]>
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 | |
0de7a4b4 | 327 | static int ofproto_expire(struct ofproto *); |
4a4cdb3b | 328 | |
0193b2af JG |
329 | static void update_stats(struct ofproto *, struct rule *, |
330 | const struct odp_flow_stats *); | |
064af421 BP |
331 | static bool revalidate_rule(struct ofproto *p, struct rule *rule); |
332 | static void revalidate_cb(struct cls_rule *rule_, void *p_); | |
333 | ||
334 | static void handle_odp_msg(struct ofproto *, struct ofpbuf *); | |
335 | ||
336 | static void handle_openflow(struct ofconn *, struct ofproto *, | |
337 | struct ofpbuf *); | |
338 | ||
72b06300 BP |
339 | static void refresh_port_groups(struct ofproto *); |
340 | ||
ca0f572c | 341 | static struct ofport *get_port(const struct ofproto *, uint16_t odp_port); |
064af421 BP |
342 | static void update_port(struct ofproto *, const char *devname); |
343 | static int init_ports(struct ofproto *); | |
344 | static void reinit_ports(struct ofproto *); | |
345 | ||
346 | int | |
1a6f1e2a JG |
347 | ofproto_create(const char *datapath, const char *datapath_type, |
348 | const struct ofhooks *ofhooks, void *aux, | |
064af421 BP |
349 | struct ofproto **ofprotop) |
350 | { | |
064af421 BP |
351 | struct odp_stats stats; |
352 | struct ofproto *p; | |
c228a364 | 353 | struct dpif *dpif; |
064af421 BP |
354 | int error; |
355 | ||
356 | *ofprotop = NULL; | |
357 | ||
358 | /* Connect to datapath and start listening for messages. */ | |
1a6f1e2a | 359 | error = dpif_open(datapath, datapath_type, &dpif); |
064af421 BP |
360 | if (error) { |
361 | VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error)); | |
362 | return error; | |
363 | } | |
c228a364 | 364 | error = dpif_get_dp_stats(dpif, &stats); |
064af421 BP |
365 | if (error) { |
366 | VLOG_ERR("failed to obtain stats for datapath %s: %s", | |
367 | datapath, strerror(error)); | |
c228a364 | 368 | dpif_close(dpif); |
064af421 BP |
369 | return error; |
370 | } | |
72b06300 | 371 | error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW); |
064af421 BP |
372 | if (error) { |
373 | VLOG_ERR("failed to listen on datapath %s: %s", | |
374 | datapath, strerror(error)); | |
c228a364 | 375 | dpif_close(dpif); |
064af421 BP |
376 | return error; |
377 | } | |
c228a364 | 378 | dpif_flow_flush(dpif); |
8f24562a | 379 | dpif_recv_purge(dpif); |
064af421 BP |
380 | |
381 | /* Initialize settings. */ | |
ec6fde61 | 382 | p = xzalloc(sizeof *p); |
064af421 | 383 | p->fallback_dpid = pick_fallback_dpid(); |
fa60c019 | 384 | p->datapath_id = p->fallback_dpid; |
5a719c38 JP |
385 | p->mfr_desc = xstrdup(DEFAULT_MFR_DESC); |
386 | p->hw_desc = xstrdup(DEFAULT_HW_DESC); | |
387 | p->sw_desc = xstrdup(DEFAULT_SW_DESC); | |
388 | p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC); | |
23ff2821 | 389 | p->dp_desc = xstrdup(DEFAULT_DP_DESC); |
064af421 BP |
390 | |
391 | /* Initialize datapath. */ | |
392 | p->dpif = dpif; | |
8b61709d | 393 | p->netdev_monitor = netdev_monitor_create(); |
ca0f572c | 394 | hmap_init(&p->ports); |
064af421 BP |
395 | shash_init(&p->port_by_name); |
396 | p->max_ports = stats.max_ports; | |
397 | ||
398 | /* Initialize submodules. */ | |
399 | p->switch_status = switch_status_create(p); | |
400 | p->in_band = NULL; | |
064af421 | 401 | p->fail_open = NULL; |
064af421 | 402 | p->netflow = NULL; |
72b06300 | 403 | p->sflow = NULL; |
064af421 BP |
404 | |
405 | /* Initialize flow table. */ | |
406 | classifier_init(&p->cls); | |
407 | p->need_revalidate = false; | |
408 | p->next_expiration = time_msec() + 1000; | |
409 | tag_set_init(&p->revalidate_set); | |
410 | ||
411 | /* Initialize OpenFlow connections. */ | |
412 | list_init(&p->all_conns); | |
76ce9432 | 413 | hmap_init(&p->controllers); |
7d674866 | 414 | hmap_init(&p->services); |
064af421 BP |
415 | p->snoops = NULL; |
416 | p->n_snoops = 0; | |
417 | ||
418 | /* Initialize hooks. */ | |
419 | if (ofhooks) { | |
420 | p->ofhooks = ofhooks; | |
421 | p->aux = aux; | |
422 | p->ml = NULL; | |
423 | } else { | |
424 | p->ofhooks = &default_ofhooks; | |
425 | p->aux = p; | |
426 | p->ml = mac_learning_create(); | |
427 | } | |
428 | ||
fa60c019 BP |
429 | /* Pick final datapath ID. */ |
430 | p->datapath_id = pick_datapath_id(p); | |
b123cc3c | 431 | VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id); |
fa60c019 | 432 | |
064af421 BP |
433 | *ofprotop = p; |
434 | return 0; | |
435 | } | |
436 | ||
437 | void | |
438 | ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id) | |
439 | { | |
440 | uint64_t old_dpid = p->datapath_id; | |
fa60c019 | 441 | p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p); |
064af421 | 442 | if (p->datapath_id != old_dpid) { |
b123cc3c | 443 | VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id); |
76ce9432 BP |
444 | |
445 | /* Force all active connections to reconnect, since there is no way to | |
446 | * notify a controller that the datapath ID has changed. */ | |
fa05809b | 447 | ofproto_reconnect_controllers(p); |
064af421 BP |
448 | } |
449 | } | |
450 | ||
76ce9432 BP |
451 | static bool |
452 | is_discovery_controller(const struct ofproto_controller *c) | |
453 | { | |
454 | return !strcmp(c->target, "discover"); | |
455 | } | |
456 | ||
457 | static bool | |
458 | is_in_band_controller(const struct ofproto_controller *c) | |
459 | { | |
460 | return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND; | |
461 | } | |
462 | ||
463 | /* Creates a new controller in 'ofproto'. Some of the settings are initially | |
464 | * drawn from 'c', but update_controller() needs to be called later to finish | |
465 | * the new ofconn's configuration. */ | |
466 | static void | |
467 | add_controller(struct ofproto *ofproto, const struct ofproto_controller *c) | |
468 | { | |
469 | struct discovery *discovery; | |
470 | struct ofconn *ofconn; | |
471 | ||
472 | if (is_discovery_controller(c)) { | |
473 | int error = discovery_create(c->accept_re, c->update_resolv_conf, | |
474 | ofproto->dpif, ofproto->switch_status, | |
475 | &discovery); | |
476 | if (error) { | |
477 | return; | |
478 | } | |
479 | } else { | |
480 | discovery = NULL; | |
481 | } | |
482 | ||
5899143f | 483 | ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY); |
76ce9432 BP |
484 | ofconn->pktbuf = pktbuf_create(); |
485 | ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN; | |
486 | if (discovery) { | |
487 | ofconn->discovery = discovery; | |
488 | } else { | |
eb15cdbb BP |
489 | char *name = ofconn_make_name(ofproto, c->target); |
490 | rconn_connect(ofconn->rconn, c->target, name); | |
491 | free(name); | |
76ce9432 BP |
492 | } |
493 | hmap_insert(&ofproto->controllers, &ofconn->hmap_node, | |
494 | hash_string(c->target, 0)); | |
495 | } | |
496 | ||
497 | /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's | |
498 | * target or turn discovery on or off (these are done by creating new ofconns | |
499 | * and deleting old ones), but it can update the rest of an ofconn's | |
500 | * settings. */ | |
501 | static void | |
502 | update_controller(struct ofconn *ofconn, const struct ofproto_controller *c) | |
064af421 | 503 | { |
76ce9432 | 504 | int probe_interval; |
79c9f2ee | 505 | |
d2ede7bc BP |
506 | ofconn->band = (is_in_band_controller(c) |
507 | ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND); | |
508 | ||
76ce9432 | 509 | rconn_set_max_backoff(ofconn->rconn, c->max_backoff); |
79c9f2ee | 510 | |
76ce9432 BP |
511 | probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0; |
512 | rconn_set_probe_interval(ofconn->rconn, probe_interval); | |
79c9f2ee | 513 | |
76ce9432 BP |
514 | if (ofconn->discovery) { |
515 | discovery_set_update_resolv_conf(ofconn->discovery, | |
516 | c->update_resolv_conf); | |
517 | discovery_set_accept_controller_re(ofconn->discovery, c->accept_re); | |
518 | } | |
79c9f2ee | 519 | |
7d674866 | 520 | ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit); |
76ce9432 | 521 | } |
79c9f2ee | 522 | |
76ce9432 BP |
523 | static const char * |
524 | ofconn_get_target(const struct ofconn *ofconn) | |
525 | { | |
eb15cdbb | 526 | return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn); |
76ce9432 BP |
527 | } |
528 | ||
529 | static struct ofconn * | |
530 | find_controller_by_target(struct ofproto *ofproto, const char *target) | |
531 | { | |
532 | struct ofconn *ofconn; | |
533 | ||
4e8e4213 | 534 | HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node, |
76ce9432 BP |
535 | hash_string(target, 0), &ofproto->controllers) { |
536 | if (!strcmp(ofconn_get_target(ofconn), target)) { | |
537 | return ofconn; | |
79c9f2ee | 538 | } |
064af421 | 539 | } |
76ce9432 BP |
540 | return NULL; |
541 | } | |
064af421 | 542 | |
d2ede7bc BP |
543 | static void |
544 | update_in_band_remotes(struct ofproto *ofproto) | |
545 | { | |
546 | const struct ofconn *ofconn; | |
547 | struct sockaddr_in *addrs; | |
917e50e1 | 548 | size_t max_addrs, n_addrs; |
d2ede7bc | 549 | bool discovery; |
917e50e1 | 550 | size_t i; |
d2ede7bc | 551 | |
917e50e1 BP |
552 | /* Allocate enough memory for as many remotes as we could possibly have. */ |
553 | max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers); | |
554 | addrs = xmalloc(max_addrs * sizeof *addrs); | |
d2ede7bc BP |
555 | n_addrs = 0; |
556 | ||
557 | /* Add all the remotes. */ | |
558 | discovery = false; | |
4e8e4213 | 559 | HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) { |
d2ede7bc BP |
560 | struct sockaddr_in *sin = &addrs[n_addrs]; |
561 | ||
487ec65f BP |
562 | if (ofconn->band == OFPROTO_OUT_OF_BAND) { |
563 | continue; | |
564 | } | |
565 | ||
d2ede7bc BP |
566 | sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn); |
567 | if (sin->sin_addr.s_addr) { | |
568 | sin->sin_port = rconn_get_remote_port(ofconn->rconn); | |
569 | n_addrs++; | |
570 | } | |
571 | if (ofconn->discovery) { | |
572 | discovery = true; | |
573 | } | |
574 | } | |
917e50e1 BP |
575 | for (i = 0; i < ofproto->n_extra_remotes; i++) { |
576 | addrs[n_addrs++] = ofproto->extra_in_band_remotes[i]; | |
577 | } | |
d2ede7bc BP |
578 | |
579 | /* Create or update or destroy in-band. | |
580 | * | |
581 | * Ordinarily we only enable in-band if there's at least one remote | |
582 | * address, but discovery needs the in-band rules for DHCP to be installed | |
583 | * even before we know any remote addresses. */ | |
584 | if (n_addrs || discovery) { | |
585 | if (!ofproto->in_band) { | |
586 | in_band_create(ofproto, ofproto->dpif, ofproto->switch_status, | |
587 | &ofproto->in_band); | |
588 | } | |
40cae670 BP |
589 | if (ofproto->in_band) { |
590 | in_band_set_remotes(ofproto->in_band, addrs, n_addrs); | |
591 | } | |
d2ede7bc BP |
592 | ofproto->next_in_band_update = time_msec() + 1000; |
593 | } else { | |
594 | in_band_destroy(ofproto->in_band); | |
595 | ofproto->in_band = NULL; | |
596 | } | |
597 | ||
598 | /* Clean up. */ | |
599 | free(addrs); | |
600 | } | |
601 | ||
31681a5d JP |
602 | static void |
603 | update_fail_open(struct ofproto *p) | |
604 | { | |
605 | struct ofconn *ofconn; | |
606 | ||
607 | if (!hmap_is_empty(&p->controllers) | |
608 | && p->fail_mode == OFPROTO_FAIL_STANDALONE) { | |
609 | struct rconn **rconns; | |
610 | size_t n; | |
611 | ||
612 | if (!p->fail_open) { | |
613 | p->fail_open = fail_open_create(p, p->switch_status); | |
614 | } | |
615 | ||
616 | n = 0; | |
617 | rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns); | |
4e8e4213 | 618 | HMAP_FOR_EACH (ofconn, hmap_node, &p->controllers) { |
31681a5d JP |
619 | rconns[n++] = ofconn->rconn; |
620 | } | |
621 | ||
622 | fail_open_set_controllers(p->fail_open, rconns, n); | |
623 | /* p->fail_open takes ownership of 'rconns'. */ | |
624 | } else { | |
625 | fail_open_destroy(p->fail_open); | |
626 | p->fail_open = NULL; | |
627 | } | |
628 | } | |
629 | ||
76ce9432 BP |
630 | void |
631 | ofproto_set_controllers(struct ofproto *p, | |
632 | const struct ofproto_controller *controllers, | |
633 | size_t n_controllers) | |
634 | { | |
635 | struct shash new_controllers; | |
7d674866 BP |
636 | struct ofconn *ofconn, *next_ofconn; |
637 | struct ofservice *ofservice, *next_ofservice; | |
76ce9432 | 638 | bool ss_exists; |
76ce9432 | 639 | size_t i; |
79c9f2ee | 640 | |
7d674866 BP |
641 | /* Create newly configured controllers and services. |
642 | * Create a name to ofproto_controller mapping in 'new_controllers'. */ | |
76ce9432 BP |
643 | shash_init(&new_controllers); |
644 | for (i = 0; i < n_controllers; i++) { | |
645 | const struct ofproto_controller *c = &controllers[i]; | |
646 | ||
7d674866 BP |
647 | if (!vconn_verify_name(c->target) || !strcmp(c->target, "discover")) { |
648 | if (!find_controller_by_target(p, c->target)) { | |
649 | add_controller(p, c); | |
650 | } | |
651 | } else if (!pvconn_verify_name(c->target)) { | |
652 | if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) { | |
653 | continue; | |
654 | } | |
655 | } else { | |
656 | VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"", | |
657 | dpif_name(p->dpif), c->target); | |
658 | continue; | |
76ce9432 | 659 | } |
7d674866 BP |
660 | |
661 | shash_add_once(&new_controllers, c->target, &controllers[i]); | |
76ce9432 BP |
662 | } |
663 | ||
7d674866 BP |
664 | /* Delete controllers that are no longer configured. |
665 | * Update configuration of all now-existing controllers. */ | |
76ce9432 | 666 | ss_exists = false; |
4e8e4213 | 667 | HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &p->controllers) { |
76ce9432 BP |
668 | struct ofproto_controller *c; |
669 | ||
670 | c = shash_find_data(&new_controllers, ofconn_get_target(ofconn)); | |
671 | if (!c) { | |
672 | ofconn_destroy(ofconn); | |
79c9f2ee | 673 | } else { |
76ce9432 | 674 | update_controller(ofconn, c); |
76ce9432 BP |
675 | if (ofconn->ss) { |
676 | ss_exists = true; | |
677 | } | |
76ce9432 BP |
678 | } |
679 | } | |
7d674866 BP |
680 | |
681 | /* Delete services that are no longer configured. | |
682 | * Update configuration of all now-existing services. */ | |
4e8e4213 | 683 | HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) { |
7d674866 BP |
684 | struct ofproto_controller *c; |
685 | ||
686 | c = shash_find_data(&new_controllers, | |
687 | pvconn_get_name(ofservice->pvconn)); | |
688 | if (!c) { | |
689 | ofservice_destroy(p, ofservice); | |
690 | } else { | |
691 | ofservice_reconfigure(ofservice, c); | |
692 | } | |
693 | } | |
694 | ||
76ce9432 BP |
695 | shash_destroy(&new_controllers); |
696 | ||
d2ede7bc | 697 | update_in_band_remotes(p); |
31681a5d | 698 | update_fail_open(p); |
79c9f2ee | 699 | |
76ce9432 BP |
700 | if (!hmap_is_empty(&p->controllers) && !ss_exists) { |
701 | ofconn = CONTAINER_OF(hmap_first(&p->controllers), | |
702 | struct ofconn, hmap_node); | |
703 | ofconn->ss = switch_status_register(p->switch_status, "remote", | |
704 | rconn_status_cb, ofconn->rconn); | |
79c9f2ee | 705 | } |
064af421 BP |
706 | } |
707 | ||
31681a5d JP |
708 | void |
709 | ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode) | |
710 | { | |
711 | p->fail_mode = fail_mode; | |
712 | update_fail_open(p); | |
713 | } | |
714 | ||
fa05809b BP |
715 | /* Drops the connections between 'ofproto' and all of its controllers, forcing |
716 | * them to reconnect. */ | |
717 | void | |
718 | ofproto_reconnect_controllers(struct ofproto *ofproto) | |
719 | { | |
720 | struct ofconn *ofconn; | |
721 | ||
4e8e4213 | 722 | LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) { |
fa05809b BP |
723 | rconn_reconnect(ofconn->rconn); |
724 | } | |
725 | } | |
726 | ||
917e50e1 BP |
727 | static bool |
728 | any_extras_changed(const struct ofproto *ofproto, | |
729 | const struct sockaddr_in *extras, size_t n) | |
730 | { | |
731 | size_t i; | |
732 | ||
733 | if (n != ofproto->n_extra_remotes) { | |
734 | return true; | |
735 | } | |
736 | ||
737 | for (i = 0; i < n; i++) { | |
738 | const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i]; | |
739 | const struct sockaddr_in *new = &extras[i]; | |
740 | ||
741 | if (old->sin_addr.s_addr != new->sin_addr.s_addr || | |
742 | old->sin_port != new->sin_port) { | |
743 | return true; | |
744 | } | |
745 | } | |
746 | ||
747 | return false; | |
748 | } | |
749 | ||
750 | /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s | |
751 | * in-band control should guarantee access, in the same way that in-band | |
752 | * control guarantees access to OpenFlow controllers. */ | |
753 | void | |
754 | ofproto_set_extra_in_band_remotes(struct ofproto *ofproto, | |
755 | const struct sockaddr_in *extras, size_t n) | |
756 | { | |
757 | if (!any_extras_changed(ofproto, extras, n)) { | |
758 | return; | |
759 | } | |
760 | ||
761 | free(ofproto->extra_in_band_remotes); | |
762 | ofproto->n_extra_remotes = n; | |
763 | ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras); | |
764 | ||
765 | update_in_band_remotes(ofproto); | |
766 | } | |
767 | ||
064af421 BP |
768 | void |
769 | ofproto_set_desc(struct ofproto *p, | |
5a719c38 JP |
770 | const char *mfr_desc, const char *hw_desc, |
771 | const char *sw_desc, const char *serial_desc, | |
8abc4ed7 | 772 | const char *dp_desc) |
064af421 | 773 | { |
5a719c38 JP |
774 | struct ofp_desc_stats *ods; |
775 | ||
776 | if (mfr_desc) { | |
777 | if (strlen(mfr_desc) >= sizeof ods->mfr_desc) { | |
778 | VLOG_WARN("truncating mfr_desc, must be less than %zu characters", | |
779 | sizeof ods->mfr_desc); | |
780 | } | |
781 | free(p->mfr_desc); | |
782 | p->mfr_desc = xstrdup(mfr_desc); | |
064af421 | 783 | } |
5a719c38 JP |
784 | if (hw_desc) { |
785 | if (strlen(hw_desc) >= sizeof ods->hw_desc) { | |
786 | VLOG_WARN("truncating hw_desc, must be less than %zu characters", | |
787 | sizeof ods->hw_desc); | |
788 | } | |
789 | free(p->hw_desc); | |
790 | p->hw_desc = xstrdup(hw_desc); | |
064af421 | 791 | } |
5a719c38 JP |
792 | if (sw_desc) { |
793 | if (strlen(sw_desc) >= sizeof ods->sw_desc) { | |
794 | VLOG_WARN("truncating sw_desc, must be less than %zu characters", | |
795 | sizeof ods->sw_desc); | |
796 | } | |
797 | free(p->sw_desc); | |
798 | p->sw_desc = xstrdup(sw_desc); | |
799 | } | |
800 | if (serial_desc) { | |
801 | if (strlen(serial_desc) >= sizeof ods->serial_num) { | |
802 | VLOG_WARN("truncating serial_desc, must be less than %zu " | |
803 | "characters", | |
804 | sizeof ods->serial_num); | |
805 | } | |
806 | free(p->serial_desc); | |
807 | p->serial_desc = xstrdup(serial_desc); | |
064af421 | 808 | } |
8abc4ed7 | 809 | if (dp_desc) { |
5a719c38 JP |
810 | if (strlen(dp_desc) >= sizeof ods->dp_desc) { |
811 | VLOG_WARN("truncating dp_desc, must be less than %zu characters", | |
812 | sizeof ods->dp_desc); | |
813 | } | |
8abc4ed7 JP |
814 | free(p->dp_desc); |
815 | p->dp_desc = xstrdup(dp_desc); | |
816 | } | |
064af421 BP |
817 | } |
818 | ||
064af421 BP |
819 | static int |
820 | set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp, | |
821 | const struct svec *svec) | |
822 | { | |
823 | struct pvconn **pvconns = *pvconnsp; | |
824 | size_t n_pvconns = *n_pvconnsp; | |
825 | int retval = 0; | |
826 | size_t i; | |
827 | ||
828 | for (i = 0; i < n_pvconns; i++) { | |
829 | pvconn_close(pvconns[i]); | |
830 | } | |
831 | free(pvconns); | |
832 | ||
833 | pvconns = xmalloc(svec->n * sizeof *pvconns); | |
834 | n_pvconns = 0; | |
835 | for (i = 0; i < svec->n; i++) { | |
836 | const char *name = svec->names[i]; | |
837 | struct pvconn *pvconn; | |
838 | int error; | |
839 | ||
840 | error = pvconn_open(name, &pvconn); | |
841 | if (!error) { | |
842 | pvconns[n_pvconns++] = pvconn; | |
843 | } else { | |
844 | VLOG_ERR("failed to listen on %s: %s", name, strerror(error)); | |
845 | if (!retval) { | |
846 | retval = error; | |
847 | } | |
848 | } | |
849 | } | |
850 | ||
851 | *pvconnsp = pvconns; | |
852 | *n_pvconnsp = n_pvconns; | |
853 | ||
854 | return retval; | |
855 | } | |
856 | ||
064af421 BP |
857 | int |
858 | ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops) | |
859 | { | |
860 | return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops); | |
861 | } | |
862 | ||
863 | int | |
0193b2af JG |
864 | ofproto_set_netflow(struct ofproto *ofproto, |
865 | const struct netflow_options *nf_options) | |
064af421 | 866 | { |
76343538 | 867 | if (nf_options && nf_options->collectors.n) { |
064af421 BP |
868 | if (!ofproto->netflow) { |
869 | ofproto->netflow = netflow_create(); | |
870 | } | |
0193b2af | 871 | return netflow_set_options(ofproto->netflow, nf_options); |
064af421 BP |
872 | } else { |
873 | netflow_destroy(ofproto->netflow); | |
874 | ofproto->netflow = NULL; | |
875 | return 0; | |
876 | } | |
877 | } | |
878 | ||
72b06300 BP |
879 | void |
880 | ofproto_set_sflow(struct ofproto *ofproto, | |
881 | const struct ofproto_sflow_options *oso) | |
882 | { | |
883 | struct ofproto_sflow *os = ofproto->sflow; | |
884 | if (oso) { | |
885 | if (!os) { | |
886 | struct ofport *ofport; | |
72b06300 BP |
887 | |
888 | os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif); | |
889 | refresh_port_groups(ofproto); | |
4e8e4213 | 890 | HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) { |
ca0f572c | 891 | ofproto_sflow_add_port(os, ofport->odp_port, |
72b06300 BP |
892 | netdev_get_name(ofport->netdev)); |
893 | } | |
894 | } | |
895 | ofproto_sflow_set_options(os, oso); | |
896 | } else { | |
897 | ofproto_sflow_destroy(os); | |
898 | ofproto->sflow = NULL; | |
