]>
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 | 32 | #include "fail-open.h" |
ca0f572c BP |
33 | #include "hash.h" |
34 | #include "hmap.h" | |
064af421 BP |
35 | #include "in-band.h" |
36 | #include "mac-learning.h" | |
37 | #include "netdev.h" | |
38 | #include "netflow.h" | |
39 | #include "odp-util.h" | |
40 | #include "ofp-print.h" | |
fa37b408 | 41 | #include "ofp-util.h" |
72b06300 | 42 | #include "ofproto-sflow.h" |
064af421 BP |
43 | #include "ofpbuf.h" |
44 | #include "openflow/nicira-ext.h" | |
45 | #include "openflow/openflow.h" | |
064af421 BP |
46 | #include "openvswitch/datapath-protocol.h" |
47 | #include "packets.h" | |
48 | #include "pinsched.h" | |
49 | #include "pktbuf.h" | |
50 | #include "poll-loop.h" | |
064af421 BP |
51 | #include "rconn.h" |
52 | #include "shash.h" | |
53 | #include "status.h" | |
fe55ad15 | 54 | #include "stream-ssl.h" |
064af421 BP |
55 | #include "svec.h" |
56 | #include "tag.h" | |
57 | #include "timeval.h" | |
4f2cad2c | 58 | #include "unixctl.h" |
064af421 | 59 | #include "vconn.h" |
5136ce49 | 60 | #include "vlog.h" |
064af421 BP |
61 | #include "xtoxll.h" |
62 | ||
5136ce49 | 63 | VLOG_DEFINE_THIS_MODULE(ofproto) |
064af421 | 64 | |
72b06300 | 65 | #include "sflow_api.h" |
064af421 BP |
66 | |
67 | enum { | |
68 | TABLEID_HASH = 0, | |
69 | TABLEID_CLASSIFIER = 1 | |
70 | }; | |
71 | ||
ca0f572c | 72 | |
064af421 | 73 | struct ofport { |
ca0f572c | 74 | struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */ |
064af421 BP |
75 | struct netdev *netdev; |
76 | struct ofp_phy_port opp; /* In host byte order. */ | |
ca0f572c | 77 | uint16_t odp_port; |
064af421 BP |
78 | }; |
79 | ||
80 | static void ofport_free(struct ofport *); | |
81 | static void hton_ofp_phy_port(struct ofp_phy_port *); | |
82 | ||
83 | static int xlate_actions(const union ofp_action *in, size_t n_in, | |
84 | const flow_t *flow, struct ofproto *ofproto, | |
85 | const struct ofpbuf *packet, | |
86 | struct odp_actions *out, tag_type *tags, | |
6a07af36 | 87 | bool *may_set_up_flow, uint16_t *nf_output_iface); |
064af421 BP |
88 | |
89 | struct rule { | |
90 | struct cls_rule cr; | |
91 | ||
d295e8e9 | 92 | uint64_t flow_cookie; /* Controller-issued identifier. |
39997502 | 93 | (Kept in network-byte order.) */ |
064af421 BP |
94 | uint16_t idle_timeout; /* In seconds from time of last use. */ |
95 | uint16_t hard_timeout; /* In seconds from time of creation. */ | |
ca069229 | 96 | bool send_flow_removed; /* Send a flow removed message? */ |
064af421 BP |
97 | long long int used; /* Last-used time (0 if never used). */ |
98 | long long int created; /* Creation time. */ | |
99 | uint64_t packet_count; /* Number of packets received. */ | |
100 | uint64_t byte_count; /* Number of bytes received. */ | |
101 | uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */ | |
064af421 | 102 | tag_type tags; /* Tags (set only by hooks). */ |
0193b2af | 103 | struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ |
064af421 BP |
104 | |
105 | /* If 'super' is non-NULL, this rule is a subrule, that is, it is an | |
106 | * exact-match rule (having cr.wc.wildcards of 0) generated from the | |
107 | * wildcard rule 'super'. In this case, 'list' is an element of the | |
108 | * super-rule's list. | |
109 | * | |
110 | * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of | |
111 | * a list of subrules. A super-rule with no wildcards (where | |
112 | * cr.wc.wildcards is 0) will never have any subrules. */ | |
113 | struct rule *super; | |
114 | struct list list; | |
115 | ||
116 | /* OpenFlow actions. | |
79eee1eb BP |
117 | * |
118 | * 'n_actions' is the number of elements in the 'actions' array. A single | |
119 | * action may take up more more than one element's worth of space. | |
064af421 BP |
120 | * |
121 | * A subrule has no actions (it uses the super-rule's actions). */ | |
122 | int n_actions; | |
123 | union ofp_action *actions; | |
124 | ||
125 | /* Datapath actions. | |
126 | * | |
127 | * A super-rule with wildcard fields never has ODP actions (since the | |
128 | * datapath only supports exact-match flows). */ | |
129 | bool installed; /* Installed in datapath? */ | |
130 | bool may_install; /* True ordinarily; false if actions must | |
131 | * be reassessed for every packet. */ | |
132 | int n_odp_actions; | |
133 | union odp_action *odp_actions; | |
134 | }; | |
135 | ||
136 | static inline bool | |
137 | rule_is_hidden(const struct rule *rule) | |
138 | { | |
139 | /* Subrules are merely an implementation detail, so hide them from the | |
140 | * controller. */ | |
141 | if (rule->super != NULL) { | |
142 | return true; | |
143 | } | |
144 | ||
8cd4882f | 145 | /* Rules with priority higher than UINT16_MAX are set up by ofproto itself |
064af421 BP |
146 | * (e.g. by in-band control) and are intentionally hidden from the |
147 | * controller. */ | |
148 | if (rule->cr.priority > UINT16_MAX) { | |
149 | return true; | |
150 | } | |
151 | ||
152 | return false; | |
153 | } | |
154 | ||
0193b2af JG |
155 | static struct rule *rule_create(struct ofproto *, struct rule *super, |
156 | const union ofp_action *, size_t n_actions, | |
ca069229 | 157 | uint16_t idle_timeout, uint16_t hard_timeout, |
39997502 | 158 | uint64_t flow_cookie, bool send_flow_removed); |
064af421 BP |
159 | static void rule_free(struct rule *); |
160 | static void rule_destroy(struct ofproto *, struct rule *); | |
161 | static struct rule *rule_from_cls_rule(const struct cls_rule *); | |
162 | static void rule_insert(struct ofproto *, struct rule *, | |
163 | struct ofpbuf *packet, uint16_t in_port); | |
164 | static void rule_remove(struct ofproto *, struct rule *); | |
165 | static bool rule_make_actions(struct ofproto *, struct rule *, | |
166 | const struct ofpbuf *packet); | |
167 | static void rule_install(struct ofproto *, struct rule *, | |
168 | struct rule *displaced_rule); | |
169 | static void rule_uninstall(struct ofproto *, struct rule *); | |
170 | static void rule_post_uninstall(struct ofproto *, struct rule *); | |
ca069229 JP |
171 | static void send_flow_removed(struct ofproto *p, struct rule *rule, |
172 | long long int now, uint8_t reason); | |
064af421 | 173 | |
76ce9432 BP |
174 | /* ofproto supports two kinds of OpenFlow connections: |
175 | * | |
5899143f BP |
176 | * - "Primary" connections to ordinary OpenFlow controllers. ofproto |
177 | * maintains persistent connections to these controllers and by default | |
178 | * sends them asynchronous messages such as packet-ins. | |
76ce9432 | 179 | * |
5899143f | 180 | * - "Service" connections, e.g. from ovs-ofctl. When these connections |
76ce9432 BP |
181 | * drop, it is the other side's responsibility to reconnect them if |
182 | * necessary. ofproto does not send them asynchronous messages by default. | |
7d674866 BP |
183 | * |
184 | * Currently, active (tcp, ssl, unix) connections are always "primary" | |
185 | * connections and passive (ptcp, pssl, punix) connections are always "service" | |
186 | * connections. There is no inherent reason for this, but it reflects the | |
187 | * common case. | |
76ce9432 BP |
188 | */ |
189 | enum ofconn_type { | |
5899143f BP |
190 | OFCONN_PRIMARY, /* An ordinary OpenFlow controller. */ |
191 | OFCONN_SERVICE /* A service connection, e.g. "ovs-ofctl". */ | |
76ce9432 | 192 | }; |
064af421 | 193 | |
7d674866 BP |
194 | /* A listener for incoming OpenFlow "service" connections. */ |
195 | struct ofservice { | |
196 | struct hmap_node node; /* In struct ofproto's "services" hmap. */ | |
197 | struct pvconn *pvconn; /* OpenFlow connection listener. */ | |
198 | ||
199 | /* These are not used by ofservice directly. They are settings for | |
200 | * accepted "struct ofconn"s from the pvconn. */ | |
201 | int probe_interval; /* Max idle time before probing, in seconds. */ | |
202 | int rate_limit; /* Max packet-in rate in packets per second. */ | |
203 | int burst_limit; /* Limit on accumulating packet credits. */ | |
204 | }; | |
205 | ||
206 | static struct ofservice *ofservice_lookup(struct ofproto *, | |
207 | const char *target); | |
208 | static int ofservice_create(struct ofproto *, | |
209 | const struct ofproto_controller *); | |
210 | static void ofservice_reconfigure(struct ofservice *, | |
211 | const struct ofproto_controller *); | |
212 | static void ofservice_destroy(struct ofproto *, struct ofservice *); | |
213 | ||
76ce9432 BP |
214 | /* An OpenFlow connection. */ |
215 | struct ofconn { | |
216 | struct ofproto *ofproto; /* The ofproto that owns this connection. */ | |
217 | struct list node; /* In struct ofproto's "all_conns" list. */ | |
218 | struct rconn *rconn; /* OpenFlow connection. */ | |
219 | enum ofconn_type type; /* Type. */ | |
220 | ||
221 | /* OFPT_PACKET_IN related data. */ | |
222 | struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */ | |
223 | struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */ | |
224 | struct pktbuf *pktbuf; /* OpenFlow packet buffers. */ | |
225 | int miss_send_len; /* Bytes to send of buffered packets. */ | |
226 | ||
227 | /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow | |
228 | * requests, and the maximum number before we stop reading OpenFlow | |
229 | * requests. */ | |
064af421 BP |
230 | #define OFCONN_REPLY_MAX 100 |
231 | struct rconn_packet_counter *reply_counter; | |
76ce9432 | 232 | |
5899143f | 233 | /* type == OFCONN_PRIMARY only. */ |
9deba63b | 234 | enum nx_role role; /* Role. */ |
76ce9432 BP |
235 | struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */ |
236 | struct discovery *discovery; /* Controller discovery object, if enabled. */ | |
237 | struct status_category *ss; /* Switch status category. */ | |
d2ede7bc | 238 | enum ofproto_band band; /* In-band or out-of-band? */ |
064af421 BP |
239 | }; |
240 | ||
76ce9432 BP |
241 | /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's |
242 | * "schedulers" array. Their values are 0 and 1, and their meanings and values | |
243 | * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In | |
244 | * case anything ever changes, check their values here. */ | |
245 | #define N_SCHEDULERS 2 | |
246 | BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0); | |
247 | BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR); | |
248 | BUILD_ASSERT_DECL(OFPR_ACTION == 1); | |
249 | BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR); | |
250 | ||
251 | static struct ofconn *ofconn_create(struct ofproto *, struct rconn *, | |
252 | enum ofconn_type); | |
c475ae67 | 253 | static void ofconn_destroy(struct ofconn *); |
064af421 BP |
254 | static void ofconn_run(struct ofconn *, struct ofproto *); |
255 | static void ofconn_wait(struct ofconn *); | |
c91248b3 | 256 | static bool ofconn_receives_async_msgs(const struct ofconn *); |
eb15cdbb | 257 | static char *ofconn_make_name(const struct ofproto *, const char *target); |
7d674866 | 258 | static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst); |
c91248b3 | 259 | |
064af421 BP |
260 | static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn, |
261 | struct rconn_packet_counter *counter); | |
262 | ||
76ce9432 BP |
263 | static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg); |
264 | static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn); | |
265 | ||
064af421 BP |
266 | struct ofproto { |
267 | /* Settings. */ | |
268 | uint64_t datapath_id; /* Datapath ID. */ | |
269 | uint64_t fallback_dpid; /* Datapath ID if no better choice found. */ | |
5a719c38 JP |
270 | char *mfr_desc; /* Manufacturer. */ |
271 | char *hw_desc; /* Hardware. */ | |
272 | char *sw_desc; /* Software version. */ | |
273 | char *serial_desc; /* Serial number. */ | |
8abc4ed7 | 274 | char *dp_desc; /* Datapath description. */ |
064af421 BP |
275 | |
276 | /* Datapath. */ | |
c228a364 | 277 | struct dpif *dpif; |
e9e28be3 | 278 | struct netdev_monitor *netdev_monitor; |
ca0f572c | 279 | struct hmap ports; /* Contains "struct ofport"s. */ |
064af421 BP |
280 | struct shash port_by_name; |
281 | uint32_t max_ports; | |
282 | ||
283 | /* Configuration. */ | |
284 | struct switch_status *switch_status; | |
064af421 | 285 | struct fail_open *fail_open; |
064af421 | 286 | struct netflow *netflow; |
72b06300 | 287 | struct ofproto_sflow *sflow; |
064af421 | 288 | |
d2ede7bc BP |
289 | /* In-band control. */ |
290 | struct in_band *in_band; | |
291 | long long int next_in_band_update; | |
917e50e1 BP |
292 | struct sockaddr_in *extra_in_band_remotes; |
293 | size_t n_extra_remotes; | |
294 | ||
064af421 BP |
295 | /* Flow table. */ |
296 | struct classifier cls; | |
297 | bool need_revalidate; | |
298 | long long int next_expiration; | |
299 | struct tag_set revalidate_set; | |
659586ef | 300 | bool tun_id_from_cookie; |
064af421 BP |
301 | |
302 | /* OpenFlow connections. */ | |
76ce9432 BP |
303 | struct hmap controllers; /* Controller "struct ofconn"s. */ |
304 | struct list all_conns; /* Contains "struct ofconn"s. */ | |
31681a5d | 305 | enum ofproto_fail_mode fail_mode; |
7d674866 BP |
306 | |
307 | /* OpenFlow listeners. */ | |
308 | struct hmap services; /* Contains "struct ofservice"s. */ | |
064af421 BP |
309 | struct pvconn **snoops; |
310 | size_t n_snoops; | |
311 | ||
312 | /* Hooks for ovs-vswitchd. */ | |
313 | const struct ofhooks *ofhooks; | |
314 | void *aux; | |
315 | ||
316 | /* Used by default ofhooks. */ | |
317 | struct mac_learning *ml; | |
318 | }; | |
319 | ||
320 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
321 | ||
322 | static const struct ofhooks default_ofhooks; | |
323 | ||
fa60c019 | 324 | static uint64_t pick_datapath_id(const struct ofproto *); |
064af421 | 325 | static uint64_t pick_fallback_dpid(void); |
76ce9432 | 326 | |
064af421 | 327 | static void update_used(struct ofproto *); |
0193b2af JG |
328 | static void update_stats(struct ofproto *, struct rule *, |
329 | const struct odp_flow_stats *); | |
064af421 | 330 | static void expire_rule(struct cls_rule *, void *ofproto); |
0193b2af | 331 | static void active_timeout(struct ofproto *ofproto, struct rule *rule); |
064af421 BP |
332 | static bool revalidate_rule(struct ofproto *p, struct rule *rule); |
333 | static void revalidate_cb(struct cls_rule *rule_, void *p_); | |
334 | ||
335 | static void handle_odp_msg(struct ofproto *, struct ofpbuf *); | |
336 | ||
337 | static void handle_openflow(struct ofconn *, struct ofproto *, | |
338 | struct ofpbuf *); | |
339 | ||
72b06300 BP |
340 | static void refresh_port_groups(struct ofproto *); |
341 | ||
ca0f572c | 342 | static struct ofport *get_port(const struct ofproto *, uint16_t odp_port); |
064af421 BP |
343 | static void update_port(struct ofproto *, const char *devname); |
344 | static int init_ports(struct ofproto *); | |
345 | static void reinit_ports(struct ofproto *); | |
346 | ||
347 | int | |
1a6f1e2a JG |
348 | ofproto_create(const char *datapath, const char *datapath_type, |
349 | const struct ofhooks *ofhooks, void *aux, | |
064af421 BP |
350 | struct ofproto **ofprotop) |
351 | { | |
064af421 BP |
352 | struct odp_stats stats; |
353 | struct ofproto *p; | |
c228a364 | 354 | struct dpif *dpif; |
064af421 BP |
355 | int error; |
356 | ||
357 | *ofprotop = NULL; | |
358 | ||
359 | /* Connect to datapath and start listening for messages. */ | |
1a6f1e2a | 360 | error = dpif_open(datapath, datapath_type, &dpif); |
064af421 BP |
361 | if (error) { |
362 | VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error)); | |
363 | return error; | |
364 | } | |
c228a364 | 365 | error = dpif_get_dp_stats(dpif, &stats); |
064af421 BP |
366 | if (error) { |
367 | VLOG_ERR("failed to obtain stats for datapath %s: %s", | |
368 | datapath, strerror(error)); | |
c228a364 | 369 | dpif_close(dpif); |
064af421 BP |
370 | return error; |
371 | } | |
72b06300 | 372 | error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW); |
064af421 BP |
373 | if (error) { |
374 | VLOG_ERR("failed to listen on datapath %s: %s", | |
375 | datapath, strerror(error)); | |
c228a364 | 376 | dpif_close(dpif); |
064af421 BP |
377 | return error; |
378 | } | |
c228a364 | 379 | dpif_flow_flush(dpif); |
8f24562a | 380 | dpif_recv_purge(dpif); |
064af421 BP |
381 | |
382 | /* Initialize settings. */ | |
ec6fde61 | 383 | p = xzalloc(sizeof *p); |
064af421 | 384 | p->fallback_dpid = pick_fallback_dpid(); |
fa60c019 | 385 | p->datapath_id = p->fallback_dpid; |
5a719c38 JP |
386 | p->mfr_desc = xstrdup(DEFAULT_MFR_DESC); |
387 | p->hw_desc = xstrdup(DEFAULT_HW_DESC); | |
388 | p->sw_desc = xstrdup(DEFAULT_SW_DESC); | |
389 | p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC); | |
23ff2821 | 390 | p->dp_desc = xstrdup(DEFAULT_DP_DESC); |
064af421 BP |
391 | |
392 | /* Initialize datapath. */ | |
393 | p->dpif = dpif; | |
8b61709d | 394 | p->netdev_monitor = netdev_monitor_create(); |
ca0f572c | 395 | hmap_init(&p->ports); |
064af421 BP |
396 | shash_init(&p->port_by_name); |
397 | p->max_ports = stats.max_ports; | |
398 | ||
399 | /* Initialize submodules. */ | |
400 | p->switch_status = switch_status_create(p); | |
401 | p->in_band = NULL; | |
064af421 | 402 | p->fail_open = NULL; |
064af421 | 403 | p->netflow = NULL; |
72b06300 | 404 | p->sflow = NULL; |
064af421 BP |
405 | |
406 | /* Initialize flow table. */ | |
407 | classifier_init(&p->cls); | |
408 | p->need_revalidate = false; | |
409 | p->next_expiration = time_msec() + 1000; | |
410 | tag_set_init(&p->revalidate_set); | |
411 | ||
412 | /* Initialize OpenFlow connections. */ | |
413 | list_init(&p->all_conns); | |
76ce9432 | 414 | hmap_init(&p->controllers); |
7d674866 | 415 | hmap_init(&p->services); |
064af421 BP |
416 | p->snoops = NULL; |
417 | p->n_snoops = 0; | |
418 | ||
419 | /* Initialize hooks. */ | |
420 | if (ofhooks) { | |
421 | p->ofhooks = ofhooks; | |
422 | p->aux = aux; | |
423 | p->ml = NULL; | |
424 | } else { | |
425 | p->ofhooks = &default_ofhooks; | |
426 | p->aux = p; | |
427 | p->ml = mac_learning_create(); | |
428 | } | |
429 | ||
fa60c019 BP |
430 | /* Pick final datapath ID. */ |
431 | p->datapath_id = pick_datapath_id(p); | |
b123cc3c | 432 | VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id); |
fa60c019 | 433 | |
064af421 BP |
434 | *ofprotop = p; |
435 | return 0; | |
436 | } | |
437 | ||
438 | void | |
439 | ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id) | |
440 | { | |
441 | uint64_t old_dpid = p->datapath_id; | |
fa60c019 | 442 | p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p); |
064af421 | 443 | if (p->datapath_id != old_dpid) { |
b123cc3c | 444 | VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id); |
76ce9432 BP |
445 | |
446 | /* Force all active connections to reconnect, since there is no way to | |
447 | * notify a controller that the datapath ID has changed. */ | |
fa05809b | 448 | ofproto_reconnect_controllers(p); |
064af421 BP |
449 | } |
450 | } | |
451 | ||
76ce9432 BP |
452 | static bool |
453 | is_discovery_controller(const struct ofproto_controller *c) | |
454 | { | |
455 | return !strcmp(c->target, "discover"); | |
456 | } | |
457 | ||
458 | static bool | |
459 | is_in_band_controller(const struct ofproto_controller *c) | |
460 | { | |
461 | return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND; | |
462 | } | |
463 | ||
464 | /* Creates a new controller in 'ofproto'. Some of the settings are initially | |
465 | * drawn from 'c', but update_controller() needs to be called later to finish | |
466 | * the new ofconn's configuration. */ | |
467 | static void | |
468 | add_controller(struct ofproto *ofproto, const struct ofproto_controller *c) | |
469 | { | |
470 | struct discovery *discovery; | |
471 | struct ofconn *ofconn; | |
472 | ||
473 | if (is_discovery_controller(c)) { | |
474 | int error = discovery_create(c->accept_re, c->update_resolv_conf, | |
475 | ofproto->dpif, ofproto->switch_status, | |
476 | &discovery); | |
477 | if (error) { | |
478 | return; | |
479 | } | |
480 | } else { | |
481 | discovery = NULL; | |
482 | } | |
483 | ||
5899143f | 484 | ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY); |
76ce9432 BP |
485 | ofconn->pktbuf = pktbuf_create(); |
486 | ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN; | |
487 | if (discovery) { | |
488 | ofconn->discovery = discovery; | |
489 | } else { | |
eb15cdbb BP |
490 | char *name = ofconn_make_name(ofproto, c->target); |
491 | rconn_connect(ofconn->rconn, c->target, name); | |
492 | free(name); | |
76ce9432 BP |
493 | } |
494 | hmap_insert(&ofproto->controllers, &ofconn->hmap_node, | |
495 | hash_string(c->target, 0)); | |
496 | } | |
497 | ||
498 | /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's | |
499 | * target or turn discovery on or off (these are done by creating new ofconns | |
500 | * and deleting old ones), but it can update the rest of an ofconn's | |
501 | * settings. */ | |
502 | static void | |
503 | update_controller(struct ofconn *ofconn, const struct ofproto_controller *c) | |
064af421 | 504 | { |
76ce9432 | 505 | int probe_interval; |
79c9f2ee | 506 | |
d2ede7bc BP |
507 | ofconn->band = (is_in_band_controller(c) |
508 | ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND); | |
509 | ||
76ce9432 | 510 | rconn_set_max_backoff(ofconn->rconn, c->max_backoff); |
79c9f2ee | 511 | |
76ce9432 BP |
512 | probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0; |
513 | rconn_set_probe_interval(ofconn->rconn, probe_interval); | |
79c9f2ee | 514 | |
76ce9432 BP |
515 | if (ofconn->discovery) { |
516 | discovery_set_update_resolv_conf(ofconn->discovery, | |
517 | c->update_resolv_conf); | |
518 | discovery_set_accept_controller_re(ofconn->discovery, c->accept_re); | |
519 | } | |
79c9f2ee | 520 | |
7d674866 | 521 | ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit); |
76ce9432 | 522 | } |
79c9f2ee | 523 | |
76ce9432 BP |
524 | static const char * |
525 | ofconn_get_target(const struct ofconn *ofconn) | |
526 | { | |
eb15cdbb | 527 | return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn); |
76ce9432 BP |
528 | } |
529 | ||
530 | static struct ofconn * | |
531 | find_controller_by_target(struct ofproto *ofproto, const char *target) | |
532 | { | |
533 | struct ofconn *ofconn; | |
534 | ||
4e8e4213 | 535 | HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node, |
76ce9432 BP |
536 | hash_string(target, 0), &ofproto->controllers) { |
537 | if (!strcmp(ofconn_get_target(ofconn), target)) { | |
