281 | } else { |
4282 | rule_uninstall(p, rule); | |
064af421 BP |
4283 | } |
4284 | ||
8fe1a59d | 4285 | if (!rule_is_hidden(rule)) { |
ca069229 JP |
4286 | send_flow_removed(p, rule, now, |
4287 | (now >= hard_expire | |
4288 | ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT)); | |
8fe1a59d | 4289 | } |
064af421 BP |
4290 | rule_remove(p, rule); |
4291 | } | |
4292 | ||
0193b2af JG |
4293 | static void |
4294 | active_timeout(struct ofproto *ofproto, struct rule *rule) | |
4295 | { | |
4296 | if (ofproto->netflow && !is_controller_rule(rule) && | |
4297 | netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) { | |
4298 | struct ofexpired expired; | |
4299 | struct odp_flow odp_flow; | |
4300 | ||
4301 | /* Get updated flow stats. */ | |
4302 | memset(&odp_flow, 0, sizeof odp_flow); | |
094e1514 JG |
4303 | if (rule->installed) { |
4304 | odp_flow.key = rule->cr.flow; | |
4305 | odp_flow.flags = ODPFF_ZERO_TCP_FLAGS; | |
d65349ea | 4306 | dpif_flow_get(ofproto->dpif, &odp_flow); |
094e1514 JG |
4307 | |
4308 | if (odp_flow.stats.n_packets) { | |
4309 | update_time(ofproto, rule, &odp_flow.stats); | |
abfec865 | 4310 | netflow_flow_update_flags(&rule->nf_flow, |
094e1514 JG |
4311 | odp_flow.stats.tcp_flags); |
4312 | } | |
0193b2af JG |
4313 | } |
4314 | ||
4315 | expired.flow = rule->cr.flow; | |
4316 | expired.packet_count = rule->packet_count + | |
4317 | odp_flow.stats.n_packets; | |
4318 | expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes; | |
4319 | expired.used = rule->used; | |
4320 | ||
4321 | netflow_expire(ofproto->netflow, &rule->nf_flow, &expired); | |
4322 | ||
4323 | /* Schedule us to send the accumulated records once we have | |
4324 | * collected all of them. */ | |
4325 | poll_immediate_wake(); | |
4326 | } | |
4327 | } | |
4328 | ||
064af421 BP |
4329 | static void |
4330 | update_used(struct ofproto *p) | |
4331 | { | |
4332 | struct odp_flow *flows; | |
4333 | size_t n_flows; | |
4334 | size_t i; | |
4335 | int error; | |
4336 | ||
c228a364 | 4337 | error = dpif_flow_list_all(p->dpif, &flows, &n_flows); |
064af421 BP |
4338 | if (error) { |
4339 | return; | |
4340 | } | |
4341 | ||
4342 | for (i = 0; i < n_flows; i++) { | |
4343 | struct odp_flow *f = &flows[i]; | |
4344 | struct rule *rule; | |
4345 | ||
4346 | rule = rule_from_cls_rule( | |
4347 | classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX)); | |
4348 | if (!rule || !rule->installed) { | |
4349 | COVERAGE_INC(ofproto_unexpected_rule); | |
c228a364 | 4350 | dpif_flow_del(p->dpif, f); |
064af421 BP |
4351 | continue; |
4352 | } | |
4353 | ||
0193b2af | 4354 | update_time(p, rule, &f->stats); |
064af421 BP |
4355 | rule_account(p, rule, f->stats.n_bytes); |
4356 | } | |
4357 | free(flows); | |
4358 | } | |
4359 | ||
43253595 | 4360 | /* pinsched callback for sending 'packet' on 'ofconn'. */ |
064af421 | 4361 | static void |
76ce9432 | 4362 | do_send_packet_in(struct ofpbuf *packet, void *ofconn_) |
064af421 | 4363 | { |
76ce9432 | 4364 | struct ofconn *ofconn = ofconn_; |
43253595 BP |
4365 | |
4366 | rconn_send_with_limit(ofconn->rconn, packet, | |
4367 | ofconn->packet_in_counter, 100); | |
4368 | } | |
4369 | ||
4370 | /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and | |
4371 | * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s | |
4372 | * packet scheduler for sending. | |
4373 | * | |
30ea5d93 BP |
4374 | * 'max_len' specifies the maximum number of bytes of the packet to send on |
4375 | * 'ofconn' (INT_MAX specifies no limit). | |
4376 | * | |
43253595 BP |
4377 | * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise, |
4378 | * ownership is transferred to this function. */ | |
4379 | static void | |
30ea5d93 BP |
4380 | schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len, |
4381 | bool clone) | |
43253595 | 4382 | { |
76ce9432 | 4383 | struct ofproto *ofproto = ofconn->ofproto; |
43253595 BP |
4384 | struct ofp_packet_in *opi = packet->data; |
4385 | uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port)); | |