899 | } | |
900 | } | |
901 | ||
064af421 BP |
902 | uint64_t |
903 | ofproto_get_datapath_id(const struct ofproto *ofproto) | |
904 | { | |
905 | return ofproto->datapath_id; | |
906 | } | |
907 | ||
76ce9432 | 908 | bool |
7d674866 | 909 | ofproto_has_primary_controller(const struct ofproto *ofproto) |
064af421 | 910 | { |
76ce9432 | 911 | return !hmap_is_empty(&ofproto->controllers); |
064af421 BP |
912 | } |
913 | ||
abdfe474 JP |
914 | enum ofproto_fail_mode |
915 | ofproto_get_fail_mode(const struct ofproto *p) | |
916 | { | |
917 | return p->fail_mode; | |
918 | } | |
919 | ||
064af421 BP |
920 | void |
921 | ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops) | |
922 | { | |
923 | size_t i; | |
924 | ||
925 | for (i = 0; i < ofproto->n_snoops; i++) { | |
926 | svec_add(snoops, pvconn_get_name(ofproto->snoops[i])); | |
927 | } | |
928 | } | |
929 | ||
930 | void | |
931 | ofproto_destroy(struct ofproto *p) | |
932 | { | |
7d674866 | 933 | struct ofservice *ofservice, *next_ofservice; |
064af421 | 934 | struct ofconn *ofconn, *next_ofconn; |
ca0f572c | 935 | struct ofport *ofport, *next_ofport; |
064af421 BP |
936 | size_t i; |
937 | ||
938 | if (!p) { | |
939 | return; | |
940 | } | |
941 | ||
f7de2cdf | 942 | /* Destroy fail-open and in-band early, since they touch the classifier. */ |
79c9f2ee BP |
943 | fail_open_destroy(p->fail_open); |
944 | p->fail_open = NULL; | |
945 | ||
946 | in_band_destroy(p->in_band); | |
947 | p->in_band = NULL; | |
917e50e1 | 948 | free(p->extra_in_band_remotes); |
2f6d3445 | 949 | |
064af421 BP |
950 | ofproto_flush_flows(p); |
951 | classifier_destroy(&p->cls); | |
952 | ||
4e8e4213 | 953 | LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) { |
c475ae67 | 954 | ofconn_destroy(ofconn); |
064af421 | 955 | } |
76ce9432 | 956 | hmap_destroy(&p->controllers); |
064af421 | 957 | |
c228a364 | 958 | dpif_close(p->dpif); |
e9e28be3 | 959 | netdev_monitor_destroy(p->netdev_monitor); |
4e8e4213 | 960 | HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) { |
ca0f572c | 961 | hmap_remove(&p->ports, &ofport->hmap_node); |
064af421 BP |
962 | ofport_free(ofport); |
963 | } | |
964 | shash_destroy(&p->port_by_name); | |
965 | ||
966 | switch_status_destroy(p->switch_status); | |
064af421 | 967 | netflow_destroy(p->netflow); |
72b06300 | 968 | ofproto_sflow_destroy(p->sflow); |
064af421 | 969 | |
4e8e4213 | 970 | HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) { |
7d674866 | 971 | ofservice_destroy(p, ofservice); |
064af421 | 972 | } |
7d674866 | 973 | hmap_destroy(&p->services); |
064af421 BP |
974 | |
975 | for (i = 0; i < p->n_snoops; i++) { | |
976 | pvconn_close(p->snoops[i]); | |
977 | } | |
978 | free(p->snoops); | |
979 | ||
980 | mac_learning_destroy(p->ml); | |
981 | ||
5a719c38 JP |
982 | free(p->mfr_desc); |
983 | free(p->hw_desc); | |
984 | free(p->sw_desc); | |
985 | free(p->serial_desc); | |
cb871ae0 JP |
986 | free(p->dp_desc); |
987 | ||
ca0f572c | 988 | hmap_destroy(&p->ports); |
3b917492 | 989 | |
064af421 BP |
990 | free(p); |
991 | } | |
992 | ||
993 | int | |
994 | ofproto_run(struct ofproto *p) | |
995 | { | |
996 | int error = ofproto_run1(p); | |
997 | if (!error) { | |
998 | error = ofproto_run2(p, false); | |
999 | } | |
1000 | return error; | |
1001 | } | |
1002 | ||
e9e28be3 BP |
1003 | static void |
1004 | process_port_change(struct ofproto *ofproto, int error, char *devname) | |
1005 | { | |
1006 | if (error == ENOBUFS) { | |
1007 | reinit_ports(ofproto); | |
1008 | } else if (!error) { | |
1009 | update_port(ofproto, devname); | |
1010 | free(devname); | |
1011 | } | |
1012 | } | |
1013 | ||
e2bfacb6 BP |
1014 | /* Returns a "preference level" for snooping 'ofconn'. A higher return value |
1015 | * means that 'ofconn' is more interesting for monitoring than a lower return | |
1016 | * value. */ | |
1017 | static int | |
1018 | snoop_preference(const struct ofconn *ofconn) | |
1019 | { | |
1020 | switch (ofconn->role) { | |
1021 | case NX_ROLE_MASTER: | |
1022 | return 3; | |
1023 | case NX_ROLE_OTHER: | |
1024 | return 2; | |
1025 | case NX_ROLE_SLAVE: | |
1026 | return 1; | |
1027 | default: | |
1028 | /* Shouldn't happen. */ | |
1029 | return 0; | |
1030 | } | |
1031 | } | |
1032 | ||
76ce9432 BP |
1033 | /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'. |
1034 | * Connects this vconn to a controller. */ | |
1035 | static void | |
1036 | add_snooper(struct ofproto *ofproto, struct vconn *vconn) | |
1037 | { | |
e2bfacb6 | 1038 | struct ofconn *ofconn, *best; |
76ce9432 | 1039 | |
e2bfacb6 BP |
1040 | /* Pick a controller for monitoring. */ |
1041 | best = NULL; | |
4e8e4213 | 1042 | LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) { |
5899143f | 1043 | if (ofconn->type == OFCONN_PRIMARY |
e2bfacb6 BP |
1044 | && (!best || snoop_preference(ofconn) > snoop_preference(best))) { |
1045 | best = ofconn; | |
76ce9432 | 1046 | } |
e2bfacb6 | 1047 | } |
76ce9432 | 1048 | |
e2bfacb6 BP |
1049 | if (best) { |
1050 | rconn_add_monitor(best->rconn, vconn); | |
1051 | } else { | |
1052 | VLOG_INFO_RL(&rl, "no controller connection to snoop"); | |
1053 | vconn_close(vconn); | |
76ce9432 | 1054 | } |
76ce9432 BP |
1055 | } |
1056 | ||
064af421 BP |
1057 | int |
1058 | ofproto_run1(struct ofproto *p) | |
1059 | { | |
1060 | struct ofconn *ofconn, *next_ofconn; | |
7d674866 | 1061 | struct ofservice *ofservice; |
064af421 BP |
1062 | char *devname; |
1063 | int error; | |
1064 | int i; | |
1065 | ||
149f577a JG |
1066 | if (shash_is_empty(&p->port_by_name)) { |
1067 | init_ports(p); | |
1068 | } | |
1069 | ||
064af421 BP |
1070 | for (i = 0; i < 50; i++) { |
1071 | struct ofpbuf *buf; | |
064af421 | 1072 | |
c228a364 | 1073 | error = dpif_recv(p->dpif, &buf); |
064af421 BP |
1074 | if (error) { |
1075 | if (error == ENODEV) { | |
1076 | /* Someone destroyed the datapath behind our back. The caller | |
1077 | * better destroy us and give up, because we're just going to | |
1078 | * spin from here on out. */ | |
39a559f2 BP |
1079 | static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5); |
1080 | VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally", | |
c228a364 | 1081 | dpif_name(p->dpif)); |
064af421 BP |
1082 | return ENODEV; |
1083 | } | |
1084 | break; | |
1085 | } | |
1086 | ||
1087 | handle_odp_msg(p, buf); | |
1088 | } | |
1089 | ||
e9e28be3 BP |
1090 | while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) { |
1091 | process_port_change(p, error, devname); | |
1092 | } | |
1093 | while ((error = netdev_monitor_poll(p->netdev_monitor, | |
1094 | &devname)) != EAGAIN) { | |
1095 | process_port_change(p, error, devname); | |
064af421 BP |
1096 | } |
1097 | ||
1098 | if (p->in_band) { | |
d2ede7bc BP |
1099 | if (time_msec() >= p->next_in_band_update) { |
1100 | update_in_band_remotes(p); | |
1101 | } | |
064af421 BP |
1102 | in_band_run(p->in_band); |
1103 | } | |
064af421 | 1104 | |
4e8e4213 | 1105 | LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) { |
064af421 BP |
1106 | ofconn_run(ofconn, p); |
1107 | } | |
1108 | ||
7778bd15 BP |
1109 | /* Fail-open maintenance. Do this after processing the ofconns since |
1110 | * fail-open checks the status of the controller rconn. */ | |
1111 | if (p->fail_open) { | |
1112 | fail_open_run(p->fail_open); | |
1113 | } | |
1114 | ||
4e8e4213 | 1115 | HMAP_FOR_EACH (ofservice, node, &p->services) { |
064af421 BP |
1116 | struct vconn *vconn; |
1117 | int retval; | |
1118 | ||
7d674866 | 1119 | retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn); |
064af421 | 1120 | if (!retval) { |
9794e806 | 1121 | struct rconn *rconn; |
eb15cdbb | 1122 | char *name; |
9794e806 | 1123 | |
7d674866 | 1124 | rconn = rconn_create(ofservice->probe_interval, 0); |
eb15cdbb BP |
1125 | name = ofconn_make_name(p, vconn_get_name(vconn)); |
1126 | rconn_connect_unreliably(rconn, vconn, name); | |
1127 | free(name); | |
1128 | ||
7d674866 BP |
1129 | ofconn = ofconn_create(p, rconn, OFCONN_SERVICE); |
1130 | ofconn_set_rate_limit(ofconn, ofservice->rate_limit, | |
1131 | ofservice->burst_limit); | |
064af421 BP |
1132 | } else if (retval != EAGAIN) { |
1133 | VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval)); | |
1134 | } | |
1135 | } | |
1136 | ||
1137 | for (i = 0; i < p->n_snoops; i++) { | |
1138 | struct vconn *vconn; | |
1139 | int retval; | |
1140 | ||
1141 | retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn); | |
1142 | if (!retval) { | |
76ce9432 | 1143 | add_snooper(p, vconn); |
064af421 BP |
1144 | } else if (retval != EAGAIN) { |
1145 | VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval)); | |
1146 | } | |
1147 | } | |
1148 | ||
1149 | if (time_msec() >= p->next_expiration) { | |
0de7a4b4 BP |
1150 | int delay = ofproto_expire(p); |
1151 | p->next_expiration = time_msec() + delay; | |
064af421 | 1152 | COVERAGE_INC(ofproto_expiration); |
064af421 BP |
1153 | } |
1154 | ||
1155 | if (p->netflow) { | |
1156 | netflow_run(p->netflow); | |
1157 | } | |
72b06300 BP |
1158 | if (p->sflow) { |
1159 | ofproto_sflow_run(p->sflow); | |
1160 | } | |
064af421 BP |
1161 | |
1162 | return 0; | |
1163 | } | |
1164 | ||
1165 | struct revalidate_cbdata { | |
1166 | struct ofproto *ofproto; | |
1167 | bool revalidate_all; /* Revalidate all exact-match rules? */ | |
1168 | bool revalidate_subrules; /* Revalidate all exact-match subrules? */ | |
1169 | struct tag_set revalidate_set; /* Set of tags to revalidate. */ | |
1170 | }; | |
1171 | ||
1172 | int | |
1173 | ofproto_run2(struct ofproto *p, bool revalidate_all) | |
1174 | { | |
1175 | if (p->need_revalidate || revalidate_all | |
1176 | || !tag_set_is_empty(&p->revalidate_set)) { | |
1177 | struct revalidate_cbdata cbdata; | |
1178 | cbdata.ofproto = p; | |
1179 | cbdata.revalidate_all = revalidate_all; | |
1180 | cbdata.revalidate_subrules = p->need_revalidate; | |
1181 | cbdata.revalidate_set = p->revalidate_set; | |
1182 | tag_set_init(&p->revalidate_set); | |
1183 | COVERAGE_INC(ofproto_revalidate); | |
1184 | classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata); | |
1185 | p->need_revalidate = false; | |
1186 | } | |
1187 | ||
1188 | return 0; | |
1189 | } | |
1190 | ||
1191 | void | |
1192 | ofproto_wait(struct ofproto *p) | |
1193 | { | |
7d674866 | 1194 | struct ofservice *ofservice; |
064af421 BP |
1195 | struct ofconn *ofconn; |
1196 | size_t i; | |
1197 | ||
c228a364 | 1198 | dpif_recv_wait(p->dpif); |
e9e28be3 BP |
1199 | dpif_port_poll_wait(p->dpif); |
1200 | netdev_monitor_poll_wait(p->netdev_monitor); | |
4e8e4213 | 1201 | LIST_FOR_EACH (ofconn, node, &p->all_conns) { |
064af421 BP |
1202 | ofconn_wait(ofconn); |
1203 | } | |
1204 | if (p->in_band) { | |
7cf8b266 | 1205 | poll_timer_wait_until(p->next_in_band_update); |
064af421 BP |
1206 | in_band_wait(p->in_band); |
1207 | } | |
064af421 BP |
1208 | if (p->fail_open) { |
1209 | fail_open_wait(p->fail_open); | |
1210 | } | |
72b06300 BP |
1211 | if (p->sflow) { |
1212 | ofproto_sflow_wait(p->sflow); | |
1213 | } | |
064af421 BP |
1214 | if (!tag_set_is_empty(&p->revalidate_set)) { |
1215 | poll_immediate_wake(); | |
1216 | } | |
1217 | if (p->need_revalidate) { | |
1218 | /* Shouldn't happen, but if it does just go around again. */ | |
1219 | VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); | |
1220 | poll_immediate_wake(); | |
1221 | } else if (p->next_expiration != LLONG_MAX) { | |
7cf8b266 | 1222 | poll_timer_wait_until(p->next_expiration); |
064af421 | 1223 | } |
4e8e4213 | 1224 | HMAP_FOR_EACH (ofservice, node, &p->services) { |
7d674866 | 1225 | pvconn_wait(ofservice->pvconn); |
064af421 BP |
1226 | } |
1227 | for (i = 0; i < p->n_snoops; i++) { | |
1228 | pvconn_wait(p->snoops[i]); | |
1229 | } | |
1230 | } | |
1231 | ||
1232 | void | |
1233 | ofproto_revalidate(struct ofproto *ofproto, tag_type tag) | |
1234 | { | |
1235 | tag_set_add(&ofproto->revalidate_set, tag); | |
1236 | } | |
1237 | ||
1238 | struct tag_set * | |
1239 | ofproto_get_revalidate_set(struct ofproto *ofproto) | |
1240 | { | |
1241 | return &ofproto->revalidate_set; | |
1242 | } | |
1243 | ||
1244 | bool | |
1245 | ofproto_is_alive(const struct ofproto *p) | |
1246 | { | |
76ce9432 | 1247 | return !hmap_is_empty(&p->controllers); |
064af421 BP |
1248 | } |
1249 | ||
3a6ccc8c BP |
1250 | /* Deletes port number 'odp_port' from the datapath for 'ofproto'. |
1251 | * | |
1252 | * This is almost the same as calling dpif_port_del() directly on the | |
1253 | * datapath, but it also makes 'ofproto' close its open netdev for the port | |
1254 | * (if any). This makes it possible to create a new netdev of a different | |
1255 | * type under the same name, which otherwise the netdev library would refuse | |
1256 | * to do because of the conflict. (The netdev would eventually get closed on | |
1257 | * the next trip through ofproto_run(), but this interface is more direct.) | |
1258 | * | |
3a6ccc8c BP |
1259 | * Returns 0 if successful, otherwise a positive errno. */ |
1260 | int | |
1261 | ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port) | |
1262 | { | |
1263 | struct ofport *ofport = get_port(ofproto, odp_port); | |
1264 | const char *name = ofport ? (char *) ofport->opp.name : "<unknown>"; | |
1265 | int error; | |
1266 | ||
1267 | error = dpif_port_del(ofproto->dpif, odp_port); | |
1268 | if (error) { | |
1269 | VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)", | |
1270 | dpif_name(ofproto->dpif), odp_port, name, strerror(error)); | |
1271 | } else if (ofport) { | |
1272 | /* 'name' is ofport->opp.name and update_port() is going to destroy | |
1273 | * 'ofport'. Just in case update_port() refers to 'name' after it | |
1274 | * destroys 'ofport', make a copy of it around the update_port() | |
1275 | * call. */ | |
1276 | char *devname = xstrdup(name); | |
1277 | update_port(ofproto, devname); | |
1278 | free(devname); | |
1279 | } | |
1280 | return error; | |
1281 | } | |
1282 | ||
064af421 BP |
1283 | int |
1284 | ofproto_send_packet(struct ofproto *p, const flow_t *flow, | |
1285 | const union ofp_action *actions, size_t n_actions, | |
1286 | const struct ofpbuf *packet) | |
1287 | { | |
1288 | struct odp_actions odp_actions; | |
1289 | int error; | |
1290 | ||
1291 | error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions, | |
6a07af36 | 1292 | NULL, NULL, NULL); |
064af421 BP |
1293 | if (error) { |
1294 | return error; | |
1295 | } | |
1296 | ||
1297 | /* XXX Should we translate the dpif_execute() errno value into an OpenFlow | |
1298 | * error code? */ | |
c228a364 | 1299 | dpif_execute(p->dpif, flow->in_port, odp_actions.actions, |
064af421 BP |
1300 | odp_actions.n_actions, packet); |
1301 | return 0; | |
1302 | } | |
1303 | ||
1304 | void | |
1305 | ofproto_add_flow(struct ofproto *p, | |
1306 | const flow_t *flow, uint32_t wildcards, unsigned int priority, | |
1307 | const union ofp_action *actions, size_t n_actions, | |
1308 | int idle_timeout) | |
1309 | { | |
1310 | struct rule *rule; | |
0193b2af | 1311 | rule = rule_create(p, NULL, actions, n_actions, |
d295e8e9 | 1312 | idle_timeout >= 0 ? idle_timeout : 5 /* XXX */, |
39997502 | 1313 | 0, 0, false); |
659586ef | 1314 | cls_rule_from_flow(flow, wildcards, priority, &rule->cr); |
064af421 BP |
1315 | rule_insert(p, rule, NULL, 0); |
1316 | } | |
1317 | ||
1318 | void | |
1319 | ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow, | |
1320 | uint32_t wildcards, unsigned int priority) | |
1321 | { | |
1322 | struct rule *rule; | |
1323 | ||
1324 | rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls, | |
1325 | flow, wildcards, | |
1326 | priority)); | |
1327 | if (rule) { | |
1328 | rule_remove(ofproto, rule); | |
1329 | } | |
1330 | } | |
1331 | ||
1332 | static void | |
1333 | destroy_rule(struct cls_rule *rule_, void *ofproto_) | |
1334 | { | |
1335 | struct rule *rule = rule_from_cls_rule(rule_); | |
1336 | struct ofproto *ofproto = ofproto_; | |
1337 | ||
1338 | /* Mark the flow as not installed, even though it might really be | |
1339 | * installed, so that rule_remove() doesn't bother trying to uninstall it. | |
1340 | * There is no point in uninstalling it individually since we are about to | |
1341 | * blow away all the flows with dpif_flow_flush(). */ | |
1342 | rule->installed = false; | |
1343 | ||
1344 | rule_remove(ofproto, rule); | |
1345 | } | |
1346 | ||
1347 | void | |
1348 | ofproto_flush_flows(struct ofproto *ofproto) | |
1349 | { | |
1350 | COVERAGE_INC(ofproto_flush); | |
1351 | classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto); | |
c228a364 | 1352 | dpif_flow_flush(ofproto->dpif); |
064af421 BP |
1353 | if (ofproto->in_band) { |
1354 | in_band_flushed(ofproto->in_band); | |
1355 | } | |
1356 | if (ofproto->fail_open) { | |
1357 | fail_open_flushed(ofproto->fail_open); | |
1358 | } | |
1359 | } | |
1360 | \f | |
1361 | static void | |
1362 | reinit_ports(struct ofproto *p) | |
1363 | { | |
1364 | struct svec devnames; | |
1365 | struct ofport *ofport; | |
064af421 BP |
1366 | struct odp_port *odp_ports; |
1367 | size_t n_odp_ports; | |
1368 | size_t i; | |
1369 | ||
1370 | svec_init(&devnames); | |
4e8e4213 | 1371 | HMAP_FOR_EACH (ofport, hmap_node, &p->ports) { |
064af421 BP |
1372 | svec_add (&devnames, (char *) ofport->opp.name); |
1373 | } | |
c228a364 | 1374 | dpif_port_list(p->dpif, &odp_ports, &n_odp_ports); |
064af421 BP |
1375 | for (i = 0; i < n_odp_ports; i++) { |
1376 | svec_add (&devnames, odp_ports[i].devname); | |
1377 | } | |
1378 | free(odp_ports); | |
1379 | ||
1380 | svec_sort_unique(&devnames); | |
1381 | for (i = 0; i < devnames.n; i++) { | |
1382 | update_port(p, devnames.names[i]); | |
1383 | } | |
1384 | svec_destroy(&devnames); | |
1385 | } | |
1386 | ||
72b06300 | 1387 | static size_t |
064af421 BP |
1388 | refresh_port_group(struct ofproto *p, unsigned int group) |
1389 | { | |
1390 | uint16_t *ports; | |
1391 | size_t n_ports; | |
1392 | struct ofport *port; | |
064af421 BP |
1393 | |
1394 | assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD); | |
1395 | ||
ca0f572c | 1396 | ports = xmalloc(hmap_count(&p->ports) * sizeof *ports); |
064af421 | 1397 | n_ports = 0; |
4e8e4213 | 1398 | HMAP_FOR_EACH (port, hmap_node, &p->ports) { |
064af421 | 1399 | if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) { |
ca0f572c | 1400 | ports[n_ports++] = port->odp_port; |
064af421 BP |
1401 | } |
1402 | } | |
c228a364 | 1403 | dpif_port_group_set(p->dpif, group, ports, n_ports); |
064af421 | 1404 | free(ports); |
72b06300 BP |
1405 | |
1406 | return n_ports; | |
064af421 BP |
1407 | } |
1408 | ||
1409 | static void | |
1410 | refresh_port_groups(struct ofproto *p) | |
1411 | { | |
72b06300 BP |
1412 | size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD); |
1413 | size_t n_all = refresh_port_group(p, DP_GROUP_ALL); | |
1414 | if (p->sflow) { | |
1415 | ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all); | |
1416 | } | |
064af421 BP |
1417 | } |
1418 | ||
1419 | static struct ofport * | |
1420 | make_ofport(const struct odp_port *odp_port) | |
1421 | { | |
149f577a | 1422 | struct netdev_options netdev_options; |
064af421 BP |
1423 | enum netdev_flags flags; |
1424 | struct ofport *ofport; | |
1425 | struct netdev *netdev; | |
1426 | bool carrier; | |
1427 | int error; | |
1428 | ||
149f577a JG |
1429 | memset(&netdev_options, 0, sizeof netdev_options); |
1430 | netdev_options.name = odp_port->devname; | |
1431 | netdev_options.ethertype = NETDEV_ETH_TYPE_NONE; | |
149f577a JG |
1432 | |
1433 | error = netdev_open(&netdev_options, &netdev); | |
064af421 BP |
1434 | if (error) { |
1435 | VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s " | |
1436 | "cannot be opened (%s)", | |
1437 | odp_port->devname, odp_port->port, | |
1438 | odp_port->devname, strerror(error)); | |
1439 | return NULL; | |
1440 | } | |
1441 | ||
1442 | ofport = xmalloc(sizeof *ofport); | |
1443 | ofport->netdev = netdev; | |
ca0f572c | 1444 | ofport->odp_port = odp_port->port; |
064af421 | 1445 | ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port); |
80992a35 | 1446 | netdev_get_etheraddr(netdev, ofport->opp.hw_addr); |
064af421 BP |
1447 | memcpy(ofport->opp.name, odp_port->devname, |
1448 | MIN(sizeof ofport->opp.name, sizeof odp_port->devname)); | |
1449 | ofport->opp.name[sizeof ofport->opp.name - 1] = '\0'; | |
1450 | ||
1451 | netdev_get_flags(netdev, &flags); | |
1452 | ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN; | |
1453 | ||
1454 | netdev_get_carrier(netdev, &carrier); | |
1455 | ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN; | |
1456 | ||
1457 | netdev_get_features(netdev, | |
1458 | &ofport->opp.curr, &ofport->opp.advertised, | |
1459 | &ofport->opp.supported, &ofport->opp.peer); | |
1460 | return ofport; | |
1461 | } | |
1462 | ||
1463 | static bool | |
1464 | ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port) | |
1465 | { | |
ca0f572c | 1466 | if (get_port(p, odp_port->port)) { |
064af421 BP |
1467 | VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath", |
1468 | odp_port->port); | |
1469 | return true; | |
1470 | } else if (shash_find(&p->port_by_name, odp_port->devname)) { | |
1471 | VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath", | |
1472 | odp_port->devname); | |
1473 | return true; | |
1474 | } else { | |
1475 | return false; | |
1476 | } | |
1477 | } | |
1478 | ||
1479 | static int | |