538 | return ofconn; | |
79c9f2ee | 539 | } |
064af421 | 540 | } |
76ce9432 BP |
541 | return NULL; |
542 | } | |
064af421 | 543 | |
d2ede7bc BP |
544 | static void |
545 | update_in_band_remotes(struct ofproto *ofproto) | |
546 | { | |
547 | const struct ofconn *ofconn; | |
548 | struct sockaddr_in *addrs; | |
917e50e1 | 549 | size_t max_addrs, n_addrs; |
d2ede7bc | 550 | bool discovery; |
917e50e1 | 551 | size_t i; |
d2ede7bc | 552 | |
917e50e1 BP |
553 | /* Allocate enough memory for as many remotes as we could possibly have. */ |
554 | max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers); | |
555 | addrs = xmalloc(max_addrs * sizeof *addrs); | |
d2ede7bc BP |
556 | n_addrs = 0; |
557 | ||
558 | /* Add all the remotes. */ | |
559 | discovery = false; | |
4e8e4213 | 560 | HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) { |
d2ede7bc BP |
561 | struct sockaddr_in *sin = &addrs[n_addrs]; |
562 | ||
487ec65f BP |
563 | if (ofconn->band == OFPROTO_OUT_OF_BAND) { |
564 | continue; | |
565 | } | |
566 | ||
d2ede7bc BP |
567 | sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn); |
568 | if (sin->sin_addr.s_addr) { | |
569 | sin->sin_port = rconn_get_remote_port(ofconn->rconn); | |
570 | n_addrs++; | |
571 | } | |
572 | if (ofconn->discovery) { | |
573 | discovery = true; | |
574 | } | |
575 | } | |
917e50e1 BP |
576 | for (i = 0; i < ofproto->n_extra_remotes; i++) { |
577 | addrs[n_addrs++] = ofproto->extra_in_band_remotes[i]; | |
578 | } | |
d2ede7bc BP |
579 | |
580 | /* Create or update or destroy in-band. | |
581 | * | |
582 | * Ordinarily we only enable in-band if there's at least one remote | |
583 | * address, but discovery needs the in-band rules for DHCP to be installed | |
584 | * even before we know any remote addresses. */ | |
585 | if (n_addrs || discovery) { | |
586 | if (!ofproto->in_band) { | |
587 | in_band_create(ofproto, ofproto->dpif, ofproto->switch_status, | |
588 | &ofproto->in_band); | |
589 | } | |
40cae670 BP |
590 | if (ofproto->in_band) { |
591 | in_band_set_remotes(ofproto->in_band, addrs, n_addrs); | |
592 | } | |
d2ede7bc BP |
593 | ofproto->next_in_band_update = time_msec() + 1000; |
594 | } else { | |
595 | in_band_destroy(ofproto->in_band); | |
596 | ofproto->in_band = NULL; | |
597 | } | |
598 | ||
599 | /* Clean up. */ | |
600 | free(addrs); | |
601 | } | |
602 | ||
31681a5d JP |
603 | static void |
604 | update_fail_open(struct ofproto *p) | |
605 | { | |
606 | struct ofconn *ofconn; | |
607 | ||
608 | if (!hmap_is_empty(&p->controllers) | |
609 | && p->fail_mode == OFPROTO_FAIL_STANDALONE) { | |
610 | struct rconn **rconns; | |
611 | size_t n; | |
612 | ||
613 | if (!p->fail_open) { | |
614 | p->fail_open = fail_open_create(p, p->switch_status); | |
615 | } | |
616 | ||
617 | n = 0; | |
618 | rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns); | |
4e8e4213 | 619 | HMAP_FOR_EACH (ofconn, hmap_node, &p->controllers) { |
31681a5d JP |
620 | rconns[n++] = ofconn->rconn; |
621 | } | |
622 | ||
623 | fail_open_set_controllers(p->fail_open, rconns, n); | |
624 | /* p->fail_open takes ownership of 'rconns'. */ | |
625 | } else { | |
626 | fail_open_destroy(p->fail_open); | |
627 | p->fail_open = NULL; | |
628 | } | |
629 | } | |
630 | ||
76ce9432 BP |
631 | void |
632 | ofproto_set_controllers(struct ofproto *p, | |
633 | const struct ofproto_controller *controllers, | |
634 | size_t n_controllers) | |
635 | { | |
636 | struct shash new_controllers; | |
7d674866 BP |
637 | struct ofconn *ofconn, *next_ofconn; |
638 | struct ofservice *ofservice, *next_ofservice; | |
76ce9432 | 639 | bool ss_exists; |
76ce9432 | 640 | size_t i; |
79c9f2ee | 641 | |
7d674866 BP |
642 | /* Create newly configured controllers and services. |
643 | * Create a name to ofproto_controller mapping in 'new_controllers'. */ | |
76ce9432 BP |
644 | shash_init(&new_controllers); |
645 | for (i = 0; i < n_controllers; i++) { | |
646 | const struct ofproto_controller *c = &controllers[i]; | |
647 | ||
7d674866 BP |
648 | if (!vconn_verify_name(c->target) || !strcmp(c->target, "discover")) { |
649 | if (!find_controller_by_target(p, c->target)) { | |
650 | add_controller(p, c); | |
651 | } | |
652 | } else if (!pvconn_verify_name(c->target)) { | |
653 | if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) { | |
654 | continue; | |
655 | } | |
656 | } else { | |
657 | VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"", | |
658 | dpif_name(p->dpif), c->target); | |
659 | continue; | |
76ce9432 | 660 | } |
7d674866 BP |
661 | |
662 | shash_add_once(&new_controllers, c->target, &controllers[i]); | |
76ce9432 BP |
663 | } |
664 | ||
7d674866 BP |
665 | /* Delete controllers that are no longer configured. |
666 | * Update configuration of all now-existing controllers. */ | |
76ce9432 | 667 | ss_exists = false; |
4e8e4213 | 668 | HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &p->controllers) { |
76ce9432 BP |
669 | struct ofproto_controller *c; |
670 | ||
671 | c = shash_find_data(&new_controllers, ofconn_get_target(ofconn)); | |
672 | if (!c) { | |
673 | ofconn_destroy(ofconn); | |
79c9f2ee | 674 | } else { |
76ce9432 | 675 | update_controller(ofconn, c); |
76ce9432 BP |
676 | if (ofconn->ss) { |
677 | ss_exists = true; | |
678 | } | |
76ce9432 BP |
679 | } |
680 | } | |
7d674866 BP |
681 | |
682 | /* Delete services that are no longer configured. | |
683 | * Update configuration of all now-existing services. */ | |
4e8e4213 | 684 | HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) { |
7d674866 BP |
685 | struct ofproto_controller *c; |
686 | ||
687 | c = shash_find_data(&new_controllers, | |
688 | pvconn_get_name(ofservice->pvconn)); | |
689 | if (!c) { | |
690 | ofservice_destroy(p, ofservice); | |
691 | } else { | |
692 | ofservice_reconfigure(ofservice, c); | |
693 | } | |
694 | } | |
695 | ||
76ce9432 BP |
696 | shash_destroy(&new_controllers); |
697 | ||
d2ede7bc | 698 | update_in_band_remotes(p); |
31681a5d | 699 | update_fail_open(p); |
79c9f2ee | 700 | |
76ce9432 BP |
701 | if (!hmap_is_empty(&p->controllers) && !ss_exists) { |
702 | ofconn = CONTAINER_OF(hmap_first(&p->controllers), | |
703 | struct ofconn, hmap_node); | |
704 | ofconn->ss = switch_status_register(p->switch_status, "remote", | |
705 | rconn_status_cb, ofconn->rconn); | |
79c9f2ee | 706 | } |
064af421 BP |
707 | } |
708 | ||
31681a5d JP |
709 | void |
710 | ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode) | |
711 | { | |
712 | p->fail_mode = fail_mode; | |
713 | update_fail_open(p); | |
714 | } | |
715 | ||
fa05809b BP |
716 | /* Drops the connections between 'ofproto' and all of its controllers, forcing |
717 | * them to reconnect. */ | |
718 | void | |
719 | ofproto_reconnect_controllers(struct ofproto *ofproto) | |
720 | { | |
721 | struct ofconn *ofconn; | |
722 | ||
4e8e4213 | 723 | LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) { |
fa05809b BP |
724 | rconn_reconnect(ofconn->rconn); |
725 | } | |
726 | } | |
727 | ||
917e50e1 BP |
728 | static bool |
729 | any_extras_changed(const struct ofproto *ofproto, | |
730 | const struct sockaddr_in *extras, size_t n) | |
731 | { | |
732 | size_t i; | |
733 | ||
734 | if (n != ofproto->n_extra_remotes) { | |
735 | return true; | |
736 | } | |
737 | ||
738 | for (i = 0; i < n; i++) { | |
739 | const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i]; | |
740 | const struct sockaddr_in *new = &extras[i]; | |
741 | ||
742 | if (old->sin_addr.s_addr != new->sin_addr.s_addr || | |
743 | old->sin_port != new->sin_port) { | |
744 | return true; | |
745 | } | |
746 | } | |
747 | ||
748 | return false; | |
749 | } | |
750 | ||
751 | /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s | |
752 | * in-band control should guarantee access, in the same way that in-band | |
753 | * control guarantees access to OpenFlow controllers. */ | |
754 | void | |
755 | ofproto_set_extra_in_band_remotes(struct ofproto *ofproto, | |
756 | const struct sockaddr_in *extras, size_t n) | |
757 | { | |
758 | if (!any_extras_changed(ofproto, extras, n)) { | |
759 | return; | |
760 | } | |
761 | ||
762 | free(ofproto->extra_in_band_remotes); | |
763 | ofproto->n_extra_remotes = n; | |
764 | ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras); | |
765 | ||
766 | update_in_band_remotes(ofproto); | |
767 | } | |
768 | ||
064af421 BP |
769 | void |
770 | ofproto_set_desc(struct ofproto *p, | |
5a719c38 JP |
771 | const char *mfr_desc, const char *hw_desc, |
772 | const char *sw_desc, const char *serial_desc, | |
8abc4ed7 | 773 | const char *dp_desc) |
064af421 | 774 | { |
5a719c38 JP |
775 | struct ofp_desc_stats *ods; |
776 | ||
777 | if (mfr_desc) { | |
778 | if (strlen(mfr_desc) >= sizeof ods->mfr_desc) { | |
779 | VLOG_WARN("truncating mfr_desc, must be less than %zu characters", | |
780 | sizeof ods->mfr_desc); | |
781 | } | |
782 | free(p->mfr_desc); | |
783 | p->mfr_desc = xstrdup(mfr_desc); | |
064af421 | 784 | } |
5a719c38 JP |
785 | if (hw_desc) { |
786 | if (strlen(hw_desc) >= sizeof ods->hw_desc) { | |
787 | VLOG_WARN("truncating hw_desc, must be less than %zu characters", | |
788 | sizeof ods->hw_desc); | |
789 | } | |
790 | free(p->hw_desc); | |
791 | p->hw_desc = xstrdup(hw_desc); | |
064af421 | 792 | } |
5a719c38 JP |
793 | if (sw_desc) { |
794 | if (strlen(sw_desc) >= sizeof ods->sw_desc) { | |
795 | VLOG_WARN("truncating sw_desc, must be less than %zu characters", | |
796 | sizeof ods->sw_desc); | |
797 | } | |
798 | free(p->sw_desc); | |
799 | p->sw_desc = xstrdup(sw_desc); | |
800 | } | |
801 | if (serial_desc) { | |
802 | if (strlen(serial_desc) >= sizeof ods->serial_num) { | |
803 | VLOG_WARN("truncating serial_desc, must be less than %zu " | |
804 | "characters", | |
805 | sizeof ods->serial_num); | |
806 | } | |
807 | free(p->serial_desc); | |
808 | p->serial_desc = xstrdup(serial_desc); | |
064af421 | 809 | } |
8abc4ed7 | 810 | if (dp_desc) { |
5a719c38 JP |
811 | if (strlen(dp_desc) >= sizeof ods->dp_desc) { |
812 | VLOG_WARN("truncating dp_desc, must be less than %zu characters", | |
813 | sizeof ods->dp_desc); | |
814 | } | |
8abc4ed7 JP |
815 | free(p->dp_desc); |
816 | p->dp_desc = xstrdup(dp_desc); | |
817 | } | |
064af421 BP |
818 | } |
819 | ||
064af421 BP |
820 | static int |
821 | set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp, | |
822 | const struct svec *svec) | |
823 | { | |
824 | struct pvconn **pvconns = *pvconnsp; | |
825 | size_t n_pvconns = *n_pvconnsp; | |
826 | int retval = 0; | |
827 | size_t i; | |
828 | ||
829 | for (i = 0; i < n_pvconns; i++) { | |
830 | pvconn_close(pvconns[i]); | |
831 | } | |
832 | free(pvconns); | |
833 | ||
834 | pvconns = xmalloc(svec->n * sizeof *pvconns); | |
835 | n_pvconns = 0; | |
836 | for (i = 0; i < svec->n; i++) { | |
837 | const char *name = svec->names[i]; | |
838 | struct pvconn *pvconn; | |
839 | int error; | |
840 | ||
841 | error = pvconn_open(name, &pvconn); | |
842 | if (!error) { | |
843 | pvconns[n_pvconns++] = pvconn; | |
844 | } else { | |
845 | VLOG_ERR("failed to listen on %s: %s", name, strerror(error)); | |
846 | if (!retval) { | |
847 | retval = error; | |
848 | } | |
849 | } | |
850 | } | |
851 | ||
852 | *pvconnsp = pvconns; | |
853 | *n_pvconnsp = n_pvconns; | |
854 | ||
855 | return retval; | |
856 | } | |
857 | ||
064af421 BP |
858 | int |
859 | ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops) | |
860 | { | |
861 | return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops); | |
862 | } | |
863 | ||
864 | int | |
0193b2af JG |
865 | ofproto_set_netflow(struct ofproto *ofproto, |
866 | const struct netflow_options *nf_options) | |
064af421 | 867 | { |
76343538 | 868 | if (nf_options && nf_options->collectors.n) { |
064af421 BP |
869 | if (!ofproto->netflow) { |
870 | ofproto->netflow = netflow_create(); | |
871 | } | |
0193b2af | 872 | return netflow_set_options(ofproto->netflow, nf_options); |
064af421 BP |
873 | } else { |
874 | netflow_destroy(ofproto->netflow); | |
875 | ofproto->netflow = NULL; | |
876 | return 0; | |
877 | } | |
878 | } | |
879 | ||
72b06300 BP |
880 | void |
881 | ofproto_set_sflow(struct ofproto *ofproto, | |
882 | const struct ofproto_sflow_options *oso) | |
883 | { | |
884 | struct ofproto_sflow *os = ofproto->sflow; | |
885 | if (oso) { | |
886 | if (!os) { | |
887 | struct ofport *ofport; | |
72b06300 BP |
888 | |
889 | os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif); | |
890 | refresh_port_groups(ofproto); | |
4e8e4213 | 891 | HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) { |
ca0f572c | 892 | ofproto_sflow_add_port(os, ofport->odp_port, |
72b06300 BP |
893 | netdev_get_name(ofport->netdev)); |
894 | } | |
895 | } | |
896 | ofproto_sflow_set_options(os, oso); | |
897 | } else { | |
898 | ofproto_sflow_destroy(os); | |
899 | ofproto->sflow = NULL; | |
900 | } | |
901 | } | |
902 | ||
064af421 BP |
903 | uint64_t |
904 | ofproto_get_datapath_id(const struct ofproto *ofproto) | |
905 | { | |
906 | return ofproto->datapath_id; | |
907 | } | |
908 | ||
76ce9432 | 909 | bool |
7d674866 | 910 | ofproto_has_primary_controller(const struct ofproto *ofproto) |
064af421 | 911 | { |
76ce9432 | 912 | return !hmap_is_empty(&ofproto->controllers); |
064af421 BP |
913 | } |
914 | ||
abdfe474 JP |
915 | enum ofproto_fail_mode |
916 | ofproto_get_fail_mode(const struct ofproto *p) | |
917 | { | |
918 | return p->fail_mode; | |
919 | } | |
920 | ||
064af421 BP |
921 | void |
922 | ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops) | |
923 | { | |
924 | size_t i; | |
925 | ||
926 | for (i = 0; i < ofproto->n_snoops; i++) { | |
927 | svec_add(snoops, pvconn_get_name(ofproto->snoops[i])); | |
928 | } | |
929 | } | |
930 | ||
931 | void | |
932 | ofproto_destroy(struct ofproto *p) | |
933 | { | |
7d674866 | 934 | struct ofservice *ofservice, *next_ofservice; |
064af421 | 935 | struct ofconn *ofconn, *next_ofconn; |
ca0f572c | 936 | struct ofport *ofport, *next_ofport; |
064af421 BP |
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 | ||
4e8e4213 | 954 | LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) { |
c475ae67 | 955 | ofconn_destroy(ofconn); |
064af421 | 956 | } |
76ce9432 | 957 | hmap_destroy(&p->controllers); |
064af421 | 958 | |
c228a364 | 959 | dpif_close(p->dpif); |
e9e28be3 | 960 | netdev_monitor_destroy(p->netdev_monitor); |
4e8e4213 | 961 | HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) { |
ca0f572c | 962 | hmap_remove(&p->ports, &ofport->hmap_node); |
064af421 BP |
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 | |
4e8e4213 | 971 | HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) { |
7d674866 | 972 | ofservice_destroy(p, ofservice); |
064af421 | 973 | } |
7d674866 | 974 | hmap_destroy(&p->services); |
064af421 BP |
975 | |
976 | for (i = 0; i < p->n_snoops; i++) { | |
977 | pvconn_close(p->snoops[i]); | |
978 | } | |
979 | free(p->snoops); | |
980 | ||
981 | mac_learning_destroy(p->ml); | |
982 | ||
5a719c38 JP |
983 | free(p->mfr_desc); |
984 | free(p->hw_desc); | |
985 | free(p->sw_desc); | |
986 | free(p->serial_desc); | |
cb871ae0 JP |
987 | free(p->dp_desc); |
988 | ||
ca0f572c | 989 | hmap_destroy(&p->ports); |
3b917492 | 990 | |
064af421 BP |
991 | free(p); |
992 | } | |
993 | ||
994 | int | |
995 | ofproto_run(struct ofproto *p) | |
996 | { | |
997 | int error = ofproto_run1(p); | |
998 | if (!error) { | |
999 | error = ofproto_run2(p, false); | |
1000 | } | |
1001 | return error; | |
1002 | } | |
1003 | ||
e9e28be3 BP |
1004 | static void |
1005 | process_port_change(struct ofproto *ofproto, int error, char *devname) | |
1006 | { | |
1007 | if (error == ENOBUFS) { | |
1008 | reinit_ports(ofproto); | |
1009 | } else if (!error) { | |
1010 | update_port(ofproto, devname); | |
1011 | free(devname); | |
1012 | } | |
1013 | } | |
1014 | ||
e2bfacb6 BP |
1015 | /* Returns a "preference level" for snooping 'ofconn'. A higher return value |
1016 | * means that 'ofconn' is more interesting for monitoring than a lower return | |
1017 | * value. */ | |
1018 | static int | |
1019 | snoop_preference(const struct ofconn *ofconn) | |
1020 | { | |
1021 | switch (ofconn->role) { | |
1022 | case NX_ROLE_MASTER: | |
1023 | return 3; | |
1024 | case NX_ROLE_OTHER: | |
1025 | return 2; | |
1026 | case NX_ROLE_SLAVE: | |
1027 | return 1; | |
1028 | default: | |
1029 | /* Shouldn't happen. */ | |
1030 | return 0; | |
1031 | } | |
1032 | } | |
1033 | ||
76ce9432 BP |
1034 | /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'. |
1035 | * Connects this vconn to a controller. */ | |
1036 | static void | |
1037 | add_snooper(struct ofproto *ofproto, struct vconn *vconn) | |
1038 | { | |
e2bfacb6 | 1039 | struct ofconn *ofconn, *best; |
76ce9432 | 1040 | |
e2bfacb6 BP |
1041 | /* Pick a controller for monitoring. */ |
1042 | best = NULL; | |
4e8e4213 | 1043 | LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) { |
5899143f | 1044 | if (ofconn->type == OFCONN_PRIMARY |
e2bfacb6 BP |
1045 | && (!best || snoop_preference(ofconn) > snoop_preference(best))) { |
1046 | best = ofconn; | |
76ce9432 | 1047 | } |
e2bfacb6 | 1048 | } |
76ce9432 | 1049 | |
e2bfacb6 BP |
1050 | if (best) { |
1051 | rconn_add_monitor(best->rconn, vconn); | |
1052 | } else { | |
1053 | VLOG_INFO_RL(&rl, "no controller connection to snoop"); | |
1054 | vconn_close(vconn); | |
76ce9432 | 1055 | } |
76ce9432 BP |
1056 | } |
1057 | ||
064af421 BP |
1058 | int |
1059 | ofproto_run1(struct ofproto *p) | |
1060 | { | |
1061 | struct ofconn *ofconn, *next_ofconn; | |
7d674866 | 1062 | struct ofservice *ofservice; |
064af421 BP |
1063 | char *devname; |
1064 | int error; | |
1065 | int i; | |
1066 | ||
149f577a JG |
1067 | if (shash_is_empty(&p->port_by_name)) { |
1068 | init_ports(p); | |
1069 | } | |
1070 | ||
064af421 BP |
1071 | for (i = 0; i < 50; i++) { |
1072 | struct ofpbuf *buf; | |
064af421 | 1073 | |
c228a364 | 1074 | error = dpif_recv(p->dpif, &buf); |
064af421 BP |
1075 | if (error) { |
1076 | if (error == ENODEV) { | |
1077 | /* Someone destroyed the datapath behind our back. The caller | |
1078 | * better destroy us and give up, because we're just going to | |
1079 | * spin from here on out. */ | |
39a559f2 BP |
1080 | static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5); |
1081 | VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally", | |
c228a364 | 1082 | dpif_name(p->dpif)); |
064af421 BP |
1083 | return ENODEV; |
1084 | } | |
1085 | break; | |
1086 | } | |
1087 | ||
1088 | handle_odp_msg(p, buf); | |
1089 | } | |
1090 | ||
e9e28be3 BP |
1091 | while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) { |
1092 | process_port_change(p, error, devname); | |
1093 | } | |
1094 | while ((error = netdev_monitor_poll(p->netdev_monitor, | |
1095 | &devname)) != EAGAIN) { | |
1096 | process_port_change(p, error, devname); | |
064af421 BP |
1097 | } |
1098 | ||
1099 | if (p->in_band) { | |
d2ede7bc BP |
1100 | if (time_msec() >= p->next_in_band_update) { |
1101 | update_in_band_remotes(p); | |
1102 | } | |
064af421 BP |
1103 | in_band_run(p->in_band); |
1104 | } | |
064af421 | 1105 | |
4e8e4213 | 1106 | LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) { |
064af421 BP |
1107 | ofconn_run(ofconn, p); |
1108 | } | |
1109 | ||
7778bd15 BP |
1110 | /* Fail-open maintenance. Do this after processing the ofconns since |
1111 | * fail-open checks the status of the controller rconn. */ | |
1112 | if (p->fail_open) { | |
1113 | fail_open_run(p->fail_open); | |
1114 | } | |
1115 | ||
4e8e4213 | 1116 | HMAP_FOR_EACH (ofservice, node, &p->services) { |
064af421 BP |
1117 | struct vconn *vconn; |
1118 | int retval; | |
1119 | ||
7d674866 | 1120 | retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn); |
064af421 | 1121 | if (!retval) { |
9794e806 | 1122 | struct rconn *rconn; |
eb15cdbb | 1123 | char *name; |
9794e806 | 1124 | |
7d674866 | 1125 | rconn = rconn_create(ofservice->probe_interval, 0); |
eb15cdbb BP |
1126 | name = ofconn_make_name(p, vconn_get_name(vconn)); |
1127 | rconn_connect_unreliably(rconn, vconn, name); | |
1128 | free(name); | |
1129 | ||
7d674866 BP |
1130 | ofconn = ofconn_create(p, rconn, OFCONN_SERVICE); |
1131 | ofconn_set_rate_limit(ofconn, ofservice->rate_limit, | |
1132 | ofservice->burst_limit); | |
064af421 BP |
1133 | } else if (retval != EAGAIN) { |
1134 | VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval)); | |
1135 | } | |
1136 | } | |
1137 | ||
1138 | for (i = 0; i < p->n_snoops; i++) { | |
1139 | struct vconn *vconn; | |
1140 | int retval; | |
1141 | ||
1142 | retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn); | |
1143 | if (!retval) { | |
76ce9432 | 1144 | add_snooper(p, vconn); |
064af421 BP |
1145 | } else if (retval != EAGAIN) { |
1146 | VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval)); | |
1147 | } | |
1148 | } | |
1149 | ||
1150 | if (time_msec() >= p->next_expiration) { | |
1151 | COVERAGE_INC(ofproto_expiration); | |
1152 | p->next_expiration = time_msec() + 1000; | |
1153 | update_used(p); | |
1154 | ||
1155 | classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p); | |
1156 | ||
1157 | /* Let the hook know that we're at a stable point: all outstanding data | |
1158 | * in existing flows has been accounted to the account_cb. Thus, the | |
1159 | * hook can now reasonably do operations that depend on having accurate | |
1160 | * flow volume accounting (currently, that's just bond rebalancing). */ | |
1161 | if (p->ofhooks->account_checkpoint_cb) { | |
1162 | p->ofhooks->account_checkpoint_cb(p->aux); | |
1163 | } | |
1164 | } | |
1165 | ||
1166 | if (p->netflow) { | |
1167 | netflow_run(p->netflow); | |
1168 | } | |
72b06300 BP |
1169 | if (p->sflow) { |
1170 | ofproto_sflow_run(p->sflow); | |
1171 | } | |
064af421 BP |
1172 | |
1173 | return 0; | |
1174 | } | |
1175 | ||
1176 | struct revalidate_cbdata { | |
1177 | struct ofproto *ofproto; | |
1178 | bool revalidate_all; /* Revalidate all exact-match rules? */ | |
1179 | bool revalidate_subrules; /* Revalidate all exact-match subrules? */ | |
1180 | struct tag_set revalidate_set; /* Set of tags to revalidate. */ | |
1181 | }; | |
1182 | ||
1183 | int | |
1184 | ofproto_run2(struct ofproto *p, bool revalidate_all) | |
1185 | { | |
1186 | if (p->need_revalidate || revalidate_all | |
1187 | || !tag_set_is_empty(&p->revalidate_set)) { | |
1188 | struct revalidate_cbdata cbdata; | |