4386 | int send_len, trim_size; | |
76ce9432 | 4387 | uint32_t buffer_id; |
064af421 | 4388 | |
43253595 BP |
4389 | /* Get buffer. */ |
4390 | if (opi->reason == OFPR_ACTION) { | |
76ce9432 | 4391 | buffer_id = UINT32_MAX; |
43253595 BP |
4392 | } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) { |
4393 | buffer_id = pktbuf_get_null(); | |
89b9612d BP |
4394 | } else if (!ofconn->pktbuf) { |
4395 | buffer_id = UINT32_MAX; | |
76ce9432 | 4396 | } else { |
43253595 BP |
4397 | struct ofpbuf payload; |
4398 | payload.data = opi->data; | |
4399 | payload.size = packet->size - offsetof(struct ofp_packet_in, data); | |
4400 | buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port); | |
76ce9432 | 4401 | } |
372179d4 | 4402 | |
43253595 BP |
4403 | /* Figure out how much of the packet to send. */ |
4404 | send_len = ntohs(opi->total_len); | |
4405 | if (buffer_id != UINT32_MAX) { | |
4406 | send_len = MIN(send_len, ofconn->miss_send_len); | |
4407 | } | |
30ea5d93 | 4408 | send_len = MIN(send_len, max_len); |
064af421 | 4409 | |
43253595 BP |
4410 | /* Adjust packet length and clone if necessary. */ |
4411 | trim_size = offsetof(struct ofp_packet_in, data) + send_len; | |
4412 | if (clone) { | |
4413 | packet = ofpbuf_clone_data(packet->data, trim_size); | |
4414 | opi = packet->data; | |
4415 | } else { | |
4416 | packet->size = trim_size; | |
4417 | } | |
4418 | ||
4419 | /* Update packet headers. */ | |
4420 | opi->buffer_id = htonl(buffer_id); | |
4421 | update_openflow_length(packet); | |
4422 | ||
4423 | /* Hand over to packet scheduler. It might immediately call into | |
4424 | * do_send_packet_in() or it might buffer it for a while (until a later | |
4425 | * call to pinsched_run()). */ | |
4426 | pinsched_send(ofconn->schedulers[opi->reason], in_port, | |
4427 | packet, do_send_packet_in, ofconn); | |
064af421 BP |
4428 | } |
4429 | ||
43253595 BP |
4430 | /* Replace struct odp_msg header in 'packet' by equivalent struct |
4431 | * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as | |
4432 | * returned by dpif_recv()). | |
4433 | * | |
4434 | * The conversion is not complete: the caller still needs to trim any unneeded | |
4435 | * payload off the end of the buffer, set the length in the OpenFlow header, | |
4436 | * and set buffer_id. Those require us to know the controller settings and so | |
30ea5d93 BP |
4437 | * must be done on a per-controller basis. |
4438 | * | |
4439 | * Returns the maximum number of bytes of the packet that should be sent to | |
4440 | * the controller (INT_MAX if no limit). */ | |
4441 | static int | |
43253595 | 4442 | do_convert_to_packet_in(struct ofpbuf *packet) |
064af421 | 4443 | { |
76ce9432 | 4444 | struct odp_msg *msg = packet->data; |
43253595 BP |
4445 | struct ofp_packet_in *opi; |
4446 | uint8_t reason; | |
4447 | uint16_t total_len; | |
4448 | uint16_t in_port; | |
30ea5d93 | 4449 | int max_len; |
43253595 BP |
4450 | |
4451 | /* Extract relevant header fields */ | |
30ea5d93 BP |
4452 | if (msg->type == _ODPL_ACTION_NR) { |
4453 | reason = OFPR_ACTION; | |
4454 | max_len = msg->arg; | |
4455 | } else { | |
4456 | reason = OFPR_NO_MATCH; | |
4457 | max_len = INT_MAX; | |
4458 | } | |
43253595 BP |
4459 | total_len = msg->length - sizeof *msg; |
4460 | in_port = odp_port_to_ofp_port(msg->port); | |
4461 | ||
4462 | /* Repurpose packet buffer by overwriting header. */ | |
4463 | ofpbuf_pull(packet, sizeof(struct odp_msg)); | |
4464 | opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data)); | |
4465 | opi->header.version = OFP_VERSION; | |
4466 | opi->header.type = OFPT_PACKET_IN; | |
4467 | opi->total_len = htons(total_len); | |
4468 | opi->in_port = htons(in_port); | |
4469 | opi->reason = reason; | |
30ea5d93 BP |
4470 | |
4471 | return max_len; | |
43253595 BP |
4472 | } |
4473 | ||
4474 | /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or | |
4475 | * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller | |
4476 | * as necessary according to their individual configurations. | |