1480 | ofport_equal(const struct ofport *a_, const struct ofport *b_) | |
1481 | { | |
1482 | const struct ofp_phy_port *a = &a_->opp; | |
1483 | const struct ofp_phy_port *b = &b_->opp; | |
1484 | ||
1485 | BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */ | |
1486 | return (a->port_no == b->port_no | |
1487 | && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr) | |
1488 | && !strcmp((char *) a->name, (char *) b->name) | |
1489 | && a->state == b->state | |
1490 | && a->config == b->config | |
1491 | && a->curr == b->curr | |
1492 | && a->advertised == b->advertised | |
1493 | && a->supported == b->supported | |
1494 | && a->peer == b->peer); | |
1495 | } | |
1496 | ||
1497 | static void | |
1498 | send_port_status(struct ofproto *p, const struct ofport *ofport, | |
1499 | uint8_t reason) | |
1500 | { | |
1501 | /* XXX Should limit the number of queued port status change messages. */ | |
1502 | struct ofconn *ofconn; | |
4e8e4213 | 1503 | LIST_FOR_EACH (ofconn, node, &p->all_conns) { |
064af421 BP |
1504 | struct ofp_port_status *ops; |
1505 | struct ofpbuf *b; | |
1506 | ||
c91248b3 | 1507 | if (!ofconn_receives_async_msgs(ofconn)) { |
9deba63b BP |
1508 | continue; |
1509 | } | |
1510 | ||
064af421 BP |
1511 | ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b); |
1512 | ops->reason = reason; | |
1513 | ops->desc = ofport->opp; | |
1514 | hton_ofp_phy_port(&ops->desc); | |
1515 | queue_tx(b, ofconn, NULL); | |
1516 | } | |
064af421 BP |
1517 | } |
1518 | ||
1519 | static void | |
1520 | ofport_install(struct ofproto *p, struct ofport *ofport) | |
1521 | { | |
72b06300 BP |
1522 | const char *netdev_name = (const char *) ofport->opp.name; |
1523 | ||
e9e28be3 | 1524 | netdev_monitor_add(p->netdev_monitor, ofport->netdev); |
ca0f572c | 1525 | hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0)); |
72b06300 BP |
1526 | shash_add(&p->port_by_name, netdev_name, ofport); |
1527 | if (p->sflow) { | |
ca0f572c | 1528 | ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name); |
72b06300 | 1529 | } |
064af421 BP |
1530 | } |
1531 | ||
1532 | static void | |
1533 | ofport_remove(struct ofproto *p, struct ofport *ofport) | |
1534 | { | |
e9e28be3 | 1535 | netdev_monitor_remove(p->netdev_monitor, ofport->netdev); |
ca0f572c | 1536 | hmap_remove(&p->ports, &ofport->hmap_node); |
064af421 BP |
1537 | shash_delete(&p->port_by_name, |
1538 | shash_find(&p->port_by_name, (char *) ofport->opp.name)); | |
72b06300 | 1539 | if (p->sflow) { |
ca0f572c | 1540 | ofproto_sflow_del_port(p->sflow, ofport->odp_port); |
72b06300 | 1541 | } |
064af421 BP |
1542 | } |
1543 | ||
1544 | static void | |
1545 | ofport_free(struct ofport *ofport) | |
1546 | { | |
1547 | if (ofport) { | |
1548 | netdev_close(ofport->netdev); | |
1549 | free(ofport); | |
1550 | } | |
1551 | } | |
1552 | ||
ca0f572c BP |
1553 | static struct ofport * |
1554 | get_port(const struct ofproto *ofproto, uint16_t odp_port) | |
1555 | { | |
1556 | struct ofport *port; | |
1557 | ||
4e8e4213 | 1558 | HMAP_FOR_EACH_IN_BUCKET (port, hmap_node, |
ca0f572c BP |
1559 | hash_int(odp_port, 0), &ofproto->ports) { |
1560 | if (port->odp_port == odp_port) { | |
1561 | return port; | |
1562 | } | |
1563 | } | |
1564 | return NULL; | |
1565 | } | |
1566 | ||
064af421 BP |
1567 | static void |
1568 | update_port(struct ofproto *p, const char *devname) | |
1569 | { | |
1570 | struct odp_port odp_port; | |
c874dc6d BP |
1571 | struct ofport *old_ofport; |
1572 | struct ofport *new_ofport; | |
064af421 BP |
1573 | int error; |
1574 | ||
1575 | COVERAGE_INC(ofproto_update_port); | |
c874dc6d BP |
1576 | |
1577 | /* Query the datapath for port information. */ | |
c228a364 | 1578 | error = dpif_port_query_by_name(p->dpif, devname, &odp_port); |
064af421 | 1579 | |
c874dc6d BP |
1580 | /* Find the old ofport. */ |
1581 | old_ofport = shash_find_data(&p->port_by_name, devname); | |
1582 | if (!error) { | |
1583 | if (!old_ofport) { | |
1584 | /* There's no port named 'devname' but there might be a port with | |
1585 | * the same port number. This could happen if a port is deleted | |
1586 | * and then a new one added in its place very quickly, or if a port | |
1587 | * is renamed. In the former case we want to send an OFPPR_DELETE | |
1588 | * and an OFPPR_ADD, and in the latter case we want to send a | |
1589 | * single OFPPR_MODIFY. We can distinguish the cases by comparing | |
1590 | * the old port's ifindex against the new port, or perhaps less | |
1591 | * reliably but more portably by comparing the old port's MAC | |
1592 | * against the new port's MAC. However, this code isn't that smart | |
1593 | * and always sends an OFPPR_MODIFY (XXX). */ | |
ca0f572c | 1594 | old_ofport = get_port(p, odp_port.port); |
064af421 | 1595 | } |
c874dc6d | 1596 | } else if (error != ENOENT && error != ENODEV) { |
064af421 BP |
1597 | VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error " |
1598 | "%s", strerror(error)); | |
1599 | return; | |
1600 | } | |
c874dc6d BP |
1601 | |
1602 | /* Create a new ofport. */ | |
1603 | new_ofport = !error ? make_ofport(&odp_port) : NULL; | |
1604 | ||
1605 | /* Eliminate a few pathological cases. */ | |
1606 | if (!old_ofport && !new_ofport) { | |
1607 | return; | |
1608 | } else if (old_ofport && new_ofport) { | |
1609 | /* Most of the 'config' bits are OpenFlow soft state, but | |
1610 | * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow | |
1611 | * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and | |
1612 | * leaves the other bits 0.) */ | |
1613 | new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN; | |
1614 | ||
1615 | if (ofport_equal(old_ofport, new_ofport)) { | |
1616 | /* False alarm--no change. */ | |
1617 | ofport_free(new_ofport); | |
1618 | return; | |
1619 | } | |
1620 | } | |
1621 | ||
1622 | /* Now deal with the normal cases. */ | |
1623 | if (old_ofport) { | |
1624 | ofport_remove(p, old_ofport); | |
1625 | } | |
1626 | if (new_ofport) { | |
1627 | ofport_install(p, new_ofport); | |
1628 | } | |
1629 | send_port_status(p, new_ofport ? new_ofport : old_ofport, | |
1630 | (!old_ofport ? OFPPR_ADD | |
1631 | : !new_ofport ? OFPPR_DELETE | |
1632 | : OFPPR_MODIFY)); | |
1633 | ofport_free(old_ofport); | |
1634 | ||
1635 | /* Update port groups. */ | |
064af421 BP |
1636 | refresh_port_groups(p); |
1637 | } | |
1638 | ||
1639 | static int | |
1640 | init_ports(struct ofproto *p) | |
1641 | { | |
1642 | struct odp_port *ports; | |
1643 | size_t n_ports; | |
1644 | size_t i; | |
1645 | int error; | |
1646 | ||
c228a364 | 1647 | error = dpif_port_list(p->dpif, &ports, &n_ports); |
064af421 BP |
1648 | if (error) { |
1649 | return error; | |
1650 | } | |
1651 | ||
1652 | for (i = 0; i < n_ports; i++) { | |
1653 | const struct odp_port *odp_port = &ports[i]; | |
1654 | if (!ofport_conflicts(p, odp_port)) { | |
1655 | struct ofport *ofport = make_ofport(odp_port); | |
1656 | if (ofport) { | |
1657 | ofport_install(p, ofport); | |
1658 | } | |
1659 | } | |
1660 | } | |
1661 | free(ports); | |
1662 | refresh_port_groups(p); | |
1663 | return 0; | |
1664 | } | |
1665 | \f | |
1666 | static struct ofconn * | |
76ce9432 | 1667 | ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type) |
064af421 | 1668 | { |
76ce9432 BP |
1669 | struct ofconn *ofconn = xzalloc(sizeof *ofconn); |
1670 | ofconn->ofproto = p; | |
064af421 BP |
1671 | list_push_back(&p->all_conns, &ofconn->node); |
1672 | ofconn->rconn = rconn; | |
76ce9432 | 1673 | ofconn->type = type; |
9deba63b | 1674 | ofconn->role = NX_ROLE_OTHER; |
76ce9432 | 1675 | ofconn->packet_in_counter = rconn_packet_counter_create (); |
064af421 | 1676 | ofconn->pktbuf = NULL; |
064af421 | 1677 | ofconn->miss_send_len = 0; |
064af421 BP |
1678 | ofconn->reply_counter = rconn_packet_counter_create (); |
1679 | return ofconn; | |
1680 | } | |
1681 | ||
1682 | static void | |
c475ae67 | 1683 | ofconn_destroy(struct ofconn *ofconn) |
064af421 | 1684 | { |
5899143f | 1685 | if (ofconn->type == OFCONN_PRIMARY) { |
76ce9432 BP |
1686 | hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node); |
1687 | } | |
1688 | discovery_destroy(ofconn->discovery); | |
1689 | ||
064af421 | 1690 | list_remove(&ofconn->node); |
76ce9432 | 1691 | switch_status_unregister(ofconn->ss); |
064af421 BP |
1692 | rconn_destroy(ofconn->rconn); |
1693 | rconn_packet_counter_destroy(ofconn->packet_in_counter); | |
1694 | rconn_packet_counter_destroy(ofconn->reply_counter); | |
1695 | pktbuf_destroy(ofconn->pktbuf); | |
1696 | free(ofconn); | |
1697 | } | |
1698 | ||
1699 | static void | |
1700 | ofconn_run(struct ofconn *ofconn, struct ofproto *p) | |
1701 | { | |
1702 | int iteration; | |
76ce9432 BP |
1703 | size_t i; |
1704 | ||
1705 | if (ofconn->discovery) { | |
1706 | char *controller_name; | |
1707 | if (rconn_is_connectivity_questionable(ofconn->rconn)) { | |
1708 | discovery_question_connectivity(ofconn->discovery); | |
1709 | } | |
1710 | if (discovery_run(ofconn->discovery, &controller_name)) { | |
1711 | if (controller_name) { | |
eb15cdbb BP |
1712 | char *ofconn_name = ofconn_make_name(p, controller_name); |
1713 | rconn_connect(ofconn->rconn, controller_name, ofconn_name); | |
1714 | free(ofconn_name); | |
76ce9432 BP |
1715 | } else { |
1716 | rconn_disconnect(ofconn->rconn); | |
1717 | } | |
1718 | } | |
1719 | } | |
1720 | ||
1721 | for (i = 0; i < N_SCHEDULERS; i++) { | |
1722 | pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn); | |
1723 | } | |
064af421 BP |
1724 | |
1725 | rconn_run(ofconn->rconn); | |
1726 | ||
1727 | if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) { | |
1728 | /* Limit the number of iterations to prevent other tasks from | |
1729 | * starving. */ | |
1730 | for (iteration = 0; iteration < 50; iteration++) { | |
1731 | struct ofpbuf *of_msg = rconn_recv(ofconn->rconn); | |
1732 | if (!of_msg) { | |
1733 | break; | |
1734 | } | |
7778bd15 BP |
1735 | if (p->fail_open) { |
1736 | fail_open_maybe_recover(p->fail_open); | |
1737 | } | |
064af421 BP |
1738 | handle_openflow(ofconn, p, of_msg); |
1739 | ofpbuf_delete(of_msg); | |
1740 | } | |
1741 | } | |
1742 | ||
76ce9432 | 1743 | if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) { |
c475ae67 | 1744 | ofconn_destroy(ofconn); |
064af421 BP |
1745 | } |
1746 | } | |
1747 | ||
1748 | static void | |
1749 | ofconn_wait(struct ofconn *ofconn) | |
1750 | { | |
76ce9432 BP |
1751 | int i; |
1752 | ||
1753 | if (ofconn->discovery) { | |
1754 | discovery_wait(ofconn->discovery); | |
1755 | } | |
1756 | for (i = 0; i < N_SCHEDULERS; i++) { | |
1757 | pinsched_wait(ofconn->schedulers[i]); | |
1758 | } | |
064af421 BP |
1759 | rconn_run_wait(ofconn->rconn); |
1760 | if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) { | |
1761 | rconn_recv_wait(ofconn->rconn); | |
1762 | } else { | |
1763 | COVERAGE_INC(ofproto_ofconn_stuck); | |
1764 | } | |
1765 | } | |
c91248b3 BP |
1766 | |
1767 | /* Returns true if 'ofconn' should receive asynchronous messages. */ | |
1768 | static bool | |
1769 | ofconn_receives_async_msgs(const struct ofconn *ofconn) | |
1770 | { | |
5899143f BP |
1771 | if (ofconn->type == OFCONN_PRIMARY) { |
1772 | /* Primary controllers always get asynchronous messages unless they | |
c91248b3 BP |
1773 | * have configured themselves as "slaves". */ |
1774 | return ofconn->role != NX_ROLE_SLAVE; | |
1775 | } else { | |
5899143f BP |
1776 | /* Service connections don't get asynchronous messages unless they have |
1777 | * explicitly asked for them by setting a nonzero miss send length. */ | |
c91248b3 BP |
1778 | return ofconn->miss_send_len > 0; |
1779 | } | |
1780 | } | |
eb15cdbb BP |
1781 | |
1782 | /* Returns a human-readable name for an OpenFlow connection between 'ofproto' | |
1783 | * and 'target', suitable for use in log messages for identifying the | |
1784 | * connection. | |
1785 | * | |
1786 | * The name is dynamically allocated. The caller should free it (with free()) | |
1787 | * when it is no longer needed. */ | |
1788 | static char * | |
1789 | ofconn_make_name(const struct ofproto *ofproto, const char *target) | |
1790 | { | |
1791 | return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target); | |
1792 | } | |
7d674866 BP |
1793 | |
1794 | static void | |
1795 | ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst) | |
1796 | { | |
1797 | int i; | |
1798 | ||
1799 | for (i = 0; i < N_SCHEDULERS; i++) { | |
1800 | struct pinsched **s = &ofconn->schedulers[i]; | |
1801 | ||
1802 | if (rate > 0) { | |
1803 | if (!*s) { | |
1804 | *s = pinsched_create(rate, burst, | |
1805 | ofconn->ofproto->switch_status); | |
1806 | } else { | |
1807 | pinsched_set_limits(*s, rate, burst); | |
1808 | } | |
1809 | } else { | |
1810 | pinsched_destroy(*s); | |
1811 | *s = NULL; | |
1812 | } | |
1813 | } | |
1814 | } | |
1815 | \f | |
1816 | static void | |
1817 | ofservice_reconfigure(struct ofservice *ofservice, | |
1818 | const struct ofproto_controller *c) | |
1819 | { | |
1820 | ofservice->probe_interval = c->probe_interval; | |
1821 | ofservice->rate_limit = c->rate_limit; | |
1822 | ofservice->burst_limit = c->burst_limit; | |
1823 | } | |
1824 | ||
1825 | /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a | |
1826 | * positive errno value. */ | |
1827 | static int | |
1828 | ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c) | |
1829 | { | |
1830 | struct ofservice *ofservice; | |
1831 | struct pvconn *pvconn; | |
1832 | int error; | |
1833 | ||
1834 | error = pvconn_open(c->target, &pvconn); | |
1835 | if (error) { | |
1836 | return error; | |
1837 | } | |
1838 | ||
1839 | ofservice = xzalloc(sizeof *ofservice); | |
1840 | hmap_insert(&ofproto->services, &ofservice->node, | |
1841 | hash_string(c->target, 0)); | |
1842 | ofservice->pvconn = pvconn; | |
1843 | ||
1844 | ofservice_reconfigure(ofservice, c); | |
1845 | ||
1846 | return 0; | |
1847 | } | |
1848 | ||
1849 | static void | |
1850 | ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice) | |
1851 | { | |
1852 | hmap_remove(&ofproto->services, &ofservice->node); | |
1853 | pvconn_close(ofservice->pvconn); | |
1854 | free(ofservice); | |
1855 | } | |
1856 | ||
1857 | /* Finds and returns the ofservice within 'ofproto' that has the given | |
1858 | * 'target', or a null pointer if none exists. */ | |
1859 | static struct ofservice * | |
1860 | ofservice_lookup(struct ofproto *ofproto, const char *target) | |
1861 | { | |
1862 | struct ofservice *ofservice; | |
1863 | ||
4e8e4213 BP |
1864 | HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0), |
1865 | &ofproto->services) { | |
7d674866 BP |
1866 | if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) { |
1867 | return ofservice; | |
1868 | } | |
1869 | } | |
1870 | return NULL; | |
1871 | } | |
064af421 BP |
1872 | \f |
1873 | /* Caller is responsible for initializing the 'cr' member of the returned | |
1874 | * rule. */ | |
1875 | static struct rule * | |
0193b2af | 1876 | rule_create(struct ofproto *ofproto, struct rule *super, |
064af421 | 1877 | const union ofp_action *actions, size_t n_actions, |
ca069229 | 1878 | uint16_t idle_timeout, uint16_t hard_timeout, |
39997502 | 1879 | uint64_t flow_cookie, bool send_flow_removed) |
064af421 | 1880 | { |
ec6fde61 | 1881 | struct rule *rule = xzalloc(sizeof *rule); |
064af421 BP |
1882 | rule->idle_timeout = idle_timeout; |
1883 | rule->hard_timeout = hard_timeout; | |
39997502 | 1884 | rule->flow_cookie = flow_cookie; |
064af421 | 1885 | rule->used = rule->created = time_msec(); |
ca069229 | 1886 | rule->send_flow_removed = send_flow_removed; |
064af421 BP |
1887 | rule->super = super; |
1888 | if (super) { | |
1889 | list_push_back(&super->list, &rule->list); | |
1890 | } else { | |
1891 | list_init(&rule->list); | |
1892 | } | |
3dffcf07 BP |
1893 | if (n_actions > 0) { |
1894 | rule->n_actions = n_actions; | |
1895 | rule->actions = xmemdup(actions, n_actions * sizeof *actions); | |
1896 | } | |
0193b2af JG |
1897 | netflow_flow_clear(&rule->nf_flow); |
1898 | netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created); | |
1899 | ||
064af421 BP |
1900 | return rule; |
1901 | } | |
1902 | ||
1903 | static struct rule * | |
1904 | rule_from_cls_rule(const struct cls_rule *cls_rule) | |
1905 | { | |
1906 | return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL; | |
1907 | } | |
1908 | ||
1909 | static void | |
1910 | rule_free(struct rule *rule) | |
1911 | { | |
1912 | free(rule->actions); | |
1913 | free(rule->odp_actions); | |
1914 | free(rule); | |
1915 | } | |
1916 | ||
1917 | /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its | |
1918 | * super-rule's list of subrules. If 'rule' is a super-rule, also iterates | |
1919 | * through all of its subrules and revalidates them, destroying any that no | |
1920 | * longer has a super-rule (which is probably all of them). | |
1921 | * | |
1922 | * Before calling this function, the caller must make have removed 'rule' from | |
1923 | * the classifier. If 'rule' is an exact-match rule, the caller is also | |
1924 | * responsible for ensuring that it has been uninstalled from the datapath. */ | |
1925 | static void | |
1926 | rule_destroy(struct ofproto *ofproto, struct rule *rule) | |
1927 | { | |
1928 | if (!rule->super) { | |
1929 | struct rule *subrule, *next; | |
4e8e4213 | 1930 | LIST_FOR_EACH_SAFE (subrule, next, list, &rule->list) { |
064af421 BP |
1931 | revalidate_rule(ofproto, subrule); |
1932 | } | |
1933 | } else { | |
1934 | list_remove(&rule->list); | |
1935 | } | |
1936 | rule_free(rule); | |
1937 | } | |
1938 | ||
1939 | static bool | |
1940 | rule_has_out_port(const struct rule *rule, uint16_t out_port) | |
1941 | { | |
1942 | const union ofp_action *oa; | |
1943 | struct actions_iterator i; | |
1944 | ||
1945 | if (out_port == htons(OFPP_NONE)) { | |
1946 | return true; | |
1947 | } | |
1948 | for (oa = actions_first(&i, rule->actions, rule->n_actions); oa; | |
1949 | oa = actions_next(&i)) { | |
c1c9c9c4 | 1950 | if (action_outputs_to_port(oa, out_port)) { |
064af421 BP |
1951 | return true; |
1952 | } | |
1953 | } | |
1954 | return false; | |
1955 | } | |
1956 | ||
750638bb BP |
1957 | /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on |
1958 | * 'packet', which arrived on 'in_port'. | |
1959 | * | |
1960 | * Takes ownership of 'packet'. */ | |
9dbb9d5e BP |
1961 | static bool |
1962 | execute_odp_actions(struct ofproto *ofproto, uint16_t in_port, | |
1963 | const union odp_action *actions, size_t n_actions, | |
750638bb | 1964 | struct ofpbuf *packet) |
9dbb9d5e | 1965 | { |
750638bb | 1966 | if (n_actions == 1 && actions[0].type == ODPAT_CONTROLLER) { |
9dbb9d5e BP |
1967 | /* As an optimization, avoid a round-trip from userspace to kernel to |
1968 | * userspace. This also avoids possibly filling up kernel packet | |
1969 | * buffers along the way. */ | |
9dbb9d5e BP |
1970 | struct odp_msg *msg; |
1971 | ||
750638bb | 1972 | msg = ofpbuf_push_uninit(packet, sizeof *msg); |
9dbb9d5e BP |
1973 | msg->type = _ODPL_ACTION_NR; |
1974 | msg->length = sizeof(struct odp_msg) + packet->size; | |
1975 | msg->port = in_port; | |
1976 | msg->reserved = 0; | |
1977 | msg->arg = actions[0].controller.arg; | |
9dbb9d5e | 1978 | |
750638bb | 1979 | send_packet_in(ofproto, packet); |
9dbb9d5e | 1980 | |
750638bb BP |
1981 | return true; |
1982 | } else { | |
1983 | int error; | |
9dbb9d5e | 1984 | |
750638bb BP |
1985 | error = dpif_execute(ofproto->dpif, in_port, |
1986 | actions, n_actions, packet); | |
1987 | ofpbuf_delete(packet); | |
1988 | return !error; | |
1989 | } | |
9dbb9d5e BP |
1990 | } |
1991 | ||
064af421 | 1992 | /* Executes the actions indicated by 'rule' on 'packet', which is in flow |
750638bb BP |
1993 | * 'flow' and is considered to have arrived on ODP port 'in_port'. 'packet' |
1994 | * must have at least sizeof(struct ofp_packet_in) bytes of headroom. | |
064af421 BP |
1995 | * |
1996 | * The flow that 'packet' actually contains does not need to actually match | |
1997 | * 'rule'; the actions in 'rule' will be applied to it either way. Likewise, | |
1998 | * the packet and byte counters for 'rule' will be credited for the packet sent | |
1999 | * out whether or not the packet actually matches 'rule'. | |
2000 | * | |
2001 | * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow, | |
2002 | * the caller must already have accurately composed ODP actions for it given | |
2003 | * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if | |
2004 | * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this | |
2005 | * function will compose a set of ODP actions based on 'rule''s OpenFlow | |
750638bb BP |
2006 | * actions and apply them to 'packet'. |
2007 | * | |
2008 | * Takes ownership of 'packet'. */ | |
064af421 BP |
2009 | static void |
2010 | rule_execute(struct ofproto *ofproto, struct rule *rule, | |
2011 | struct ofpbuf *packet, const flow_t *flow) | |
2012 | { | |
2013 | const union odp_action *actions; | |
750638bb | 2014 | struct odp_flow_stats stats; |
064af421 BP |
2015 | size_t n_actions; |
2016 | struct odp_actions a; | |
2017 | ||
750638bb BP |
2018 | assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in)); |
2019 | ||
064af421 BP |
2020 | /* Grab or compose the ODP actions. |
2021 | * | |
2022 | * The special case for an exact-match 'rule' where 'flow' is not the | |
2023 | * rule's flow is important to avoid, e.g., sending a packet out its input | |