1189 | cbdata.ofproto = p; | |
1190 | cbdata.revalidate_all = revalidate_all; | |
1191 | cbdata.revalidate_subrules = p->need_revalidate; | |
1192 | cbdata.revalidate_set = p->revalidate_set; | |
1193 | tag_set_init(&p->revalidate_set); | |
1194 | COVERAGE_INC(ofproto_revalidate); | |
1195 | classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata); | |
1196 | p->need_revalidate = false; | |
1197 | } | |
1198 | ||
1199 | return 0; | |
1200 | } | |
1201 | ||
1202 | void | |
1203 | ofproto_wait(struct ofproto *p) | |
1204 | { | |
7d674866 | 1205 | struct ofservice *ofservice; |
064af421 BP |
1206 | struct ofconn *ofconn; |
1207 | size_t i; | |
1208 | ||
c228a364 | 1209 | dpif_recv_wait(p->dpif); |
e9e28be3 BP |
1210 | dpif_port_poll_wait(p->dpif); |
1211 | netdev_monitor_poll_wait(p->netdev_monitor); | |
4e8e4213 | 1212 | LIST_FOR_EACH (ofconn, node, &p->all_conns) { |
064af421 BP |
1213 | ofconn_wait(ofconn); |
1214 | } | |
1215 | if (p->in_band) { | |
7cf8b266 | 1216 | poll_timer_wait_until(p->next_in_band_update); |
064af421 BP |
1217 | in_band_wait(p->in_band); |
1218 | } | |
064af421 BP |
1219 | if (p->fail_open) { |
1220 | fail_open_wait(p->fail_open); | |
1221 | } | |
72b06300 BP |
1222 | if (p->sflow) { |
1223 | ofproto_sflow_wait(p->sflow); | |
1224 | } | |
064af421 BP |
1225 | if (!tag_set_is_empty(&p->revalidate_set)) { |
1226 | poll_immediate_wake(); | |
1227 | } | |
1228 | if (p->need_revalidate) { | |
1229 | /* Shouldn't happen, but if it does just go around again. */ | |
1230 | VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); | |
1231 | poll_immediate_wake(); | |
1232 | } else if (p->next_expiration != LLONG_MAX) { | |
7cf8b266 | 1233 | poll_timer_wait_until(p->next_expiration); |
064af421 | 1234 | } |
4e8e4213 | 1235 | HMAP_FOR_EACH (ofservice, node, &p->services) { |
7d674866 | 1236 | pvconn_wait(ofservice->pvconn); |
064af421 BP |
1237 | } |
1238 | for (i = 0; i < p->n_snoops; i++) { | |
1239 | pvconn_wait(p->snoops[i]); | |
1240 | } | |
1241 | } | |
1242 | ||
1243 | void | |
1244 | ofproto_revalidate(struct ofproto *ofproto, tag_type tag) | |
1245 | { | |
1246 | tag_set_add(&ofproto->revalidate_set, tag); | |
1247 | } | |
1248 | ||
1249 | struct tag_set * | |
1250 | ofproto_get_revalidate_set(struct ofproto *ofproto) | |
1251 | { | |
1252 | return &ofproto->revalidate_set; | |
1253 | } | |
1254 | ||
1255 | bool | |
1256 | ofproto_is_alive(const struct ofproto *p) | |
1257 | { | |
76ce9432 | 1258 | return !hmap_is_empty(&p->controllers); |
064af421 BP |
1259 | } |
1260 | ||
1261 | int | |
1262 | ofproto_send_packet(struct ofproto *p, const flow_t *flow, | |
1263 | const union ofp_action *actions, size_t n_actions, | |
1264 | const struct ofpbuf *packet) | |
1265 | { | |
1266 | struct odp_actions odp_actions; | |
1267 | int error; | |
1268 | ||
1269 | error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions, | |
6a07af36 | 1270 | NULL, NULL, NULL); |
064af421 BP |
1271 | if (error) { |
1272 | return error; | |
1273 | } | |
1274 | ||
1275 | /* XXX Should we translate the dpif_execute() errno value into an OpenFlow | |
1276 | * error code? */ | |
c228a364 | 1277 | dpif_execute(p->dpif, flow->in_port, odp_actions.actions, |
064af421 BP |
1278 | odp_actions.n_actions, packet); |
1279 | return 0; | |
1280 | } | |
1281 | ||
1282 | void | |
1283 | ofproto_add_flow(struct ofproto *p, | |
1284 | const flow_t *flow, uint32_t wildcards, unsigned int priority, | |
1285 | const union ofp_action *actions, size_t n_actions, | |
1286 | int idle_timeout) | |
1287 | { | |
1288 | struct rule *rule; | |
0193b2af | 1289 | rule = rule_create(p, NULL, actions, n_actions, |
d295e8e9 | 1290 | idle_timeout >= 0 ? idle_timeout : 5 /* XXX */, |
39997502 | 1291 | 0, 0, false); |
659586ef | 1292 | cls_rule_from_flow(flow, wildcards, priority, &rule->cr); |
064af421 BP |
1293 | rule_insert(p, rule, NULL, 0); |
1294 | } | |
1295 | ||
1296 | void | |
1297 | ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow, | |
1298 | uint32_t wildcards, unsigned int priority) | |
1299 | { | |
1300 | struct rule *rule; | |
1301 | ||
1302 | rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls, | |
1303 | flow, wildcards, | |
1304 | priority)); | |
1305 | if (rule) { | |
1306 | rule_remove(ofproto, rule); | |
1307 | } | |
1308 | } | |
1309 | ||
1310 | static void | |
1311 | destroy_rule(struct cls_rule *rule_, void *ofproto_) | |
1312 | { | |
1313 | struct rule *rule = rule_from_cls_rule(rule_); | |
1314 | struct ofproto *ofproto = ofproto_; | |
1315 | ||
1316 | /* Mark the flow as not installed, even though it might really be | |
1317 | * installed, so that rule_remove() doesn't bother trying to uninstall it. | |
1318 | * There is no point in uninstalling it individually since we are about to | |
1319 | * blow away all the flows with dpif_flow_flush(). */ | |
1320 | rule->installed = false; | |
1321 | ||
1322 | rule_remove(ofproto, rule); | |
1323 | } | |
1324 | ||
1325 | void | |
1326 | ofproto_flush_flows(struct ofproto *ofproto) | |
1327 | { | |
1328 | COVERAGE_INC(ofproto_flush); | |
1329 | classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto); | |
c228a364 | 1330 | dpif_flow_flush(ofproto->dpif); |
064af421 BP |
1331 | if (ofproto->in_band) { |
1332 | in_band_flushed(ofproto->in_band); | |
1333 | } | |
1334 | if (ofproto->fail_open) { | |
1335 | fail_open_flushed(ofproto->fail_open); | |
1336 | } | |
1337 | } | |
1338 | \f | |
1339 | static void | |
1340 | reinit_ports(struct ofproto *p) | |
1341 | { | |
1342 | struct svec devnames; | |
1343 | struct ofport *ofport; | |
064af421 BP |
1344 | struct odp_port *odp_ports; |
1345 | size_t n_odp_ports; | |
1346 | size_t i; | |
1347 | ||
1348 | svec_init(&devnames); | |
4e8e4213 | 1349 | HMAP_FOR_EACH (ofport, hmap_node, &p->ports) { |
064af421 BP |
1350 | svec_add (&devnames, (char *) ofport->opp.name); |
1351 | } | |
c228a364 | 1352 | dpif_port_list(p->dpif, &odp_ports, &n_odp_ports); |
064af421 BP |
1353 | for (i = 0; i < n_odp_ports; i++) { |
1354 | svec_add (&devnames, odp_ports[i].devname); | |
1355 | } | |
1356 | free(odp_ports); | |
1357 | ||
1358 | svec_sort_unique(&devnames); | |
1359 | for (i = 0; i < devnames.n; i++) { | |
1360 | update_port(p, devnames.names[i]); | |
1361 | } | |
1362 | svec_destroy(&devnames); | |
1363 | } | |
1364 | ||
72b06300 | 1365 | static size_t |
064af421 BP |
1366 | refresh_port_group(struct ofproto *p, unsigned int group) |
1367 | { | |
1368 | uint16_t *ports; | |
1369 | size_t n_ports; | |
1370 | struct ofport *port; | |
064af421 BP |
1371 | |
1372 | assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD); | |
1373 | ||
ca0f572c | 1374 | ports = xmalloc(hmap_count(&p->ports) * sizeof *ports); |
064af421 | 1375 | n_ports = 0; |
4e8e4213 | 1376 | HMAP_FOR_EACH (port, hmap_node, &p->ports) { |
064af421 | 1377 | if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) { |
ca0f572c | 1378 | ports[n_ports++] = port->odp_port; |
064af421 BP |
1379 | } |
1380 | } | |
c228a364 | 1381 | dpif_port_group_set(p->dpif, group, ports, n_ports); |
064af421 | 1382 | free(ports); |
72b06300 BP |
1383 | |
1384 | return n_ports; | |
064af421 BP |
1385 | } |
1386 | ||
1387 | static void | |
1388 | refresh_port_groups(struct ofproto *p) | |
1389 | { | |
72b06300 BP |
1390 | size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD); |
1391 | size_t n_all = refresh_port_group(p, DP_GROUP_ALL); | |
1392 | if (p->sflow) { | |
1393 | ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all); | |
1394 | } | |
064af421 BP |
1395 | } |
1396 | ||
1397 | static struct ofport * | |
1398 | make_ofport(const struct odp_port *odp_port) | |
1399 | { | |
149f577a | 1400 | struct netdev_options netdev_options; |
064af421 BP |
1401 | enum netdev_flags flags; |
1402 | struct ofport *ofport; | |
1403 | struct netdev *netdev; | |
1404 | bool carrier; | |
1405 | int error; | |
1406 | ||
149f577a JG |
1407 | memset(&netdev_options, 0, sizeof netdev_options); |
1408 | netdev_options.name = odp_port->devname; | |
1409 | netdev_options.ethertype = NETDEV_ETH_TYPE_NONE; | |
149f577a JG |
1410 | |
1411 | error = netdev_open(&netdev_options, &netdev); | |
064af421 BP |
1412 | if (error) { |
1413 | VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s " | |
1414 | "cannot be opened (%s)", | |
1415 | odp_port->devname, odp_port->port, | |
1416 | odp_port->devname, strerror(error)); | |
1417 | return NULL; | |
1418 | } | |
1419 | ||
1420 | ofport = xmalloc(sizeof *ofport); | |
1421 | ofport->netdev = netdev; | |
ca0f572c | 1422 | ofport->odp_port = odp_port->port; |
064af421 | 1423 | ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port); |
80992a35 | 1424 | netdev_get_etheraddr(netdev, ofport->opp.hw_addr); |
064af421 BP |
1425 | memcpy(ofport->opp.name, odp_port->devname, |
1426 | MIN(sizeof ofport->opp.name, sizeof odp_port->devname)); | |
1427 | ofport->opp.name[sizeof ofport->opp.name - 1] = '\0'; | |
1428 | ||
1429 | netdev_get_flags(netdev, &flags); | |
1430 | ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN; | |
1431 | ||
1432 | netdev_get_carrier(netdev, &carrier); | |
1433 | ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN; | |
1434 | ||
1435 | netdev_get_features(netdev, | |
1436 | &ofport->opp.curr, &ofport->opp.advertised, | |
1437 | &ofport->opp.supported, &ofport->opp.peer); | |
1438 | return ofport; | |
1439 | } | |
1440 | ||
1441 | static bool | |
1442 | ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port) | |
1443 | { | |
ca0f572c | 1444 | if (get_port(p, odp_port->port)) { |
064af421 BP |
1445 | VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath", |
1446 | odp_port->port); | |
1447 | return true; | |
1448 | } else if (shash_find(&p->port_by_name, odp_port->devname)) { | |
1449 | VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath", | |
1450 | odp_port->devname); | |
1451 | return true; | |
1452 | } else { | |
1453 | return false; | |
1454 | } | |
1455 | } | |
1456 | ||
1457 | static int | |
1458 | ofport_equal(const struct ofport *a_, const struct ofport *b_) | |
1459 | { | |
1460 | const struct ofp_phy_port *a = &a_->opp; | |
1461 | const struct ofp_phy_port *b = &b_->opp; | |
1462 | ||
1463 | BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */ | |
1464 | return (a->port_no == b->port_no | |
1465 | && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr) | |
1466 | && !strcmp((char *) a->name, (char *) b->name) | |
1467 | && a->state == b->state | |
1468 | && a->config == b->config | |
1469 | && a->curr == b->curr | |
1470 | && a->advertised == b->advertised | |
1471 | && a->supported == b->supported | |
1472 | && a->peer == b->peer); | |
1473 | } | |
1474 | ||
1475 | static void | |
1476 | send_port_status(struct ofproto *p, const struct ofport *ofport, | |
1477 | uint8_t reason) | |
1478 | { | |
1479 | /* XXX Should limit the number of queued port status change messages. */ | |
1480 | struct ofconn *ofconn; | |
4e8e4213 | 1481 | LIST_FOR_EACH (ofconn, node, &p->all_conns) { |
064af421 BP |
1482 | struct ofp_port_status *ops; |
1483 | struct ofpbuf *b; | |
1484 | ||
c91248b3 | 1485 | if (!ofconn_receives_async_msgs(ofconn)) { |
9deba63b BP |
1486 | continue; |
1487 | } | |
1488 | ||
064af421 BP |
1489 | ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b); |
1490 | ops->reason = reason; | |
1491 | ops->desc = ofport->opp; | |
1492 | hton_ofp_phy_port(&ops->desc); | |
1493 | queue_tx(b, ofconn, NULL); | |
1494 | } | |
1495 | if (p->ofhooks->port_changed_cb) { | |
1496 | p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux); | |
1497 | } | |
1498 | } | |
1499 | ||
1500 | static void | |
1501 | ofport_install(struct ofproto *p, struct ofport *ofport) | |
1502 | { | |
72b06300 BP |
1503 | const char *netdev_name = (const char *) ofport->opp.name; |
1504 | ||
e9e28be3 | 1505 | netdev_monitor_add(p->netdev_monitor, ofport->netdev); |
ca0f572c | 1506 | hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0)); |
72b06300 BP |
1507 | shash_add(&p->port_by_name, netdev_name, ofport); |
1508 | if (p->sflow) { | |
ca0f572c | 1509 | ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name); |
72b06300 | 1510 | } |
064af421 BP |
1511 | } |
1512 | ||
1513 | static void | |
1514 | ofport_remove(struct ofproto *p, struct ofport *ofport) | |
1515 | { | |
e9e28be3 | 1516 | netdev_monitor_remove(p->netdev_monitor, ofport->netdev); |
ca0f572c | 1517 | hmap_remove(&p->ports, &ofport->hmap_node); |
064af421 BP |
1518 | shash_delete(&p->port_by_name, |
1519 | shash_find(&p->port_by_name, (char *) ofport->opp.name)); | |
72b06300 | 1520 | if (p->sflow) { |
ca0f572c | 1521 | ofproto_sflow_del_port(p->sflow, ofport->odp_port); |
72b06300 | 1522 | } |
064af421 BP |
1523 | } |
1524 | ||
1525 | static void | |
1526 | ofport_free(struct ofport *ofport) | |
1527 | { | |
1528 | if (ofport) { | |
1529 | netdev_close(ofport->netdev); | |
1530 | free(ofport); | |
1531 | } | |
1532 | } | |
1533 | ||
ca0f572c BP |
1534 | static struct ofport * |
1535 | get_port(const struct ofproto *ofproto, uint16_t odp_port) | |
1536 | { | |
1537 | struct ofport *port; | |
1538 | ||
4e8e4213 | 1539 | HMAP_FOR_EACH_IN_BUCKET (port, hmap_node, |
ca0f572c BP |
1540 | hash_int(odp_port, 0), &ofproto->ports) { |
1541 | if (port->odp_port == odp_port) { | |
1542 | return port; | |
1543 | } | |
1544 | } | |
1545 | return NULL; | |
1546 | } | |
1547 | ||
064af421 BP |
1548 | static void |
1549 | update_port(struct ofproto *p, const char *devname) | |
1550 | { | |
1551 | struct odp_port odp_port; | |
c874dc6d BP |
1552 | struct ofport *old_ofport; |
1553 | struct ofport *new_ofport; | |
064af421 BP |
1554 | int error; |
1555 | ||
1556 | COVERAGE_INC(ofproto_update_port); | |
c874dc6d BP |
1557 | |
1558 | /* Query the datapath for port information. */ | |
c228a364 | 1559 | error = dpif_port_query_by_name(p->dpif, devname, &odp_port); |
064af421 | 1560 | |
c874dc6d BP |
1561 | /* Find the old ofport. */ |
1562 | old_ofport = shash_find_data(&p->port_by_name, devname); | |
1563 | if (!error) { | |
1564 | if (!old_ofport) { | |
1565 | /* There's no port named 'devname' but there might be a port with | |
1566 | * the same port number. This could happen if a port is deleted | |
1567 | * and then a new one added in its place very quickly, or if a port | |
1568 | * is renamed. In the former case we want to send an OFPPR_DELETE | |
1569 | * and an OFPPR_ADD, and in the latter case we want to send a | |
1570 | * single OFPPR_MODIFY. We can distinguish the cases by comparing | |
1571 | * the old port's ifindex against the new port, or perhaps less | |
1572 | * reliably but more portably by comparing the old port's MAC | |
1573 | * against the new port's MAC. However, this code isn't that smart | |
1574 | * and always sends an OFPPR_MODIFY (XXX). */ | |
ca0f572c | 1575 | old_ofport = get_port(p, odp_port.port); |
064af421 | 1576 | } |
c874dc6d | 1577 | } else if (error != ENOENT && error != ENODEV) { |
064af421 BP |
1578 | VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error " |
1579 | "%s", strerror(error)); | |
1580 | return; | |
1581 | } | |
c874dc6d BP |
1582 | |
1583 | /* Create a new ofport. */ | |
1584 | new_ofport = !error ? make_ofport(&odp_port) : NULL; | |
1585 | ||
1586 | /* Eliminate a few pathological cases. */ | |
1587 | if (!old_ofport && !new_ofport) { | |
1588 | return; | |
1589 | } else if (old_ofport && new_ofport) { | |
1590 | /* Most of the 'config' bits are OpenFlow soft state, but | |
1591 | * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow | |
1592 | * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and | |
1593 | * leaves the other bits 0.) */ | |
1594 | new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN; | |
1595 | ||
1596 | if (ofport_equal(old_ofport, new_ofport)) { | |
1597 | /* False alarm--no change. */ | |
1598 | ofport_free(new_ofport); | |
1599 | return; | |
1600 | } | |
1601 | } | |
1602 | ||
1603 | /* Now deal with the normal cases. */ | |
1604 | if (old_ofport) { | |
1605 | ofport_remove(p, old_ofport); | |
1606 | } | |
1607 | if (new_ofport) { | |
1608 | ofport_install(p, new_ofport); | |
1609 | } | |
1610 | send_port_status(p, new_ofport ? new_ofport : old_ofport, | |
1611 | (!old_ofport ? OFPPR_ADD | |
1612 | : !new_ofport ? OFPPR_DELETE | |
1613 | : OFPPR_MODIFY)); | |
1614 | ofport_free(old_ofport); | |
1615 | ||
1616 | /* Update port groups. */ | |
064af421 BP |
1617 | refresh_port_groups(p); |
1618 | } | |
1619 | ||
1620 | static int | |
1621 | init_ports(struct ofproto *p) | |
1622 | { | |
1623 | struct odp_port *ports; | |
1624 | size_t n_ports; | |
1625 | size_t i; | |
1626 | int error; | |
1627 | ||
c228a364 | 1628 | error = dpif_port_list(p->dpif, &ports, &n_ports); |
064af421 BP |
1629 | if (error) { |
1630 | return error; | |
1631 | } | |
1632 | ||
1633 | for (i = 0; i < n_ports; i++) { | |
1634 | const struct odp_port *odp_port = &ports[i]; | |
1635 | if (!ofport_conflicts(p, odp_port)) { | |
1636 | struct ofport *ofport = make_ofport(odp_port); | |
1637 | if (ofport) { | |
1638 | ofport_install(p, ofport); | |
1639 | } | |
1640 | } | |
1641 | } | |
1642 | free(ports); | |
1643 | refresh_port_groups(p); | |
1644 | return 0; | |
1645 | } | |
1646 | \f | |
1647 | static struct ofconn * | |
76ce9432 | 1648 | ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type) |
064af421 | 1649 | { |
76ce9432 BP |
1650 | struct ofconn *ofconn = xzalloc(sizeof *ofconn); |
1651 | ofconn->ofproto = p; | |
064af421 BP |
1652 | list_push_back(&p->all_conns, &ofconn->node); |
1653 | ofconn->rconn = rconn; | |
76ce9432 | 1654 | ofconn->type = type; |
9deba63b | 1655 | ofconn->role = NX_ROLE_OTHER; |
76ce9432 | 1656 | ofconn->packet_in_counter = rconn_packet_counter_create (); |
064af421 | 1657 | ofconn->pktbuf = NULL; |
064af421 | 1658 | ofconn->miss_send_len = 0; |
064af421 BP |
1659 | ofconn->reply_counter = rconn_packet_counter_create (); |
1660 | return ofconn; | |
1661 | } | |
1662 | ||
1663 | static void | |
c475ae67 | 1664 | ofconn_destroy(struct ofconn *ofconn) |
064af421 | 1665 | { |
5899143f | 1666 | if (ofconn->type == OFCONN_PRIMARY) { |
76ce9432 BP |
1667 | hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node); |
1668 | } | |
1669 | discovery_destroy(ofconn->discovery); | |
1670 | ||
064af421 | 1671 | list_remove(&ofconn->node); |
76ce9432 | 1672 | switch_status_unregister(ofconn->ss); |
064af421 BP |
1673 | rconn_destroy(ofconn->rconn); |
1674 | rconn_packet_counter_destroy(ofconn->packet_in_counter); | |
1675 | rconn_packet_counter_destroy(ofconn->reply_counter); | |
1676 | pktbuf_destroy(ofconn->pktbuf); | |
1677 | free(ofconn); | |
1678 | } | |
1679 | ||
1680 | static void | |
1681 | ofconn_run(struct ofconn *ofconn, struct ofproto *p) | |
1682 | { | |
1683 | int iteration; | |
76ce9432 BP |
1684 | size_t i; |
1685 | ||
1686 | if (ofconn->discovery) { | |
1687 | char *controller_name; | |
1688 | if (rconn_is_connectivity_questionable(ofconn->rconn)) { | |
1689 | discovery_question_connectivity(ofconn->discovery); | |
1690 | } | |
1691 | if (discovery_run(ofconn->discovery, &controller_name)) { | |
1692 | if (controller_name) { | |
eb15cdbb BP |
1693 | char *ofconn_name = ofconn_make_name(p, controller_name); |
1694 | rconn_connect(ofconn->rconn, controller_name, ofconn_name); | |
1695 | free(ofconn_name); | |
76ce9432 BP |
1696 | } else { |
1697 | rconn_disconnect(ofconn->rconn); | |
1698 | } | |
1699 | } | |
1700 | } | |
1701 | ||
1702 | for (i = 0; i < N_SCHEDULERS; i++) { | |
1703 | pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn); | |
1704 | } | |
064af421 BP |
1705 | |
1706 | rconn_run(ofconn->rconn); | |
1707 | ||
1708 | if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) { | |
1709 | /* Limit the number of iterations to prevent other tasks from | |
1710 | * starving. */ | |
1711 | for (iteration = 0; iteration < 50; iteration++) { | |
1712 | struct ofpbuf *of_msg = rconn_recv(ofconn->rconn); | |
1713 | if (!of_msg) { | |
1714 | break; | |
1715 | } | |
7778bd15 BP |
1716 | if (p->fail_open) { |
1717 | fail_open_maybe_recover(p->fail_open); | |
1718 | } | |
064af421 BP |
1719 | handle_openflow(ofconn, p, of_msg); |
1720 | ofpbuf_delete(of_msg); | |
1721 | } | |
1722 | } | |
1723 | ||
76ce9432 | 1724 | if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) { |
c475ae67 | 1725 | ofconn_destroy(ofconn); |
064af421 BP |
1726 | } |
1727 | } | |
1728 | ||
1729 | static void | |
1730 | ofconn_wait(struct ofconn *ofconn) | |
1731 | { | |
76ce9432 BP |
1732 | int i; |
1733 | ||
1734 | if (ofconn->discovery) { | |
1735 | discovery_wait(ofconn->discovery); | |
1736 | } | |
1737 | for (i = 0; i < N_SCHEDULERS; i++) { | |
1738 | pinsched_wait(ofconn->schedulers[i]); | |
1739 | } | |
064af421 BP |
1740 | rconn_run_wait(ofconn->rconn); |
1741 | if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) { | |
1742 | rconn_recv_wait(ofconn->rconn); | |
1743 | } else { | |
1744 | COVERAGE_INC(ofproto_ofconn_stuck); | |
1745 | } | |
1746 | } | |
c91248b3 BP |
1747 | |
1748 | /* Returns true if 'ofconn' should receive asynchronous messages. */ | |
1749 | static bool | |
1750 | ofconn_receives_async_msgs(const struct ofconn *ofconn) | |
1751 | { | |
5899143f BP |
1752 | if (ofconn->type == OFCONN_PRIMARY) { |
1753 | /* Primary controllers always get asynchronous messages unless they | |
c91248b3 BP |
1754 | * have configured themselves as "slaves". */ |
1755 | return ofconn->role != NX_ROLE_SLAVE; | |
1756 | } else { | |
5899143f BP |
1757 | /* Service connections don't get asynchronous messages unless they have |
1758 | * explicitly asked for them by setting a nonzero miss send length. */ | |
c91248b3 BP |
1759 | return ofconn->miss_send_len > 0; |
1760 | } | |
1761 | } | |
eb15cdbb BP |
1762 | |
1763 | /* Returns a human-readable name for an OpenFlow connection between 'ofproto' | |
1764 | * and 'target', suitable for use in log messages for identifying the | |
1765 | * connection. | |
1766 | * | |
1767 | * The name is dynamically allocated. The caller should free it (with free()) | |
1768 | * when it is no longer needed. */ | |
1769 | static char * | |
1770 | ofconn_make_name(const struct ofproto *ofproto, const char *target) | |
1771 | { | |
1772 | return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target); | |
1773 | } | |
7d674866 BP |
1774 | |
1775 | static void | |