4477 | * | |
4478 | * 'packet' must have sufficient headroom to convert it into a struct | |
4479 | * ofp_packet_in (e.g. as returned by dpif_recv()). | |
4480 | * | |
4481 | * Takes ownership of 'packet'. */ | |
4482 | static void | |
4483 | send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet) | |
4484 | { | |
76ce9432 | 4485 | struct ofconn *ofconn, *prev; |
30ea5d93 | 4486 | int max_len; |
064af421 | 4487 | |
30ea5d93 | 4488 | max_len = do_convert_to_packet_in(packet); |
76ce9432 BP |
4489 | |
4490 | prev = NULL; | |
4491 | LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) { | |
c91248b3 | 4492 | if (ofconn_receives_async_msgs(ofconn)) { |
9deba63b | 4493 | if (prev) { |
30ea5d93 | 4494 | schedule_packet_in(prev, packet, max_len, true); |
9deba63b BP |
4495 | } |
4496 | prev = ofconn; | |
064af421 | 4497 | } |
76ce9432 BP |
4498 | } |
4499 | if (prev) { | |
30ea5d93 | 4500 | schedule_packet_in(prev, packet, max_len, false); |
76ce9432 BP |
4501 | } else { |
4502 | ofpbuf_delete(packet); | |
064af421 | 4503 | } |
064af421 BP |
4504 | } |
4505 | ||
4506 | static uint64_t | |
fa60c019 | 4507 | pick_datapath_id(const struct ofproto *ofproto) |
064af421 | 4508 | { |
fa60c019 | 4509 | const struct ofport *port; |
064af421 | 4510 | |
fa60c019 BP |
4511 | port = port_array_get(&ofproto->ports, ODPP_LOCAL); |
4512 | if (port) { | |
4513 | uint8_t ea[ETH_ADDR_LEN]; | |
4514 | int error; | |
4515 | ||
4516 | error = netdev_get_etheraddr(port->netdev, ea); | |
064af421 BP |
4517 | if (!error) { |
4518 | return eth_addr_to_uint64(ea); | |
4519 | } | |
4520 | VLOG_WARN("could not get MAC address for %s (%s)", | |
fa60c019 | 4521 | netdev_get_name(port->netdev), strerror(error)); |
064af421 | 4522 | } |
fa60c019 | 4523 | return ofproto->fallback_dpid; |
064af421 BP |
4524 | } |
4525 | ||
4526 | static uint64_t | |
4527 | pick_fallback_dpid(void) | |
4528 | { | |
4529 | uint8_t ea[ETH_ADDR_LEN]; | |
70150daf | 4530 | eth_addr_nicira_random(ea); |
064af421 BP |
4531 | return eth_addr_to_uint64(ea); |
4532 | } | |
4533 | \f | |
4534 | static bool | |
4535 | default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet, | |
4536 | struct odp_actions *actions, tag_type *tags, | |
6a07af36 | 4537 | uint16_t *nf_output_iface, void *ofproto_) |
064af421 BP |
4538 | { |
4539 | struct ofproto *ofproto = ofproto_; | |
4540 | int out_port; | |
4541 | ||
4542 | /* Drop frames for reserved multicast addresses. */ | |
4543 | if (eth_addr_is_reserved(flow->dl_dst)) { | |
4544 | return true; | |
4545 | } | |
4546 | ||
4547 | /* Learn source MAC (but don't try to learn from revalidation). */ | |
4548 | if (packet != NULL) { | |
4549 | tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src, | |
7febb910 JG |
4550 | 0, flow->in_port, |
4551 | GRAT_ARP_LOCK_NONE); | |
064af421 BP |
4552 | if (rev_tag) { |
4553 | /* The log messages here could actually be useful in debugging, | |
4554 | * so keep the rate limit relatively high. */ | |
4555 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); | |
4556 | VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16, | |
4557 | ETH_ADDR_ARGS(flow->dl_src), flow->in_port); | |
4558 | ofproto_revalidate(ofproto, rev_tag); | |
4559 | } | |
4560 | } | |
4561 | ||
4562 | /* Determine output port. */ | |
7febb910 JG |
4563 | out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags, |
4564 | NULL); | |
064af421 | 4565 | if (out_port < 0) { |
6a07af36 | 4566 | add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface); |
064af421 BP |
4567 | } else if (out_port != flow->in_port) { |
4568 | odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port; | |
6a07af36 | 4569 | *nf_output_iface = out_port; |
064af421 BP |
4570 | } else { |
4571 | /* Drop. */ | |
4572 | } | |
4573 | ||
4574 | return true; | |
4575 | } | |
4576 | ||
4577 | static const struct ofhooks default_ofhooks = { | |
4578 | NULL, | |
4579 | default_normal_ofhook_cb, | |
4580 | NULL, | |
4581 | NULL | |
4582 | }; |