2024 | * port simply because the ODP actions were composed for the wrong | |
2025 | * scenario. */ | |
2026 | if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) { | |
2027 | struct rule *super = rule->super ? rule->super : rule; | |
2028 | if (xlate_actions(super->actions, super->n_actions, flow, ofproto, | |
6a07af36 | 2029 | packet, &a, NULL, 0, NULL)) { |
750638bb | 2030 | ofpbuf_delete(packet); |
064af421 BP |
2031 | return; |
2032 | } | |
2033 | actions = a.actions; | |
2034 | n_actions = a.n_actions; | |
2035 | } else { | |
2036 | actions = rule->odp_actions; | |
2037 | n_actions = rule->n_odp_actions; | |
2038 | } | |
2039 | ||
2040 | /* Execute the ODP actions. */ | |
750638bb | 2041 | flow_extract_stats(flow, packet, &stats); |
9dbb9d5e BP |
2042 | if (execute_odp_actions(ofproto, flow->in_port, |
2043 | actions, n_actions, packet)) { | |
0193b2af | 2044 | update_stats(ofproto, rule, &stats); |
064af421 | 2045 | rule->used = time_msec(); |
0193b2af | 2046 | netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used); |
064af421 BP |
2047 | } |
2048 | } | |
2049 | ||
750638bb BP |
2050 | /* Inserts 'rule' into 'p''s flow table. |
2051 | * | |
2052 | * If 'packet' is nonnull, takes ownership of 'packet', executes 'rule''s | |
2053 | * actions on it and credits the statistics for sending the packet to 'rule'. | |
2054 | * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of | |
2055 | * headroom. */ | |
064af421 BP |
2056 | static void |
2057 | rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet, | |
2058 | uint16_t in_port) | |
2059 | { | |
2060 | struct rule *displaced_rule; | |
2061 | ||
2062 | /* Insert the rule in the classifier. */ | |
2063 | displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr)); | |
2064 | if (!rule->cr.wc.wildcards) { | |
2065 | rule_make_actions(p, rule, packet); | |
2066 | } | |
2067 | ||
2068 | /* Send the packet and credit it to the rule. */ | |
2069 | if (packet) { | |
2070 | flow_t flow; | |
659586ef | 2071 | flow_extract(packet, 0, in_port, &flow); |
064af421 BP |
2072 | rule_execute(p, rule, packet, &flow); |
2073 | } | |
2074 | ||
2075 | /* Install the rule in the datapath only after sending the packet, to | |
2076 | * avoid packet reordering. */ | |
2077 | if (rule->cr.wc.wildcards) { | |
2078 | COVERAGE_INC(ofproto_add_wc_flow); | |
2079 | p->need_revalidate = true; | |
2080 | } else { | |
2081 | rule_install(p, rule, displaced_rule); | |
2082 | } | |
2083 | ||
2084 | /* Free the rule that was displaced, if any. */ | |
2085 | if (displaced_rule) { | |
2086 | rule_destroy(p, displaced_rule); | |
2087 | } | |
2088 | } | |
2089 | ||
2090 | static struct rule * | |
2091 | rule_create_subrule(struct ofproto *ofproto, struct rule *rule, | |
2092 | const flow_t *flow) | |
2093 | { | |
0193b2af | 2094 | struct rule *subrule = rule_create(ofproto, rule, NULL, 0, |
ca069229 | 2095 | rule->idle_timeout, rule->hard_timeout, |
39997502 | 2096 | 0, false); |
064af421 | 2097 | COVERAGE_INC(ofproto_subrule_create); |
659586ef JG |
2098 | cls_rule_from_flow(flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX |
2099 | : rule->cr.priority), &subrule->cr); | |
064af421 BP |
2100 | classifier_insert_exact(&ofproto->cls, &subrule->cr); |
2101 | ||
2102 | return subrule; | |
2103 | } | |
2104 | ||
431d4707 BP |
2105 | /* Remove 'rule' from 'ofproto' and free up the associated memory: |
2106 | * | |
2107 | * - If 'rule' was installed in the datapath, uninstalls it and updates | |
2108 | * 'rule''s statistics (or its super-rule's statistics, if it is a | |
2109 | * subrule), via rule_uninstall(). | |
2110 | * | |
2111 | * - Removes 'rule' from the classifier. | |
2112 | * | |
2113 | * - If 'rule' is a super-rule that has subrules, revalidates (and possibly | |
2114 | * uninstalls and destroys) its subrules, via rule_destroy(). | |
2115 | */ | |
064af421 BP |
2116 | static void |
2117 | rule_remove(struct ofproto *ofproto, struct rule *rule) | |
2118 | { | |
2119 | if (rule->cr.wc.wildcards) { | |
2120 | COVERAGE_INC(ofproto_del_wc_flow); | |
2121 | ofproto->need_revalidate = true; | |
2122 | } else { | |
2123 | rule_uninstall(ofproto, rule); | |
2124 | } | |
2125 | classifier_remove(&ofproto->cls, &rule->cr); | |
2126 | rule_destroy(ofproto, rule); | |
2127 | } | |
2128 | ||
2129 | /* Returns true if the actions changed, false otherwise. */ | |
2130 | static bool | |
2131 | rule_make_actions(struct ofproto *p, struct rule *rule, | |
2132 | const struct ofpbuf *packet) | |
2133 | { | |
2134 | const struct rule *super; | |
2135 | struct odp_actions a; | |
2136 | size_t actions_len; | |
2137 | ||
2138 | assert(!rule->cr.wc.wildcards); | |
2139 | ||
2140 | super = rule->super ? rule->super : rule; | |
2141 | rule->tags = 0; | |
2142 | xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p, | |
6a07af36 | 2143 | packet, &a, &rule->tags, &rule->may_install, |
0193b2af | 2144 | &rule->nf_flow.output_iface); |
064af421 BP |
2145 | |
2146 | actions_len = a.n_actions * sizeof *a.actions; | |
2147 | if (rule->n_odp_actions != a.n_actions | |
2148 | || memcmp(rule->odp_actions, a.actions, actions_len)) { | |
2149 | COVERAGE_INC(ofproto_odp_unchanged); | |
2150 | free(rule->odp_actions); | |
2151 | rule->n_odp_actions = a.n_actions; | |
2152 | rule->odp_actions = xmemdup(a.actions, actions_len); | |
2153 | return true; | |
2154 | } else { | |
2155 | return false; | |
2156 | } | |
2157 | } | |
2158 | ||
2159 | static int | |
2160 | do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags, | |
2161 | struct odp_flow_put *put) | |
2162 | { | |
2163 | memset(&put->flow.stats, 0, sizeof put->flow.stats); | |
2164 | put->flow.key = rule->cr.flow; | |
2165 | put->flow.actions = rule->odp_actions; | |
2166 | put->flow.n_actions = rule->n_odp_actions; | |
ab48643b | 2167 | put->flow.flags = 0; |
064af421 | 2168 | put->flags = flags; |
c228a364 | 2169 | return dpif_flow_put(ofproto->dpif, put); |
064af421 BP |
2170 | } |
2171 | ||
2172 | static void | |
2173 | rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule) | |
2174 | { | |
2175 | assert(!rule->cr.wc.wildcards); | |
2176 | ||
2177 | if (rule->may_install) { | |
2178 | struct odp_flow_put put; | |
2179 | if (!do_put_flow(p, rule, | |
2180 | ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS, | |
2181 | &put)) { | |
2182 | rule->installed = true; | |
2183 | if (displaced_rule) { | |
14986b31 | 2184 | update_stats(p, displaced_rule, &put.flow.stats); |
064af421 BP |
2185 | rule_post_uninstall(p, displaced_rule); |
2186 | } | |
2187 | } | |
2188 | } else if (displaced_rule) { | |
2189 | rule_uninstall(p, displaced_rule); | |
2190 | } | |
2191 | } | |
2192 | ||
2193 | static void | |
2194 | rule_reinstall(struct ofproto *ofproto, struct rule *rule) | |
2195 | { | |
2196 | if (rule->installed) { | |
2197 | struct odp_flow_put put; | |
2198 | COVERAGE_INC(ofproto_dp_missed); | |
2199 | do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put); | |
2200 | } else { | |
2201 | rule_install(ofproto, rule, NULL); | |
2202 | } | |
2203 | } | |
2204 | ||
2205 | static void | |
2206 | rule_update_actions(struct ofproto *ofproto, struct rule *rule) | |
2207 | { | |
42c3641c JG |
2208 | bool actions_changed; |
2209 | uint16_t new_out_iface, old_out_iface; | |
2210 | ||
2211 | old_out_iface = rule->nf_flow.output_iface; | |
2212 | actions_changed = rule_make_actions(ofproto, rule, NULL); | |
2213 | ||
064af421 BP |
2214 | if (rule->may_install) { |
2215 | if (rule->installed) { | |
2216 | if (actions_changed) { | |
064af421 | 2217 | struct odp_flow_put put; |
42c3641c JG |
2218 | do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY |
2219 | | ODPPF_ZERO_STATS, &put); | |
2220 | update_stats(ofproto, rule, &put.flow.stats); | |
2221 | ||
2222 | /* Temporarily set the old output iface so that NetFlow | |
2223 | * messages have the correct output interface for the old | |
2224 | * stats. */ | |
2225 | new_out_iface = rule->nf_flow.output_iface; | |
2226 | rule->nf_flow.output_iface = old_out_iface; | |
2227 | rule_post_uninstall(ofproto, rule); | |
2228 | rule->nf_flow.output_iface = new_out_iface; | |
064af421 BP |
2229 | } |
2230 | } else { | |
2231 | rule_install(ofproto, rule, NULL); | |
2232 | } | |
2233 | } else { | |
2234 | rule_uninstall(ofproto, rule); | |
2235 | } | |
2236 | } | |
2237 | ||
2238 | static void | |
2239 | rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes) | |
2240 | { | |
2241 | uint64_t total_bytes = rule->byte_count + extra_bytes; | |
2242 | ||
2243 | if (ofproto->ofhooks->account_flow_cb | |
2244 | && total_bytes > rule->accounted_bytes) | |
2245 | { | |
2246 | ofproto->ofhooks->account_flow_cb( | |
26d79bf2 | 2247 | &rule->cr.flow, rule->tags, rule->odp_actions, rule->n_odp_actions, |
064af421 BP |
2248 | total_bytes - rule->accounted_bytes, ofproto->aux); |
2249 | rule->accounted_bytes = total_bytes; | |
2250 | } | |
2251 | } | |
2252 | ||
431d4707 BP |
2253 | /* 'rule' must be an exact-match rule in 'p'. |
2254 | * | |
2255 | * If 'rule' is installed in the datapath, uninstalls it and updates's | |
2256 | * statistics. If 'rule' is a subrule, the statistics that are updated are | |
2257 | * actually its super-rule's statistics; otherwise 'rule''s own statistics are | |
2258 | * updated. | |
2259 | * | |
2260 | * If 'rule' is not installed, this function has no effect. */ | |
064af421 BP |
2261 | static void |
2262 | rule_uninstall(struct ofproto *p, struct rule *rule) | |
2263 | { | |
2264 | assert(!rule->cr.wc.wildcards); | |
2265 | if (rule->installed) { | |
2266 | struct odp_flow odp_flow; | |
2267 | ||
2268 | odp_flow.key = rule->cr.flow; | |
2269 | odp_flow.actions = NULL; | |
2270 | odp_flow.n_actions = 0; | |
ab48643b | 2271 | odp_flow.flags = 0; |
c228a364 | 2272 | if (!dpif_flow_del(p->dpif, &odp_flow)) { |
0193b2af | 2273 | update_stats(p, rule, &odp_flow.stats); |
064af421 BP |
2274 | } |
2275 | rule->installed = false; | |
2276 | ||
2277 | rule_post_uninstall(p, rule); | |
2278 | } | |
2279 | } | |
2280 | ||
0193b2af JG |
2281 | static bool |
2282 | is_controller_rule(struct rule *rule) | |
2283 | { | |
2284 | /* If the only action is send to the controller then don't report | |
2285 | * NetFlow expiration messages since it is just part of the control | |
2286 | * logic for the network and not real traffic. */ | |
2287 | ||
c1c9c9c4 BP |
2288 | return (rule |
2289 | && rule->super | |
2290 | && rule->super->n_actions == 1 | |
2291 | && action_outputs_to_port(&rule->super->actions[0], | |
2292 | htons(OFPP_CONTROLLER))); | |
0193b2af JG |
2293 | } |
2294 | ||
064af421 BP |
2295 | static void |
2296 | rule_post_uninstall(struct ofproto *ofproto, struct rule *rule) | |
2297 | { | |
2298 | struct rule *super = rule->super; | |
2299 | ||
2300 | rule_account(ofproto, rule, 0); | |
6a07af36 | 2301 | |
0193b2af | 2302 | if (ofproto->netflow && !is_controller_rule(rule)) { |
064af421 BP |
2303 | struct ofexpired expired; |
2304 | expired.flow = rule->cr.flow; | |
2305 | expired.packet_count = rule->packet_count; | |
2306 | expired.byte_count = rule->byte_count; | |
2307 | expired.used = rule->used; | |
0193b2af | 2308 | netflow_expire(ofproto->netflow, &rule->nf_flow, &expired); |
064af421 BP |
2309 | } |
2310 | if (super) { | |
2311 | super->packet_count += rule->packet_count; | |
2312 | super->byte_count += rule->byte_count; | |
064af421 | 2313 | |
0c0afbec JG |
2314 | /* Reset counters to prevent double counting if the rule ever gets |
2315 | * reinstalled. */ | |
2316 | rule->packet_count = 0; | |
2317 | rule->byte_count = 0; | |
2318 | rule->accounted_bytes = 0; | |
0193b2af JG |
2319 | |
2320 | netflow_flow_clear(&rule->nf_flow); | |
0c0afbec | 2321 | } |
064af421 BP |
2322 | } |
2323 | \f | |
2324 | static void | |
2325 | queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn, | |
2326 | struct rconn_packet_counter *counter) | |
2327 | { | |
2328 | update_openflow_length(msg); | |
2329 | if (rconn_send(ofconn->rconn, msg, counter)) { | |
2330 | ofpbuf_delete(msg); | |
2331 | } | |
2332 | } | |
2333 | ||
2334 | static void | |
2335 | send_error(const struct ofconn *ofconn, const struct ofp_header *oh, | |
2336 | int error, const void *data, size_t len) | |
2337 | { | |
2338 | struct ofpbuf *buf; | |
2339 | struct ofp_error_msg *oem; | |
2340 | ||
2341 | if (!(error >> 16)) { | |
2342 | VLOG_WARN_RL(&rl, "not sending bad error code %d to controller", | |
2343 | error); | |
2344 | return; | |
2345 | } | |
2346 | ||
2347 | COVERAGE_INC(ofproto_error); | |
2348 | oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR, | |
2349 | oh ? oh->xid : 0, &buf); | |
2350 | oem->type = htons((unsigned int) error >> 16); | |
2351 | oem->code = htons(error & 0xffff); | |
2352 | memcpy(oem->data, data, len); | |
2353 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
2354 | } | |
2355 | ||
2356 | static void | |
2357 | send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh, | |
2358 | int error) | |
2359 | { | |
2360 | size_t oh_length = ntohs(oh->length); | |
2361 | send_error(ofconn, oh, error, oh, MIN(oh_length, 64)); | |
2362 | } | |
2363 | ||
2364 | static void | |
2365 | hton_ofp_phy_port(struct ofp_phy_port *opp) | |
2366 | { | |
2367 | opp->port_no = htons(opp->port_no); | |
2368 | opp->config = htonl(opp->config); | |
2369 | opp->state = htonl(opp->state); | |
2370 | opp->curr = htonl(opp->curr); | |
2371 | opp->advertised = htonl(opp->advertised); | |
2372 | opp->supported = htonl(opp->supported); | |
2373 | opp->peer = htonl(opp->peer); | |
2374 | } | |
2375 | ||
2376 | static int | |
2377 | handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh) | |
2378 | { | |
2379 | struct ofp_header *rq = oh; | |
2380 | queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter); | |
2381 | return 0; | |
2382 | } | |
2383 | ||
2384 | static int | |
2385 | handle_features_request(struct ofproto *p, struct ofconn *ofconn, | |
2386 | struct ofp_header *oh) | |
2387 | { | |
2388 | struct ofp_switch_features *osf; | |
2389 | struct ofpbuf *buf; | |
064af421 BP |
2390 | struct ofport *port; |
2391 | ||
2392 | osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf); | |
2393 | osf->datapath_id = htonll(p->datapath_id); | |
2394 | osf->n_buffers = htonl(pktbuf_capacity()); | |
2395 | osf->n_tables = 2; | |
2396 | osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS | | |
0254ae23 | 2397 | OFPC_PORT_STATS | OFPC_ARP_MATCH_IP); |
064af421 BP |
2398 | osf->actions = htonl((1u << OFPAT_OUTPUT) | |
2399 | (1u << OFPAT_SET_VLAN_VID) | | |
2400 | (1u << OFPAT_SET_VLAN_PCP) | | |
2401 | (1u << OFPAT_STRIP_VLAN) | | |
2402 | (1u << OFPAT_SET_DL_SRC) | | |
2403 | (1u << OFPAT_SET_DL_DST) | | |
2404 | (1u << OFPAT_SET_NW_SRC) | | |
2405 | (1u << OFPAT_SET_NW_DST) | | |
959a2ecd | 2406 | (1u << OFPAT_SET_NW_TOS) | |
064af421 | 2407 | (1u << OFPAT_SET_TP_SRC) | |
c1c9c9c4 BP |
2408 | (1u << OFPAT_SET_TP_DST) | |
2409 | (1u << OFPAT_ENQUEUE)); | |
064af421 | 2410 | |
4e8e4213 | 2411 | HMAP_FOR_EACH (port, hmap_node, &p->ports) { |
064af421 BP |
2412 | hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp)); |
2413 | } | |
2414 | ||
2415 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
2416 | return 0; | |
2417 | } | |
2418 | ||
2419 | static int | |
2420 | handle_get_config_request(struct ofproto *p, struct ofconn *ofconn, | |
2421 | struct ofp_header *oh) | |
2422 | { | |
2423 | struct ofpbuf *buf; | |
2424 | struct ofp_switch_config *osc; | |
2425 | uint16_t flags; | |
2426 | bool drop_frags; | |
2427 | ||
2428 | /* Figure out flags. */ | |
c228a364 | 2429 | dpif_get_drop_frags(p->dpif, &drop_frags); |
064af421 | 2430 | flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL; |
064af421 BP |
2431 | |
2432 | /* Send reply. */ | |
2433 | osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf); | |
2434 | osc->flags = htons(flags); | |
2435 | osc->miss_send_len = htons(ofconn->miss_send_len); | |
2436 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
2437 | ||
2438 | return 0; | |
2439 | } | |
2440 | ||
2441 | static int | |
2442 | handle_set_config(struct ofproto *p, struct ofconn *ofconn, | |
2443 | struct ofp_switch_config *osc) | |
2444 | { | |
2445 | uint16_t flags; | |
2446 | int error; | |
2447 | ||
2448 | error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc); | |
2449 | if (error) { | |
2450 | return error; | |
2451 | } | |
2452 | flags = ntohs(osc->flags); | |
2453 | ||
5899143f | 2454 | if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) { |
064af421 BP |
2455 | switch (flags & OFPC_FRAG_MASK) { |
2456 | case OFPC_FRAG_NORMAL: | |
c228a364 | 2457 | dpif_set_drop_frags(p->dpif, false); |
064af421 BP |
2458 | break; |
2459 | case OFPC_FRAG_DROP: | |
c228a364 | 2460 | dpif_set_drop_frags(p->dpif, true); |
064af421 BP |
2461 | break; |
2462 | default: | |
2463 | VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")", | |
2464 | osc->flags); | |
2465 | break; | |
2466 | } | |
2467 | } | |
2468 | ||
064af421 BP |
2469 | ofconn->miss_send_len = ntohs(osc->miss_send_len); |
2470 | ||
2471 | return 0; | |
2472 | } | |
2473 | ||
2474 | static void | |
6a07af36 JG |
2475 | add_output_group_action(struct odp_actions *actions, uint16_t group, |
2476 | uint16_t *nf_output_iface) | |
064af421 BP |
2477 | { |
2478 | odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group; | |
6a07af36 JG |
2479 | |
2480 | if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) { | |
2481 | *nf_output_iface = NF_OUT_FLOOD; | |
2482 | } | |
064af421 BP |
2483 | } |
2484 | ||
2485 | static void | |
aae51f53 | 2486 | add_controller_action(struct odp_actions *actions, uint16_t max_len) |
064af421 BP |
2487 | { |
2488 | union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER); | |
aae51f53 | 2489 | a->controller.arg = max_len; |
064af421 BP |
2490 | } |
2491 | ||
2492 | struct action_xlate_ctx { | |
2493 | /* Input. */ | |
e18fe8a2 | 2494 | flow_t flow; /* Flow to which these actions correspond. */ |
064af421 BP |
2495 | int recurse; /* Recursion level, via xlate_table_action. */ |
2496 | struct ofproto *ofproto; | |
2497 | const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a | |
2498 | * null pointer if we are revalidating | |
2499 | * without a packet to refer to. */ | |
2500 | ||
2501 | /* Output. */ | |
2502 | struct odp_actions *out; /* Datapath actions. */ | |
2503 | tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */ | |
d6fbec6d | 2504 | bool may_set_up_flow; /* True ordinarily; false if the actions must |
064af421 | 2505 | * be reassessed for every packet. */ |
6a07af36 | 2506 | uint16_t nf_output_iface; /* Output interface index for NetFlow. */ |
064af421 BP |
2507 | }; |
2508 | ||
5f8bfd69 BP |
2509 | /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a |
2510 | * flow translation. */ | |
2511 | #define MAX_RESUBMIT_RECURSION 8 | |
2512 | ||
064af421 BP |
2513 | static void do_xlate_actions(const union ofp_action *in, size_t n_in, |
2514 | struct action_xlate_ctx *ctx); | |
2515 | ||
2516 | static void | |
2517 | add_output_action(struct action_xlate_ctx *ctx, uint16_t port) | |
2518 | { | |
ca0f572c | 2519 | const struct ofport *ofport = get_port(ctx->ofproto, port); |
6cfaf517 BP |
2520 | |
2521 | if (ofport) { | |
2522 | if (ofport->opp.config & OFPPC_NO_FWD) { | |
2523 | /* Forwarding disabled on port. */ | |
2524 | return; | |
2525 | } | |
2526 | } else { | |
2527 | /* | |
2528 | * We don't have an ofport record for this port, but it doesn't hurt to | |
2529 | * allow forwarding to it anyhow. Maybe such a port will appear later | |
2530 | * and we're pre-populating the flow table. | |
2531 | */ | |
064af421 | 2532 | } |
6cfaf517 BP |
2533 | |
2534 | odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port; | |
6a07af36 | 2535 | ctx->nf_output_iface = port; |
064af421 BP |
2536 | } |
2537 | ||
2538 | static struct rule * | |
2539 | lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow) | |
2540 | { | |
2541 | struct rule *rule; | |
2542 | rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow)); | |
2543 | ||
2544 | /* The rule we found might not be valid, since we could be in need of | |
2545 | * revalidation. If it is not valid, don't return it. */ | |
2546 | if (rule | |
2547 | && rule->super | |
2548 | && ofproto->need_revalidate | |
2549 | && !revalidate_rule(ofproto, rule)) { | |
2550 | COVERAGE_INC(ofproto_invalidated); | |
2551 | return NULL; | |
2552 | } | |
2553 | ||
2554 | return rule; | |
2555 | } | |
2556 | ||
2557 | static void | |
2558 | xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port) | |
2559 | { | |
5f8bfd69 | 2560 | if (ctx->recurse < MAX_RESUBMIT_RECURSION) { |
2c5d1389 | 2561 | uint16_t old_in_port; |
064af421 | 2562 | struct rule *rule; |
064af421 | 2563 | |
2c5d1389 BP |
2564 | /* Look up a flow with 'in_port' as the input port. Then restore the |
2565 | * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will | |
2566 | * have surprising behavior). */ | |
2567 | old_in_port = ctx->flow.in_port; | |
e18fe8a2 BP |
2568 | ctx->flow.in_port = in_port; |
2569 | rule = lookup_valid_rule(ctx->ofproto, &ctx->flow); | |
2c5d1389 BP |
2570 | ctx->flow.in_port = old_in_port; |
2571 | ||
064af421 BP |
2572 | if (rule) { |
2573 | if (rule->super) { | |
2574 | rule = rule->super; | |
2575 | } | |
2576 | ||
2577 | ctx->recurse++; | |
2578 | do_xlate_actions(rule->actions, rule->n_actions, ctx); | |
2579 | ctx->recurse--; | |