1776 | ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst) | |
1777 | { | |
1778 | int i; | |
1779 | ||
1780 | for (i = 0; i < N_SCHEDULERS; i++) { | |
1781 | struct pinsched **s = &ofconn->schedulers[i]; | |
1782 | ||
1783 | if (rate > 0) { | |
1784 | if (!*s) { | |
1785 | *s = pinsched_create(rate, burst, | |
1786 | ofconn->ofproto->switch_status); | |
1787 | } else { | |
1788 | pinsched_set_limits(*s, rate, burst); | |
1789 | } | |
1790 | } else { | |
1791 | pinsched_destroy(*s); | |
1792 | *s = NULL; | |
1793 | } | |
1794 | } | |
1795 | } | |
1796 | \f | |
1797 | static void | |
1798 | ofservice_reconfigure(struct ofservice *ofservice, | |
1799 | const struct ofproto_controller *c) | |
1800 | { | |
1801 | ofservice->probe_interval = c->probe_interval; | |
1802 | ofservice->rate_limit = c->rate_limit; | |
1803 | ofservice->burst_limit = c->burst_limit; | |
1804 | } | |
1805 | ||
1806 | /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a | |
1807 | * positive errno value. */ | |
1808 | static int | |
1809 | ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c) | |
1810 | { | |
1811 | struct ofservice *ofservice; | |
1812 | struct pvconn *pvconn; | |
1813 | int error; | |
1814 | ||
1815 | error = pvconn_open(c->target, &pvconn); | |
1816 | if (error) { | |
1817 | return error; | |
1818 | } | |
1819 | ||
1820 | ofservice = xzalloc(sizeof *ofservice); | |
1821 | hmap_insert(&ofproto->services, &ofservice->node, | |
1822 | hash_string(c->target, 0)); | |
1823 | ofservice->pvconn = pvconn; | |
1824 | ||
1825 | ofservice_reconfigure(ofservice, c); | |
1826 | ||
1827 | return 0; | |
1828 | } | |
1829 | ||
1830 | static void | |
1831 | ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice) | |
1832 | { | |
1833 | hmap_remove(&ofproto->services, &ofservice->node); | |
1834 | pvconn_close(ofservice->pvconn); | |
1835 | free(ofservice); | |
1836 | } | |
1837 | ||
1838 | /* Finds and returns the ofservice within 'ofproto' that has the given | |
1839 | * 'target', or a null pointer if none exists. */ | |
1840 | static struct ofservice * | |
1841 | ofservice_lookup(struct ofproto *ofproto, const char *target) | |
1842 | { | |
1843 | struct ofservice *ofservice; | |
1844 | ||
4e8e4213 BP |
1845 | HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0), |
1846 | &ofproto->services) { | |
7d674866 BP |
1847 | if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) { |
1848 | return ofservice; | |
1849 | } | |
1850 | } | |
1851 | return NULL; | |
1852 | } | |
064af421 BP |
1853 | \f |
1854 | /* Caller is responsible for initializing the 'cr' member of the returned | |
1855 | * rule. */ | |
1856 | static struct rule * | |
0193b2af | 1857 | rule_create(struct ofproto *ofproto, struct rule *super, |
064af421 | 1858 | const union ofp_action *actions, size_t n_actions, |
ca069229 | 1859 | uint16_t idle_timeout, uint16_t hard_timeout, |
39997502 | 1860 | uint64_t flow_cookie, bool send_flow_removed) |
064af421 | 1861 | { |
ec6fde61 | 1862 | struct rule *rule = xzalloc(sizeof *rule); |
064af421 BP |
1863 | rule->idle_timeout = idle_timeout; |
1864 | rule->hard_timeout = hard_timeout; | |
39997502 | 1865 | rule->flow_cookie = flow_cookie; |
064af421 | 1866 | rule->used = rule->created = time_msec(); |
ca069229 | 1867 | rule->send_flow_removed = send_flow_removed; |
064af421 BP |
1868 | rule->super = super; |
1869 | if (super) { | |
1870 | list_push_back(&super->list, &rule->list); | |
1871 | } else { | |
1872 | list_init(&rule->list); | |
1873 | } | |
1874 | rule->n_actions = n_actions; | |
1875 | rule->actions = xmemdup(actions, n_actions * sizeof *actions); | |
0193b2af JG |
1876 | netflow_flow_clear(&rule->nf_flow); |
1877 | netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created); | |
1878 | ||
064af421 BP |
1879 | return rule; |
1880 | } | |
1881 | ||
1882 | static struct rule * | |
1883 | rule_from_cls_rule(const struct cls_rule *cls_rule) | |
1884 | { | |
1885 | return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL; | |
1886 | } | |
1887 | ||
1888 | static void | |
1889 | rule_free(struct rule *rule) | |
1890 | { | |
1891 | free(rule->actions); | |
1892 | free(rule->odp_actions); | |
1893 | free(rule); | |
1894 | } | |
1895 | ||
1896 | /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its | |
1897 | * super-rule's list of subrules. If 'rule' is a super-rule, also iterates | |
1898 | * through all of its subrules and revalidates them, destroying any that no | |
1899 | * longer has a super-rule (which is probably all of them). | |
1900 | * | |
1901 | * Before calling this function, the caller must make have removed 'rule' from | |
1902 | * the classifier. If 'rule' is an exact-match rule, the caller is also | |
1903 | * responsible for ensuring that it has been uninstalled from the datapath. */ | |
1904 | static void | |
1905 | rule_destroy(struct ofproto *ofproto, struct rule *rule) | |
1906 | { | |
1907 | if (!rule->super) { | |
1908 | struct rule *subrule, *next; | |
4e8e4213 | 1909 | LIST_FOR_EACH_SAFE (subrule, next, list, &rule->list) { |
064af421 BP |
1910 | revalidate_rule(ofproto, subrule); |
1911 | } | |
1912 | } else { | |
1913 | list_remove(&rule->list); | |
1914 | } | |
1915 | rule_free(rule); | |
1916 | } | |
1917 | ||
1918 | static bool | |
1919 | rule_has_out_port(const struct rule *rule, uint16_t out_port) | |
1920 | { | |
1921 | const union ofp_action *oa; | |
1922 | struct actions_iterator i; | |
1923 | ||
1924 | if (out_port == htons(OFPP_NONE)) { | |
1925 | return true; | |
1926 | } | |
1927 | for (oa = actions_first(&i, rule->actions, rule->n_actions); oa; | |
1928 | oa = actions_next(&i)) { | |
c1c9c9c4 | 1929 | if (action_outputs_to_port(oa, out_port)) { |
064af421 BP |
1930 | return true; |
1931 | } | |
1932 | } | |
1933 | return false; | |
1934 | } | |
1935 | ||
750638bb BP |
1936 | /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on |
1937 | * 'packet', which arrived on 'in_port'. | |
1938 | * | |
1939 | * Takes ownership of 'packet'. */ | |
9dbb9d5e BP |
1940 | static bool |
1941 | execute_odp_actions(struct ofproto *ofproto, uint16_t in_port, | |
1942 | const union odp_action *actions, size_t n_actions, | |
750638bb | 1943 | struct ofpbuf *packet) |
9dbb9d5e | 1944 | { |
750638bb | 1945 | if (n_actions == 1 && actions[0].type == ODPAT_CONTROLLER) { |
9dbb9d5e BP |
1946 | /* As an optimization, avoid a round-trip from userspace to kernel to |
1947 | * userspace. This also avoids possibly filling up kernel packet | |
1948 | * buffers along the way. */ | |
9dbb9d5e BP |
1949 | struct odp_msg *msg; |
1950 | ||
750638bb | 1951 | msg = ofpbuf_push_uninit(packet, sizeof *msg); |
9dbb9d5e BP |
1952 | msg->type = _ODPL_ACTION_NR; |
1953 | msg->length = sizeof(struct odp_msg) + packet->size; | |
1954 | msg->port = in_port; | |
1955 | msg->reserved = 0; | |
1956 | msg->arg = actions[0].controller.arg; | |
9dbb9d5e | 1957 | |
750638bb | 1958 | send_packet_in(ofproto, packet); |
9dbb9d5e | 1959 | |
750638bb BP |
1960 | return true; |
1961 | } else { | |
1962 | int error; | |
9dbb9d5e | 1963 | |
750638bb BP |
1964 | error = dpif_execute(ofproto->dpif, in_port, |
1965 | actions, n_actions, packet); | |
1966 | ofpbuf_delete(packet); | |
1967 | return !error; | |
1968 | } | |
9dbb9d5e BP |
1969 | } |
1970 | ||
064af421 | 1971 | /* Executes the actions indicated by 'rule' on 'packet', which is in flow |
750638bb BP |
1972 | * 'flow' and is considered to have arrived on ODP port 'in_port'. 'packet' |
1973 | * must have at least sizeof(struct ofp_packet_in) bytes of headroom. | |
064af421 BP |
1974 | * |
1975 | * The flow that 'packet' actually contains does not need to actually match | |
1976 | * 'rule'; the actions in 'rule' will be applied to it either way. Likewise, | |
1977 | * the packet and byte counters for 'rule' will be credited for the packet sent | |
1978 | * out whether or not the packet actually matches 'rule'. | |
1979 | * | |
1980 | * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow, | |
1981 | * the caller must already have accurately composed ODP actions for it given | |
1982 | * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if | |
1983 | * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this | |
1984 | * function will compose a set of ODP actions based on 'rule''s OpenFlow | |
750638bb BP |
1985 | * actions and apply them to 'packet'. |
1986 | * | |
1987 | * Takes ownership of 'packet'. */ | |
064af421 BP |
1988 | static void |
1989 | rule_execute(struct ofproto *ofproto, struct rule *rule, | |
1990 | struct ofpbuf *packet, const flow_t *flow) | |
1991 | { | |
1992 | const union odp_action *actions; | |
750638bb | 1993 | struct odp_flow_stats stats; |
064af421 BP |
1994 | size_t n_actions; |
1995 | struct odp_actions a; | |
1996 | ||
750638bb BP |
1997 | assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in)); |
1998 | ||
064af421 BP |
1999 | /* Grab or compose the ODP actions. |
2000 | * | |
2001 | * The special case for an exact-match 'rule' where 'flow' is not the | |
2002 | * rule's flow is important to avoid, e.g., sending a packet out its input | |
2003 | * port simply because the ODP actions were composed for the wrong | |
2004 | * scenario. */ | |
2005 | if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) { | |
2006 | struct rule *super = rule->super ? rule->super : rule; | |
2007 | if (xlate_actions(super->actions, super->n_actions, flow, ofproto, | |
6a07af36 | 2008 | packet, &a, NULL, 0, NULL)) { |
750638bb | 2009 | ofpbuf_delete(packet); |
064af421 BP |
2010 | return; |
2011 | } | |
2012 | actions = a.actions; | |
2013 | n_actions = a.n_actions; | |
2014 | } else { | |
2015 | actions = rule->odp_actions; | |
2016 | n_actions = rule->n_odp_actions; | |
2017 | } | |
2018 | ||
2019 | /* Execute the ODP actions. */ | |
750638bb | 2020 | flow_extract_stats(flow, packet, &stats); |
9dbb9d5e BP |
2021 | if (execute_odp_actions(ofproto, flow->in_port, |
2022 | actions, n_actions, packet)) { | |
0193b2af | 2023 | update_stats(ofproto, rule, &stats); |
064af421 | 2024 | rule->used = time_msec(); |
0193b2af | 2025 | netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used); |
064af421 BP |
2026 | } |
2027 | } | |
2028 | ||
750638bb BP |
2029 | /* Inserts 'rule' into 'p''s flow table. |
2030 | * | |
2031 | * If 'packet' is nonnull, takes ownership of 'packet', executes 'rule''s | |
2032 | * actions on it and credits the statistics for sending the packet to 'rule'. | |
2033 | * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of | |
2034 | * headroom. */ | |
064af421 BP |
2035 | static void |
2036 | rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet, | |
2037 | uint16_t in_port) | |
2038 | { | |
2039 | struct rule *displaced_rule; | |
2040 | ||
2041 | /* Insert the rule in the classifier. */ | |
2042 | displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr)); | |
2043 | if (!rule->cr.wc.wildcards) { | |
2044 | rule_make_actions(p, rule, packet); | |
2045 | } | |
2046 | ||
2047 | /* Send the packet and credit it to the rule. */ | |
2048 | if (packet) { | |
2049 | flow_t flow; | |
659586ef | 2050 | flow_extract(packet, 0, in_port, &flow); |
064af421 BP |
2051 | rule_execute(p, rule, packet, &flow); |
2052 | } | |
2053 | ||
2054 | /* Install the rule in the datapath only after sending the packet, to | |
2055 | * avoid packet reordering. */ | |
2056 | if (rule->cr.wc.wildcards) { | |
2057 | COVERAGE_INC(ofproto_add_wc_flow); | |
2058 | p->need_revalidate = true; | |
2059 | } else { | |
2060 | rule_install(p, rule, displaced_rule); | |
2061 | } | |
2062 | ||
2063 | /* Free the rule that was displaced, if any. */ | |
2064 | if (displaced_rule) { | |
2065 | rule_destroy(p, displaced_rule); | |
2066 | } | |
2067 | } | |
2068 | ||
2069 | static struct rule * | |
2070 | rule_create_subrule(struct ofproto *ofproto, struct rule *rule, | |
2071 | const flow_t *flow) | |
2072 | { | |
0193b2af | 2073 | struct rule *subrule = rule_create(ofproto, rule, NULL, 0, |
ca069229 | 2074 | rule->idle_timeout, rule->hard_timeout, |
39997502 | 2075 | 0, false); |
064af421 | 2076 | COVERAGE_INC(ofproto_subrule_create); |
659586ef JG |
2077 | cls_rule_from_flow(flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX |
2078 | : rule->cr.priority), &subrule->cr); | |
064af421 BP |
2079 | classifier_insert_exact(&ofproto->cls, &subrule->cr); |
2080 | ||
2081 | return subrule; | |
2082 | } | |
2083 | ||
2084 | static void | |
2085 | rule_remove(struct ofproto *ofproto, struct rule *rule) | |
2086 | { | |
2087 | if (rule->cr.wc.wildcards) { | |
2088 | COVERAGE_INC(ofproto_del_wc_flow); | |
2089 | ofproto->need_revalidate = true; | |
2090 | } else { | |
2091 | rule_uninstall(ofproto, rule); | |
2092 | } | |
2093 | classifier_remove(&ofproto->cls, &rule->cr); | |
2094 | rule_destroy(ofproto, rule); | |
2095 | } | |
2096 | ||
2097 | /* Returns true if the actions changed, false otherwise. */ | |
2098 | static bool | |
2099 | rule_make_actions(struct ofproto *p, struct rule *rule, | |
2100 | const struct ofpbuf *packet) | |
2101 | { | |
2102 | const struct rule *super; | |
2103 | struct odp_actions a; | |
2104 | size_t actions_len; | |
2105 | ||
2106 | assert(!rule->cr.wc.wildcards); | |
2107 | ||
2108 | super = rule->super ? rule->super : rule; | |
2109 | rule->tags = 0; | |
2110 | xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p, | |
6a07af36 | 2111 | packet, &a, &rule->tags, &rule->may_install, |
0193b2af | 2112 | &rule->nf_flow.output_iface); |
064af421 BP |
2113 | |
2114 | actions_len = a.n_actions * sizeof *a.actions; | |
2115 | if (rule->n_odp_actions != a.n_actions | |
2116 | || memcmp(rule->odp_actions, a.actions, actions_len)) { | |
2117 | COVERAGE_INC(ofproto_odp_unchanged); | |
2118 | free(rule->odp_actions); | |
2119 | rule->n_odp_actions = a.n_actions; | |
2120 | rule->odp_actions = xmemdup(a.actions, actions_len); | |
2121 | return true; | |
2122 | } else { | |
2123 | return false; | |
2124 | } | |
2125 | } | |
2126 | ||
2127 | static int | |
2128 | do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags, | |
2129 | struct odp_flow_put *put) | |
2130 | { | |
2131 | memset(&put->flow.stats, 0, sizeof put->flow.stats); | |
2132 | put->flow.key = rule->cr.flow; | |
2133 | put->flow.actions = rule->odp_actions; | |
2134 | put->flow.n_actions = rule->n_odp_actions; | |
ab48643b | 2135 | put->flow.flags = 0; |
064af421 | 2136 | put->flags = flags; |
c228a364 | 2137 | return dpif_flow_put(ofproto->dpif, put); |
064af421 BP |
2138 | } |
2139 | ||
2140 | static void | |
2141 | rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule) | |
2142 | { | |
2143 | assert(!rule->cr.wc.wildcards); | |
2144 | ||
2145 | if (rule->may_install) { | |
2146 | struct odp_flow_put put; | |
2147 | if (!do_put_flow(p, rule, | |
2148 | ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS, | |
2149 | &put)) { | |
2150 | rule->installed = true; | |
2151 | if (displaced_rule) { | |
14986b31 | 2152 | update_stats(p, displaced_rule, &put.flow.stats); |
064af421 BP |
2153 | rule_post_uninstall(p, displaced_rule); |
2154 | } | |
2155 | } | |
2156 | } else if (displaced_rule) { | |
2157 | rule_uninstall(p, displaced_rule); | |
2158 | } | |
2159 | } | |
2160 | ||
2161 | static void | |
2162 | rule_reinstall(struct ofproto *ofproto, struct rule *rule) | |
2163 | { | |
2164 | if (rule->installed) { | |
2165 | struct odp_flow_put put; | |
2166 | COVERAGE_INC(ofproto_dp_missed); | |
2167 | do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put); | |
2168 | } else { | |
2169 | rule_install(ofproto, rule, NULL); | |
2170 | } | |
2171 | } | |
2172 | ||
2173 | static void | |
2174 | rule_update_actions(struct ofproto *ofproto, struct rule *rule) | |
2175 | { | |
42c3641c JG |
2176 | bool actions_changed; |
2177 | uint16_t new_out_iface, old_out_iface; | |
2178 | ||
2179 | old_out_iface = rule->nf_flow.output_iface; | |
2180 | actions_changed = rule_make_actions(ofproto, rule, NULL); | |
2181 | ||
064af421 BP |
2182 | if (rule->may_install) { |
2183 | if (rule->installed) { | |
2184 | if (actions_changed) { | |
064af421 | 2185 | struct odp_flow_put put; |
42c3641c JG |
2186 | do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY |
2187 | | ODPPF_ZERO_STATS, &put); | |
2188 | update_stats(ofproto, rule, &put.flow.stats); | |
2189 | ||
2190 | /* Temporarily set the old output iface so that NetFlow | |
2191 | * messages have the correct output interface for the old | |
2192 | * stats. */ | |
2193 | new_out_iface = rule->nf_flow.output_iface; | |
2194 | rule->nf_flow.output_iface = old_out_iface; | |
2195 | rule_post_uninstall(ofproto, rule); | |
2196 | rule->nf_flow.output_iface = new_out_iface; | |
064af421 BP |
2197 | } |
2198 | } else { | |
2199 | rule_install(ofproto, rule, NULL); | |
2200 | } | |
2201 | } else { | |
2202 | rule_uninstall(ofproto, rule); | |
2203 | } | |
2204 | } | |
2205 | ||
2206 | static void | |
2207 | rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes) | |
2208 | { | |
2209 | uint64_t total_bytes = rule->byte_count + extra_bytes; | |
2210 | ||
2211 | if (ofproto->ofhooks->account_flow_cb | |
2212 | && total_bytes > rule->accounted_bytes) | |
2213 | { | |
2214 | ofproto->ofhooks->account_flow_cb( | |
26d79bf2 | 2215 | &rule->cr.flow, rule->tags, rule->odp_actions, rule->n_odp_actions, |
064af421 BP |
2216 | total_bytes - rule->accounted_bytes, ofproto->aux); |
2217 | rule->accounted_bytes = total_bytes; | |
2218 | } | |
2219 | } | |
2220 | ||
2221 | static void | |
2222 | rule_uninstall(struct ofproto *p, struct rule *rule) | |
2223 | { | |
2224 | assert(!rule->cr.wc.wildcards); | |
2225 | if (rule->installed) { | |
2226 | struct odp_flow odp_flow; | |
2227 | ||
2228 | odp_flow.key = rule->cr.flow; | |
2229 | odp_flow.actions = NULL; | |
2230 | odp_flow.n_actions = 0; | |
ab48643b | 2231 | odp_flow.flags = 0; |
c228a364 | 2232 | if (!dpif_flow_del(p->dpif, &odp_flow)) { |
0193b2af | 2233 | update_stats(p, rule, &odp_flow.stats); |
064af421 BP |
2234 | } |
2235 | rule->installed = false; | |
2236 | ||
2237 | rule_post_uninstall(p, rule); | |
2238 | } | |
2239 | } | |
2240 | ||
0193b2af JG |
2241 | static bool |
2242 | is_controller_rule(struct rule *rule) | |
2243 | { | |
2244 | /* If the only action is send to the controller then don't report | |
2245 | * NetFlow expiration messages since it is just part of the control | |
2246 | * logic for the network and not real traffic. */ | |
2247 | ||
c1c9c9c4 BP |
2248 | return (rule |
2249 | && rule->super | |
2250 | && rule->super->n_actions == 1 | |
2251 | && action_outputs_to_port(&rule->super->actions[0], | |
2252 | htons(OFPP_CONTROLLER))); | |
0193b2af JG |
2253 | } |
2254 | ||
064af421 BP |
2255 | static void |
2256 | rule_post_uninstall(struct ofproto *ofproto, struct rule *rule) | |
2257 | { | |
2258 | struct rule *super = rule->super; | |
2259 | ||
2260 | rule_account(ofproto, rule, 0); | |
6a07af36 | 2261 | |
0193b2af | 2262 | if (ofproto->netflow && !is_controller_rule(rule)) { |
064af421 BP |
2263 | struct ofexpired expired; |
2264 | expired.flow = rule->cr.flow; | |
2265 | expired.packet_count = rule->packet_count; | |
2266 | expired.byte_count = rule->byte_count; | |
2267 | expired.used = rule->used; | |
0193b2af | 2268 | netflow_expire(ofproto->netflow, &rule->nf_flow, &expired); |
064af421 BP |
2269 | } |
2270 | if (super) { | |
2271 | super->packet_count += rule->packet_count; | |
2272 | super->byte_count += rule->byte_count; | |
064af421 | 2273 | |
0c0afbec JG |
2274 | /* Reset counters to prevent double counting if the rule ever gets |
2275 | * reinstalled. */ | |
2276 | rule->packet_count = 0; | |
2277 | rule->byte_count = 0; | |
2278 | rule->accounted_bytes = 0; | |
0193b2af JG |
2279 | |
2280 | netflow_flow_clear(&rule->nf_flow); | |
0c0afbec | 2281 | } |
064af421 BP |
2282 | } |
2283 | \f | |
2284 | static void | |
2285 | queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn, | |
2286 | struct rconn_packet_counter *counter) | |
2287 | { | |
2288 | update_openflow_length(msg); | |
2289 | if (rconn_send(ofconn->rconn, msg, counter)) { | |
2290 | ofpbuf_delete(msg); | |
2291 | } | |
2292 | } | |
2293 | ||
2294 | static void | |
2295 | send_error(const struct ofconn *ofconn, const struct ofp_header *oh, | |
2296 | int error, const void *data, size_t len) | |
2297 | { | |
2298 | struct ofpbuf *buf; | |
2299 | struct ofp_error_msg *oem; | |
2300 | ||
2301 | if (!(error >> 16)) { | |
2302 | VLOG_WARN_RL(&rl, "not sending bad error code %d to controller", | |
2303 | error); | |
2304 | return; | |
2305 | } | |
2306 | ||
2307 | COVERAGE_INC(ofproto_error); | |
2308 | oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR, | |
2309 | oh ? oh->xid : 0, &buf); | |
2310 | oem->type = htons((unsigned int) error >> 16); | |
2311 | oem->code = htons(error & 0xffff); | |
2312 | memcpy(oem->data, data, len); | |
2313 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
2314 | } | |
2315 | ||
2316 | static void | |
2317 | send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh, | |
2318 | int error) | |
2319 | { | |
2320 | size_t oh_length = ntohs(oh->length); | |
2321 | send_error(ofconn, oh, error, oh, MIN(oh_length, 64)); | |
2322 | } | |
2323 | ||
2324 | static void | |
2325 | hton_ofp_phy_port(struct ofp_phy_port *opp) | |
2326 | { | |
2327 | opp->port_no = htons(opp->port_no); | |
2328 | opp->config = htonl(opp->config); | |
2329 | opp->state = htonl(opp->state); | |
2330 | opp->curr = htonl(opp->curr); | |
2331 | opp->advertised = htonl(opp->advertised); | |
2332 | opp->supported = htonl(opp->supported); | |
2333 | opp->peer = htonl(opp->peer); | |
2334 | } | |
2335 | ||
2336 | static int | |
2337 | handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh) | |
2338 | { | |
2339 | struct ofp_header *rq = oh; | |
2340 | queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter); | |
2341 | return 0; | |
2342 | } | |
2343 | ||
2344 | static int | |
2345 | handle_features_request(struct ofproto *p, struct ofconn *ofconn, | |
2346 | struct ofp_header *oh) | |
2347 | { | |
2348 | struct ofp_switch_features *osf; | |
2349 | struct ofpbuf *buf; | |
064af421 BP |
2350 | struct ofport *port; |
2351 | ||
2352 | osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf); | |
2353 | osf->datapath_id = htonll(p->datapath_id); | |
2354 | osf->n_buffers = htonl(pktbuf_capacity()); | |
2355 | osf->n_tables = 2; | |
2356 | osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS | | |
0254ae23 | 2357 | OFPC_PORT_STATS | OFPC_ARP_MATCH_IP); |
064af421 BP |
2358 | osf->actions = htonl((1u << OFPAT_OUTPUT) | |
2359 | (1u << OFPAT_SET_VLAN_VID) | | |
2360 | (1u << OFPAT_SET_VLAN_PCP) | | |
2361 | (1u << OFPAT_STRIP_VLAN) | | |
2362 | (1u << OFPAT_SET_DL_SRC) | | |
2363 | (1u << OFPAT_SET_DL_DST) | | |
2364 | (1u << OFPAT_SET_NW_SRC) | | |
2365 | (1u << OFPAT_SET_NW_DST) | | |
959a2ecd | 2366 | (1u << OFPAT_SET_NW_TOS) | |
064af421 | 2367 | (1u << OFPAT_SET_TP_SRC) | |
c1c9c9c4 BP |
2368 | (1u << OFPAT_SET_TP_DST) | |
2369 | (1u << OFPAT_ENQUEUE)); | |
064af421 | 2370 | |
4e8e4213 | 2371 | HMAP_FOR_EACH (port, hmap_node, &p->ports) { |
064af421 BP |
2372 | hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp)); |
2373 | } | |
2374 | ||
2375 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
2376 | return 0; | |
2377 | } | |
2378 | ||
2379 | static int | |
2380 | handle_get_config_request(struct ofproto *p, struct ofconn *ofconn, | |
2381 | struct ofp_header *oh) | |
2382 | { | |
2383 | struct ofpbuf *buf; | |
2384 | struct ofp_switch_config *osc; | |
2385 | uint16_t flags; | |
2386 | bool drop_frags; | |
2387 | ||
2388 | /* Figure out flags. */ | |
c228a364 | 2389 | dpif_get_drop_frags(p->dpif, &drop_frags); |
064af421 | 2390 | flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL; |
064af421 BP |
2391 | |
2392 | /* Send reply. */ | |
2393 | osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf); | |
2394 | osc->flags = htons(flags); | |
2395 | osc->miss_send_len = htons(ofconn->miss_send_len); | |
2396 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
2397 | ||
2398 | return 0; | |
2399 | } | |
2400 | ||
2401 | static int | |
2402 | handle_set_config(struct ofproto *p, struct ofconn *ofconn, | |
2403 | struct ofp_switch_config *osc) | |
2404 | { | |
2405 | uint16_t flags; | |
2406 | int error; | |
2407 | ||
2408 | error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc); | |
2409 | if (error) { | |
2410 | return error; | |
2411 | } | |
2412 | flags = ntohs(osc->flags); | |
2413 | ||
5899143f | 2414 | if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) { |
064af421 BP |
2415 | switch (flags & OFPC_FRAG_MASK) { |
2416 | case OFPC_FRAG_NORMAL: | |
c228a364 | 2417 | dpif_set_drop_frags(p->dpif, false); |
064af421 BP |
2418 | break; |
2419 | case OFPC_FRAG_DROP: | |
c228a364 | 2420 | dpif_set_drop_frags(p->dpif, true); |
064af421 BP |
2421 | break; |
2422 | default: | |
2423 | VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")", | |
2424 | osc->flags); | |
2425 | break; | |
2426 | } | |
2427 | } | |
2428 | ||
064af421 BP |
2429 | ofconn->miss_send_len = ntohs(osc->miss_send_len); |
2430 | ||
2431 | return 0; | |
2432 | } | |
2433 | ||
2434 | static void | |
6a07af36 JG |
2435 | add_output_group_action(struct odp_actions *actions, uint16_t group, |
2436 | uint16_t *nf_output_iface) | |
064af421 BP |
2437 | { |
2438 | odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group; | |
6a07af36 JG |
2439 | |
2440 | if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) { | |
2441 | *nf_output_iface = NF_OUT_FLOOD; | |
2442 | } | |
064af421 BP |
2443 | } |
2444 | ||
2445 | static void | |
aae51f53 | 2446 | add_controller_action(struct odp_actions *actions, uint16_t max_len) |
064af421 BP |
2447 | { |
2448 | union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER); | |
aae51f53 | 2449 | a->controller.arg = max_len; |
064af421 BP |
2450 | } |
2451 | ||
2452 | struct action_xlate_ctx { | |
2453 | /* Input. */ | |
e18fe8a2 | 2454 | flow_t flow; /* Flow to which these actions correspond. */ |
064af421 BP |
2455 | int recurse; /* Recursion level, via xlate_table_action. */ |
2456 | struct ofproto *ofproto; | |
2457 | const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a | |
2458 | * null pointer if we are revalidating | |
2459 | * without a packet to refer to. */ | |
2460 | ||
2461 | /* Output. */ | |
2462 | struct odp_actions *out; /* Datapath actions. */ | |
2463 | tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */ | |
d6fbec6d | 2464 | bool may_set_up_flow; /* True ordinarily; false if the actions must |
064af421 | 2465 | * be reassessed for every packet. */ |
6a07af36 | 2466 | uint16_t nf_output_iface; /* Output interface index for NetFlow. */ |
064af421 BP |
2467 | }; |
2468 | ||
5f8bfd69 BP |
2469 | /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a |
2470 | * flow translation. */ | |
2471 | #define MAX_RESUBMIT_RECURSION 8 | |
2472 | ||
064af421 BP |
2473 | static void do_xlate_actions(const union ofp_action *in, size_t n_in, |
2474 | struct action_xlate_ctx *ctx); | |
2475 | ||
2476 | static void | |
2477 | add_output_action(struct action_xlate_ctx *ctx, uint16_t port) | |
2478 | { | |
ca0f572c | 2479 | const struct ofport *ofport = get_port(ctx->ofproto, port); |
6cfaf517 BP |
2480 | |
2481 | if (ofport) { | |
2482 | if (ofport->opp.config & OFPPC_NO_FWD) { | |
2483 | /* Forwarding disabled on port. */ | |
2484 | return; | |
2485 | } | |
2486 | } else { | |
2487 | /* | |
2488 | * We don't have an ofport record for this port, but it doesn't hurt to | |
2489 | * allow forwarding to it anyhow. Maybe such a port will appear later | |
2490 | * and we're pre-populating the flow table. | |
2491 | */ | |
064af421 | 2492 | } |
6cfaf517 BP |
2493 | |
2494 | odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port; | |
6a07af36 | 2495 | ctx->nf_output_iface = port; |
064af421 BP |
2496 | } |
2497 | ||
2498 | static struct rule * | |
2499 | lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow) | |
2500 | { | |
2501 | struct rule *rule; | |
2502 | rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow)); | |
2503 | ||
2504 | /* The rule we found might not be valid, since we could be in need of | |
2505 | * revalidation. If it is not valid, don't return it. */ | |
2506 | if (rule | |
2507 | && rule->super | |
2508 | && ofproto->need_revalidate | |
2509 | && !revalidate_rule(ofproto, rule)) { | |
2510 | COVERAGE_INC(ofproto_invalidated); | |
2511 | return NULL; | |
2512 | } | |
2513 | ||
2514 | return rule; | |
2515 | } | |
2516 | ||
2517 | static void | |
2518 | xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port) | |
2519 | { | |
5f8bfd69 | 2520 | if (ctx->recurse < MAX_RESUBMIT_RECURSION) { |
2c5d1389 | 2521 | uint16_t old_in_port; |
064af421 | 2522 | struct rule *rule; |
064af421 | 2523 | |
2c5d1389 BP |
2524 | /* Look up a flow with 'in_port' as the input port. Then restore the |
2525 | * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will | |
2526 | * have surprising behavior). */ | |
2527 | old_in_port = ctx->flow.in_port; | |
e18fe8a2 BP |
2528 | ctx->flow.in_port = in_port; |
2529 | rule = lookup_valid_rule(ctx->ofproto, &ctx->flow); | |
2c5d1389 BP |
2530 | ctx->flow.in_port = old_in_port; |
2531 | ||
064af421 BP |
2532 | if (rule) { |
2533 | if (rule->super) { | |
2534 | rule = rule->super; | |
2535 | } | |
2536 | ||
2537 | ctx->recurse++; | |
2538 | do_xlate_actions(rule->actions, rule->n_actions, ctx); | |
2539 | ctx->recurse--; | |
2540 | } | |
5f8bfd69 BP |
2541 | } else { |
2542 | struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
2543 | ||
2544 | VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times", | |
2545 | MAX_RESUBMIT_RECURSION); | |
064af421 BP |
2546 | } |
2547 | } | |
2548 | ||
2549 | static void | |
aae51f53 BP |
2550 | xlate_output_action__(struct action_xlate_ctx *ctx, |
2551 | uint16_t port, uint16_t max_len) | |
064af421 BP |
2552 | { |
2553 | uint16_t odp_port; | |
6a07af36 JG |
2554 | uint16_t prev_nf_output_iface = ctx->nf_output_iface; |
2555 | ||
2556 | ctx->nf_output_iface = NF_OUT_DROP; | |
064af421 | 2557 | |
aae51f53 | 2558 | switch (port) { |
064af421 | 2559 | case OFPP_IN_PORT: |
e18fe8a2 | 2560 | add_output_action(ctx, ctx->flow.in_port); |
064af421 BP |
2561 | break; |
2562 | case OFPP_TABLE: | |
e18fe8a2 | 2563 | xlate_table_action(ctx, ctx->flow.in_port); |
064af421 BP |
2564 | break; |
2565 | case OFPP_NORMAL: | |
e18fe8a2 | 2566 | if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet, |
064af421 | 2567 | ctx->out, ctx->tags, |
6a07af36 | 2568 | &ctx->nf_output_iface, |
064af421 BP |
2569 | ctx->ofproto->aux)) { |
2570 | COVERAGE_INC(ofproto_uninstallable); | |
d6fbec6d | 2571 | ctx->may_set_up_flow = false; |
064af421 BP |
2572 | } |
2573 | break; | |
2574 | case OFPP_FLOOD: | |
6a07af36 JG |
2575 | add_output_group_action(ctx->out, DP_GROUP_FLOOD, |
2576 | &ctx->nf_output_iface); | |
064af421 BP |
2577 | break; |
2578 | case OFPP_ALL: | |
6a07af36 | 2579 | add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface); |
064af421 BP |
2580 | break; |
2581 | case OFPP_CONTROLLER: | |
aae51f53 | 2582 | add_controller_action(ctx->out, max_len); |
064af421 BP |
2583 | break; |
2584 | case OFPP_LOCAL: | |
2585 | add_output_action(ctx, ODPP_LOCAL); | |
2586 | break; | |
2587 | default: | |
aae51f53 | 2588 | odp_port = ofp_port_to_odp_port(port); |
e18fe8a2 | 2589 | if (odp_port != ctx->flow.in_port) { |
064af421 BP |
2590 | add_output_action(ctx, odp_port); |
2591 | } | |
2592 | break; | |
2593 | } | |
6a07af36 JG |
2594 | |
2595 | if (prev_nf_output_iface == NF_OUT_FLOOD) { | |
2596 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
2597 | } else if (ctx->nf_output_iface == NF_OUT_DROP) { | |
2598 | ctx->nf_output_iface = prev_nf_output_iface; | |
2599 | } else if (prev_nf_output_iface != NF_OUT_DROP && | |
2600 | ctx->nf_output_iface != NF_OUT_FLOOD) { | |
2601 | ctx->nf_output_iface = NF_OUT_MULTI; | |
2602 | } | |
064af421 BP |
2603 | } |
2604 | ||
aae51f53 BP |
2605 | static void |
2606 | xlate_output_action(struct action_xlate_ctx *ctx, | |
2607 | const struct ofp_action_output *oao) | |
2608 | { | |
2609 | xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len)); | |
2610 | } | |
2611 | ||
c1c9c9c4 BP |
2612 | /* If the final ODP action in 'ctx' is "pop priority", drop it, as an |
2613 | * optimization, because we're going to add another action that sets the | |
2614 | * priority immediately after, or because there are no actions following the | |
2615 | * pop. */ | |
2616 | static void | |
2617 | remove_pop_action(struct action_xlate_ctx *ctx) | |
2618 | { | |
2619 | size_t n = ctx->out->n_actions; | |
2620 | if (n > 0 && ctx->out->actions[n - 1].type == ODPAT_POP_PRIORITY) { | |
2621 | ctx->out->n_actions--; | |
2622 | } | |
2623 | } | |
2624 | ||
2625 | static void | |
2626 | xlate_enqueue_action(struct action_xlate_ctx *ctx, | |
2627 | const struct ofp_action_enqueue *oae) | |
2628 | { | |
2629 | uint16_t ofp_port, odp_port; | |
aae51f53 BP |
2630 | uint32_t priority; |
2631 | int error; | |
2632 | ||
2633 | error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id), | |
2634 | &priority); | |
2635 | if (error) { | |
2636 | /* Fall back to ordinary output action. */ | |
2637 | xlate_output_action__(ctx, ntohs(oae->port), 0); | |
2638 | return; | |
2639 | } | |
c1c9c9c4 BP |
2640 | |
2641 | /* Figure out ODP output port. */ | |
2642 | ofp_port = ntohs(oae->port); | |
2643 | if (ofp_port != OFPP_IN_PORT) { | |
2644 | odp_port = ofp_port_to_odp_port(ofp_port); | |
2645 | } else { | |
2646 | odp_port = ctx->flow.in_port; | |
2647 | } | |
2648 | ||
2649 | /* Add ODP actions. */ | |
2650 | remove_pop_action(ctx); | |
2651 | odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority | |
aae51f53 | 2652 | = priority; |
c1c9c9c4 BP |
2653 | add_output_action(ctx, odp_port); |
2654 | odp_actions_add(ctx->out, ODPAT_POP_PRIORITY); | |
2655 | ||
2656 | /* Update NetFlow output port. */ | |
2657 | if (ctx->nf_output_iface == NF_OUT_DROP) { | |
2658 | ctx->nf_output_iface = odp_port; | |
2659 | } else if (ctx->nf_output_iface != NF_OUT_FLOOD) { | |
2660 | ctx->nf_output_iface = NF_OUT_MULTI; | |
2661 | } | |
2662 | } | |
2663 | ||
064af421 BP |
2664 | static void |
2665 | xlate_nicira_action(struct action_xlate_ctx *ctx, | |
2666 | const struct nx_action_header *nah) | |
2667 | { | |
2668 | const struct nx_action_resubmit *nar; | |
659586ef JG |
2669 | const struct nx_action_set_tunnel *nast; |
2670 | union odp_action *oa; | |
064af421 BP |
2671 | int subtype = ntohs(nah->subtype); |
2672 | ||
2673 | assert(nah->vendor == htonl(NX_VENDOR_ID)); | |
2674 | switch (subtype) { | |
2675 | case NXAST_RESUBMIT: | |
2676 | nar = (const struct nx_action_resubmit *) nah; | |
2677 | xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port))); | |
2678 | break; | |
2679 | ||
659586ef JG |
2680 | case NXAST_SET_TUNNEL: |
2681 | nast = (const struct nx_action_set_tunnel *) nah; | |
2682 | oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL); | |
2683 | ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id; | |
2684 | break; | |
2685 | ||
401eeb92 BP |
2686 | case NXAST_DROP_SPOOFED_ARP: |
2687 | if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) { | |
2688 | odp_actions_add(ctx->out, ODPAT_DROP_SPOOFED_ARP); | |
2689 | } | |
2690 | break; | |
2691 | ||
999f0d45 | 2692 | /* If you add a new action here that modifies flow data, don't forget to |
c1c9c9c4 | 2693 | * update the flow key in ctx->flow at the same time. */ |
999f0d45 | 2694 | |
064af421 BP |
2695 | default: |
2696 | VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype); | |
2697 | break; | |
2698 | } | |
2699 | } | |
2700 | ||
2701 | static void | |
2702 | do_xlate_actions(const union ofp_action *in, size_t n_in, | |
2703 | struct action_xlate_ctx *ctx) | |
2704 | { | |
2705 | struct actions_iterator iter; | |
2706 | const union ofp_action *ia; | |
2707 | const struct ofport *port; | |
2708 | ||
ca0f572c | 2709 | port = get_port(ctx->ofproto, ctx->flow.in_port); |
064af421 | 2710 | if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) && |
ba186119 | 2711 | port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp) |
064af421 BP |
2712 | ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) { |
2713 | /* Drop this flow. */ | |
2714 | return; | |
2715 | } | |
2716 | ||
2717 | for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) { | |
2718 | uint16_t type = ntohs(ia->type); | |
2719 | union odp_action *oa; | |
2720 | ||
2721 | switch (type) { | |
2722 | case OFPAT_OUTPUT: | |
2723 | xlate_output_action(ctx, &ia->output); | |
2724 | break; | |
2725 | ||
2726 | case OFPAT_SET_VLAN_VID: | |
2727 | oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID); | |
999f0d45 | 2728 | ctx->flow.dl_vlan = oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid; |
064af421 BP |
2729 | break; |
2730 | ||
2731 | case OFPAT_SET_VLAN_PCP: | |
2732 | oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP); | |
999f0d45 | 2733 | ctx->flow.dl_vlan_pcp = oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp; |
064af421 BP |
2734 | break; |
2735 | ||
2736 | case OFPAT_STRIP_VLAN: | |
2737 | odp_actions_add(ctx->out, ODPAT_STRIP_VLAN); | |
7ac0f8db | 2738 | ctx->flow.dl_vlan = htons(OFP_VLAN_NONE); |
999f0d45 | 2739 | ctx->flow.dl_vlan_pcp = 0; |
064af421 BP |
2740 | break; |
2741 | ||
2742 | case OFPAT_SET_DL_SRC: | |
2743 | oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC); | |
2744 | memcpy(oa->dl_addr.dl_addr, | |
2745 | ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); | |
999f0d45 BP |
2746 | memcpy(ctx->flow.dl_src, |
2747 | ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); | |
064af421 BP |
2748 | break; |
2749 | ||
2750 | case OFPAT_SET_DL_DST: | |
2751 | oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST); | |
2752 | memcpy(oa->dl_addr.dl_addr, | |
2753 | ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); | |
999f0d45 BP |
2754 | memcpy(ctx->flow.dl_dst, |
2755 | ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); | |
064af421 BP |
2756 | break; |
2757 | ||
2758 | case OFPAT_SET_NW_SRC: | |
2759 | oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC); | |
999f0d45 | 2760 | ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr; |
064af421 BP |
2761 | break; |
2762 | ||
2d70a31a JP |
2763 | case OFPAT_SET_NW_DST: |
2764 | oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST); | |
999f0d45 | 2765 | ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr; |
2d38e234 | 2766 | break; |
959a2ecd JP |
2767 | |
2768 | case OFPAT_SET_NW_TOS: | |
2769 | oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS); | |
999f0d45 | 2770 | ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos; |
2d70a31a JP |
2771 | break; |
2772 | ||
064af421 BP |
2773 | case OFPAT_SET_TP_SRC: |
2774 | oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC); | |
999f0d45 | 2775 | ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port; |
064af421 BP |
2776 | break; |
2777 | ||
2d70a31a JP |
2778 | case OFPAT_SET_TP_DST: |
2779 | oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST); | |
999f0d45 | 2780 | ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port; |
2d70a31a JP |
2781 | break; |
2782 | ||
064af421 BP |
2783 | case OFPAT_VENDOR: |
2784 | xlate_nicira_action(ctx, (const struct nx_action_header *) ia); | |
2785 | break; | |
2786 | ||
c1c9c9c4 BP |
2787 | case OFPAT_ENQUEUE: |
2788 | xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia); | |
2789 | break; | |
2790 | ||
064af421 BP |
2791 | default: |
2792 | VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type); | |
2793 | break; | |
2794 | } | |
2795 | } | |
2796 | } | |
2797 | ||
2798 | static int | |
2799 | xlate_actions(const union ofp_action *in, size_t n_in, | |
2800 | const flow_t *flow, struct ofproto *ofproto, | |
2801 | const struct ofpbuf *packet, | |
6a07af36 JG |
2802 | struct odp_actions *out, tag_type *tags, bool *may_set_up_flow, |
2803 | uint16_t *nf_output_iface) | |
064af421 BP |
2804 | { |
2805 | tag_type no_tags = 0; | |
2806 | struct action_xlate_ctx ctx; | |
2807 | COVERAGE_INC(ofproto_ofp2odp); | |
2808 | odp_actions_init(out); | |
e18fe8a2 | 2809 | ctx.flow = *flow; |
064af421 BP |
2810 | ctx.recurse = 0; |
2811 | ctx.ofproto = ofproto; | |
2812 | ctx.packet = packet; | |
2813 | ctx.out = out; | |
2814 | ctx.tags = tags ? tags : &no_tags; | |
d6fbec6d | 2815 | ctx.may_set_up_flow = true; |
6a07af36 | 2816 | ctx.nf_output_iface = NF_OUT_DROP; |
064af421 | 2817 | do_xlate_actions(in, n_in, &ctx); |
c1c9c9c4 | 2818 | remove_pop_action(&ctx); |
0ad9b732 | 2819 | |
d6fbec6d | 2820 | /* Check with in-band control to see if we're allowed to set up this |
0ad9b732 JP |
2821 | * flow. */ |
2822 | if (!in_band_rule_check(ofproto->in_band, flow, out)) { | |
d6fbec6d | 2823 | ctx.may_set_up_flow = false; |
0ad9b732 JP |
2824 | } |
2825 | ||
d6fbec6d BP |
2826 | if (may_set_up_flow) { |
2827 | *may_set_up_flow = ctx.may_set_up_flow; | |
064af421 | 2828 | } |
6a07af36 JG |
2829 | if (nf_output_iface) { |
2830 | *nf_output_iface = ctx.nf_output_iface; | |
064af421 BP |
2831 | } |
2832 | if (odp_actions_overflow(out)) { | |
9026bf34 | 2833 | COVERAGE_INC(odp_overflow); |
064af421 BP |
2834 | odp_actions_init(out); |
2835 | return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY); | |
2836 | } | |
2837 | return 0; | |
2838 | } | |
2839 | ||
9deba63b BP |
2840 | /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow |
2841 | * error message code (composed with ofp_mkerr()) for the caller to propagate | |
2842 | * upward. Otherwise, returns 0. | |
2843 | * | |
2844 | * 'oh' is used to make log messages more informative. */ | |
2845 | static int | |
2846 | reject_slave_controller(struct ofconn *ofconn, const struct ofp_header *oh) | |
2847 | { | |
5899143f | 2848 | if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) { |
9deba63b BP |
2849 | static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5); |
2850 | char *type_name; | |
2851 | ||
2852 | type_name = ofp_message_type_to_string(oh->type); | |
2853 | VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller", | |
2854 | type_name); | |
2855 | free(type_name); | |
2856 | ||
2857 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM); | |
2858 | } else { | |
2859 | return 0; | |
2860 | } | |
2861 | } | |
2862 | ||
064af421 BP |
2863 | static int |
2864 | handle_packet_out(struct ofproto *p, struct ofconn *ofconn, | |
2865 | struct ofp_header *oh) | |
2866 | { | |
2867 | struct ofp_packet_out *opo; | |
2868 | struct ofpbuf payload, *buffer; | |
2869 | struct odp_actions actions; | |
2870 | int n_actions; | |
2871 | uint16_t in_port; | |
2872 | flow_t flow; | |
2873 | int error; | |
2874 | ||
9deba63b BP |
2875 | error = reject_slave_controller(ofconn, oh); |
2876 | if (error) { | |
2877 | return error; | |
2878 | } | |
2879 | ||
064af421 BP |
2880 | error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports); |
2881 | if (error) { | |
2882 | return error; | |
2883 | } | |
2884 | opo = (struct ofp_packet_out *) oh; | |
2885 | ||
2886 | COVERAGE_INC(ofproto_packet_out); | |
2887 | if (opo->buffer_id != htonl(UINT32_MAX)) { | |
2888 | error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id), | |
2889 | &buffer, &in_port); | |
7778bd15 | 2890 | if (error || !buffer) { |
064af421 BP |
2891 | return error; |
2892 | } | |
2893 | payload = *buffer; | |
2894 | } else { | |
2895 | buffer = NULL; | |
2896 | } | |
2897 | ||
659586ef | 2898 | flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow); |
064af421 | 2899 | error = xlate_actions((const union ofp_action *) opo->actions, n_actions, |
6a07af36 | 2900 | &flow, p, &payload, &actions, NULL, NULL, NULL); |
064af421 BP |
2901 | if (error) { |
2902 | return error; | |
2903 | } | |
2904 | ||
c228a364 | 2905 | dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions, |
064af421 BP |
2906 | &payload); |
2907 | ofpbuf_delete(buffer); | |
2908 | ||
2909 | return 0; | |
2910 | } | |
2911 | ||
2912 | static void | |
2913 | update_port_config(struct ofproto *p, struct ofport *port, | |
2914 | uint32_t config, uint32_t mask) | |
2915 | { | |
2916 | mask &= config ^ port->opp.config; | |
2917 | if (mask & OFPPC_PORT_DOWN) { | |
2918 | if (config & OFPPC_PORT_DOWN) { | |
2919 | netdev_turn_flags_off(port->netdev, NETDEV_UP, true); | |
2920 | } else { | |
2921 | netdev_turn_flags_on(port->netdev, NETDEV_UP, true); | |
2922 | } | |
2923 | } | |
2924 | #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD) | |
2925 | if (mask & REVALIDATE_BITS) { | |
2926 | COVERAGE_INC(ofproto_costly_flags); | |
2927 | port->opp.config ^= mask & REVALIDATE_BITS; | |
2928 | p->need_revalidate = true; | |
2929 | } | |