2580 | } | |
5f8bfd69 BP |
2581 | } else { |
2582 | struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
2583 | ||
2584 | VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times", | |
2585 | MAX_RESUBMIT_RECURSION); | |
064af421 BP |
2586 | } |
2587 | } | |
2588 | ||
2589 | static void | |
aae51f53 BP |
2590 | xlate_output_action__(struct action_xlate_ctx *ctx, |
2591 | uint16_t port, uint16_t max_len) | |
064af421 BP |
2592 | { |
2593 | uint16_t odp_port; | |
6a07af36 JG |
2594 | uint16_t prev_nf_output_iface = ctx->nf_output_iface; |
2595 | ||
2596 | ctx->nf_output_iface = NF_OUT_DROP; | |
064af421 | 2597 | |
aae51f53 | 2598 | switch (port) { |
064af421 | 2599 | case OFPP_IN_PORT: |
e18fe8a2 | 2600 | add_output_action(ctx, ctx->flow.in_port); |
064af421 BP |
2601 | break; |
2602 | case OFPP_TABLE: | |
e18fe8a2 | 2603 | xlate_table_action(ctx, ctx->flow.in_port); |
064af421 BP |
2604 | break; |
2605 | case OFPP_NORMAL: | |
e18fe8a2 | 2606 | if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet, |
064af421 | 2607 | ctx->out, ctx->tags, |
6a07af36 | 2608 | &ctx->nf_output_iface, |
064af421 BP |
2609 | ctx->ofproto->aux)) { |
2610 | COVERAGE_INC(ofproto_uninstallable); | |
d6fbec6d | 2611 | ctx->may_set_up_flow = false; |
064af421 BP |
2612 | } |
2613 | break; | |
2614 | case OFPP_FLOOD: | |
6a07af36 JG |
2615 | add_output_group_action(ctx->out, DP_GROUP_FLOOD, |
2616 | &ctx->nf_output_iface); | |
064af421 BP |
2617 | break; |
2618 | case OFPP_ALL: | |
6a07af36 | 2619 | add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface); |
064af421 BP |
2620 | break; |
2621 | case OFPP_CONTROLLER: | |
aae51f53 | 2622 | add_controller_action(ctx->out, max_len); |
064af421 BP |
2623 | break; |
2624 | case OFPP_LOCAL: | |
2625 | add_output_action(ctx, ODPP_LOCAL); | |
2626 | break; | |
2627 | default: | |
aae51f53 | 2628 | odp_port = ofp_port_to_odp_port(port); |
e18fe8a2 | 2629 | if (odp_port != ctx->flow.in_port) { |
064af421 BP |
2630 | add_output_action(ctx, odp_port); |
2631 | } | |
2632 | break; | |
2633 | } | |
6a07af36 JG |
2634 | |
2635 | if (prev_nf_output_iface == NF_OUT_FLOOD) { | |
2636 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
2637 | } else if (ctx->nf_output_iface == NF_OUT_DROP) { | |
2638 | ctx->nf_output_iface = prev_nf_output_iface; | |
2639 | } else if (prev_nf_output_iface != NF_OUT_DROP && | |
2640 | ctx->nf_output_iface != NF_OUT_FLOOD) { | |
2641 | ctx->nf_output_iface = NF_OUT_MULTI; | |
2642 | } | |
064af421 BP |
2643 | } |
2644 | ||
aae51f53 BP |
2645 | static void |
2646 | xlate_output_action(struct action_xlate_ctx *ctx, | |
2647 | const struct ofp_action_output *oao) | |
2648 | { | |
2649 | xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len)); | |
2650 | } | |
2651 | ||
c1c9c9c4 BP |
2652 | /* If the final ODP action in 'ctx' is "pop priority", drop it, as an |
2653 | * optimization, because we're going to add another action that sets the | |
2654 | * priority immediately after, or because there are no actions following the | |
2655 | * pop. */ | |
2656 | static void | |
2657 | remove_pop_action(struct action_xlate_ctx *ctx) | |
2658 | { | |
2659 | size_t n = ctx->out->n_actions; | |
2660 | if (n > 0 && ctx->out->actions[n - 1].type == ODPAT_POP_PRIORITY) { | |
2661 | ctx->out->n_actions--; | |
2662 | } | |
2663 | } | |
2664 | ||
2665 | static void | |
2666 | xlate_enqueue_action(struct action_xlate_ctx *ctx, | |
2667 | const struct ofp_action_enqueue *oae) | |
2668 | { | |
2669 | uint16_t ofp_port, odp_port; | |
aae51f53 BP |
2670 | uint32_t priority; |
2671 | int error; | |
2672 | ||
2673 | error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id), | |
2674 | &priority); | |
2675 | if (error) { | |
2676 | /* Fall back to ordinary output action. */ | |
2677 | xlate_output_action__(ctx, ntohs(oae->port), 0); | |
2678 | return; | |
2679 | } | |
c1c9c9c4 BP |
2680 | |
2681 | /* Figure out ODP output port. */ | |
2682 | ofp_port = ntohs(oae->port); | |
2683 | if (ofp_port != OFPP_IN_PORT) { | |
2684 | odp_port = ofp_port_to_odp_port(ofp_port); | |
2685 | } else { | |
2686 | odp_port = ctx->flow.in_port; | |
2687 | } | |
2688 | ||
2689 | /* Add ODP actions. */ | |
2690 | remove_pop_action(ctx); | |
2691 | odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority | |
aae51f53 | 2692 | = priority; |
c1c9c9c4 BP |
2693 | add_output_action(ctx, odp_port); |
2694 | odp_actions_add(ctx->out, ODPAT_POP_PRIORITY); | |
2695 | ||
2696 | /* Update NetFlow output port. */ | |
2697 | if (ctx->nf_output_iface == NF_OUT_DROP) { | |
2698 | ctx->nf_output_iface = odp_port; | |
2699 | } else if (ctx->nf_output_iface != NF_OUT_FLOOD) { | |
2700 | ctx->nf_output_iface = NF_OUT_MULTI; | |
2701 | } | |
2702 | } | |
2703 | ||
eedc0097 JP |
2704 | static void |
2705 | xlate_set_queue_action(struct action_xlate_ctx *ctx, | |
2706 | const struct nx_action_set_queue *nasq) | |
2707 | { | |
2708 | uint32_t priority; | |
2709 | int error; | |
2710 | ||
2711 | error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id), | |
2712 | &priority); | |
2713 | if (error) { | |
2714 | /* Couldn't translate queue to a priority, so ignore. A warning | |
2715 | * has already been logged. */ | |
2716 | return; | |
2717 | } | |
2718 | ||
2719 | remove_pop_action(ctx); | |
2720 | odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority | |
2721 | = priority; | |
2722 | } | |
2723 | ||
064af421 BP |
2724 | static void |
2725 | xlate_nicira_action(struct action_xlate_ctx *ctx, | |
2726 | const struct nx_action_header *nah) | |
2727 | { | |
2728 | const struct nx_action_resubmit *nar; | |
659586ef | 2729 | const struct nx_action_set_tunnel *nast; |
eedc0097 | 2730 | const struct nx_action_set_queue *nasq; |
659586ef | 2731 | union odp_action *oa; |
064af421 BP |
2732 | int subtype = ntohs(nah->subtype); |
2733 | ||
2734 | assert(nah->vendor == htonl(NX_VENDOR_ID)); | |
2735 | switch (subtype) { | |
2736 | case NXAST_RESUBMIT: | |
2737 | nar = (const struct nx_action_resubmit *) nah; | |
2738 | xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port))); | |
2739 | break; | |
2740 | ||
659586ef JG |
2741 | case NXAST_SET_TUNNEL: |
2742 | nast = (const struct nx_action_set_tunnel *) nah; | |
2743 | oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL); | |
2744 | ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id; | |
2745 | break; | |
2746 | ||
401eeb92 BP |
2747 | case NXAST_DROP_SPOOFED_ARP: |
2748 | if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) { | |
2749 | odp_actions_add(ctx->out, ODPAT_DROP_SPOOFED_ARP); | |
2750 | } | |
2751 | break; | |
2752 | ||
eedc0097 JP |
2753 | case NXAST_SET_QUEUE: |
2754 | nasq = (const struct nx_action_set_queue *) nah; | |
2755 | xlate_set_queue_action(ctx, nasq); | |
2756 | break; | |
2757 | ||
2758 | case NXAST_POP_QUEUE: | |
2759 | odp_actions_add(ctx->out, ODPAT_POP_PRIORITY); | |
2760 | break; | |
2761 | ||
999f0d45 | 2762 | /* If you add a new action here that modifies flow data, don't forget to |
c1c9c9c4 | 2763 | * update the flow key in ctx->flow at the same time. */ |
999f0d45 | 2764 | |
064af421 BP |
2765 | default: |
2766 | VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype); | |
2767 | break; | |
2768 | } | |
2769 | } | |
2770 | ||
2771 | static void | |
2772 | do_xlate_actions(const union ofp_action *in, size_t n_in, | |
2773 | struct action_xlate_ctx *ctx) | |
2774 | { | |
2775 | struct actions_iterator iter; | |
2776 | const union ofp_action *ia; | |
2777 | const struct ofport *port; | |
2778 | ||
ca0f572c | 2779 | port = get_port(ctx->ofproto, ctx->flow.in_port); |
064af421 | 2780 | if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) && |
ba186119 | 2781 | port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp) |
064af421 BP |
2782 | ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) { |
2783 | /* Drop this flow. */ | |
2784 | return; | |
2785 | } | |
2786 | ||
2787 | for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) { | |
2788 | uint16_t type = ntohs(ia->type); | |
2789 | union odp_action *oa; | |
2790 | ||
2791 | switch (type) { | |
2792 | case OFPAT_OUTPUT: | |
2793 | xlate_output_action(ctx, &ia->output); | |
2794 | break; | |
2795 | ||
2796 | case OFPAT_SET_VLAN_VID: | |
2797 | oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID); | |
999f0d45 | 2798 | ctx->flow.dl_vlan = oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid; |
064af421 BP |
2799 | break; |
2800 | ||
2801 | case OFPAT_SET_VLAN_PCP: | |
2802 | oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP); | |
999f0d45 | 2803 | ctx->flow.dl_vlan_pcp = oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp; |
064af421 BP |
2804 | break; |
2805 | ||
2806 | case OFPAT_STRIP_VLAN: | |
2807 | odp_actions_add(ctx->out, ODPAT_STRIP_VLAN); | |
7ac0f8db | 2808 | ctx->flow.dl_vlan = htons(OFP_VLAN_NONE); |
999f0d45 | 2809 | ctx->flow.dl_vlan_pcp = 0; |
064af421 BP |
2810 | break; |
2811 | ||
2812 | case OFPAT_SET_DL_SRC: | |
2813 | oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC); | |
2814 | memcpy(oa->dl_addr.dl_addr, | |
2815 | ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); | |
999f0d45 BP |
2816 | memcpy(ctx->flow.dl_src, |
2817 | ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); | |
064af421 BP |
2818 | break; |
2819 | ||
2820 | case OFPAT_SET_DL_DST: | |
2821 | oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST); | |
2822 | memcpy(oa->dl_addr.dl_addr, | |
2823 | ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); | |
999f0d45 BP |
2824 | memcpy(ctx->flow.dl_dst, |
2825 | ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); | |
064af421 BP |
2826 | break; |
2827 | ||
2828 | case OFPAT_SET_NW_SRC: | |
2829 | oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC); | |
999f0d45 | 2830 | ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr; |
064af421 BP |
2831 | break; |
2832 | ||
2d70a31a JP |
2833 | case OFPAT_SET_NW_DST: |
2834 | oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST); | |
999f0d45 | 2835 | ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr; |
2d38e234 | 2836 | break; |
959a2ecd JP |
2837 | |
2838 | case OFPAT_SET_NW_TOS: | |
2839 | oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS); | |
999f0d45 | 2840 | ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos; |
2d70a31a JP |
2841 | break; |
2842 | ||
064af421 BP |
2843 | case OFPAT_SET_TP_SRC: |
2844 | oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC); | |
999f0d45 | 2845 | ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port; |
064af421 BP |
2846 | break; |
2847 | ||
2d70a31a JP |
2848 | case OFPAT_SET_TP_DST: |
2849 | oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST); | |
999f0d45 | 2850 | ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port; |
2d70a31a JP |
2851 | break; |
2852 | ||
064af421 BP |
2853 | case OFPAT_VENDOR: |
2854 | xlate_nicira_action(ctx, (const struct nx_action_header *) ia); | |
2855 | break; | |
2856 | ||
c1c9c9c4 BP |
2857 | case OFPAT_ENQUEUE: |
2858 | xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia); | |
2859 | break; | |
2860 | ||
064af421 BP |
2861 | default: |
2862 | VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type); | |
2863 | break; | |
2864 | } | |
2865 | } | |
2866 | } | |
2867 | ||
2868 | static int | |
2869 | xlate_actions(const union ofp_action *in, size_t n_in, | |
2870 | const flow_t *flow, struct ofproto *ofproto, | |
2871 | const struct ofpbuf *packet, | |
6a07af36 JG |
2872 | struct odp_actions *out, tag_type *tags, bool *may_set_up_flow, |
2873 | uint16_t *nf_output_iface) | |
064af421 BP |
2874 | { |
2875 | tag_type no_tags = 0; | |
2876 | struct action_xlate_ctx ctx; | |
2877 | COVERAGE_INC(ofproto_ofp2odp); | |
2878 | odp_actions_init(out); | |
e18fe8a2 | 2879 | ctx.flow = *flow; |
064af421 BP |
2880 | ctx.recurse = 0; |
2881 | ctx.ofproto = ofproto; | |
2882 | ctx.packet = packet; | |
2883 | ctx.out = out; | |
2884 | ctx.tags = tags ? tags : &no_tags; | |
d6fbec6d | 2885 | ctx.may_set_up_flow = true; |
6a07af36 | 2886 | ctx.nf_output_iface = NF_OUT_DROP; |
064af421 | 2887 | do_xlate_actions(in, n_in, &ctx); |
c1c9c9c4 | 2888 | remove_pop_action(&ctx); |
0ad9b732 | 2889 | |
d6fbec6d | 2890 | /* Check with in-band control to see if we're allowed to set up this |
0ad9b732 JP |
2891 | * flow. */ |
2892 | if (!in_band_rule_check(ofproto->in_band, flow, out)) { | |
d6fbec6d | 2893 | ctx.may_set_up_flow = false; |
0ad9b732 JP |
2894 | } |
2895 | ||
d6fbec6d BP |
2896 | if (may_set_up_flow) { |
2897 | *may_set_up_flow = ctx.may_set_up_flow; | |
064af421 | 2898 | } |
6a07af36 JG |
2899 | if (nf_output_iface) { |
2900 | *nf_output_iface = ctx.nf_output_iface; | |
064af421 BP |
2901 | } |
2902 | if (odp_actions_overflow(out)) { | |
9026bf34 | 2903 | COVERAGE_INC(odp_overflow); |
064af421 BP |
2904 | odp_actions_init(out); |
2905 | return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY); | |
2906 | } | |
2907 | return 0; | |
2908 | } | |
2909 | ||
9deba63b BP |
2910 | /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow |
2911 | * error message code (composed with ofp_mkerr()) for the caller to propagate | |
2912 | * upward. Otherwise, returns 0. | |
2913 | * | |
2914 | * 'oh' is used to make log messages more informative. */ | |
2915 | static int | |
2916 | reject_slave_controller(struct ofconn *ofconn, const struct ofp_header *oh) | |
2917 | { | |
5899143f | 2918 | if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) { |
9deba63b BP |
2919 | static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5); |
2920 | char *type_name; | |
2921 | ||
2922 | type_name = ofp_message_type_to_string(oh->type); | |
2923 | VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller", | |
2924 | type_name); | |
2925 | free(type_name); | |
2926 | ||
2927 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM); | |
2928 | } else { | |
2929 | return 0; | |
2930 | } | |
2931 | } | |
2932 | ||
064af421 BP |
2933 | static int |
2934 | handle_packet_out(struct ofproto *p, struct ofconn *ofconn, | |
2935 | struct ofp_header *oh) | |
2936 | { | |
2937 | struct ofp_packet_out *opo; | |
2938 | struct ofpbuf payload, *buffer; | |
2939 | struct odp_actions actions; | |
2940 | int n_actions; | |
2941 | uint16_t in_port; | |
2942 | flow_t flow; | |
2943 | int error; | |
2944 | ||
9deba63b BP |
2945 | error = reject_slave_controller(ofconn, oh); |
2946 | if (error) { | |
2947 | return error; | |
2948 | } | |
2949 | ||
064af421 BP |
2950 | error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports); |
2951 | if (error) { | |
2952 | return error; | |
2953 | } | |
2954 | opo = (struct ofp_packet_out *) oh; | |
2955 | ||
2956 | COVERAGE_INC(ofproto_packet_out); | |
2957 | if (opo->buffer_id != htonl(UINT32_MAX)) { | |
2958 | error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id), | |
2959 | &buffer, &in_port); | |
7778bd15 | 2960 | if (error || !buffer) { |
064af421 BP |
2961 | return error; |
2962 | } | |
2963 | payload = *buffer; | |
2964 | } else { | |
2965 | buffer = NULL; | |
2966 | } | |
2967 | ||
659586ef | 2968 | flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow); |
064af421 | 2969 | error = xlate_actions((const union ofp_action *) opo->actions, n_actions, |
6a07af36 | 2970 | &flow, p, &payload, &actions, NULL, NULL, NULL); |
064af421 BP |
2971 | if (error) { |
2972 | return error; | |
2973 | } | |
2974 | ||
c228a364 | 2975 | dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions, |
064af421 BP |
2976 | &payload); |
2977 | ofpbuf_delete(buffer); | |
2978 | ||
2979 | return 0; | |
2980 | } | |
2981 | ||
2982 | static void | |
2983 | update_port_config(struct ofproto *p, struct ofport *port, | |
2984 | uint32_t config, uint32_t mask) | |
2985 | { | |
2986 | mask &= config ^ port->opp.config; | |
2987 | if (mask & OFPPC_PORT_DOWN) { | |
2988 | if (config & OFPPC_PORT_DOWN) { | |
2989 | netdev_turn_flags_off(port->netdev, NETDEV_UP, true); | |
2990 | } else { | |
2991 | netdev_turn_flags_on(port->netdev, NETDEV_UP, true); | |
2992 | } | |
2993 | } | |
2994 | #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD) | |
2995 | if (mask & REVALIDATE_BITS) { | |
2996 | COVERAGE_INC(ofproto_costly_flags); | |
2997 | port->opp.config ^= mask & REVALIDATE_BITS; | |
2998 | p->need_revalidate = true; | |
2999 | } | |
3000 | #undef REVALIDATE_BITS | |
3001 | if (mask & OFPPC_NO_FLOOD) { | |
3002 | port->opp.config ^= OFPPC_NO_FLOOD; | |
72b06300 | 3003 | refresh_port_groups(p); |
064af421 BP |
3004 | } |
3005 | if (mask & OFPPC_NO_PACKET_IN) { | |
3006 | port->opp.config ^= OFPPC_NO_PACKET_IN; | |
3007 | } | |
3008 | } | |
3009 | ||
3010 | static int | |
9deba63b BP |
3011 | handle_port_mod(struct ofproto *p, struct ofconn *ofconn, |
3012 | struct ofp_header *oh) | |
064af421 BP |
3013 | { |
3014 | const struct ofp_port_mod *opm; | |
3015 | struct ofport *port; | |
3016 | int error; | |
3017 | ||
9deba63b BP |
3018 | error = reject_slave_controller(ofconn, oh); |
3019 | if (error) { | |
3020 | return error; | |
3021 | } | |
064af421 BP |
3022 | error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm); |
3023 | if (error) { | |
3024 | return error; | |
3025 | } | |
3026 | opm = (struct ofp_port_mod *) oh; | |
3027 | ||
ca0f572c | 3028 | port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no))); |
064af421 BP |
3029 | if (!port) { |
3030 | return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT); | |
3031 | } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) { | |
3032 | return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR); | |
3033 | } else { | |
3034 | update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask)); | |
3035 | if (opm->advertise) { | |
3036 | netdev_set_advertisements(port->netdev, ntohl(opm->advertise)); | |
3037 | } | |
3038 | } | |
3039 | return 0; | |
3040 | } | |
3041 | ||
3042 | static struct ofpbuf * | |
3043 | make_stats_reply(uint32_t xid, uint16_t type, size_t body_len) | |
3044 | { | |
3045 | struct ofp_stats_reply *osr; | |
3046 | struct ofpbuf *msg; | |
3047 | ||
3048 | msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX)); | |
3049 | osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg); | |
3050 | osr->type = type; | |
3051 | osr->flags = htons(0); | |
3052 | return msg; | |
3053 | } | |
3054 | ||
3055 | static struct ofpbuf * | |
3056 | start_stats_reply(const struct ofp_stats_request *request, size_t body_len) | |
3057 | { | |
3058 | return make_stats_reply(request->header.xid, request->type, body_len); | |
3059 | } | |
3060 | ||
3061 | static void * | |
3062 | append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp) | |
3063 | { | |
3064 | struct ofpbuf *msg = *msgp; | |
3065 | assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply)); | |
3066 | if (nbytes + msg->size > UINT16_MAX) { | |
3067 | struct ofp_stats_reply *reply = msg->data; | |
3068 | reply->flags = htons(OFPSF_REPLY_MORE); | |
3069 | *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes); | |
3070 | queue_tx(msg, ofconn, ofconn->reply_counter); | |
3071 | } | |
3072 | return ofpbuf_put_uninit(*msgp, nbytes); | |
3073 | } | |
3074 | ||
3075 | static int | |
3076 | handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn, | |
3077 | struct ofp_stats_request *request) | |
3078 | { | |
3079 | struct ofp_desc_stats *ods; | |
3080 | struct ofpbuf *msg; | |
3081 | ||
3082 | msg = start_stats_reply(request, sizeof *ods); | |
3083 | ods = append_stats_reply(sizeof *ods, ofconn, &msg); | |
5a719c38 JP |
3084 | memset(ods, 0, sizeof *ods); |
3085 | ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc); | |
3086 | ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc); | |
3087 | ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc); | |
3088 | ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num); | |
3089 | ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc); | |
064af421 BP |
3090 | queue_tx(msg, ofconn, ofconn->reply_counter); |
3091 | ||
3092 | return 0; | |
3093 | } | |
3094 | ||
064af421 BP |
3095 | static int |
3096 | handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn, | |
3097 | struct ofp_stats_request *request) | |
3098 | { | |
3099 | struct ofp_table_stats *ots; | |
3100 | struct ofpbuf *msg; | |
3101 | struct odp_stats dpstats; | |
3102 | int n_exact, n_subrules, n_wild; | |
35950f0c | 3103 | struct rule *rule; |
064af421 BP |
3104 | |
3105 | msg = start_stats_reply(request, sizeof *ots * 2); | |
3106 | ||
3107 | /* Count rules of various kinds. */ | |
3108 | n_subrules = 0; | |
35950f0c BP |
3109 | CLASSIFIER_FOR_EACH_EXACT_RULE (rule, cr, &p->cls) { |
3110 | if (rule->super) { | |
3111 | n_subrules++; | |
3112 | } | |
3113 | } | |
064af421 BP |
3114 | n_exact = classifier_count_exact(&p->cls) - n_subrules; |
3115 | n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls); | |
3116 | ||
3117 | /* Hash table. */ | |
c228a364 | 3118 | dpif_get_dp_stats(p->dpif, &dpstats); |
064af421 BP |
3119 | ots = append_stats_reply(sizeof *ots, ofconn, &msg); |
3120 | memset(ots, 0, sizeof *ots); | |
3121 | ots->table_id = TABLEID_HASH; | |
3122 | strcpy(ots->name, "hash"); | |
3123 | ots->wildcards = htonl(0); | |
3124 | ots->max_entries = htonl(dpstats.max_capacity); | |
3125 | ots->active_count = htonl(n_exact); | |
3126 | ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit + | |
3127 | dpstats.n_missed); | |
3128 | ots->matched_count = htonll(dpstats.n_hit); /* XXX */ | |
3129 | ||
3130 | /* Classifier table. */ | |
3131 | ots = append_stats_reply(sizeof *ots, ofconn, &msg); | |
3132 | memset(ots, 0, sizeof *ots); | |
3133 | ots->table_id = TABLEID_CLASSIFIER; | |
3134 | strcpy(ots->name, "classifier"); | |
659586ef JG |
3135 | ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL) |
3136 | : htonl(OFPFW_ALL); | |
064af421 BP |
3137 | ots->max_entries = htonl(65536); |
3138 | ots->active_count = htonl(n_wild); | |
3139 | ots->lookup_count = htonll(0); /* XXX */ | |
3140 | ots->matched_count = htonll(0); /* XXX */ | |
3141 | ||
3142 | queue_tx(msg, ofconn, ofconn->reply_counter); | |
3143 | return 0; | |
3144 | } | |
3145 | ||
abaad8cf | 3146 | static void |
ca0f572c | 3147 | append_port_stat(struct ofport *port, struct ofconn *ofconn, |
a4948b95 | 3148 | struct ofpbuf **msgp) |
abaad8cf JP |
3149 | { |
3150 | struct netdev_stats stats; | |
3151 | struct ofp_port_stats *ops; | |
3152 | ||
d295e8e9 JP |
3153 | /* Intentionally ignore return value, since errors will set |
3154 | * 'stats' to all-1s, which is correct for OpenFlow, and | |
abaad8cf JP |
3155 | * netdev_get_stats() will log errors. */ |
3156 | netdev_get_stats(port->netdev, &stats); | |
3157 | ||
a4948b95 | 3158 | ops = append_stats_reply(sizeof *ops, ofconn, msgp); |
ca0f572c | 3159 | ops->port_no = htons(port->opp.port_no); |
abaad8cf JP |
3160 | memset(ops->pad, 0, sizeof ops->pad); |
3161 | ops->rx_packets = htonll(stats.rx_packets); | |
3162 | ops->tx_packets = htonll(stats.tx_packets); | |
3163 | ops->rx_bytes = htonll(stats.rx_bytes); | |
3164 | ops->tx_bytes = htonll(stats.tx_bytes); | |
3165 | ops->rx_dropped = htonll(stats.rx_dropped); | |
3166 | ops->tx_dropped = htonll(stats.tx_dropped); | |
3167 | ops->rx_errors = htonll(stats.rx_errors); | |
3168 | ops->tx_errors = htonll(stats.tx_errors); | |
3169 | ops->rx_frame_err = htonll(stats.rx_frame_errors); | |
3170 | ops->rx_over_err = htonll(stats.rx_over_errors); | |
3171 | ops->rx_crc_err = htonll(stats.rx_crc_errors); | |
3172 | ops->collisions = htonll(stats.collisions); | |
3173 | } | |
3174 | ||
064af421 BP |
3175 | static int |
3176 | handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn, | |
abaad8cf JP |
3177 | struct ofp_stats_request *osr, |
3178 | size_t arg_size) | |
064af421 | 3179 | { |
abaad8cf | 3180 | struct ofp_port_stats_request *psr; |
064af421 BP |
3181 | struct ofp_port_stats *ops; |
3182 | struct ofpbuf *msg; | |
3183 | struct ofport *port; | |
064af421 | 3184 | |
abaad8cf JP |
3185 | if (arg_size != sizeof *psr) { |
3186 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
3187 | } | |
3188 | psr = (struct ofp_port_stats_request *) osr->body; | |
3189 | ||
3190 | msg = start_stats_reply(osr, sizeof *ops * 16); | |
3191 | if (psr->port_no != htons(OFPP_NONE)) { | |
ca0f572c | 3192 | port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no))); |
abaad8cf | 3193 | if (port) { |
ca0f572c | 3194 | append_port_stat(port, ofconn, &msg); |
abaad8cf JP |
3195 | } |
3196 | } else { | |
4e8e4213 | 3197 | HMAP_FOR_EACH (port, hmap_node, &p->ports) { |
ca0f572c | 3198 | append_port_stat(port, ofconn, &msg); |
abaad8cf | 3199 | } |
064af421 BP |
3200 | } |
3201 | ||
3202 | queue_tx(msg, ofconn, ofconn->reply_counter); | |
3203 | return 0; | |
3204 | } | |
3205 | ||
3206 | struct flow_stats_cbdata { | |
3207 | struct ofproto *ofproto; | |
3208 | struct ofconn *ofconn; | |
3209 | uint16_t out_port; | |
3210 | struct ofpbuf *msg; | |
3211 | }; | |
3212 | ||
01149cfd BP |
3213 | /* Obtains statistic counters for 'rule' within 'p' and stores them into |
3214 | * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the | |
3215 | * returned statistic include statistics for all of 'rule''s subrules. */ | |
064af421 BP |
3216 | static void |
3217 | query_stats(struct ofproto *p, struct rule *rule, | |
3218 | uint64_t *packet_countp, uint64_t *byte_countp) | |
3219 | { | |
3220 | uint64_t packet_count, byte_count; | |
3221 | struct rule *subrule; | |
3222 | struct odp_flow *odp_flows; | |
3223 | size_t n_odp_flows; | |
3224 | ||
01149cfd BP |
3225 | /* Start from historical data for 'rule' itself that are no longer tracked |
3226 | * by the datapath. This counts, for example, subrules that have | |
3227 | * expired. */ | |
b3137fe8 JG |
3228 | packet_count = rule->packet_count; |
3229 | byte_count = rule->byte_count; | |
3230 | ||
01149cfd BP |
3231 | /* Prepare to ask the datapath for statistics on 'rule', or if it is |
3232 | * wildcarded then on all of its subrules. | |
3233 | * | |
3234 | * Also, add any statistics that are not tracked by the datapath for each | |
3235 | * subrule. This includes, for example, statistics for packets that were | |
3236 | * executed "by hand" by ofproto via dpif_execute() but must be accounted | |
3237 | * to a flow. */ | |
064af421 | 3238 | n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1; |
ec6fde61 | 3239 | odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows); |
064af421 BP |
3240 | if (rule->cr.wc.wildcards) { |
3241 | size_t i = 0; | |
4e8e4213 | 3242 | LIST_FOR_EACH (subrule, list, &rule->list) { |
064af421 | 3243 | odp_flows[i++].key = subrule->cr.flow; |
b3137fe8 JG |
3244 | packet_count += subrule->packet_count; |
3245 | byte_count += subrule->byte_count; | |
064af421 BP |
3246 | } |
3247 | } else { | |
3248 | odp_flows[0].key = rule->cr.flow; | |
3249 | } | |
3250 | ||
28998b22 | 3251 | /* Fetch up-to-date statistics from the datapath and add them in. */ |
c228a364 | 3252 | if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) { |
064af421 BP |
3253 | size_t i; |
3254 | for (i = 0; i < n_odp_flows; i++) { | |
3255 | struct odp_flow *odp_flow = &odp_flows[i]; | |
3256 | packet_count += odp_flow->stats.n_packets; | |
3257 | byte_count += odp_flow->stats.n_bytes; | |
3258 | } | |
3259 | } | |
3260 | free(odp_flows); | |
3261 | ||
01149cfd | 3262 | /* Return the stats to the caller. */ |
064af421 BP |
3263 | *packet_countp = packet_count; |
3264 | *byte_countp = byte_count; | |
3265 | } | |
3266 | ||
3267 | static void | |
3268 | flow_stats_cb(struct cls_rule *rule_, void *cbdata_) | |
3269 | { | |
3270 | struct rule *rule = rule_from_cls_rule(rule_); | |
3271 | struct flow_stats_cbdata *cbdata = cbdata_; | |
3272 | struct ofp_flow_stats *ofs; | |
3273 | uint64_t packet_count, byte_count; | |
3274 | size_t act_len, len; | |
26c3f94a JP |
3275 | long long int tdiff = time_msec() - rule->created; |
3276 | uint32_t sec = tdiff / 1000; | |
3277 | uint32_t msec = tdiff - (sec * 1000); | |
064af421 BP |
3278 | |
3279 | if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) { | |
3280 | return; | |
3281 | } | |
3282 | ||
3283 | act_len = sizeof *rule->actions * rule->n_actions; | |
3284 | len = offsetof(struct ofp_flow_stats, actions) + act_len; | |
3285 | ||
3286 | query_stats(cbdata->ofproto, rule, &packet_count, &byte_count); | |
3287 | ||
3288 | ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg); | |
3289 | ofs->length = htons(len); | |
3290 | ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH; | |
3291 | ofs->pad = 0; | |
659586ef JG |
3292 | flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, |
3293 | cbdata->ofproto->tun_id_from_cookie, &ofs->match); | |
26c3f94a JP |
3294 | ofs->duration_sec = htonl(sec); |
3295 | ofs->duration_nsec = htonl(msec * 1000000); | |
39997502 | 3296 | ofs->cookie = rule->flow_cookie; |
064af421 BP |
3297 | ofs->priority = htons(rule->cr.priority); |
3298 | ofs->idle_timeout = htons(rule->idle_timeout); | |
3299 | ofs->hard_timeout = htons(rule->hard_timeout); | |
39997502 | 3300 | memset(ofs->pad2, 0, sizeof ofs->pad2); |
064af421 BP |
3301 | ofs->packet_count = htonll(packet_count); |
3302 | ofs->byte_count = htonll(byte_count); | |
3dffcf07 BP |
3303 | if (rule->n_actions > 0) { |
3304 | memcpy(ofs->actions, rule->actions, act_len); | |
3305 | } | |
064af421 BP |
3306 | } |
3307 | ||
3308 | static int | |
3309 | table_id_to_include(uint8_t table_id) | |
3310 | { | |
3311 | return (table_id == TABLEID_HASH ? CLS_INC_EXACT | |
3312 | : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD | |
3313 | : table_id == 0xff ? CLS_INC_ALL | |
3314 | : 0); | |
3315 | } | |
3316 | ||
3317 | static int | |
3318 | handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn, | |
3319 | const struct ofp_stats_request *osr, | |
3320 | size_t arg_size) | |
3321 | { | |
3322 | struct ofp_flow_stats_request *fsr; | |
3323 | struct flow_stats_cbdata cbdata; | |
3324 | struct cls_rule target; | |
3325 | ||
3326 | if (arg_size != sizeof *fsr) { | |
49bdc010 | 3327 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); |
064af421 BP |
3328 | } |
3329 | fsr = (struct ofp_flow_stats_request *) osr->body; | |
3330 | ||
3331 | COVERAGE_INC(ofproto_flows_req); | |
3332 | cbdata.ofproto = p; | |
3333 | cbdata.ofconn = ofconn; | |
3334 | cbdata.out_port = fsr->out_port; | |
3335 | cbdata.msg = start_stats_reply(osr, 1024); | |
659586ef | 3336 | cls_rule_from_match(&fsr->match, 0, false, 0, &target); |
064af421 BP |
3337 | classifier_for_each_match(&p->cls, &target, |
3338 | table_id_to_include(fsr->table_id), | |
3339 | flow_stats_cb, &cbdata); | |
3340 | queue_tx(cbdata.msg, ofconn, ofconn->reply_counter); | |
3341 | return 0; | |
3342 | } | |
3343 | ||
4f2cad2c JP |
3344 | struct flow_stats_ds_cbdata { |
3345 | struct ofproto *ofproto; | |
3346 | struct ds *results; | |
3347 | }; | |
3348 | ||
3349 | static void | |
3350 | flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_) | |
3351 | { | |
3352 | struct rule *rule = rule_from_cls_rule(rule_); | |
3353 | struct flow_stats_ds_cbdata *cbdata = cbdata_; | |
3354 | struct ds *results = cbdata->results; | |
3355 | struct ofp_match match; | |
3356 | uint64_t packet_count, byte_count; | |
3357 | size_t act_len = sizeof *rule->actions * rule->n_actions; | |
3358 | ||
3359 | /* Don't report on subrules. */ | |
3360 | if (rule->super != NULL) { | |
3361 | return; | |
3362 | } | |
3363 | ||
3364 | query_stats(cbdata->ofproto, rule, &packet_count, &byte_count); | |
659586ef JG |
3365 | flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, |
3366 | cbdata->ofproto->tun_id_from_cookie, &match); | |
4f2cad2c JP |
3367 | |
3368 | ds_put_format(results, "duration=%llds, ", | |
3369 | (time_msec() - rule->created) / 1000); | |
52ae00b3 | 3370 | ds_put_format(results, "priority=%u, ", rule->cr.priority); |
4f2cad2c JP |
3371 | ds_put_format(results, "n_packets=%"PRIu64", ", packet_count); |
3372 | ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count); | |
3373 | ofp_print_match(results, &match, true); | |
3dffcf07 BP |
3374 | if (act_len > 0) { |
3375 | ofp_print_actions(results, &rule->actions->header, act_len); | |
3376 | } | |
4f2cad2c JP |
3377 | ds_put_cstr(results, "\n"); |
3378 | } | |
3379 | ||
d295e8e9 | 3380 | /* Adds a pretty-printed description of all flows to 'results', including |
4f2cad2c JP |
3381 | * those marked hidden by secchan (e.g., by in-band control). */ |
3382 | void | |
3383 | ofproto_get_all_flows(struct ofproto *p, struct ds *results) | |
3384 | { | |
3385 | struct ofp_match match; | |
3386 | struct cls_rule target; | |
3387 | struct flow_stats_ds_cbdata cbdata; | |
3388 | ||
3389 | memset(&match, 0, sizeof match); | |
659586ef | 3390 | match.wildcards = htonl(OVSFW_ALL); |
4f2cad2c JP |
3391 | |
3392 | cbdata.ofproto = p; | |
3393 | cbdata.results = results; | |
3394 | ||
659586ef | 3395 | cls_rule_from_match(&match, 0, false, 0, &target); |
4f2cad2c JP |
3396 | classifier_for_each_match(&p->cls, &target, CLS_INC_ALL, |
3397 | flow_stats_ds_cb, &cbdata); | |
3398 | } | |
3399 | ||
064af421 BP |
3400 | struct aggregate_stats_cbdata { |
3401 | struct ofproto *ofproto; | |
3402 | uint16_t out_port; | |
3403 | uint64_t packet_count; | |
3404 | uint64_t byte_count; | |
3405 | uint32_t n_flows; | |
3406 | }; | |
3407 | ||
3408 | static void | |
3409 | aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_) | |
3410 | { | |
3411 | struct rule *rule = rule_from_cls_rule(rule_); | |
3412 | struct aggregate_stats_cbdata *cbdata = cbdata_; | |
3413 | uint64_t packet_count, byte_count; | |
3414 | ||
3415 | if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) { | |
3416 | return; | |
3417 | } | |
3418 | ||
3419 | query_stats(cbdata->ofproto, rule, &packet_count, &byte_count); | |
3420 | ||
3421 | cbdata->packet_count += packet_count; | |
3422 | cbdata->byte_count += byte_count; | |
3423 | cbdata->n_flows++; | |
3424 | } | |
3425 | ||
3426 | static int | |
3427 | handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn, | |
3428 | const struct ofp_stats_request *osr, | |
3429 | size_t arg_size) | |
3430 | { | |
3431 | struct ofp_aggregate_stats_request *asr; | |
3432 | struct ofp_aggregate_stats_reply *reply; | |
3433 | struct aggregate_stats_cbdata cbdata; | |
3434 | struct cls_rule target; | |
3435 | struct ofpbuf *msg; | |
3436 | ||
3437 | if (arg_size != sizeof *asr) { | |
49bdc010 | 3438 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); |
064af421 BP |
3439 | } |
3440 | asr = (struct ofp_aggregate_stats_request *) osr->body; | |
3441 | ||
3442 | COVERAGE_INC(ofproto_agg_request); | |
3443 | cbdata.ofproto = p; | |
3444 | cbdata.out_port = asr->out_port; | |
3445 | cbdata.packet_count = 0; | |
3446 | cbdata.byte_count = 0; | |
3447 | cbdata.n_flows = 0; | |
659586ef | 3448 | cls_rule_from_match(&asr->match, 0, false, 0, &target); |
064af421 BP |
3449 | classifier_for_each_match(&p->cls, &target, |
3450 | table_id_to_include(asr->table_id), | |
3451 | aggregate_stats_cb, &cbdata); | |
3452 | ||
3453 | msg = start_stats_reply(osr, sizeof *reply); | |
3454 | reply = append_stats_reply(sizeof *reply, ofconn, &msg); | |
3455 | reply->flow_count = htonl(cbdata.n_flows); | |
3456 | reply->packet_count = htonll(cbdata.packet_count); | |
3457 | reply->byte_count = htonll(cbdata.byte_count); | |
3458 | queue_tx(msg, ofconn, ofconn->reply_counter); | |
3459 | return 0; | |
3460 | } | |
3461 | ||
c1c9c9c4 BP |
3462 | struct queue_stats_cbdata { |
3463 | struct ofconn *ofconn; | |
ca0f572c | 3464 | struct ofport *ofport; |
c1c9c9c4 | 3465 | struct ofpbuf *msg; |
c1c9c9c4 BP |
3466 | }; |
3467 | ||
3468 | static void | |
db9220c3 | 3469 | put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id, |
c1c9c9c4 BP |
3470 | const struct netdev_queue_stats *stats) |
3471 | { | |
3472 | struct ofp_queue_stats *reply; | |
3473 | ||
3474 | reply = append_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg); | |
ca0f572c | 3475 | reply->port_no = htons(cbdata->ofport->opp.port_no); |
c1c9c9c4 BP |
3476 | memset(reply->pad, 0, sizeof reply->pad); |
3477 | reply->queue_id = htonl(queue_id); | |
3478 | reply->tx_bytes = htonll(stats->tx_bytes); | |
3479 | reply->tx_packets = htonll(stats->tx_packets); | |
3480 | reply->tx_errors = htonll(stats->tx_errors); | |
3481 | } | |
3482 | ||
3483 | static void | |
db9220c3 | 3484 | handle_queue_stats_dump_cb(uint32_t queue_id, |
c1c9c9c4 BP |
3485 | struct netdev_queue_stats *stats, |
3486 | void *cbdata_) | |
3487 | { | |
3488 | struct queue_stats_cbdata *cbdata = cbdata_; | |
3489 | ||
3490 | put_queue_stats(cbdata, queue_id, stats); | |
3491 | } | |
3492 | ||
3493 | static void | |
ca0f572c | 3494 | handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id, |
c1c9c9c4 BP |
3495 | struct queue_stats_cbdata *cbdata) |
3496 | { | |
ca0f572c | 3497 | cbdata->ofport = port; |
c1c9c9c4 BP |
3498 | if (queue_id == OFPQ_ALL) { |
3499 | netdev_dump_queue_stats(port->netdev, | |
3500 | handle_queue_stats_dump_cb, cbdata); | |
3501 | } else { | |
3502 | struct netdev_queue_stats stats; | |
3503 | ||
1ac788f6 BP |
3504 | if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) { |
3505 | put_queue_stats(cbdata, queue_id, &stats); | |
3506 | } | |
c1c9c9c4 BP |
3507 | } |
3508 | } | |
3509 | ||
3510 | static int | |
3511 | handle_queue_stats_request(struct ofproto *ofproto, struct ofconn *ofconn, | |
3512 | const struct ofp_stats_request *osr, | |
3513 | size_t arg_size) | |
3514 | { | |
3515 | struct ofp_queue_stats_request *qsr; | |
3516 | struct queue_stats_cbdata cbdata; | |
3517 | struct ofport *port; | |
3518 | unsigned int port_no; | |
3519 | uint32_t queue_id; | |
3520 | ||
3521 | if (arg_size != sizeof *qsr) { | |
3522 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
3523 | } | |
3524 | qsr = (struct ofp_queue_stats_request *) osr->body; | |
3525 | ||
3526 | COVERAGE_INC(ofproto_queue_req); | |
3527 | ||
3528 | cbdata.ofconn = ofconn; | |
3529 | cbdata.msg = start_stats_reply(osr, 128); | |
3530 | ||
3531 | port_no = ntohs(qsr->port_no); | |
3532 | queue_id = ntohl(qsr->queue_id); | |
3533 | if (port_no == OFPP_ALL) { | |
4e8e4213 | 3534 | HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) { |
ca0f572c | 3535 | handle_queue_stats_for_port(port, queue_id, &cbdata); |
c1c9c9c4 BP |
3536 | } |
3537 | } else if (port_no < ofproto->max_ports) { | |
ca0f572c | 3538 | port = get_port(ofproto, ofp_port_to_odp_port(port_no)); |
c1c9c9c4 | 3539 | if (port) { |
ca0f572c | 3540 | handle_queue_stats_for_port(port, queue_id, &cbdata); |
c1c9c9c4 BP |
3541 | } |
3542 | } else { | |
3543 | ofpbuf_delete(cbdata.msg); | |
3544 | return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT); | |
3545 | } | |
3546 | queue_tx(cbdata.msg, ofconn, ofconn->reply_counter); | |
3547 | ||
3548 | return 0; | |
3549 | } | |
3550 | ||
064af421 BP |
3551 | static int |
3552 | handle_stats_request(struct ofproto *p, struct ofconn *ofconn, | |
3553 | struct ofp_header *oh) | |
3554 | { | |
3555 | struct ofp_stats_request *osr; | |
3556 | size_t arg_size; | |
3557 | int error; | |
3558 | ||
3559 | error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr, | |
3560 | 1, &arg_size); | |
3561 | if (error) { | |
3562 | return error; | |
3563 | } | |
3564 | osr = (struct ofp_stats_request *) oh; | |
3565 | ||
3566 | switch (ntohs(osr->type)) { | |
3567 | case OFPST_DESC: | |
3568 | return handle_desc_stats_request(p, ofconn, osr); | |
3569 | ||
3570 | case OFPST_FLOW: | |
3571 | return handle_flow_stats_request(p, ofconn, osr, arg_size); | |
3572 | ||
3573 | case OFPST_AGGREGATE: | |
3574 | return handle_aggregate_stats_request(p, ofconn, osr, arg_size); | |
3575 | ||
3576 | case OFPST_TABLE: | |
3577 | return handle_table_stats_request(p, ofconn, osr); | |
3578 | ||
3579 | case OFPST_PORT: | |
abaad8cf | 3580 | return handle_port_stats_request(p, ofconn, osr, arg_size); |
064af421 | 3581 | |
c1c9c9c4 BP |
3582 | case OFPST_QUEUE: |
3583 | return handle_queue_stats_request(p, ofconn, osr, arg_size); | |
3584 | ||
064af421 BP |
3585 | case OFPST_VENDOR: |
3586 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR); | |
3587 | ||
3588 | default: | |
3589 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT); | |
3590 | } | |
3591 | } | |
3592 | ||
3593 | static long long int | |
3594 | msec_from_nsec(uint64_t sec, uint32_t nsec) | |
3595 | { | |
3596 | return !sec ? 0 : sec * 1000 + nsec / 1000000; | |
3597 | } | |
3598 | ||
3599 | static void | |
0193b2af JG |
3600 | update_time(struct ofproto *ofproto, struct rule *rule, |
3601 | const struct odp_flow_stats *stats) | |
064af421 BP |
3602 | { |
3603 | long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec); | |
3604 | if (used > rule->used) { | |
3605 | rule->used = used; | |
4836f9f2 JP |
3606 | if (rule->super && used > rule->super->used) { |
3607 | rule->super->used = used; | |
3608 | } | |
0193b2af | 3609 | netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used); |
064af421 BP |
3610 | } |
3611 | } | |
3612 | ||
3613 | static void | |
0193b2af JG |
3614 | update_stats(struct ofproto *ofproto, struct rule *rule, |
3615 | const struct odp_flow_stats *stats) | |
064af421 | 3616 | { |
064af421 | 3617 | if (stats->n_packets) { |
0193b2af JG |
3618 | update_time(ofproto, rule, stats); |
3619 | rule->packet_count += stats->n_packets; | |
3620 | rule->byte_count += stats->n_bytes; | |
abfec865 | 3621 | netflow_flow_update_flags(&rule->nf_flow, stats->tcp_flags); |
064af421 BP |
3622 | } |
3623 | } | |
3624 | ||
79eee1eb BP |
3625 | /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT |
3626 | * in which no matching flow already exists in the flow table. | |
3627 | * | |
3628 | * Adds the flow specified by 'ofm', which is followed by 'n_actions' | |
3629 | * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error | |
3630 | * code as encoded by ofp_mkerr() on failure. | |
3631 | * | |
3632 | * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, | |
3633 | * if any. */ | |
064af421 BP |
3634 | static int |
3635 | add_flow(struct ofproto *p, struct ofconn *ofconn, | |
79eee1eb | 3636 | const struct ofp_flow_mod *ofm, size_t n_actions) |
064af421 BP |
3637 | { |
3638 | struct ofpbuf *packet; | |
3639 | struct rule *rule; | |
3640 | uint16_t in_port; | |
3641 | int error; | |
3642 | ||
49bdc010 JP |
3643 | if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) { |
3644 | flow_t flow; | |
3645 | uint32_t wildcards; | |
3646 | ||
659586ef JG |
3647 | flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie, |
3648 | &flow, &wildcards); | |
49bdc010 JP |
3649 | if (classifier_rule_overlaps(&p->cls, &flow, wildcards, |
3650 | ntohs(ofm->priority))) { | |
3651 | return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP); | |
3652 | } | |
3653 | } | |
3654 | ||