2930 | #undef REVALIDATE_BITS | |
2931 | if (mask & OFPPC_NO_FLOOD) { | |
2932 | port->opp.config ^= OFPPC_NO_FLOOD; | |
72b06300 | 2933 | refresh_port_groups(p); |
064af421 BP |
2934 | } |
2935 | if (mask & OFPPC_NO_PACKET_IN) { | |
2936 | port->opp.config ^= OFPPC_NO_PACKET_IN; | |
2937 | } | |
2938 | } | |
2939 | ||
2940 | static int | |
9deba63b BP |
2941 | handle_port_mod(struct ofproto *p, struct ofconn *ofconn, |
2942 | struct ofp_header *oh) | |
064af421 BP |
2943 | { |
2944 | const struct ofp_port_mod *opm; | |
2945 | struct ofport *port; | |
2946 | int error; | |
2947 | ||
9deba63b BP |
2948 | error = reject_slave_controller(ofconn, oh); |
2949 | if (error) { | |
2950 | return error; | |
2951 | } | |
064af421 BP |
2952 | error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm); |
2953 | if (error) { | |
2954 | return error; | |
2955 | } | |
2956 | opm = (struct ofp_port_mod *) oh; | |
2957 | ||
ca0f572c | 2958 | port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no))); |
064af421 BP |
2959 | if (!port) { |
2960 | return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT); | |
2961 | } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) { | |
2962 | return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR); | |
2963 | } else { | |
2964 | update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask)); | |
2965 | if (opm->advertise) { | |
2966 | netdev_set_advertisements(port->netdev, ntohl(opm->advertise)); | |
2967 | } | |
2968 | } | |
2969 | return 0; | |
2970 | } | |
2971 | ||
2972 | static struct ofpbuf * | |
2973 | make_stats_reply(uint32_t xid, uint16_t type, size_t body_len) | |
2974 | { | |
2975 | struct ofp_stats_reply *osr; | |
2976 | struct ofpbuf *msg; | |
2977 | ||
2978 | msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX)); | |
2979 | osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg); | |
2980 | osr->type = type; | |
2981 | osr->flags = htons(0); | |
2982 | return msg; | |
2983 | } | |
2984 | ||
2985 | static struct ofpbuf * | |
2986 | start_stats_reply(const struct ofp_stats_request *request, size_t body_len) | |
2987 | { | |
2988 | return make_stats_reply(request->header.xid, request->type, body_len); | |
2989 | } | |
2990 | ||
2991 | static void * | |
2992 | append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp) | |
2993 | { | |
2994 | struct ofpbuf *msg = *msgp; | |
2995 | assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply)); | |
2996 | if (nbytes + msg->size > UINT16_MAX) { | |
2997 | struct ofp_stats_reply *reply = msg->data; | |
2998 | reply->flags = htons(OFPSF_REPLY_MORE); | |
2999 | *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes); | |
3000 | queue_tx(msg, ofconn, ofconn->reply_counter); | |
3001 | } | |
3002 | return ofpbuf_put_uninit(*msgp, nbytes); | |
3003 | } | |
3004 | ||
3005 | static int | |
3006 | handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn, | |
3007 | struct ofp_stats_request *request) | |
3008 | { | |
3009 | struct ofp_desc_stats *ods; | |
3010 | struct ofpbuf *msg; | |
3011 | ||
3012 | msg = start_stats_reply(request, sizeof *ods); | |
3013 | ods = append_stats_reply(sizeof *ods, ofconn, &msg); | |
5a719c38 JP |
3014 | memset(ods, 0, sizeof *ods); |
3015 | ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc); | |
3016 | ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc); | |
3017 | ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc); | |
3018 | ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num); | |
3019 | ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc); | |
064af421 BP |
3020 | queue_tx(msg, ofconn, ofconn->reply_counter); |
3021 | ||
3022 | return 0; | |
3023 | } | |
3024 | ||
3025 | static void | |
3026 | count_subrules(struct cls_rule *cls_rule, void *n_subrules_) | |
3027 | { | |
3028 | struct rule *rule = rule_from_cls_rule(cls_rule); | |
3029 | int *n_subrules = n_subrules_; | |
3030 | ||
3031 | if (rule->super) { | |
3032 | (*n_subrules)++; | |
3033 | } | |
3034 | } | |
3035 | ||
3036 | static int | |
3037 | handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn, | |
3038 | struct ofp_stats_request *request) | |
3039 | { | |
3040 | struct ofp_table_stats *ots; | |
3041 | struct ofpbuf *msg; | |
3042 | struct odp_stats dpstats; | |
3043 | int n_exact, n_subrules, n_wild; | |
3044 | ||
3045 | msg = start_stats_reply(request, sizeof *ots * 2); | |
3046 | ||
3047 | /* Count rules of various kinds. */ | |
3048 | n_subrules = 0; | |
3049 | classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules); | |
3050 | n_exact = classifier_count_exact(&p->cls) - n_subrules; | |
3051 | n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls); | |
3052 | ||
3053 | /* Hash table. */ | |
c228a364 | 3054 | dpif_get_dp_stats(p->dpif, &dpstats); |
064af421 BP |
3055 | ots = append_stats_reply(sizeof *ots, ofconn, &msg); |
3056 | memset(ots, 0, sizeof *ots); | |
3057 | ots->table_id = TABLEID_HASH; | |
3058 | strcpy(ots->name, "hash"); | |
3059 | ots->wildcards = htonl(0); | |
3060 | ots->max_entries = htonl(dpstats.max_capacity); | |
3061 | ots->active_count = htonl(n_exact); | |
3062 | ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit + | |
3063 | dpstats.n_missed); | |
3064 | ots->matched_count = htonll(dpstats.n_hit); /* XXX */ | |
3065 | ||
3066 | /* Classifier table. */ | |
3067 | ots = append_stats_reply(sizeof *ots, ofconn, &msg); | |
3068 | memset(ots, 0, sizeof *ots); | |
3069 | ots->table_id = TABLEID_CLASSIFIER; | |
3070 | strcpy(ots->name, "classifier"); | |
659586ef JG |
3071 | ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL) |
3072 | : htonl(OFPFW_ALL); | |
064af421 BP |
3073 | ots->max_entries = htonl(65536); |
3074 | ots->active_count = htonl(n_wild); | |
3075 | ots->lookup_count = htonll(0); /* XXX */ | |
3076 | ots->matched_count = htonll(0); /* XXX */ | |
3077 | ||
3078 | queue_tx(msg, ofconn, ofconn->reply_counter); | |
3079 | return 0; | |
3080 | } | |
3081 | ||
abaad8cf | 3082 | static void |
ca0f572c | 3083 | append_port_stat(struct ofport *port, struct ofconn *ofconn, |
a4948b95 | 3084 | struct ofpbuf **msgp) |
abaad8cf JP |
3085 | { |
3086 | struct netdev_stats stats; | |
3087 | struct ofp_port_stats *ops; | |
3088 | ||
d295e8e9 JP |
3089 | /* Intentionally ignore return value, since errors will set |
3090 | * 'stats' to all-1s, which is correct for OpenFlow, and | |
abaad8cf JP |
3091 | * netdev_get_stats() will log errors. */ |
3092 | netdev_get_stats(port->netdev, &stats); | |
3093 | ||
a4948b95 | 3094 | ops = append_stats_reply(sizeof *ops, ofconn, msgp); |
ca0f572c | 3095 | ops->port_no = htons(port->opp.port_no); |
abaad8cf JP |
3096 | memset(ops->pad, 0, sizeof ops->pad); |
3097 | ops->rx_packets = htonll(stats.rx_packets); | |
3098 | ops->tx_packets = htonll(stats.tx_packets); | |
3099 | ops->rx_bytes = htonll(stats.rx_bytes); | |
3100 | ops->tx_bytes = htonll(stats.tx_bytes); | |
3101 | ops->rx_dropped = htonll(stats.rx_dropped); | |
3102 | ops->tx_dropped = htonll(stats.tx_dropped); | |
3103 | ops->rx_errors = htonll(stats.rx_errors); | |
3104 | ops->tx_errors = htonll(stats.tx_errors); | |
3105 | ops->rx_frame_err = htonll(stats.rx_frame_errors); | |
3106 | ops->rx_over_err = htonll(stats.rx_over_errors); | |
3107 | ops->rx_crc_err = htonll(stats.rx_crc_errors); | |
3108 | ops->collisions = htonll(stats.collisions); | |
3109 | } | |
3110 | ||
064af421 BP |
3111 | static int |
3112 | handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn, | |
abaad8cf JP |
3113 | struct ofp_stats_request *osr, |
3114 | size_t arg_size) | |
064af421 | 3115 | { |
abaad8cf | 3116 | struct ofp_port_stats_request *psr; |
064af421 BP |
3117 | struct ofp_port_stats *ops; |
3118 | struct ofpbuf *msg; | |
3119 | struct ofport *port; | |
064af421 | 3120 | |
abaad8cf JP |
3121 | if (arg_size != sizeof *psr) { |
3122 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
3123 | } | |
3124 | psr = (struct ofp_port_stats_request *) osr->body; | |
3125 | ||
3126 | msg = start_stats_reply(osr, sizeof *ops * 16); | |
3127 | if (psr->port_no != htons(OFPP_NONE)) { | |
ca0f572c | 3128 | port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no))); |
abaad8cf | 3129 | if (port) { |
ca0f572c | 3130 | append_port_stat(port, ofconn, &msg); |
abaad8cf JP |
3131 | } |
3132 | } else { | |
4e8e4213 | 3133 | HMAP_FOR_EACH (port, hmap_node, &p->ports) { |
ca0f572c | 3134 | append_port_stat(port, ofconn, &msg); |
abaad8cf | 3135 | } |
064af421 BP |
3136 | } |
3137 | ||
3138 | queue_tx(msg, ofconn, ofconn->reply_counter); | |
3139 | return 0; | |
3140 | } | |
3141 | ||
3142 | struct flow_stats_cbdata { | |
3143 | struct ofproto *ofproto; | |
3144 | struct ofconn *ofconn; | |
3145 | uint16_t out_port; | |
3146 | struct ofpbuf *msg; | |
3147 | }; | |
3148 | ||
01149cfd BP |
3149 | /* Obtains statistic counters for 'rule' within 'p' and stores them into |
3150 | * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the | |
3151 | * returned statistic include statistics for all of 'rule''s subrules. */ | |
064af421 BP |
3152 | static void |
3153 | query_stats(struct ofproto *p, struct rule *rule, | |
3154 | uint64_t *packet_countp, uint64_t *byte_countp) | |
3155 | { | |
3156 | uint64_t packet_count, byte_count; | |
3157 | struct rule *subrule; | |
3158 | struct odp_flow *odp_flows; | |
3159 | size_t n_odp_flows; | |
3160 | ||
01149cfd BP |
3161 | /* Start from historical data for 'rule' itself that are no longer tracked |
3162 | * by the datapath. This counts, for example, subrules that have | |
3163 | * expired. */ | |
b3137fe8 JG |
3164 | packet_count = rule->packet_count; |
3165 | byte_count = rule->byte_count; | |
3166 | ||
01149cfd BP |
3167 | /* Prepare to ask the datapath for statistics on 'rule', or if it is |
3168 | * wildcarded then on all of its subrules. | |
3169 | * | |
3170 | * Also, add any statistics that are not tracked by the datapath for each | |
3171 | * subrule. This includes, for example, statistics for packets that were | |
3172 | * executed "by hand" by ofproto via dpif_execute() but must be accounted | |
3173 | * to a flow. */ | |
064af421 | 3174 | n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1; |
ec6fde61 | 3175 | odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows); |
064af421 BP |
3176 | if (rule->cr.wc.wildcards) { |
3177 | size_t i = 0; | |
4e8e4213 | 3178 | LIST_FOR_EACH (subrule, list, &rule->list) { |
064af421 | 3179 | odp_flows[i++].key = subrule->cr.flow; |
b3137fe8 JG |
3180 | packet_count += subrule->packet_count; |
3181 | byte_count += subrule->byte_count; | |
064af421 BP |
3182 | } |
3183 | } else { | |
3184 | odp_flows[0].key = rule->cr.flow; | |
3185 | } | |
3186 | ||
28998b22 | 3187 | /* Fetch up-to-date statistics from the datapath and add them in. */ |
c228a364 | 3188 | if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) { |
064af421 BP |
3189 | size_t i; |
3190 | for (i = 0; i < n_odp_flows; i++) { | |
3191 | struct odp_flow *odp_flow = &odp_flows[i]; | |
3192 | packet_count += odp_flow->stats.n_packets; | |
3193 | byte_count += odp_flow->stats.n_bytes; | |
3194 | } | |
3195 | } | |
3196 | free(odp_flows); | |
3197 | ||
01149cfd | 3198 | /* Return the stats to the caller. */ |
064af421 BP |
3199 | *packet_countp = packet_count; |
3200 | *byte_countp = byte_count; | |
3201 | } | |
3202 | ||
3203 | static void | |
3204 | flow_stats_cb(struct cls_rule *rule_, void *cbdata_) | |
3205 | { | |
3206 | struct rule *rule = rule_from_cls_rule(rule_); | |
3207 | struct flow_stats_cbdata *cbdata = cbdata_; | |
3208 | struct ofp_flow_stats *ofs; | |
3209 | uint64_t packet_count, byte_count; | |
3210 | size_t act_len, len; | |
26c3f94a JP |
3211 | long long int tdiff = time_msec() - rule->created; |
3212 | uint32_t sec = tdiff / 1000; | |
3213 | uint32_t msec = tdiff - (sec * 1000); | |
064af421 BP |
3214 | |
3215 | if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) { | |
3216 | return; | |
3217 | } | |
3218 | ||
3219 | act_len = sizeof *rule->actions * rule->n_actions; | |
3220 | len = offsetof(struct ofp_flow_stats, actions) + act_len; | |
3221 | ||
3222 | query_stats(cbdata->ofproto, rule, &packet_count, &byte_count); | |
3223 | ||
3224 | ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg); | |
3225 | ofs->length = htons(len); | |
3226 | ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH; | |
3227 | ofs->pad = 0; | |
659586ef JG |
3228 | flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, |
3229 | cbdata->ofproto->tun_id_from_cookie, &ofs->match); | |
26c3f94a JP |
3230 | ofs->duration_sec = htonl(sec); |
3231 | ofs->duration_nsec = htonl(msec * 1000000); | |
39997502 | 3232 | ofs->cookie = rule->flow_cookie; |
064af421 BP |
3233 | ofs->priority = htons(rule->cr.priority); |
3234 | ofs->idle_timeout = htons(rule->idle_timeout); | |
3235 | ofs->hard_timeout = htons(rule->hard_timeout); | |
39997502 | 3236 | memset(ofs->pad2, 0, sizeof ofs->pad2); |
064af421 BP |
3237 | ofs->packet_count = htonll(packet_count); |
3238 | ofs->byte_count = htonll(byte_count); | |
3239 | memcpy(ofs->actions, rule->actions, act_len); | |
3240 | } | |
3241 | ||
3242 | static int | |
3243 | table_id_to_include(uint8_t table_id) | |
3244 | { | |
3245 | return (table_id == TABLEID_HASH ? CLS_INC_EXACT | |
3246 | : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD | |
3247 | : table_id == 0xff ? CLS_INC_ALL | |
3248 | : 0); | |
3249 | } | |
3250 | ||
3251 | static int | |
3252 | handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn, | |
3253 | const struct ofp_stats_request *osr, | |
3254 | size_t arg_size) | |
3255 | { | |
3256 | struct ofp_flow_stats_request *fsr; | |
3257 | struct flow_stats_cbdata cbdata; | |
3258 | struct cls_rule target; | |
3259 | ||
3260 | if (arg_size != sizeof *fsr) { | |
49bdc010 | 3261 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); |
064af421 BP |
3262 | } |
3263 | fsr = (struct ofp_flow_stats_request *) osr->body; | |
3264 | ||
3265 | COVERAGE_INC(ofproto_flows_req); | |
3266 | cbdata.ofproto = p; | |
3267 | cbdata.ofconn = ofconn; | |
3268 | cbdata.out_port = fsr->out_port; | |
3269 | cbdata.msg = start_stats_reply(osr, 1024); | |
659586ef | 3270 | cls_rule_from_match(&fsr->match, 0, false, 0, &target); |
064af421 BP |
3271 | classifier_for_each_match(&p->cls, &target, |
3272 | table_id_to_include(fsr->table_id), | |
3273 | flow_stats_cb, &cbdata); | |
3274 | queue_tx(cbdata.msg, ofconn, ofconn->reply_counter); | |
3275 | return 0; | |
3276 | } | |
3277 | ||
4f2cad2c JP |
3278 | struct flow_stats_ds_cbdata { |
3279 | struct ofproto *ofproto; | |
3280 | struct ds *results; | |
3281 | }; | |
3282 | ||
3283 | static void | |
3284 | flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_) | |
3285 | { | |
3286 | struct rule *rule = rule_from_cls_rule(rule_); | |
3287 | struct flow_stats_ds_cbdata *cbdata = cbdata_; | |
3288 | struct ds *results = cbdata->results; | |
3289 | struct ofp_match match; | |
3290 | uint64_t packet_count, byte_count; | |
3291 | size_t act_len = sizeof *rule->actions * rule->n_actions; | |
3292 | ||
3293 | /* Don't report on subrules. */ | |
3294 | if (rule->super != NULL) { | |
3295 | return; | |
3296 | } | |
3297 | ||
3298 | query_stats(cbdata->ofproto, rule, &packet_count, &byte_count); | |
659586ef JG |
3299 | flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, |
3300 | cbdata->ofproto->tun_id_from_cookie, &match); | |
4f2cad2c JP |
3301 | |
3302 | ds_put_format(results, "duration=%llds, ", | |
3303 | (time_msec() - rule->created) / 1000); | |
52ae00b3 | 3304 | ds_put_format(results, "priority=%u, ", rule->cr.priority); |
4f2cad2c JP |
3305 | ds_put_format(results, "n_packets=%"PRIu64", ", packet_count); |
3306 | ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count); | |
3307 | ofp_print_match(results, &match, true); | |
3308 | ofp_print_actions(results, &rule->actions->header, act_len); | |
3309 | ds_put_cstr(results, "\n"); | |
3310 | } | |
3311 | ||
d295e8e9 | 3312 | /* Adds a pretty-printed description of all flows to 'results', including |
4f2cad2c JP |
3313 | * those marked hidden by secchan (e.g., by in-band control). */ |
3314 | void | |
3315 | ofproto_get_all_flows(struct ofproto *p, struct ds *results) | |
3316 | { | |
3317 | struct ofp_match match; | |
3318 | struct cls_rule target; | |
3319 | struct flow_stats_ds_cbdata cbdata; | |
3320 | ||
3321 | memset(&match, 0, sizeof match); | |
659586ef | 3322 | match.wildcards = htonl(OVSFW_ALL); |
4f2cad2c JP |
3323 | |
3324 | cbdata.ofproto = p; | |
3325 | cbdata.results = results; | |
3326 | ||
659586ef | 3327 | cls_rule_from_match(&match, 0, false, 0, &target); |
4f2cad2c JP |
3328 | classifier_for_each_match(&p->cls, &target, CLS_INC_ALL, |
3329 | flow_stats_ds_cb, &cbdata); | |
3330 | } | |
3331 | ||
064af421 BP |
3332 | struct aggregate_stats_cbdata { |
3333 | struct ofproto *ofproto; | |
3334 | uint16_t out_port; | |
3335 | uint64_t packet_count; | |
3336 | uint64_t byte_count; | |
3337 | uint32_t n_flows; | |
3338 | }; | |
3339 | ||
3340 | static void | |
3341 | aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_) | |
3342 | { | |
3343 | struct rule *rule = rule_from_cls_rule(rule_); | |
3344 | struct aggregate_stats_cbdata *cbdata = cbdata_; | |
3345 | uint64_t packet_count, byte_count; | |
3346 | ||
3347 | if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) { | |
3348 | return; | |
3349 | } | |
3350 | ||
3351 | query_stats(cbdata->ofproto, rule, &packet_count, &byte_count); | |
3352 | ||
3353 | cbdata->packet_count += packet_count; | |
3354 | cbdata->byte_count += byte_count; | |
3355 | cbdata->n_flows++; | |
3356 | } | |
3357 | ||
3358 | static int | |
3359 | handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn, | |
3360 | const struct ofp_stats_request *osr, | |
3361 | size_t arg_size) | |
3362 | { | |
3363 | struct ofp_aggregate_stats_request *asr; | |
3364 | struct ofp_aggregate_stats_reply *reply; | |
3365 | struct aggregate_stats_cbdata cbdata; | |
3366 | struct cls_rule target; | |
3367 | struct ofpbuf *msg; | |
3368 | ||
3369 | if (arg_size != sizeof *asr) { | |
49bdc010 | 3370 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); |
064af421 BP |
3371 | } |
3372 | asr = (struct ofp_aggregate_stats_request *) osr->body; | |
3373 | ||
3374 | COVERAGE_INC(ofproto_agg_request); | |
3375 | cbdata.ofproto = p; | |
3376 | cbdata.out_port = asr->out_port; | |
3377 | cbdata.packet_count = 0; | |
3378 | cbdata.byte_count = 0; | |
3379 | cbdata.n_flows = 0; | |
659586ef | 3380 | cls_rule_from_match(&asr->match, 0, false, 0, &target); |
064af421 BP |
3381 | classifier_for_each_match(&p->cls, &target, |
3382 | table_id_to_include(asr->table_id), | |
3383 | aggregate_stats_cb, &cbdata); | |
3384 | ||
3385 | msg = start_stats_reply(osr, sizeof *reply); | |
3386 | reply = append_stats_reply(sizeof *reply, ofconn, &msg); | |
3387 | reply->flow_count = htonl(cbdata.n_flows); | |
3388 | reply->packet_count = htonll(cbdata.packet_count); | |
3389 | reply->byte_count = htonll(cbdata.byte_count); | |
3390 | queue_tx(msg, ofconn, ofconn->reply_counter); | |
3391 | return 0; | |
3392 | } | |
3393 | ||
c1c9c9c4 BP |
3394 | struct queue_stats_cbdata { |
3395 | struct ofconn *ofconn; | |
ca0f572c | 3396 | struct ofport *ofport; |
c1c9c9c4 | 3397 | struct ofpbuf *msg; |
c1c9c9c4 BP |
3398 | }; |
3399 | ||
3400 | static void | |
db9220c3 | 3401 | put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id, |
c1c9c9c4 BP |
3402 | const struct netdev_queue_stats *stats) |
3403 | { | |
3404 | struct ofp_queue_stats *reply; | |
3405 | ||
3406 | reply = append_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg); | |
ca0f572c | 3407 | reply->port_no = htons(cbdata->ofport->opp.port_no); |
c1c9c9c4 BP |
3408 | memset(reply->pad, 0, sizeof reply->pad); |
3409 | reply->queue_id = htonl(queue_id); | |
3410 | reply->tx_bytes = htonll(stats->tx_bytes); | |
3411 | reply->tx_packets = htonll(stats->tx_packets); | |
3412 | reply->tx_errors = htonll(stats->tx_errors); | |
3413 | } | |
3414 | ||
3415 | static void | |
db9220c3 | 3416 | handle_queue_stats_dump_cb(uint32_t queue_id, |
c1c9c9c4 BP |
3417 | struct netdev_queue_stats *stats, |
3418 | void *cbdata_) | |
3419 | { | |
3420 | struct queue_stats_cbdata *cbdata = cbdata_; | |
3421 | ||
3422 | put_queue_stats(cbdata, queue_id, stats); | |
3423 | } | |
3424 | ||
3425 | static void | |
ca0f572c | 3426 | handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id, |
c1c9c9c4 BP |
3427 | struct queue_stats_cbdata *cbdata) |
3428 | { | |
ca0f572c | 3429 | cbdata->ofport = port; |
c1c9c9c4 BP |
3430 | if (queue_id == OFPQ_ALL) { |
3431 | netdev_dump_queue_stats(port->netdev, | |
3432 | handle_queue_stats_dump_cb, cbdata); | |
3433 | } else { | |
3434 | struct netdev_queue_stats stats; | |
3435 | ||
3436 | netdev_get_queue_stats(port->netdev, queue_id, &stats); | |
3437 | put_queue_stats(cbdata, queue_id, &stats); | |
3438 | } | |
3439 | } | |
3440 | ||
3441 | static int | |
3442 | handle_queue_stats_request(struct ofproto *ofproto, struct ofconn *ofconn, | |
3443 | const struct ofp_stats_request *osr, | |
3444 | size_t arg_size) | |
3445 | { | |
3446 | struct ofp_queue_stats_request *qsr; | |
3447 | struct queue_stats_cbdata cbdata; | |
3448 | struct ofport *port; | |
3449 | unsigned int port_no; | |
3450 | uint32_t queue_id; | |
3451 | ||
3452 | if (arg_size != sizeof *qsr) { | |
3453 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
3454 | } | |
3455 | qsr = (struct ofp_queue_stats_request *) osr->body; | |
3456 | ||
3457 | COVERAGE_INC(ofproto_queue_req); | |
3458 | ||
3459 | cbdata.ofconn = ofconn; | |
3460 | cbdata.msg = start_stats_reply(osr, 128); | |
3461 | ||
3462 | port_no = ntohs(qsr->port_no); | |
3463 | queue_id = ntohl(qsr->queue_id); | |
3464 | if (port_no == OFPP_ALL) { | |
4e8e4213 | 3465 | HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) { |
ca0f572c | 3466 | handle_queue_stats_for_port(port, queue_id, &cbdata); |
c1c9c9c4 BP |
3467 | } |
3468 | } else if (port_no < ofproto->max_ports) { | |
ca0f572c | 3469 | port = get_port(ofproto, ofp_port_to_odp_port(port_no)); |
c1c9c9c4 | 3470 | if (port) { |
ca0f572c | 3471 | handle_queue_stats_for_port(port, queue_id, &cbdata); |
c1c9c9c4 BP |
3472 | } |
3473 | } else { | |
3474 | ofpbuf_delete(cbdata.msg); | |
3475 | return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT); | |
3476 | } | |