0193b2af | 3655 | rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions, |
064af421 | 3656 | n_actions, ntohs(ofm->idle_timeout), |
39997502 | 3657 | ntohs(ofm->hard_timeout), ofm->cookie, |
ca069229 | 3658 | ofm->flags & htons(OFPFF_SEND_FLOW_REM)); |
659586ef JG |
3659 | cls_rule_from_match(&ofm->match, ntohs(ofm->priority), |
3660 | p->tun_id_from_cookie, ofm->cookie, &rule->cr); | |
064af421 | 3661 | |
064af421 BP |
3662 | error = 0; |
3663 | if (ofm->buffer_id != htonl(UINT32_MAX)) { | |
3664 | error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id), | |
3665 | &packet, &in_port); | |
212fe71c BP |
3666 | } else { |
3667 | packet = NULL; | |
165cd8a3 | 3668 | in_port = UINT16_MAX; |
064af421 BP |
3669 | } |
3670 | ||
3671 | rule_insert(p, rule, packet, in_port); | |
064af421 BP |
3672 | return error; |
3673 | } | |
3674 | ||
79eee1eb BP |
3675 | static struct rule * |
3676 | find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm) | |
064af421 | 3677 | { |
064af421 BP |
3678 | uint32_t wildcards; |
3679 | flow_t flow; | |
3680 | ||
659586ef JG |
3681 | flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie, |
3682 | &flow, &wildcards); | |
79eee1eb | 3683 | return rule_from_cls_rule(classifier_find_rule_exactly( |
064af421 BP |
3684 | &p->cls, &flow, wildcards, |
3685 | ntohs(ofm->priority))); | |
79eee1eb | 3686 | } |
064af421 | 3687 | |
79eee1eb BP |
3688 | static int |
3689 | send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn, | |
3690 | struct rule *rule, const struct ofp_flow_mod *ofm) | |
3691 | { | |
3692 | struct ofpbuf *packet; | |
3693 | uint16_t in_port; | |
3694 | flow_t flow; | |
3695 | int error; | |
064af421 | 3696 | |
79eee1eb BP |
3697 | if (ofm->buffer_id == htonl(UINT32_MAX)) { |
3698 | return 0; | |
064af421 | 3699 | } |
79eee1eb BP |
3700 | |
3701 | error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id), | |
3702 | &packet, &in_port); | |
3703 | if (error) { | |
3704 | return error; | |
3705 | } | |
3706 | ||
659586ef | 3707 | flow_extract(packet, 0, in_port, &flow); |
79eee1eb | 3708 | rule_execute(ofproto, rule, packet, &flow); |
79eee1eb | 3709 | |
064af421 BP |
3710 | return 0; |
3711 | } | |
79eee1eb BP |
3712 | \f |
3713 | /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */ | |
064af421 BP |
3714 | |
3715 | struct modify_flows_cbdata { | |
3716 | struct ofproto *ofproto; | |
3717 | const struct ofp_flow_mod *ofm; | |
064af421 | 3718 | size_t n_actions; |
79eee1eb | 3719 | struct rule *match; |
064af421 BP |
3720 | }; |
3721 | ||
79eee1eb BP |
3722 | static int modify_flow(struct ofproto *, const struct ofp_flow_mod *, |
3723 | size_t n_actions, struct rule *); | |
3724 | static void modify_flows_cb(struct cls_rule *, void *cbdata_); | |
3725 | ||
3726 | /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as | |
3727 | * encoded by ofp_mkerr() on failure. | |
3728 | * | |
3729 | * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, | |
3730 | * if any. */ | |
3731 | static int | |
3732 | modify_flows_loose(struct ofproto *p, struct ofconn *ofconn, | |
3733 | const struct ofp_flow_mod *ofm, size_t n_actions) | |
3734 | { | |
3735 | struct modify_flows_cbdata cbdata; | |
3736 | struct cls_rule target; | |
3737 | ||
3738 | cbdata.ofproto = p; | |
3739 | cbdata.ofm = ofm; | |
3740 | cbdata.n_actions = n_actions; | |
3741 | cbdata.match = NULL; | |
3742 | ||
659586ef JG |
3743 | cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie, |
3744 | &target); | |
79eee1eb BP |
3745 | |
3746 | classifier_for_each_match(&p->cls, &target, CLS_INC_ALL, | |
3747 | modify_flows_cb, &cbdata); | |
3748 | if (cbdata.match) { | |
3749 | /* This credits the packet to whichever flow happened to happened to | |
3750 | * match last. That's weird. Maybe we should do a lookup for the | |
3751 | * flow that actually matches the packet? Who knows. */ | |
3752 | send_buffered_packet(p, ofconn, cbdata.match, ofm); | |
3753 | return 0; | |
3754 | } else { | |
3755 | return add_flow(p, ofconn, ofm, n_actions); | |
3756 | } | |
3757 | } | |
3758 | ||
3759 | /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error | |
3760 | * code as encoded by ofp_mkerr() on failure. | |
3761 | * | |
3762 | * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, | |
3763 | * if any. */ | |
3764 | static int | |
3765 | modify_flow_strict(struct ofproto *p, struct ofconn *ofconn, | |
3766 | struct ofp_flow_mod *ofm, size_t n_actions) | |
3767 | { | |
3768 | struct rule *rule = find_flow_strict(p, ofm); | |
3769 | if (rule && !rule_is_hidden(rule)) { | |
3770 | modify_flow(p, ofm, n_actions, rule); | |
3771 | return send_buffered_packet(p, ofconn, rule, ofm); | |
3772 | } else { | |
3773 | return add_flow(p, ofconn, ofm, n_actions); | |
3774 | } | |
3775 | } | |
3776 | ||
3777 | /* Callback for modify_flows_loose(). */ | |
064af421 BP |
3778 | static void |
3779 | modify_flows_cb(struct cls_rule *rule_, void *cbdata_) | |
3780 | { | |
3781 | struct rule *rule = rule_from_cls_rule(rule_); | |
3782 | struct modify_flows_cbdata *cbdata = cbdata_; | |
3783 | ||
79eee1eb BP |
3784 | if (!rule_is_hidden(rule)) { |
3785 | cbdata->match = rule; | |
3786 | modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule); | |
064af421 | 3787 | } |
064af421 BP |
3788 | } |
3789 | ||
79eee1eb BP |
3790 | /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has |
3791 | * been identified as a flow in 'p''s flow table to be modified, by changing | |
3792 | * the rule's actions to match those in 'ofm' (which is followed by 'n_actions' | |
3793 | * ofp_action[] structures). */ | |
064af421 | 3794 | static int |
79eee1eb BP |
3795 | modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm, |
3796 | size_t n_actions, struct rule *rule) | |
064af421 | 3797 | { |
79eee1eb BP |
3798 | size_t actions_len = n_actions * sizeof *rule->actions; |
3799 | ||
3800 | rule->flow_cookie = ofm->cookie; | |
3801 | ||
3802 | /* If the actions are the same, do nothing. */ | |
3803 | if (n_actions == rule->n_actions | |
3dffcf07 | 3804 | && (!n_actions || !memcmp(ofm->actions, rule->actions, actions_len))) |
79eee1eb BP |
3805 | { |
3806 | return 0; | |
3807 | } | |
3808 | ||
3809 | /* Replace actions. */ | |
3810 | free(rule->actions); | |
3dffcf07 | 3811 | rule->actions = n_actions ? xmemdup(ofm->actions, actions_len) : NULL; |
79eee1eb BP |
3812 | rule->n_actions = n_actions; |
3813 | ||
3814 | /* Make sure that the datapath gets updated properly. */ | |
3815 | if (rule->cr.wc.wildcards) { | |
3816 | COVERAGE_INC(ofproto_mod_wc_flow); | |
3817 | p->need_revalidate = true; | |
3818 | } else { | |
3819 | rule_update_actions(p, rule); | |
3820 | } | |
3821 | ||
3822 | return 0; | |
3823 | } | |
3824 | \f | |
3825 | /* OFPFC_DELETE implementation. */ | |
3826 | ||
3827 | struct delete_flows_cbdata { | |
3828 | struct ofproto *ofproto; | |
3829 | uint16_t out_port; | |
3830 | }; | |
3831 | ||
3832 | static void delete_flows_cb(struct cls_rule *, void *cbdata_); | |
3833 | static void delete_flow(struct ofproto *, struct rule *, uint16_t out_port); | |
3834 | ||
3835 | /* Implements OFPFC_DELETE. */ | |
3836 | static void | |
3837 | delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm) | |
3838 | { | |
3839 | struct delete_flows_cbdata cbdata; | |
064af421 BP |
3840 | struct cls_rule target; |
3841 | ||
3842 | cbdata.ofproto = p; | |
79eee1eb | 3843 | cbdata.out_port = ofm->out_port; |
064af421 | 3844 | |
659586ef JG |
3845 | cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie, |
3846 | &target); | |
064af421 BP |
3847 | |
3848 | classifier_for_each_match(&p->cls, &target, CLS_INC_ALL, | |
79eee1eb | 3849 | delete_flows_cb, &cbdata); |
064af421 BP |
3850 | } |
3851 | ||
79eee1eb BP |
3852 | /* Implements OFPFC_DELETE_STRICT. */ |
3853 | static void | |
3854 | delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm) | |
3855 | { | |
3856 | struct rule *rule = find_flow_strict(p, ofm); | |
3857 | if (rule) { | |
3858 | delete_flow(p, rule, ofm->out_port); | |
3859 | } | |
3860 | } | |
3861 | ||
3862 | /* Callback for delete_flows_loose(). */ | |
3863 | static void | |
3864 | delete_flows_cb(struct cls_rule *rule_, void *cbdata_) | |
3865 | { | |
3866 | struct rule *rule = rule_from_cls_rule(rule_); | |
3867 | struct delete_flows_cbdata *cbdata = cbdata_; | |
3868 | ||
3869 | delete_flow(cbdata->ofproto, rule, cbdata->out_port); | |
3870 | } | |
3871 | ||
3872 | /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has | |
3873 | * been identified as a flow to delete from 'p''s flow table, by deleting the | |
3874 | * flow and sending out a OFPT_FLOW_REMOVED message to any interested | |
3875 | * controller. | |
3876 | * | |
3877 | * Will not delete 'rule' if it is hidden. Will delete 'rule' only if | |
3878 | * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the | |
3879 | * specified 'out_port'. */ | |
3880 | static void | |
3881 | delete_flow(struct ofproto *p, struct rule *rule, uint16_t out_port) | |
3882 | { | |
3883 | if (rule_is_hidden(rule)) { | |
3884 | return; | |
3885 | } | |
3886 | ||
3887 | if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) { | |
3888 | return; | |
3889 | } | |
3890 | ||
3891 | send_flow_removed(p, rule, time_msec(), OFPRR_DELETE); | |
3892 | rule_remove(p, rule); | |
3893 | } | |
3894 | \f | |
064af421 BP |
3895 | static int |
3896 | handle_flow_mod(struct ofproto *p, struct ofconn *ofconn, | |
3897 | struct ofp_flow_mod *ofm) | |
3898 | { | |
3f09c339 | 3899 | struct ofp_match orig_match; |
064af421 BP |
3900 | size_t n_actions; |
3901 | int error; | |
3902 | ||
9deba63b BP |
3903 | error = reject_slave_controller(ofconn, &ofm->header); |
3904 | if (error) { | |
3905 | return error; | |
3906 | } | |
064af421 BP |
3907 | error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm, |
3908 | sizeof *ofm->actions, &n_actions); | |
3909 | if (error) { | |
3910 | return error; | |
3911 | } | |
3912 | ||
49bdc010 JP |
3913 | /* We do not support the emergency flow cache. It will hopefully |
3914 | * get dropped from OpenFlow in the near future. */ | |
3915 | if (ofm->flags & htons(OFPFF_EMERG)) { | |
3916 | /* There isn't a good fit for an error code, so just state that the | |
3917 | * flow table is full. */ | |
3918 | return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL); | |
3919 | } | |
3920 | ||
3f09c339 BP |
3921 | /* Normalize ofp->match. If normalization actually changes anything, then |
3922 | * log the differences. */ | |
3923 | ofm->match.pad1[0] = ofm->match.pad2[0] = 0; | |
3924 | orig_match = ofm->match; | |
064af421 | 3925 | normalize_match(&ofm->match); |
3f09c339 BP |
3926 | if (memcmp(&ofm->match, &orig_match, sizeof orig_match)) { |
3927 | static struct vlog_rate_limit normal_rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
3928 | if (!VLOG_DROP_INFO(&normal_rl)) { | |
3929 | char *old = ofp_match_to_literal_string(&orig_match); | |
3930 | char *new = ofp_match_to_literal_string(&ofm->match); | |
3931 | VLOG_INFO("%s: normalization changed ofp_match, details:", | |
3932 | rconn_get_name(ofconn->rconn)); | |
3933 | VLOG_INFO(" pre: %s", old); | |
3934 | VLOG_INFO("post: %s", new); | |
3935 | free(old); | |
3936 | free(new); | |
3937 | } | |
3938 | } | |
3939 | ||
064af421 BP |
3940 | if (!ofm->match.wildcards) { |
3941 | ofm->priority = htons(UINT16_MAX); | |
3942 | } | |
3943 | ||
3944 | error = validate_actions((const union ofp_action *) ofm->actions, | |
3945 | n_actions, p->max_ports); | |
3946 | if (error) { | |
3947 | return error; | |
3948 | } | |
3949 | ||
3950 | switch (ntohs(ofm->command)) { | |
3951 | case OFPFC_ADD: | |
3952 | return add_flow(p, ofconn, ofm, n_actions); | |
3953 | ||
3954 | case OFPFC_MODIFY: | |
79eee1eb | 3955 | return modify_flows_loose(p, ofconn, ofm, n_actions); |
064af421 BP |
3956 | |
3957 | case OFPFC_MODIFY_STRICT: | |
79eee1eb | 3958 | return modify_flow_strict(p, ofconn, ofm, n_actions); |
064af421 BP |
3959 | |
3960 | case OFPFC_DELETE: | |
79eee1eb BP |
3961 | delete_flows_loose(p, ofm); |
3962 | return 0; | |
064af421 BP |
3963 | |
3964 | case OFPFC_DELETE_STRICT: | |
79eee1eb BP |
3965 | delete_flow_strict(p, ofm); |
3966 | return 0; | |
064af421 BP |
3967 | |
3968 | default: | |
3969 | return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND); | |
3970 | } | |
3971 | } | |
3972 | ||
659586ef JG |
3973 | static int |
3974 | handle_tun_id_from_cookie(struct ofproto *p, struct nxt_tun_id_cookie *msg) | |
3975 | { | |
3976 | int error; | |
3977 | ||
3978 | error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg); | |
3979 | if (error) { | |
3980 | return error; | |
3981 | } | |
3982 | ||
3983 | p->tun_id_from_cookie = !!msg->set; | |
3984 | return 0; | |
3985 | } | |
3986 | ||
9deba63b BP |
3987 | static int |
3988 | handle_role_request(struct ofproto *ofproto, | |
3989 | struct ofconn *ofconn, struct nicira_header *msg) | |
3990 | { | |
3991 | struct nx_role_request *nrr; | |
3992 | struct nx_role_request *reply; | |
3993 | struct ofpbuf *buf; | |
3994 | uint32_t role; | |
3995 | ||
3996 | if (ntohs(msg->header.length) != sizeof *nrr) { | |
100e95db | 3997 | VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)", |
9deba63b BP |
3998 | ntohs(msg->header.length), sizeof *nrr); |
3999 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
4000 | } | |
4001 | nrr = (struct nx_role_request *) msg; | |
4002 | ||
5899143f | 4003 | if (ofconn->type != OFCONN_PRIMARY) { |
9deba63b BP |
4004 | VLOG_WARN_RL(&rl, "ignoring role request on non-controller " |
4005 | "connection"); | |
4006 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM); | |
4007 | } | |
4008 | ||
4009 | role = ntohl(nrr->role); | |
4010 | if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER | |
4011 | && role != NX_ROLE_SLAVE) { | |
4012 | VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role); | |
4013 | ||
4014 | /* There's no good error code for this. */ | |
4015 | return ofp_mkerr(OFPET_BAD_REQUEST, -1); | |
4016 | } | |
4017 | ||
4018 | if (role == NX_ROLE_MASTER) { | |
4019 | struct ofconn *other; | |
4020 | ||
4e8e4213 | 4021 | HMAP_FOR_EACH (other, hmap_node, &ofproto->controllers) { |
9deba63b BP |
4022 | if (other->role == NX_ROLE_MASTER) { |
4023 | other->role = NX_ROLE_SLAVE; | |
4024 | } | |
4025 | } | |
4026 | } | |
4027 | ofconn->role = role; | |
4028 | ||
4029 | reply = make_openflow_xid(sizeof *reply, OFPT_VENDOR, msg->header.xid, | |
4030 | &buf); | |
4031 | reply->nxh.vendor = htonl(NX_VENDOR_ID); | |
4032 | reply->nxh.subtype = htonl(NXT_ROLE_REPLY); | |
4033 | reply->role = htonl(role); | |
4034 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
4035 | ||
4036 | return 0; | |
4037 | } | |
4038 | ||
064af421 BP |
4039 | static int |
4040 | handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg) | |
4041 | { | |
4042 | struct ofp_vendor_header *ovh = msg; | |
4043 | struct nicira_header *nh; | |
4044 | ||
4045 | if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) { | |
100e95db | 4046 | VLOG_WARN_RL(&rl, "received vendor message of length %u " |
659586ef JG |
4047 | "(expected at least %zu)", |
4048 | ntohs(ovh->header.length), sizeof(struct ofp_vendor_header)); | |
49bdc010 | 4049 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); |
064af421 BP |
4050 | } |
4051 | if (ovh->vendor != htonl(NX_VENDOR_ID)) { | |
4052 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR); | |
4053 | } | |
4054 | if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) { | |
100e95db | 4055 | VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u " |
659586ef JG |
4056 | "(expected at least %zu)", |
4057 | ntohs(ovh->header.length), sizeof(struct nicira_header)); | |
49bdc010 | 4058 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); |
064af421 BP |
4059 | } |
4060 | ||
4061 | nh = msg; | |
4062 | switch (ntohl(nh->subtype)) { | |
4063 | case NXT_STATUS_REQUEST: | |
4064 | return switch_status_handle_request(p->switch_status, ofconn->rconn, | |
4065 | msg); | |
659586ef JG |
4066 | |
4067 | case NXT_TUN_ID_FROM_COOKIE: | |
4068 | return handle_tun_id_from_cookie(p, msg); | |
9deba63b BP |
4069 | |
4070 | case NXT_ROLE_REQUEST: | |
4071 | return handle_role_request(p, ofconn, msg); | |
064af421 BP |
4072 | } |
4073 | ||
4074 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); | |
4075 | } | |
4076 | ||
246e61ea JP |
4077 | static int |
4078 | handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh) | |
4079 | { | |
4080 | struct ofp_header *ob; | |
4081 | struct ofpbuf *buf; | |
4082 | ||
4083 | /* Currently, everything executes synchronously, so we can just | |
4084 | * immediately send the barrier reply. */ | |
4085 | ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf); | |
4086 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
4087 | return 0; | |
4088 | } | |
4089 | ||
064af421 BP |
4090 | static void |
4091 | handle_openflow(struct ofconn *ofconn, struct ofproto *p, | |
4092 | struct ofpbuf *ofp_msg) | |
4093 | { | |
4094 | struct ofp_header *oh = ofp_msg->data; | |
4095 | int error; | |
4096 | ||
4097 | COVERAGE_INC(ofproto_recv_openflow); | |
4098 | switch (oh->type) { | |
4099 | case OFPT_ECHO_REQUEST: | |
4100 | error = handle_echo_request(ofconn, oh); | |
4101 | break; | |
4102 | ||
4103 | case OFPT_ECHO_REPLY: | |
4104 | error = 0; | |
4105 | break; | |
4106 | ||
4107 | case OFPT_FEATURES_REQUEST: | |
4108 | error = handle_features_request(p, ofconn, oh); | |
4109 | break; | |
4110 | ||
4111 | case OFPT_GET_CONFIG_REQUEST: | |
4112 | error = handle_get_config_request(p, ofconn, oh); | |
4113 | break; | |
4114 | ||
4115 | case OFPT_SET_CONFIG: | |
4116 | error = handle_set_config(p, ofconn, ofp_msg->data); | |
4117 | break; | |
4118 | ||
4119 | case OFPT_PACKET_OUT: | |
4120 | error = handle_packet_out(p, ofconn, ofp_msg->data); | |
4121 | break; | |
4122 | ||
4123 | case OFPT_PORT_MOD: | |
9deba63b | 4124 | error = handle_port_mod(p, ofconn, oh); |
064af421 BP |
4125 | break; |
4126 | ||
4127 | case OFPT_FLOW_MOD: | |
4128 | error = handle_flow_mod(p, ofconn, ofp_msg->data); | |
4129 | break; | |
4130 | ||
4131 | case OFPT_STATS_REQUEST: | |
4132 | error = handle_stats_request(p, ofconn, oh); | |
4133 | break; | |
4134 | ||
4135 | case OFPT_VENDOR: | |
4136 | error = handle_vendor(p, ofconn, ofp_msg->data); | |
4137 | break; | |
4138 | ||
246e61ea JP |
4139 | case OFPT_BARRIER_REQUEST: |
4140 | error = handle_barrier_request(ofconn, oh); | |
4141 | break; | |
4142 | ||
064af421 BP |
4143 | default: |
4144 | if (VLOG_IS_WARN_ENABLED()) { | |
4145 | char *s = ofp_to_string(oh, ntohs(oh->length), 2); | |
4146 | VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s); | |
4147 | free(s); | |
4148 | } | |
4149 | error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE); | |
4150 | break; | |
4151 | } | |
4152 | ||
4153 | if (error) { | |
4154 | send_error_oh(ofconn, ofp_msg->data, error); | |
4155 | } | |
4156 | } | |
4157 | \f | |
4158 | static void | |
72b06300 | 4159 | handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet) |
064af421 BP |
4160 | { |
4161 | struct odp_msg *msg = packet->data; | |
064af421 BP |
4162 | struct rule *rule; |
4163 | struct ofpbuf payload; | |
4164 | flow_t flow; | |
4165 | ||
064af421 BP |
4166 | payload.data = msg + 1; |
4167 | payload.size = msg->length - sizeof *msg; | |
659586ef | 4168 | flow_extract(&payload, msg->arg, msg->port, &flow); |
064af421 | 4169 | |
0ad9b732 JP |
4170 | /* Check with in-band control to see if this packet should be sent |
4171 | * to the local port regardless of the flow table. */ | |
4172 | if (in_band_msg_in_hook(p->in_band, &flow, &payload)) { | |
4173 | union odp_action action; | |
4174 | ||
4175 | memset(&action, 0, sizeof(action)); | |
4176 | action.output.type = ODPAT_OUTPUT; | |
4177 | action.output.port = ODPP_LOCAL; | |
f1acd62b | 4178 | dpif_execute(p->dpif, flow.in_port, &action, 1, &payload); |
0ad9b732 JP |
4179 | } |
4180 | ||
064af421 BP |
4181 | rule = lookup_valid_rule(p, &flow); |
4182 | if (!rule) { | |
4183 | /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */ | |
ca0f572c | 4184 | struct ofport *port = get_port(p, msg->port); |
064af421 BP |
4185 | if (port) { |
4186 | if (port->opp.config & OFPPC_NO_PACKET_IN) { | |
4187 | COVERAGE_INC(ofproto_no_packet_in); | |
4188 | /* XXX install 'drop' flow entry */ | |
4189 | ofpbuf_delete(packet); | |
4190 | return; | |
4191 | } | |
4192 | } else { | |
4193 | VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port); | |
4194 | } | |
4195 | ||
4196 | COVERAGE_INC(ofproto_packet_in); | |
76ce9432 | 4197 | send_packet_in(p, packet); |
064af421 BP |
4198 | return; |
4199 | } | |
4200 | ||
4201 | if (rule->cr.wc.wildcards) { | |
4202 | rule = rule_create_subrule(p, rule, &flow); | |
4203 | rule_make_actions(p, rule, packet); | |
4204 | } else { | |
4205 | if (!rule->may_install) { | |
4206 | /* The rule is not installable, that is, we need to process every | |
4207 | * packet, so process the current packet and set its actions into | |
4208 | * 'subrule'. */ | |
4209 | rule_make_actions(p, rule, packet); | |
4210 | } else { | |
4211 | /* XXX revalidate rule if it needs it */ | |
4212 | } | |
4213 | } | |
4214 | ||
76ce9432 | 4215 | if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY) { |
7778bd15 BP |
4216 | /* |
4217 | * Extra-special case for fail-open mode. | |
4218 | * | |