3477 | queue_tx(cbdata.msg, ofconn, ofconn->reply_counter); | |
3478 | ||
3479 | return 0; | |
3480 | } | |
3481 | ||
064af421 BP |
3482 | static int |
3483 | handle_stats_request(struct ofproto *p, struct ofconn *ofconn, | |
3484 | struct ofp_header *oh) | |
3485 | { | |
3486 | struct ofp_stats_request *osr; | |
3487 | size_t arg_size; | |
3488 | int error; | |
3489 | ||
3490 | error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr, | |
3491 | 1, &arg_size); | |
3492 | if (error) { | |
3493 | return error; | |
3494 | } | |
3495 | osr = (struct ofp_stats_request *) oh; | |
3496 | ||
3497 | switch (ntohs(osr->type)) { | |
3498 | case OFPST_DESC: | |
3499 | return handle_desc_stats_request(p, ofconn, osr); | |
3500 | ||
3501 | case OFPST_FLOW: | |
3502 | return handle_flow_stats_request(p, ofconn, osr, arg_size); | |
3503 | ||
3504 | case OFPST_AGGREGATE: | |
3505 | return handle_aggregate_stats_request(p, ofconn, osr, arg_size); | |
3506 | ||
3507 | case OFPST_TABLE: | |
3508 | return handle_table_stats_request(p, ofconn, osr); | |
3509 | ||
3510 | case OFPST_PORT: | |
abaad8cf | 3511 | return handle_port_stats_request(p, ofconn, osr, arg_size); |
064af421 | 3512 | |
c1c9c9c4 BP |
3513 | case OFPST_QUEUE: |
3514 | return handle_queue_stats_request(p, ofconn, osr, arg_size); | |
3515 | ||
064af421 BP |
3516 | case OFPST_VENDOR: |
3517 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR); | |
3518 | ||
3519 | default: | |
3520 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT); | |
3521 | } | |
3522 | } | |
3523 | ||
3524 | static long long int | |
3525 | msec_from_nsec(uint64_t sec, uint32_t nsec) | |
3526 | { | |
3527 | return !sec ? 0 : sec * 1000 + nsec / 1000000; | |
3528 | } | |
3529 | ||
3530 | static void | |
0193b2af JG |
3531 | update_time(struct ofproto *ofproto, struct rule *rule, |
3532 | const struct odp_flow_stats *stats) | |
064af421 BP |
3533 | { |
3534 | long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec); | |
3535 | if (used > rule->used) { | |
3536 | rule->used = used; | |
4836f9f2 JP |
3537 | if (rule->super && used > rule->super->used) { |
3538 | rule->super->used = used; | |
3539 | } | |
0193b2af | 3540 | netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used); |
064af421 BP |
3541 | } |
3542 | } | |
3543 | ||
3544 | static void | |
0193b2af JG |
3545 | update_stats(struct ofproto *ofproto, struct rule *rule, |
3546 | const struct odp_flow_stats *stats) | |
064af421 | 3547 | { |
064af421 | 3548 | if (stats->n_packets) { |
0193b2af JG |
3549 | update_time(ofproto, rule, stats); |
3550 | rule->packet_count += stats->n_packets; | |
3551 | rule->byte_count += stats->n_bytes; | |
abfec865 | 3552 | netflow_flow_update_flags(&rule->nf_flow, stats->tcp_flags); |
064af421 BP |
3553 | } |
3554 | } | |
3555 | ||
79eee1eb BP |
3556 | /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT |
3557 | * in which no matching flow already exists in the flow table. | |
3558 | * | |
3559 | * Adds the flow specified by 'ofm', which is followed by 'n_actions' | |
3560 | * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error | |
3561 | * code as encoded by ofp_mkerr() on failure. | |
3562 | * | |
3563 | * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, | |
3564 | * if any. */ | |
064af421 BP |
3565 | static int |
3566 | add_flow(struct ofproto *p, struct ofconn *ofconn, | |
79eee1eb | 3567 | const struct ofp_flow_mod *ofm, size_t n_actions) |
064af421 BP |
3568 | { |
3569 | struct ofpbuf *packet; | |
3570 | struct rule *rule; | |
3571 | uint16_t in_port; | |
3572 | int error; | |
3573 | ||
49bdc010 JP |
3574 | if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) { |
3575 | flow_t flow; | |
3576 | uint32_t wildcards; | |
3577 | ||
659586ef JG |
3578 | flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie, |
3579 | &flow, &wildcards); | |
49bdc010 JP |
3580 | if (classifier_rule_overlaps(&p->cls, &flow, wildcards, |
3581 | ntohs(ofm->priority))) { | |
3582 | return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP); | |
3583 | } | |
3584 | } | |
3585 | ||
0193b2af | 3586 | rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions, |
064af421 | 3587 | n_actions, ntohs(ofm->idle_timeout), |
39997502 | 3588 | ntohs(ofm->hard_timeout), ofm->cookie, |
ca069229 | 3589 | ofm->flags & htons(OFPFF_SEND_FLOW_REM)); |
659586ef JG |
3590 | cls_rule_from_match(&ofm->match, ntohs(ofm->priority), |
3591 | p->tun_id_from_cookie, ofm->cookie, &rule->cr); | |
064af421 | 3592 | |
064af421 BP |
3593 | error = 0; |
3594 | if (ofm->buffer_id != htonl(UINT32_MAX)) { | |
3595 | error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id), | |
3596 | &packet, &in_port); | |
212fe71c BP |
3597 | } else { |
3598 | packet = NULL; | |
165cd8a3 | 3599 | in_port = UINT16_MAX; |
064af421 BP |
3600 | } |
3601 | ||
3602 | rule_insert(p, rule, packet, in_port); | |
064af421 BP |
3603 | return error; |
3604 | } | |
3605 | ||
79eee1eb BP |
3606 | static struct rule * |
3607 | find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm) | |
064af421 | 3608 | { |
064af421 BP |
3609 | uint32_t wildcards; |
3610 | flow_t flow; | |
3611 | ||
659586ef JG |
3612 | flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie, |
3613 | &flow, &wildcards); | |
79eee1eb | 3614 | return rule_from_cls_rule(classifier_find_rule_exactly( |
064af421 BP |
3615 | &p->cls, &flow, wildcards, |
3616 | ntohs(ofm->priority))); | |
79eee1eb | 3617 | } |
064af421 | 3618 | |
79eee1eb BP |
3619 | static int |
3620 | send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn, | |
3621 | struct rule *rule, const struct ofp_flow_mod *ofm) | |
3622 | { | |
3623 | struct ofpbuf *packet; | |
3624 | uint16_t in_port; | |
3625 | flow_t flow; | |
3626 | int error; | |
064af421 | 3627 | |
79eee1eb BP |
3628 | if (ofm->buffer_id == htonl(UINT32_MAX)) { |
3629 | return 0; | |
064af421 | 3630 | } |
79eee1eb BP |
3631 | |
3632 | error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id), | |
3633 | &packet, &in_port); | |
3634 | if (error) { | |
3635 | return error; | |
3636 | } | |
3637 | ||
659586ef | 3638 | flow_extract(packet, 0, in_port, &flow); |
79eee1eb | 3639 | rule_execute(ofproto, rule, packet, &flow); |
79eee1eb | 3640 | |
064af421 BP |
3641 | return 0; |
3642 | } | |
79eee1eb BP |
3643 | \f |
3644 | /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */ | |
064af421 BP |
3645 | |
3646 | struct modify_flows_cbdata { | |
3647 | struct ofproto *ofproto; | |
3648 | const struct ofp_flow_mod *ofm; | |
064af421 | 3649 | size_t n_actions; |
79eee1eb | 3650 | struct rule *match; |
064af421 BP |
3651 | }; |
3652 | ||
79eee1eb BP |
3653 | static int modify_flow(struct ofproto *, const struct ofp_flow_mod *, |
3654 | size_t n_actions, struct rule *); | |
3655 | static void modify_flows_cb(struct cls_rule *, void *cbdata_); | |
3656 | ||
3657 | /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as | |
3658 | * encoded by ofp_mkerr() on failure. | |
3659 | * | |
3660 | * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, | |
3661 | * if any. */ | |
3662 | static int | |
3663 | modify_flows_loose(struct ofproto *p, struct ofconn *ofconn, | |
3664 | const struct ofp_flow_mod *ofm, size_t n_actions) | |
3665 | { | |
3666 | struct modify_flows_cbdata cbdata; | |
3667 | struct cls_rule target; | |
3668 | ||
3669 | cbdata.ofproto = p; | |
3670 | cbdata.ofm = ofm; | |
3671 | cbdata.n_actions = n_actions; | |
3672 | cbdata.match = NULL; | |
3673 | ||
659586ef JG |
3674 | cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie, |
3675 | &target); | |
79eee1eb BP |
3676 | |
3677 | classifier_for_each_match(&p->cls, &target, CLS_INC_ALL, | |
3678 | modify_flows_cb, &cbdata); | |
3679 | if (cbdata.match) { | |
3680 | /* This credits the packet to whichever flow happened to happened to | |
3681 | * match last. That's weird. Maybe we should do a lookup for the | |
3682 | * flow that actually matches the packet? Who knows. */ | |
3683 | send_buffered_packet(p, ofconn, cbdata.match, ofm); | |
3684 | return 0; | |
3685 | } else { | |
3686 | return add_flow(p, ofconn, ofm, n_actions); | |
3687 | } | |
3688 | } | |
3689 | ||
3690 | /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error | |
3691 | * code as encoded by ofp_mkerr() on failure. | |
3692 | * | |
3693 | * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, | |
3694 | * if any. */ | |
3695 | static int | |
3696 | modify_flow_strict(struct ofproto *p, struct ofconn *ofconn, | |
3697 | struct ofp_flow_mod *ofm, size_t n_actions) | |
3698 | { | |
3699 | struct rule *rule = find_flow_strict(p, ofm); | |
3700 | if (rule && !rule_is_hidden(rule)) { | |
3701 | modify_flow(p, ofm, n_actions, rule); | |
3702 | return send_buffered_packet(p, ofconn, rule, ofm); | |
3703 | } else { | |
3704 | return add_flow(p, ofconn, ofm, n_actions); | |
3705 | } | |
3706 | } | |
3707 | ||
3708 | /* Callback for modify_flows_loose(). */ | |
064af421 BP |
3709 | static void |
3710 | modify_flows_cb(struct cls_rule *rule_, void *cbdata_) | |
3711 | { | |
3712 | struct rule *rule = rule_from_cls_rule(rule_); | |
3713 | struct modify_flows_cbdata *cbdata = cbdata_; | |
3714 | ||
79eee1eb BP |
3715 | if (!rule_is_hidden(rule)) { |
3716 | cbdata->match = rule; | |
3717 | modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule); | |
064af421 | 3718 | } |
064af421 BP |
3719 | } |
3720 | ||
79eee1eb BP |
3721 | /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has |
3722 | * been identified as a flow in 'p''s flow table to be modified, by changing | |
3723 | * the rule's actions to match those in 'ofm' (which is followed by 'n_actions' | |
3724 | * ofp_action[] structures). */ | |
064af421 | 3725 | static int |
79eee1eb BP |
3726 | modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm, |
3727 | size_t n_actions, struct rule *rule) | |
064af421 | 3728 | { |
79eee1eb BP |
3729 | size_t actions_len = n_actions * sizeof *rule->actions; |
3730 | ||
3731 | rule->flow_cookie = ofm->cookie; | |
3732 | ||
3733 | /* If the actions are the same, do nothing. */ | |
3734 | if (n_actions == rule->n_actions | |
3735 | && !memcmp(ofm->actions, rule->actions, actions_len)) | |
3736 | { | |
3737 | return 0; | |
3738 | } | |
3739 | ||
3740 | /* Replace actions. */ | |
3741 | free(rule->actions); | |
3742 | rule->actions = xmemdup(ofm->actions, actions_len); | |
3743 | rule->n_actions = n_actions; | |
3744 | ||
3745 | /* Make sure that the datapath gets updated properly. */ | |
3746 | if (rule->cr.wc.wildcards) { | |
3747 | COVERAGE_INC(ofproto_mod_wc_flow); | |
3748 | p->need_revalidate = true; | |
3749 | } else { | |
3750 | rule_update_actions(p, rule); | |
3751 | } | |
3752 | ||
3753 | return 0; | |
3754 | } | |
3755 | \f | |
3756 | /* OFPFC_DELETE implementation. */ | |
3757 | ||
3758 | struct delete_flows_cbdata { | |
3759 | struct ofproto *ofproto; | |
3760 | uint16_t out_port; | |
3761 | }; | |
3762 | ||
3763 | static void delete_flows_cb(struct cls_rule *, void *cbdata_); | |
3764 | static void delete_flow(struct ofproto *, struct rule *, uint16_t out_port); | |
3765 | ||
3766 | /* Implements OFPFC_DELETE. */ | |
3767 | static void | |
3768 | delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm) | |
3769 | { | |
3770 | struct delete_flows_cbdata cbdata; | |
064af421 BP |
3771 | struct cls_rule target; |
3772 | ||
3773 | cbdata.ofproto = p; | |
79eee1eb | 3774 | cbdata.out_port = ofm->out_port; |
064af421 | 3775 | |
659586ef JG |
3776 | cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie, |
3777 | &target); | |
064af421 BP |
3778 | |
3779 | classifier_for_each_match(&p->cls, &target, CLS_INC_ALL, | |
79eee1eb | 3780 | delete_flows_cb, &cbdata); |
064af421 BP |
3781 | } |
3782 | ||
79eee1eb BP |
3783 | /* Implements OFPFC_DELETE_STRICT. */ |
3784 | static void | |
3785 | delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm) | |
3786 | { | |
3787 | struct rule *rule = find_flow_strict(p, ofm); | |
3788 | if (rule) { | |
3789 | delete_flow(p, rule, ofm->out_port); | |
3790 | } | |
3791 | } | |
3792 | ||
3793 | /* Callback for delete_flows_loose(). */ | |
3794 | static void | |
3795 | delete_flows_cb(struct cls_rule *rule_, void *cbdata_) | |
3796 | { | |
3797 | struct rule *rule = rule_from_cls_rule(rule_); | |
3798 | struct delete_flows_cbdata *cbdata = cbdata_; | |
3799 | ||
3800 | delete_flow(cbdata->ofproto, rule, cbdata->out_port); | |
3801 | } | |
3802 | ||
3803 | /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has | |
3804 | * been identified as a flow to delete from 'p''s flow table, by deleting the | |
3805 | * flow and sending out a OFPT_FLOW_REMOVED message to any interested | |
3806 | * controller. | |
3807 | * | |
3808 | * Will not delete 'rule' if it is hidden. Will delete 'rule' only if | |
3809 | * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the | |
3810 | * specified 'out_port'. */ | |
3811 | static void | |
3812 | delete_flow(struct ofproto *p, struct rule *rule, uint16_t out_port) | |
3813 | { | |
3814 | if (rule_is_hidden(rule)) { | |
3815 | return; | |
3816 | } | |
3817 | ||
3818 | if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) { | |
3819 | return; | |
3820 | } | |
3821 | ||
3822 | send_flow_removed(p, rule, time_msec(), OFPRR_DELETE); | |
3823 | rule_remove(p, rule); | |
3824 | } | |
3825 | \f | |
064af421 BP |
3826 | static int |
3827 | handle_flow_mod(struct ofproto *p, struct ofconn *ofconn, | |
3828 | struct ofp_flow_mod *ofm) | |
3829 | { | |
3f09c339 | 3830 | struct ofp_match orig_match; |
064af421 BP |
3831 | size_t n_actions; |
3832 | int error; | |
3833 | ||
9deba63b BP |
3834 | error = reject_slave_controller(ofconn, &ofm->header); |
3835 | if (error) { | |
3836 | return error; | |
3837 | } | |
064af421 BP |
3838 | error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm, |
3839 | sizeof *ofm->actions, &n_actions); | |
3840 | if (error) { | |
3841 | return error; | |
3842 | } | |
3843 | ||
49bdc010 JP |
3844 | /* We do not support the emergency flow cache. It will hopefully |
3845 | * get dropped from OpenFlow in the near future. */ | |
3846 | if (ofm->flags & htons(OFPFF_EMERG)) { | |
3847 | /* There isn't a good fit for an error code, so just state that the | |
3848 | * flow table is full. */ | |
3849 | return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL); | |
3850 | } | |
3851 | ||
3f09c339 BP |
3852 | /* Normalize ofp->match. If normalization actually changes anything, then |
3853 | * log the differences. */ | |
3854 | ofm->match.pad1[0] = ofm->match.pad2[0] = 0; | |
3855 | orig_match = ofm->match; | |
064af421 | 3856 | normalize_match(&ofm->match); |
3f09c339 BP |
3857 | if (memcmp(&ofm->match, &orig_match, sizeof orig_match)) { |
3858 | static struct vlog_rate_limit normal_rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
3859 | if (!VLOG_DROP_INFO(&normal_rl)) { | |
3860 | char *old = ofp_match_to_literal_string(&orig_match); | |
3861 | char *new = ofp_match_to_literal_string(&ofm->match); | |
3862 | VLOG_INFO("%s: normalization changed ofp_match, details:", | |
3863 | rconn_get_name(ofconn->rconn)); | |
3864 | VLOG_INFO(" pre: %s", old); | |
3865 | VLOG_INFO("post: %s", new); | |
3866 | free(old); | |
3867 | free(new); | |
3868 | } | |
3869 | } | |
3870 | ||
064af421 BP |
3871 | if (!ofm->match.wildcards) { |
3872 | ofm->priority = htons(UINT16_MAX); | |
3873 | } | |
3874 | ||
3875 | error = validate_actions((const union ofp_action *) ofm->actions, | |
3876 | n_actions, p->max_ports); | |
3877 | if (error) { | |
3878 | return error; | |
3879 | } | |
3880 | ||
3881 | switch (ntohs(ofm->command)) { | |
3882 | case OFPFC_ADD: | |
3883 | return add_flow(p, ofconn, ofm, n_actions); | |
3884 | ||
3885 | case OFPFC_MODIFY: | |
79eee1eb | 3886 | return modify_flows_loose(p, ofconn, ofm, n_actions); |
064af421 BP |
3887 | |
3888 | case OFPFC_MODIFY_STRICT: | |
79eee1eb | 3889 | return modify_flow_strict(p, ofconn, ofm, n_actions); |
064af421 BP |
3890 | |
3891 | case OFPFC_DELETE: | |
79eee1eb BP |
3892 | delete_flows_loose(p, ofm); |
3893 | return 0; | |
064af421 BP |
3894 | |
3895 | case OFPFC_DELETE_STRICT: | |
79eee1eb BP |
3896 | delete_flow_strict(p, ofm); |
3897 | return 0; | |
064af421 BP |
3898 | |
3899 | default: | |
3900 | return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND); | |
3901 | } | |
3902 | } | |
3903 | ||
659586ef JG |
3904 | static int |
3905 | handle_tun_id_from_cookie(struct ofproto *p, struct nxt_tun_id_cookie *msg) | |
3906 | { | |
3907 | int error; | |
3908 | ||
3909 | error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg); | |
3910 | if (error) { | |
3911 | return error; | |
3912 | } | |
3913 | ||
3914 | p->tun_id_from_cookie = !!msg->set; | |
3915 | return 0; | |
3916 | } | |
3917 | ||
9deba63b BP |
3918 | static int |
3919 | handle_role_request(struct ofproto *ofproto, | |
3920 | struct ofconn *ofconn, struct nicira_header *msg) | |
3921 | { | |
3922 | struct nx_role_request *nrr; | |
3923 | struct nx_role_request *reply; | |
3924 | struct ofpbuf *buf; | |
3925 | uint32_t role; | |
3926 | ||
3927 | if (ntohs(msg->header.length) != sizeof *nrr) { | |
100e95db | 3928 | VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)", |
9deba63b BP |
3929 | ntohs(msg->header.length), sizeof *nrr); |
3930 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
3931 | } | |
3932 | nrr = (struct nx_role_request *) msg; | |
3933 | ||
5899143f | 3934 | if (ofconn->type != OFCONN_PRIMARY) { |
9deba63b BP |
3935 | VLOG_WARN_RL(&rl, "ignoring role request on non-controller " |
3936 | "connection"); | |
3937 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM); | |
3938 | } | |
3939 | ||
3940 | role = ntohl(nrr->role); | |
3941 | if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER | |
3942 | && role != NX_ROLE_SLAVE) { | |
3943 | VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role); | |
3944 | ||
3945 | /* There's no good error code for this. */ | |
3946 | return ofp_mkerr(OFPET_BAD_REQUEST, -1); | |
3947 | } | |
3948 | ||
3949 | if (role == NX_ROLE_MASTER) { | |
3950 | struct ofconn *other; | |
3951 | ||
4e8e4213 | 3952 | HMAP_FOR_EACH (other, hmap_node, &ofproto->controllers) { |
9deba63b BP |
3953 | if (other->role == NX_ROLE_MASTER) { |
3954 | other->role = NX_ROLE_SLAVE; | |
3955 | } | |
3956 | } | |
3957 | } | |
3958 | ofconn->role = role; | |
3959 | ||
3960 | reply = make_openflow_xid(sizeof *reply, OFPT_VENDOR, msg->header.xid, | |
3961 | &buf); | |
3962 | reply->nxh.vendor = htonl(NX_VENDOR_ID); | |
3963 | reply->nxh.subtype = htonl(NXT_ROLE_REPLY); | |
3964 | reply->role = htonl(role); | |
3965 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
3966 | ||
3967 | return 0; | |
3968 | } | |
3969 | ||
064af421 BP |
3970 | static int |
3971 | handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg) | |
3972 | { | |
3973 | struct ofp_vendor_header *ovh = msg; | |
3974 | struct nicira_header *nh; | |
3975 | ||
3976 | if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) { | |
100e95db | 3977 | VLOG_WARN_RL(&rl, "received vendor message of length %u " |
659586ef JG |
3978 | "(expected at least %zu)", |
3979 | ntohs(ovh->header.length), sizeof(struct ofp_vendor_header)); | |
49bdc010 | 3980 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); |
064af421 BP |
3981 | } |
3982 | if (ovh->vendor != htonl(NX_VENDOR_ID)) { | |
3983 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR); | |
3984 | } | |
3985 | if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) { | |
100e95db | 3986 | VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u " |
659586ef JG |
3987 | "(expected at least %zu)", |
3988 | ntohs(ovh->header.length), sizeof(struct nicira_header)); | |
49bdc010 | 3989 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); |
064af421 BP |
3990 | } |
3991 | ||
3992 | nh = msg; | |
3993 | switch (ntohl(nh->subtype)) { | |
3994 | case NXT_STATUS_REQUEST: | |
3995 | return switch_status_handle_request(p->switch_status, ofconn->rconn, | |
3996 | msg); | |
659586ef JG |
3997 | |
3998 | case NXT_TUN_ID_FROM_COOKIE: | |
3999 | return handle_tun_id_from_cookie(p, msg); | |
9deba63b BP |
4000 | |
4001 | case NXT_ROLE_REQUEST: | |
4002 | return handle_role_request(p, ofconn, msg); | |
064af421 BP |
4003 | } |
4004 | ||
4005 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); | |
4006 | } | |
4007 | ||
246e61ea JP |
4008 | static int |
4009 | handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh) | |
4010 | { | |
4011 | struct ofp_header *ob; | |
4012 | struct ofpbuf *buf; | |
4013 | ||
4014 | /* Currently, everything executes synchronously, so we can just | |
4015 | * immediately send the barrier reply. */ | |
4016 | ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf); | |
4017 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
4018 | return 0; | |
4019 | } | |
4020 | ||
064af421 BP |
4021 | static void |
4022 | handle_openflow(struct ofconn *ofconn, struct ofproto *p, | |
4023 | struct ofpbuf *ofp_msg) | |
4024 | { | |
4025 | struct ofp_header *oh = ofp_msg->data; | |
4026 | int error; | |
4027 | ||
4028 | COVERAGE_INC(ofproto_recv_openflow); | |
4029 | switch (oh->type) { | |
4030 | case OFPT_ECHO_REQUEST: | |
4031 | error = handle_echo_request(ofconn, oh); | |
4032 | break; | |
4033 | ||
4034 | case OFPT_ECHO_REPLY: | |
4035 | error = 0; | |
4036 | break; | |
4037 | ||
4038 | case OFPT_FEATURES_REQUEST: | |
4039 | error = handle_features_request(p, ofconn, oh); | |
4040 | break; | |
4041 | ||
4042 | case OFPT_GET_CONFIG_REQUEST: | |
4043 | error = handle_get_config_request(p, ofconn, oh); | |
4044 | break; | |
4045 | ||
4046 | case OFPT_SET_CONFIG: | |
4047 | error = handle_set_config(p, ofconn, ofp_msg->data); | |
4048 | break; | |
4049 | ||
4050 | case OFPT_PACKET_OUT: | |
4051 | error = handle_packet_out(p, ofconn, ofp_msg->data); | |
4052 | break; | |
4053 | ||
4054 | case OFPT_PORT_MOD: | |
9deba63b | 4055 | error = handle_port_mod(p, ofconn, oh); |
064af421 BP |
4056 | break; |
4057 | ||
4058 | case OFPT_FLOW_MOD: | |
4059 | error = handle_flow_mod(p, ofconn, ofp_msg->data); | |
4060 | break; | |
4061 | ||
4062 | case OFPT_STATS_REQUEST: | |
4063 | error = handle_stats_request(p, ofconn, oh); | |
4064 | break; | |
4065 | ||
4066 | case OFPT_VENDOR: | |
4067 | error = handle_vendor(p, ofconn, ofp_msg->data); | |
4068 | break; | |
4069 | ||
246e61ea JP |
4070 | case OFPT_BARRIER_REQUEST: |
4071 | error = handle_barrier_request(ofconn, oh); | |
4072 | break; | |
4073 | ||
064af421 BP |
4074 | default: |
4075 | if (VLOG_IS_WARN_ENABLED()) { | |
4076 | char *s = ofp_to_string(oh, ntohs(oh->length), 2); | |
4077 | VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s); | |
4078 | free(s); | |
4079 | } | |
4080 | error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE); | |
4081 | break; | |
4082 | } | |
4083 | ||
4084 | if (error) { | |
4085 | send_error_oh(ofconn, ofp_msg->data, error); | |
4086 | } | |
4087 | } | |
4088 | \f | |
4089 | static void | |
72b06300 | 4090 | handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet) |
064af421 BP |
4091 | { |
4092 | struct odp_msg *msg = packet->data; | |
064af421 BP |
4093 | struct rule *rule; |
4094 | struct ofpbuf payload; | |
4095 | flow_t flow; | |
4096 | ||
064af421 BP |
4097 | payload.data = msg + 1; |
4098 | payload.size = msg->length - sizeof *msg; | |
659586ef | 4099 | flow_extract(&payload, msg->arg, msg->port, &flow); |
064af421 | 4100 | |
0ad9b732 JP |
4101 | /* Check with in-band control to see if this packet should be sent |
4102 | * to the local port regardless of the flow table. */ | |
4103 | if (in_band_msg_in_hook(p->in_band, &flow, &payload)) { | |
4104 | union odp_action action; | |
4105 | ||
4106 | memset(&action, 0, sizeof(action)); | |
4107 | action.output.type = ODPAT_OUTPUT; | |
4108 | action.output.port = ODPP_LOCAL; | |
f1acd62b | 4109 | dpif_execute(p->dpif, flow.in_port, &action, 1, &payload); |
0ad9b732 JP |
4110 | } |
4111 | ||
064af421 BP |
4112 | rule = lookup_valid_rule(p, &flow); |
4113 | if (!rule) { | |
4114 | /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */ | |
ca0f572c | 4115 | struct ofport *port = get_port(p, msg->port); |
064af421 BP |
4116 | if (port) { |
4117 | if (port->opp.config & OFPPC_NO_PACKET_IN) { | |
4118 | COVERAGE_INC(ofproto_no_packet_in); | |
4119 | /* XXX install 'drop' flow entry */ | |
4120 | ofpbuf_delete(packet); | |
4121 | return; | |
4122 | } | |
4123 | } else { | |
4124 | VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port); | |
4125 | } | |
4126 | ||
4127 | COVERAGE_INC(ofproto_packet_in); | |
76ce9432 | 4128 | send_packet_in(p, packet); |
064af421 BP |
4129 | return; |
4130 | } | |
4131 | ||
4132 | if (rule->cr.wc.wildcards) { | |
4133 | rule = rule_create_subrule(p, rule, &flow); | |
4134 | rule_make_actions(p, rule, packet); | |
4135 | } else { | |
4136 | if (!rule->may_install) { | |
4137 | /* The rule is not installable, that is, we need to process every | |
4138 | * packet, so process the current packet and set its actions into | |
4139 | * 'subrule'. */ | |
4140 | rule_make_actions(p, rule, packet); | |
4141 | } else { | |
4142 | /* XXX revalidate rule if it needs it */ | |
4143 | } | |
4144 | } | |
4145 | ||
76ce9432 | 4146 | if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY) { |
7778bd15 BP |
4147 | /* |
4148 | * Extra-special case for fail-open mode. | |
4149 | * | |
4150 | * We are in fail-open mode and the packet matched the fail-open rule, | |
4151 | * but we are connected to a controller too. We should send the packet | |
4152 | * up to the controller in the hope that it will try to set up a flow | |
4153 | * and thereby allow us to exit fail-open. | |
4154 | * | |
4155 | * See the top-level comment in fail-open.c for more information. | |
4156 | */ | |
c9b5816c BP |
4157 | send_packet_in(p, ofpbuf_clone_with_headroom(packet, |
4158 | DPIF_RECV_MSG_PADDING)); | |
7778bd15 | 4159 | } |
750638bb BP |
4160 | |
4161 | ofpbuf_pull(packet, sizeof *msg); | |
4162 | rule_execute(p, rule, packet, &flow); | |
4163 | rule_reinstall(p, rule); | |
064af421 | 4164 | } |
72b06300 BP |
4165 | |
4166 | static void | |
4167 | handle_odp_msg(struct ofproto *p, struct ofpbuf *packet) | |
4168 | { | |
4169 | struct odp_msg *msg = packet->data; | |
4170 | ||
4171 | switch (msg->type) { | |
4172 | case _ODPL_ACTION_NR: | |
4173 | COVERAGE_INC(ofproto_ctlr_action); | |
76ce9432 | 4174 | send_packet_in(p, packet); |
72b06300 BP |
4175 | break; |
4176 | ||
4177 | case _ODPL_SFLOW_NR: | |
4178 | if (p->sflow) { | |
4179 | ofproto_sflow_received(p->sflow, msg); | |
4180 | } | |
4181 | ofpbuf_delete(packet); | |
4182 | break; | |
4183 | ||
4184 | case _ODPL_MISS_NR: | |
4185 | handle_odp_miss_msg(p, packet); | |
4186 | break; | |
4187 | ||
4188 | default: | |
4189 | VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32, | |
4190 | msg->type); | |
4191 | break; | |
4192 | } | |
4193 | } | |
064af421 BP |
4194 | \f |
4195 | static void | |
4196 | revalidate_cb(struct cls_rule *sub_, void *cbdata_) | |
4197 | { | |
4198 | struct rule *sub = rule_from_cls_rule(sub_); | |
4199 | struct revalidate_cbdata *cbdata = cbdata_; | |
4200 | ||
4201 | if (cbdata->revalidate_all | |
4202 | || (cbdata->revalidate_subrules && sub->super) | |
4203 | || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) { | |
4204 | revalidate_rule(cbdata->ofproto, sub); | |
4205 | } | |
4206 | } | |
4207 | ||
4208 | static bool | |
4209 | revalidate_rule(struct ofproto *p, struct rule *rule) | |
4210 | { | |
4211 | const flow_t *flow = &rule->cr.flow; | |
4212 | ||
4213 | COVERAGE_INC(ofproto_revalidate_rule); | |
4214 | if (rule->super) { | |
4215 | struct rule *super; | |
4216 | super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow)); | |
4217 | if (!super) { | |
4218 | rule_remove(p, rule); | |
4219 | return false; | |
4220 | } else if (super != rule->super) { | |
4221 | COVERAGE_INC(ofproto_revalidate_moved); | |
4222 | list_remove(&rule->list); | |
4223 | list_push_back(&super->list, &rule->list); | |
4224 | rule->super = super; | |
4225 | rule->hard_timeout = super->hard_timeout; | |
4226 | rule->idle_timeout = super->idle_timeout; | |
4227 | rule->created = super->created; | |
4228 | rule->used = 0; | |
4229 | } | |
4230 | } | |
4231 | ||
4232 | rule_update_actions(p, rule); | |
4233 | return true; | |
4234 | } | |
4235 | ||
4236 | static struct ofpbuf * | |
659586ef JG |
4237 | compose_flow_removed(struct ofproto *p, const struct rule *rule, |
4238 | long long int now, uint8_t reason) | |
064af421 | 4239 | { |
ca069229 | 4240 | struct ofp_flow_removed *ofr; |
064af421 | 4241 | struct ofpbuf *buf; |
9ca76894 | 4242 | long long int tdiff = now - rule->created; |
26c3f94a JP |
4243 | uint32_t sec = tdiff / 1000; |
4244 | uint32_t msec = tdiff - (sec * 1000); | |
064af421 | 4245 | |
ca069229 | 4246 | ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf); |
659586ef JG |
4247 | flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, p->tun_id_from_cookie, |
4248 | &ofr->match); | |
39997502 | 4249 | ofr->cookie = rule->flow_cookie; |
ca069229 JP |
4250 | ofr->priority = htons(rule->cr.priority); |
4251 | ofr->reason = reason; | |
26c3f94a JP |
4252 | ofr->duration_sec = htonl(sec); |
4253 | ofr->duration_nsec = htonl(msec * 1000000); | |
ca069229 JP |
4254 | ofr->idle_timeout = htons(rule->idle_timeout); |
4255 | ofr->packet_count = htonll(rule->packet_count); | |
4256 | ofr->byte_count = htonll(rule->byte_count); | |
064af421 BP |
4257 | |
4258 | return buf; | |
4259 | } | |
4260 | ||
4261 | static void | |
ca069229 JP |
4262 | uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule) |
4263 | { | |
4264 | assert(rule->installed); | |
4265 | assert(!rule->cr.wc.wildcards); | |
4266 | ||
4267 | if (rule->super) { | |
4268 | rule_remove(ofproto, rule); | |
4269 | } else { | |
4270 | rule_uninstall(ofproto, rule); | |
4271 | } | |
4272 | } | |
9deba63b | 4273 | |
ca069229 JP |
4274 | static void |
4275 | send_flow_removed(struct ofproto *p, struct rule *rule, | |
4276 | long long int now, uint8_t reason) | |
064af421 BP |
4277 | { |
4278 | struct ofconn *ofconn; | |
4279 | struct ofconn *prev; | |
b9b0ce61 | 4280 | struct ofpbuf *buf = NULL; |
064af421 BP |
4281 | |
4282 | /* We limit the maximum number of queued flow expirations it by accounting | |
4283 | * them under the counter for replies. That works because preventing | |
4284 | * OpenFlow requests from being processed also prevents new flows from | |
4285 | * being added (and expiring). (It also prevents processing OpenFlow | |
4286 | * requests that would not add new flows, so it is imperfect.) */ | |
4287 | ||
4288 | prev = NULL; | |
4e8e4213 | 4289 | LIST_FOR_EACH (ofconn, node, &p->all_conns) { |
9deba63b | 4290 | if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn) |
c91248b3 | 4291 | && ofconn_receives_async_msgs(ofconn)) { |
064af421 | 4292 | if (prev) { |
431d8ad2 | 4293 | queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter); |
064af421 | 4294 | } else { |
659586ef | 4295 | buf = compose_flow_removed(p, rule, now, reason); |
064af421 BP |
4296 | } |
4297 | prev = ofconn; | |
4298 | } | |
4299 | } | |
4300 | if (prev) { | |
431d8ad2 | 4301 | queue_tx(buf, prev, prev->reply_counter); |
064af421 BP |
4302 | } |
4303 | } | |
4304 | ||
064af421 BP |
4305 | |
4306 | static void | |
4307 | expire_rule(struct cls_rule *cls_rule, void *p_) | |
4308 | { | |
4309 | struct ofproto *p = p_; | |
4310 | struct rule *rule = rule_from_cls_rule(cls_rule); | |
4311 | long long int hard_expire, idle_expire, expire, now; | |
4312 | ||
4313 | hard_expire = (rule->hard_timeout | |
4314 | ? rule->created + rule->hard_timeout * 1000 | |
4315 | : LLONG_MAX); | |
4316 | idle_expire = (rule->idle_timeout | |
4317 | && (rule->super || list_is_empty(&rule->list)) | |
4318 | ? rule->used + rule->idle_timeout * 1000 | |
4319 | : LLONG_MAX); | |
4320 | expire = MIN(hard_expire, idle_expire); | |
064af421 BP |
4321 | |
4322 | now = time_msec(); | |
4323 | if (now < expire) { | |
4324 | if (rule->installed && now >= rule->used + 5000) { | |
4325 | uninstall_idle_flow(p, rule); | |
0193b2af JG |
4326 | } else if (!rule->cr.wc.wildcards) { |
4327 | active_timeout(p, rule); | |
064af421 | 4328 | } |
0193b2af | 4329 | |
064af421 BP |
4330 | return; |
4331 | } | |
4332 | ||
4333 | COVERAGE_INC(ofproto_expired); | |
46d6f36f JG |
4334 | |
4335 | /* Update stats. This code will be a no-op if the rule expired | |
4336 | * due to an idle timeout. */ | |
064af421 | 4337 | if (rule->cr.wc.wildcards) { |
064af421 | 4338 | struct rule *subrule, *next; |
4e8e4213 | 4339 | LIST_FOR_EACH_SAFE (subrule, next, list, &rule->list) { |
064af421 BP |
4340 | rule_remove(p, subrule); |
4341 | } | |
46d6f36f JG |
4342 | } else { |
4343 | rule_uninstall(p, rule); | |
064af421 BP |
4344 | } |
4345 | ||
8fe1a59d | 4346 | if (!rule_is_hidden(rule)) { |
ca069229 JP |
4347 | send_flow_removed(p, rule, now, |
4348 | (now >= hard_expire | |
4349 | ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT)); | |
8fe1a59d | 4350 | } |
064af421 BP |
4351 | rule_remove(p, rule); |
4352 | } | |
4353 | ||
0193b2af JG |
4354 | static void |
4355 | active_timeout(struct ofproto *ofproto, struct rule *rule) | |
4356 | { | |
4357 | if (ofproto->netflow && !is_controller_rule(rule) && | |
4358 | netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) { | |
4359 | struct ofexpired expired; | |
4360 | struct odp_flow odp_flow; | |
4361 | ||
4362 | /* Get updated flow stats. */ | |
4363 | memset(&odp_flow, 0, sizeof odp_flow); | |
094e1514 JG |
4364 | if (rule->installed) { |
4365 | odp_flow.key = rule->cr.flow; | |
4366 | odp_flow.flags = ODPFF_ZERO_TCP_FLAGS; | |
d65349ea | 4367 | dpif_flow_get(ofproto->dpif, &odp_flow); |
094e1514 JG |
4368 | |
4369 | if (odp_flow.stats.n_packets) { | |
4370 | update_time(ofproto, rule, &odp_flow.stats); | |
abfec865 | 4371 | netflow_flow_update_flags(&rule->nf_flow, |
094e1514 JG |
4372 | odp_flow.stats.tcp_flags); |
4373 | } | |
0193b2af JG |
4374 | } |
4375 | ||
4376 | expired.flow = rule->cr.flow; | |
4377 | expired.packet_count = rule->packet_count + | |
4378 | odp_flow.stats.n_packets; | |
4379 | expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes; | |
4380 | expired.used = rule->used; | |
4381 | ||
4382 | netflow_expire(ofproto->netflow, &rule->nf_flow, &expired); | |
4383 | ||
4384 | /* Schedule us to send the accumulated records once we have | |
4385 | * collected all of them. */ | |
4386 | poll_immediate_wake(); | |
4387 | } | |
4388 | } | |
4389 | ||
064af421 BP |
4390 | static void |
4391 | update_used(struct ofproto *p) | |
4392 | { | |
4393 | struct odp_flow *flows; | |
4394 | size_t n_flows; | |
4395 | size_t i; | |
4396 | int error; | |
4397 | ||
c228a364 | 4398 | error = dpif_flow_list_all(p->dpif, &flows, &n_flows); |
064af421 BP |
4399 | if (error) { |
4400 | return; | |
4401 | } | |
4402 | ||
4403 | for (i = 0; i < n_flows; i++) { | |
4404 | struct odp_flow *f = &flows[i]; | |
4405 | struct rule *rule; | |
4406 | ||
4407 | rule = rule_from_cls_rule( | |
4408 | classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX)); | |
4409 | if (!rule || !rule->installed) { | |
4410 | COVERAGE_INC(ofproto_unexpected_rule); | |
c228a364 | 4411 | dpif_flow_del(p->dpif, f); |
064af421 BP |
4412 | continue; |
4413 | } | |
4414 | ||
0193b2af | 4415 | update_time(p, rule, &f->stats); |
064af421 BP |
4416 | rule_account(p, rule, f->stats.n_bytes); |
4417 | } | |
4418 | free(flows); | |
4419 | } | |
4420 | ||
43253595 | 4421 | /* pinsched callback for sending 'packet' on 'ofconn'. */ |
064af421 | 4422 | static void |
76ce9432 | 4423 | do_send_packet_in(struct ofpbuf *packet, void *ofconn_) |
064af421 | 4424 | { |
76ce9432 | 4425 | struct ofconn *ofconn = ofconn_; |
43253595 BP |
4426 | |
4427 | rconn_send_with_limit(ofconn->rconn, packet, | |
4428 | ofconn->packet_in_counter, 100); | |
4429 | } | |
4430 | ||
4431 | /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and | |
4432 | * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s | |
4433 | * packet scheduler for sending. | |
4434 | * | |
30ea5d93 BP |
4435 | * 'max_len' specifies the maximum number of bytes of the packet to send on |
4436 | * 'ofconn' (INT_MAX specifies no limit). | |
4437 | * | |
43253595 BP |
4438 | * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise, |
4439 | * ownership is transferred to this function. */ | |
4440 | static void | |
30ea5d93 BP |
4441 | schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len, |
4442 | bool clone) | |
43253595 | 4443 | { |
76ce9432 | 4444 | struct ofproto *ofproto = ofconn->ofproto; |
43253595 BP |
4445 | struct ofp_packet_in *opi = packet->data; |
4446 | uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port)); | |
4447 | int send_len, trim_size; | |
76ce9432 | 4448 | uint32_t buffer_id; |
064af421 | 4449 | |
43253595 BP |
4450 | /* Get buffer. */ |
4451 | if (opi->reason == OFPR_ACTION) { | |
76ce9432 | 4452 | buffer_id = UINT32_MAX; |
43253595 BP |
4453 | } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) { |
4454 | buffer_id = pktbuf_get_null(); | |
89b9612d BP |
4455 | } else if (!ofconn->pktbuf) { |
4456 | buffer_id = UINT32_MAX; | |
76ce9432 | 4457 | } else { |
43253595 BP |
4458 | struct ofpbuf payload; |
4459 | payload.data = opi->data; | |
4460 | payload.size = packet->size - offsetof(struct ofp_packet_in, data); | |
4461 | buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port); | |
76ce9432 | 4462 | } |
372179d4 | 4463 | |
43253595 BP |
4464 | /* Figure out how much of the packet to send. */ |
4465 | send_len = ntohs(opi->total_len); | |
4466 | if (buffer_id != UINT32_MAX) { | |
4467 | send_len = MIN(send_len, ofconn->miss_send_len); | |
4468 | } | |
30ea5d93 | 4469 | send_len = MIN(send_len, max_len); |
064af421 | 4470 | |
43253595 BP |
4471 | /* Adjust packet length and clone if necessary. */ |
4472 | trim_size = offsetof(struct ofp_packet_in, data) + send_len; | |
4473 | if (clone) { | |
4474 | packet = ofpbuf_clone_data(packet->data, trim_size); | |
4475 | opi = packet->data; | |
4476 | } else { | |
4477 | packet->size = trim_size; | |
4478 | } | |
4479 | ||
4480 | /* Update packet headers. */ | |
4481 | opi->buffer_id = htonl(buffer_id); | |
4482 | update_openflow_length(packet); | |
4483 | ||
4484 | /* Hand over to packet scheduler. It might immediately call into | |
4485 | * do_send_packet_in() or it might buffer it for a while (until a later | |
4486 | * call to pinsched_run()). */ | |
4487 | pinsched_send(ofconn->schedulers[opi->reason], in_port, | |
4488 | packet, do_send_packet_in, ofconn); | |
064af421 BP |
4489 | } |
4490 | ||
43253595 BP |
4491 | /* Replace struct odp_msg header in 'packet' by equivalent struct |
4492 | * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as | |
4493 | * returned by dpif_recv()). | |
4494 | * | |
4495 | * The conversion is not complete: the caller still needs to trim any unneeded | |
4496 | * payload off the end of the buffer, set the length in the OpenFlow header, | |
4497 | * and set buffer_id. Those require us to know the controller settings and so | |
30ea5d93 BP |
4498 | * must be done on a per-controller basis. |
4499 | * | |
4500 | * Returns the maximum number of bytes of the packet that should be sent to | |
4501 | * the controller (INT_MAX if no limit). */ | |
4502 | static int | |
43253595 | 4503 | do_convert_to_packet_in(struct ofpbuf *packet) |
064af421 | 4504 | { |
76ce9432 | 4505 | struct odp_msg *msg = packet->data; |
43253595 BP |
4506 | struct ofp_packet_in *opi; |
4507 | uint8_t reason; | |
4508 | uint16_t total_len; | |
4509 | uint16_t in_port; | |
30ea5d93 | 4510 | int max_len; |
43253595 BP |
4511 | |
4512 | /* Extract relevant header fields */ | |
30ea5d93 BP |
4513 | if (msg->type == _ODPL_ACTION_NR) { |
4514 | reason = OFPR_ACTION; | |
4515 | max_len = msg->arg; | |
4516 | } else { | |
4517 | reason = OFPR_NO_MATCH; | |
4518 | max_len = INT_MAX; | |
4519 | } | |
43253595 BP |
4520 | total_len = msg->length - sizeof *msg; |
4521 | in_port = odp_port_to_ofp_port(msg->port); | |
4522 | ||
4523 | /* Repurpose packet buffer by overwriting header. */ | |
4524 | ofpbuf_pull(packet, sizeof(struct odp_msg)); | |
4525 | opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data)); | |
4526 | opi->header.version = OFP_VERSION; | |
4527 | opi->header.type = OFPT_PACKET_IN; | |
4528 | opi->total_len = htons(total_len); | |
4529 | opi->in_port = htons(in_port); | |
4530 | opi->reason = reason; | |
30ea5d93 BP |
4531 | |
4532 | return max_len; | |
43253595 BP |
4533 | } |
4534 | ||
4535 | /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or | |
4536 | * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller | |
4537 | * as necessary according to their individual configurations. | |
4538 | * | |
4539 | * 'packet' must have sufficient headroom to convert it into a struct | |
4540 | * ofp_packet_in (e.g. as returned by dpif_recv()). | |
4541 | * | |
4542 | * Takes ownership of 'packet'. */ | |
4543 | static void | |
4544 | send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet) | |
4545 | { | |
76ce9432 | 4546 | struct ofconn *ofconn, *prev; |
30ea5d93 | 4547 | int max_len; |
064af421 | 4548 | |
30ea5d93 | 4549 | max_len = do_convert_to_packet_in(packet); |
76ce9432 BP |
4550 | |
4551 | prev = NULL; | |
4e8e4213 | 4552 | LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) { |
c91248b3 | 4553 | if (ofconn_receives_async_msgs(ofconn)) { |
9deba63b | 4554 | if (prev) { |
30ea5d93 | 4555 | schedule_packet_in(prev, packet, max_len, true); |
9deba63b BP |
4556 | } |
4557 | prev = ofconn; | |
064af421 | 4558 | } |
76ce9432 BP |
4559 | } |
4560 | if (prev) { | |
30ea5d93 | 4561 | schedule_packet_in(prev, packet, max_len, false); |
76ce9432 BP |
4562 | } else { |
4563 | ofpbuf_delete(packet); | |
064af421 | 4564 | } |
064af421 BP |
4565 | } |
4566 | ||
4567 | static uint64_t | |
fa60c019 | 4568 | pick_datapath_id(const struct ofproto *ofproto) |
064af421 | 4569 | { |
fa60c019 | 4570 | const struct ofport *port; |
064af421 | 4571 | |
ca0f572c | 4572 | port = get_port(ofproto, ODPP_LOCAL); |
fa60c019 BP |
4573 | if (port) { |
4574 | uint8_t ea[ETH_ADDR_LEN]; | |
4575 | int error; | |
4576 | ||
4577 | error = netdev_get_etheraddr(port->netdev, ea); | |
064af421 BP |
4578 | if (!error) { |
4579 | return eth_addr_to_uint64(ea); | |
4580 | } | |
4581 | VLOG_WARN("could not get MAC address for %s (%s)", | |
fa60c019 | 4582 | netdev_get_name(port->netdev), strerror(error)); |
064af421 | 4583 | } |
fa60c019 | 4584 | return ofproto->fallback_dpid; |
064af421 BP |
4585 | } |
4586 | ||
4587 | static uint64_t | |
4588 | pick_fallback_dpid(void) | |
4589 | { | |
4590 | uint8_t ea[ETH_ADDR_LEN]; | |
70150daf | 4591 | eth_addr_nicira_random(ea); |
064af421 BP |
4592 | return eth_addr_to_uint64(ea); |
4593 | } | |
4594 | \f | |
4595 | static bool | |
4596 | default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet, | |
4597 | struct odp_actions *actions, tag_type *tags, | |
6a07af36 | 4598 | uint16_t *nf_output_iface, void *ofproto_) |
064af421 BP |
4599 | { |
4600 | struct ofproto *ofproto = ofproto_; | |
4601 | int out_port; | |
4602 | ||
4603 | /* Drop frames for reserved multicast addresses. */ | |
4604 | if (eth_addr_is_reserved(flow->dl_dst)) { | |
4605 | return true; | |
4606 | } | |
4607 | ||
4608 | /* Learn source MAC (but don't try to learn from revalidation). */ | |
4609 | if (packet != NULL) { | |
4610 | tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src, | |
7febb910 JG |
4611 | 0, flow->in_port, |
4612 | GRAT_ARP_LOCK_NONE); | |
064af421 BP |
4613 | if (rev_tag) { |
4614 | /* The log messages here could actually be useful in debugging, | |
4615 | * so keep the rate limit relatively high. */ | |
4616 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); | |
4617 | VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16, | |
4618 | ETH_ADDR_ARGS(flow->dl_src), flow->in_port); | |
4619 | ofproto_revalidate(ofproto, rev_tag); | |
4620 | } | |
4621 | } | |
4622 | ||
4623 | /* Determine output port. */ | |
7febb910 JG |
4624 | out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags, |
4625 | NULL); | |
064af421 | 4626 | if (out_port < 0) { |
6a07af36 | 4627 | add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface); |
064af421 BP |
4628 | } else if (out_port != flow->in_port) { |
4629 | odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port; | |
6a07af36 | 4630 | *nf_output_iface = out_port; |
064af421 BP |
4631 | } else { |
4632 | /* Drop. */ | |
4633 | } | |
4634 | ||
4635 | return true; | |
4636 | } | |
4637 | ||
4638 | static const struct ofhooks default_ofhooks = { | |
4639 | NULL, | |
4640 | default_normal_ofhook_cb, | |
4641 | NULL, | |
4642 | NULL | |
4643 | }; |