4219 | * We are in fail-open mode and the packet matched the fail-open rule, | |
4220 | * but we are connected to a controller too. We should send the packet | |
4221 | * up to the controller in the hope that it will try to set up a flow | |
4222 | * and thereby allow us to exit fail-open. | |
4223 | * | |
4224 | * See the top-level comment in fail-open.c for more information. | |
4225 | */ | |
c9b5816c BP |
4226 | send_packet_in(p, ofpbuf_clone_with_headroom(packet, |
4227 | DPIF_RECV_MSG_PADDING)); | |
7778bd15 | 4228 | } |
750638bb BP |
4229 | |
4230 | ofpbuf_pull(packet, sizeof *msg); | |
4231 | rule_execute(p, rule, packet, &flow); | |
4232 | rule_reinstall(p, rule); | |
064af421 | 4233 | } |
72b06300 BP |
4234 | |
4235 | static void | |
4236 | handle_odp_msg(struct ofproto *p, struct ofpbuf *packet) | |
4237 | { | |
4238 | struct odp_msg *msg = packet->data; | |
4239 | ||
4240 | switch (msg->type) { | |
4241 | case _ODPL_ACTION_NR: | |
4242 | COVERAGE_INC(ofproto_ctlr_action); | |
76ce9432 | 4243 | send_packet_in(p, packet); |
72b06300 BP |
4244 | break; |
4245 | ||
4246 | case _ODPL_SFLOW_NR: | |
4247 | if (p->sflow) { | |
4248 | ofproto_sflow_received(p->sflow, msg); | |
4249 | } | |
4250 | ofpbuf_delete(packet); | |
4251 | break; | |
4252 | ||
4253 | case _ODPL_MISS_NR: | |
4254 | handle_odp_miss_msg(p, packet); | |
4255 | break; | |
4256 | ||
4257 | default: | |
4258 | VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32, | |
4259 | msg->type); | |
4260 | break; | |
4261 | } | |
4262 | } | |
064af421 | 4263 | \f |
4a4cdb3b BP |
4264 | /* Flow expiration. */ |
4265 | ||
4266 | struct expire_cbdata { | |
4267 | struct ofproto *ofproto; | |
0de7a4b4 | 4268 | int dp_max_idle; |
4a4cdb3b BP |
4269 | }; |
4270 | ||
0de7a4b4 | 4271 | static int ofproto_dp_max_idle(const struct ofproto *); |
4a4cdb3b BP |
4272 | static void ofproto_update_used(struct ofproto *); |
4273 | static void rule_expire(struct cls_rule *, void *cbdata); | |
4274 | ||
4275 | /* This function is called periodically by ofproto_run(). Its job is to | |
4276 | * collect updates for the flows that have been installed into the datapath, | |
4277 | * most importantly when they last were used, and then use that information to | |
0de7a4b4 BP |
4278 | * expire flows that have not been used recently. |
4279 | * | |
4280 | * Returns the number of milliseconds after which it should be called again. */ | |
4281 | static int | |
4a4cdb3b BP |
4282 | ofproto_expire(struct ofproto *ofproto) |
4283 | { | |
4284 | struct expire_cbdata cbdata; | |
4285 | ||
4286 | /* Update 'used' for each flow in the datapath. */ | |
4287 | ofproto_update_used(ofproto); | |
4288 | ||
da89bf9e BP |
4289 | /* Expire idle flows. |
4290 | * | |
4291 | * A wildcarded flow is idle only when all of its subrules have expired due | |
4292 | * to becoming idle, so iterate through the exact-match flows first. */ | |
4a4cdb3b | 4293 | cbdata.ofproto = ofproto; |
0de7a4b4 | 4294 | cbdata.dp_max_idle = ofproto_dp_max_idle(ofproto); |
da89bf9e BP |
4295 | classifier_for_each(&ofproto->cls, CLS_INC_EXACT, rule_expire, &cbdata); |
4296 | classifier_for_each(&ofproto->cls, CLS_INC_WILD, rule_expire, &cbdata); | |
4a4cdb3b BP |
4297 | |
4298 | /* Let the hook know that we're at a stable point: all outstanding data | |
4299 | * in existing flows has been accounted to the account_cb. Thus, the | |
4300 | * hook can now reasonably do operations that depend on having accurate | |
4301 | * flow volume accounting (currently, that's just bond rebalancing). */ | |
4302 | if (ofproto->ofhooks->account_checkpoint_cb) { | |
4303 | ofproto->ofhooks->account_checkpoint_cb(ofproto->aux); | |
4304 | } | |
0de7a4b4 BP |
4305 | |
4306 | return MIN(cbdata.dp_max_idle, 1000); | |
4a4cdb3b BP |
4307 | } |
4308 | ||
4309 | /* Update 'used' member of each flow currently installed into the datapath. */ | |
4310 | static void | |
4311 | ofproto_update_used(struct ofproto *p) | |
4312 | { | |
4313 | struct odp_flow *flows; | |
4314 | size_t n_flows; | |
4315 | size_t i; | |
4316 | int error; | |
4317 | ||
4318 | error = dpif_flow_list_all(p->dpif, &flows, &n_flows); | |
4319 | if (error) { | |
4320 | return; | |
4321 | } | |
4322 | ||
4323 | for (i = 0; i < n_flows; i++) { | |
4324 | struct odp_flow *f = &flows[i]; | |
4325 | struct rule *rule; | |
4326 | ||
4327 | rule = rule_from_cls_rule( | |
4328 | classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX)); | |
4329 | ||
4330 | if (rule && rule->installed) { | |
4331 | update_time(p, rule, &f->stats); | |
4332 | rule_account(p, rule, f->stats.n_bytes); | |
4333 | } else { | |
4334 | /* There's a flow in the datapath that we know nothing about. | |
4335 | * Delete it. */ | |
4336 | COVERAGE_INC(ofproto_unexpected_rule); | |
4337 | dpif_flow_del(p->dpif, f); | |
4338 | } | |
4339 | ||
4340 | } | |
4341 | free(flows); | |
4342 | } | |
4343 | ||
0de7a4b4 BP |
4344 | /* Calculates and returns the number of milliseconds of idle time after which |
4345 | * flows should expire from the datapath and we should fold their statistics | |
4346 | * into their parent rules in userspace. */ | |
4347 | static int | |
4348 | ofproto_dp_max_idle(const struct ofproto *ofproto) | |
4349 | { | |
4350 | /* | |
4351 | * Idle time histogram. | |
4352 | * | |
4353 | * Most of the time a switch has a relatively small number of flows. When | |
4354 | * this is the case we might as well keep statistics for all of them in | |
4355 | * userspace and to cache them in the kernel datapath for performance as | |
4356 | * well. | |
4357 | * | |
4358 | * As the number of flows increases, the memory required to maintain | |
4359 | * statistics about them in userspace and in the kernel becomes | |
4360 | * significant. However, with a large number of flows it is likely that | |
4361 | * only a few of them are "heavy hitters" that consume a large amount of | |
4362 | * bandwidth. At this point, only heavy hitters are worth caching in the | |
4363 | * kernel and maintaining in userspaces; other flows we can discard. | |
4364 | * | |
4365 | * The technique used to compute the idle time is to build a histogram with | |
4366 | * N_BUCKETS bucket whose width is BUCKET_WIDTH msecs each. Each flow that | |
4367 | * is installed in the kernel gets dropped in the appropriate bucket. | |
4368 | * After the histogram has been built, we compute the cutoff so that only | |
4369 | * the most-recently-used 1% of flows (but at least 1000 flows) are kept | |
4370 | * cached. At least the most-recently-used bucket of flows is kept, so | |
4371 | * actually an arbitrary number of flows can be kept in any given | |
4372 | * expiration run (though the next run will delete most of those unless | |
4373 | * they receive additional data). | |
4374 | * | |
4375 | * This requires a second pass through the exact-match flows, in addition | |
4376 | * to the pass made by ofproto_update_used(), because the former function | |
4377 | * never looks at uninstallable flows. | |
4378 | */ | |
4379 | enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) }; | |
4380 | enum { N_BUCKETS = 5000 / BUCKET_WIDTH }; | |
4381 | int buckets[N_BUCKETS] = { 0 }; | |
4382 | int total, bucket; | |
4383 | struct rule *rule; | |
4384 | long long int now; | |
4385 | int i; | |
4386 | ||
4387 | total = classifier_count_exact(&ofproto->cls); | |
4388 | if (total <= 1000) { | |
4389 | return N_BUCKETS * BUCKET_WIDTH; | |
4390 | } | |
4391 | ||
4392 | /* Build histogram. */ | |
4393 | now = time_msec(); | |
4394 | CLASSIFIER_FOR_EACH_EXACT_RULE (rule, cr, &ofproto->cls) { | |
4395 | long long int idle = now - rule->used; | |
4396 | int bucket = (idle <= 0 ? 0 | |
4397 | : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1 | |
4398 | : (unsigned int) idle / BUCKET_WIDTH); | |
4399 | buckets[bucket]++; | |
4400 | } | |
4401 | ||
4402 | /* Find the first bucket whose flows should be expired. */ | |
4403 | for (bucket = 0; bucket < N_BUCKETS; bucket++) { | |
4404 | if (buckets[bucket]) { | |
4405 | int subtotal = 0; | |
4406 | do { | |
4407 | subtotal += buckets[bucket++]; | |
4408 | } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100)); | |
4409 | break; | |
4410 | } | |
4411 | } | |
4412 | ||
4413 | if (VLOG_IS_DBG_ENABLED()) { | |
4414 | struct ds s; | |
4415 | ||
4416 | ds_init(&s); | |
4417 | ds_put_cstr(&s, "keep"); | |
4418 | for (i = 0; i < N_BUCKETS; i++) { | |
4419 | if (i == bucket) { | |
4420 | ds_put_cstr(&s, ", drop"); | |
4421 | } | |
4422 | if (buckets[i]) { | |
4423 | ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]); | |
4424 | } | |
4425 | } | |
4426 | VLOG_INFO("%s: %s (msec:count)", | |
4427 | dpif_name(ofproto->dpif), ds_cstr(&s)); | |
4428 | ds_destroy(&s); | |
4429 | } | |
4430 | ||
4431 | return bucket * BUCKET_WIDTH; | |
4432 | } | |
4433 | ||
4a4cdb3b BP |
4434 | static void |
4435 | rule_active_timeout(struct ofproto *ofproto, struct rule *rule) | |
4436 | { | |
4437 | if (ofproto->netflow && !is_controller_rule(rule) && | |
4438 | netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) { | |
4439 | struct ofexpired expired; | |
4440 | struct odp_flow odp_flow; | |
4441 | ||
4442 | /* Get updated flow stats. | |
4443 | * | |
4444 | * XXX We could avoid this call entirely if (1) ofproto_update_used() | |
4445 | * updated TCP flags and (2) the dpif_flow_list_all() in | |
4446 | * ofproto_update_used() zeroed TCP flags. */ | |
4447 | memset(&odp_flow, 0, sizeof odp_flow); | |
4448 | if (rule->installed) { | |
4449 | odp_flow.key = rule->cr.flow; | |
4450 | odp_flow.flags = ODPFF_ZERO_TCP_FLAGS; | |
4451 | dpif_flow_get(ofproto->dpif, &odp_flow); | |
4452 | ||
4453 | if (odp_flow.stats.n_packets) { | |
4454 | update_time(ofproto, rule, &odp_flow.stats); | |
4455 | netflow_flow_update_flags(&rule->nf_flow, | |
4456 | odp_flow.stats.tcp_flags); | |
4457 | } | |
4458 | } | |
4459 | ||
4460 | expired.flow = rule->cr.flow; | |
4461 | expired.packet_count = rule->packet_count + | |
4462 | odp_flow.stats.n_packets; | |
4463 | expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes; | |
4464 | expired.used = rule->used; | |
4465 | ||
4466 | netflow_expire(ofproto->netflow, &rule->nf_flow, &expired); | |
4a4cdb3b BP |
4467 | } |
4468 | } | |
4469 | ||
4470 | /* If 'cls_rule' is an OpenFlow rule, that has expired according to OpenFlow | |
4471 | * rules, then delete it entirely. | |
4472 | * | |
4473 | * If 'cls_rule' is a subrule, that has not been used recently, remove it from | |
4474 | * the datapath and fold its statistics back into its super-rule. | |
4475 | * | |
4476 | * (This is a callback function for classifier_for_each().) */ | |
4477 | static void | |
4478 | rule_expire(struct cls_rule *cls_rule, void *cbdata_) | |
4479 | { | |
4480 | struct expire_cbdata *cbdata = cbdata_; | |
4481 | struct ofproto *ofproto = cbdata->ofproto; | |
4482 | struct rule *rule = rule_from_cls_rule(cls_rule); | |
4483 | long long int hard_expire, idle_expire, expire, now; | |
4484 | ||
4485 | /* Calculate OpenFlow expiration times for 'rule'. */ | |
4486 | hard_expire = (rule->hard_timeout | |
4487 | ? rule->created + rule->hard_timeout * 1000 | |
4488 | : LLONG_MAX); | |
4489 | idle_expire = (rule->idle_timeout | |
4490 | && (rule->super || list_is_empty(&rule->list)) | |
4491 | ? rule->used + rule->idle_timeout * 1000 | |
4492 | : LLONG_MAX); | |
4493 | expire = MIN(hard_expire, idle_expire); | |
4494 | ||
4495 | now = time_msec(); | |
4496 | if (now < expire) { | |
4497 | /* 'rule' has not expired according to OpenFlow rules. */ | |
431d4707 | 4498 | if (!rule->cr.wc.wildcards) { |
0de7a4b4 | 4499 | if (now >= rule->used + cbdata->dp_max_idle) { |
431d4707 BP |
4500 | /* This rule is idle, so drop it to free up resources. */ |
4501 | if (rule->super) { | |
4502 | /* It's not part of the OpenFlow flow table, so we can | |
4503 | * delete it entirely and fold its statistics into its | |
4504 | * super-rule. */ | |
4505 | rule_remove(ofproto, rule); | |
4506 | } else { | |
4507 | /* It is part of the OpenFlow flow table, so we have to | |
4508 | * keep the rule but we can at least uninstall it from the | |
4509 | * datapath. */ | |
4510 | rule_uninstall(ofproto, rule); | |
4511 | } | |
4a4cdb3b | 4512 | } else { |
431d4707 BP |
4513 | /* Send NetFlow active timeout if appropriate. */ |
4514 | rule_active_timeout(cbdata->ofproto, rule); | |
4a4cdb3b | 4515 | } |
4a4cdb3b BP |
4516 | } |
4517 | } else { | |
4518 | /* 'rule' has expired according to OpenFlow rules. */ | |
4519 | COVERAGE_INC(ofproto_expired); | |
4520 | ||
4521 | /* Update stats. (This is a no-op if the rule expired due to an idle | |
4522 | * timeout, because that only happens when the rule has no subrules | |
4523 | * left.) */ | |
4524 | if (rule->cr.wc.wildcards) { | |
4525 | struct rule *subrule, *next; | |
4526 | LIST_FOR_EACH_SAFE (subrule, next, list, &rule->list) { | |
4527 | rule_remove(cbdata->ofproto, subrule); | |
4528 | } | |
4529 | } else { | |
4530 | rule_uninstall(cbdata->ofproto, rule); | |
4531 | } | |
4532 | ||
4533 | /* Get rid of the rule. */ | |
4534 | if (!rule_is_hidden(rule)) { | |
4535 | send_flow_removed(cbdata->ofproto, rule, now, | |
4536 | (now >= hard_expire | |
4537 | ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT)); | |
4538 | } | |
4539 | rule_remove(cbdata->ofproto, rule); | |
4540 | } | |
4541 | } | |
4542 | \f | |
064af421 BP |
4543 | static void |
4544 | revalidate_cb(struct cls_rule *sub_, void *cbdata_) | |
4545 | { | |
4546 | struct rule *sub = rule_from_cls_rule(sub_); | |
4547 | struct revalidate_cbdata *cbdata = cbdata_; | |
4548 | ||
4549 | if (cbdata->revalidate_all | |
4550 | || (cbdata->revalidate_subrules && sub->super) | |
4551 | || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) { | |
4552 | revalidate_rule(cbdata->ofproto, sub); | |
4553 | } | |
4554 | } | |
4555 | ||
4556 | static bool | |
4557 | revalidate_rule(struct ofproto *p, struct rule *rule) | |
4558 | { | |
4559 | const flow_t *flow = &rule->cr.flow; | |
4560 | ||
4561 | COVERAGE_INC(ofproto_revalidate_rule); | |
4562 | if (rule->super) { | |
4563 | struct rule *super; | |
4564 | super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow)); | |
4565 | if (!super) { | |
4566 | rule_remove(p, rule); | |
4567 | return false; | |
4568 | } else if (super != rule->super) { | |
4569 | COVERAGE_INC(ofproto_revalidate_moved); | |
4570 | list_remove(&rule->list); | |
4571 | list_push_back(&super->list, &rule->list); | |
4572 | rule->super = super; | |
4573 | rule->hard_timeout = super->hard_timeout; | |
4574 | rule->idle_timeout = super->idle_timeout; | |
4575 | rule->created = super->created; | |
4576 | rule->used = 0; | |
4577 | } | |
4578 | } | |
4579 | ||
4580 | rule_update_actions(p, rule); | |
4581 | return true; | |
4582 | } | |
4583 | ||
4584 | static struct ofpbuf * | |
659586ef JG |
4585 | compose_flow_removed(struct ofproto *p, const struct rule *rule, |
4586 | long long int now, uint8_t reason) | |
064af421 | 4587 | { |
ca069229 | 4588 | struct ofp_flow_removed *ofr; |
064af421 | 4589 | struct ofpbuf *buf; |
9ca76894 | 4590 | long long int tdiff = now - rule->created; |
26c3f94a JP |
4591 | uint32_t sec = tdiff / 1000; |
4592 | uint32_t msec = tdiff - (sec * 1000); | |
064af421 | 4593 | |
ca069229 | 4594 | ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf); |
659586ef JG |
4595 | flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, p->tun_id_from_cookie, |
4596 | &ofr->match); | |
39997502 | 4597 | ofr->cookie = rule->flow_cookie; |
ca069229 JP |
4598 | ofr->priority = htons(rule->cr.priority); |
4599 | ofr->reason = reason; | |
26c3f94a JP |
4600 | ofr->duration_sec = htonl(sec); |
4601 | ofr->duration_nsec = htonl(msec * 1000000); | |
ca069229 JP |
4602 | ofr->idle_timeout = htons(rule->idle_timeout); |
4603 | ofr->packet_count = htonll(rule->packet_count); | |
4604 | ofr->byte_count = htonll(rule->byte_count); | |
064af421 BP |
4605 | |
4606 | return buf; | |
4607 | } | |
4608 | ||
ca069229 JP |
4609 | static void |
4610 | send_flow_removed(struct ofproto *p, struct rule *rule, | |
4611 | long long int now, uint8_t reason) | |
064af421 BP |
4612 | { |
4613 | struct ofconn *ofconn; | |
4614 | struct ofconn *prev; | |
b9b0ce61 | 4615 | struct ofpbuf *buf = NULL; |
064af421 | 4616 | |
3b587616 BP |
4617 | if (!rule->send_flow_removed) { |
4618 | return; | |
4619 | } | |
4620 | ||
064af421 BP |
4621 | /* We limit the maximum number of queued flow expirations it by accounting |
4622 | * them under the counter for replies. That works because preventing | |
4623 | * OpenFlow requests from being processed also prevents new flows from | |
4624 | * being added (and expiring). (It also prevents processing OpenFlow | |
4625 | * requests that would not add new flows, so it is imperfect.) */ | |
4626 | ||
4627 | prev = NULL; | |
4e8e4213 | 4628 | LIST_FOR_EACH (ofconn, node, &p->all_conns) { |
3b587616 | 4629 | if (rconn_is_connected(ofconn->rconn) |
c91248b3 | 4630 | && ofconn_receives_async_msgs(ofconn)) { |
064af421 | 4631 | if (prev) { |
431d8ad2 | 4632 | queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter); |
064af421 | 4633 | } else { |
659586ef | 4634 | buf = compose_flow_removed(p, rule, now, reason); |
064af421 BP |
4635 | } |
4636 | prev = ofconn; | |
4637 | } | |
4638 | } | |
4639 | if (prev) { | |
431d8ad2 | 4640 | queue_tx(buf, prev, prev->reply_counter); |
064af421 BP |
4641 | } |
4642 | } | |
4643 | ||
43253595 | 4644 | /* pinsched callback for sending 'packet' on 'ofconn'. */ |
064af421 | 4645 | static void |
76ce9432 | 4646 | do_send_packet_in(struct ofpbuf *packet, void *ofconn_) |
064af421 | 4647 | { |
76ce9432 | 4648 | struct ofconn *ofconn = ofconn_; |
43253595 BP |
4649 | |
4650 | rconn_send_with_limit(ofconn->rconn, packet, | |
4651 | ofconn->packet_in_counter, 100); | |
4652 | } | |
4653 | ||
4654 | /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and | |
4655 | * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s | |
4656 | * packet scheduler for sending. | |
4657 | * | |
30ea5d93 BP |
4658 | * 'max_len' specifies the maximum number of bytes of the packet to send on |
4659 | * 'ofconn' (INT_MAX specifies no limit). | |
4660 | * | |
43253595 BP |
4661 | * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise, |
4662 | * ownership is transferred to this function. */ | |
4663 | static void | |
30ea5d93 BP |
4664 | schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len, |
4665 | bool clone) | |
43253595 | 4666 | { |
76ce9432 | 4667 | struct ofproto *ofproto = ofconn->ofproto; |
43253595 BP |
4668 | struct ofp_packet_in *opi = packet->data; |
4669 | uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port)); | |
4670 | int send_len, trim_size; | |
76ce9432 | 4671 | uint32_t buffer_id; |
064af421 | 4672 | |
43253595 BP |
4673 | /* Get buffer. */ |
4674 | if (opi->reason == OFPR_ACTION) { | |
76ce9432 | 4675 | buffer_id = UINT32_MAX; |
43253595 BP |
4676 | } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) { |
4677 | buffer_id = pktbuf_get_null(); | |
89b9612d BP |
4678 | } else if (!ofconn->pktbuf) { |
4679 | buffer_id = UINT32_MAX; | |
76ce9432 | 4680 | } else { |
43253595 BP |
4681 | struct ofpbuf payload; |
4682 | payload.data = opi->data; | |
4683 | payload.size = packet->size - offsetof(struct ofp_packet_in, data); | |
4684 | buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port); | |
76ce9432 | 4685 | } |
372179d4 | 4686 | |
43253595 BP |
4687 | /* Figure out how much of the packet to send. */ |
4688 | send_len = ntohs(opi->total_len); | |
4689 | if (buffer_id != UINT32_MAX) { | |
4690 | send_len = MIN(send_len, ofconn->miss_send_len); | |
4691 | } | |
30ea5d93 | 4692 | send_len = MIN(send_len, max_len); |
064af421 | 4693 | |
43253595 BP |
4694 | /* Adjust packet length and clone if necessary. */ |
4695 | trim_size = offsetof(struct ofp_packet_in, data) + send_len; | |
4696 | if (clone) { | |
4697 | packet = ofpbuf_clone_data(packet->data, trim_size); | |
4698 | opi = packet->data; | |
4699 | } else { | |
4700 | packet->size = trim_size; | |
4701 | } | |
4702 | ||
4703 | /* Update packet headers. */ | |
4704 | opi->buffer_id = htonl(buffer_id); | |
4705 | update_openflow_length(packet); | |
4706 | ||
4707 | /* Hand over to packet scheduler. It might immediately call into | |
4708 | * do_send_packet_in() or it might buffer it for a while (until a later | |
4709 | * call to pinsched_run()). */ | |
4710 | pinsched_send(ofconn->schedulers[opi->reason], in_port, | |
4711 | packet, do_send_packet_in, ofconn); | |
064af421 BP |
4712 | } |
4713 | ||
43253595 BP |
4714 | /* Replace struct odp_msg header in 'packet' by equivalent struct |
4715 | * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as | |
4716 | * returned by dpif_recv()). | |
4717 | * | |
4718 | * The conversion is not complete: the caller still needs to trim any unneeded | |
4719 | * payload off the end of the buffer, set the length in the OpenFlow header, | |
4720 | * and set buffer_id. Those require us to know the controller settings and so | |
30ea5d93 BP |
4721 | * must be done on a per-controller basis. |
4722 | * | |
4723 | * Returns the maximum number of bytes of the packet that should be sent to | |
4724 | * the controller (INT_MAX if no limit). */ | |
4725 | static int | |
43253595 | 4726 | do_convert_to_packet_in(struct ofpbuf *packet) |
064af421 | 4727 | { |
76ce9432 | 4728 | struct odp_msg *msg = packet->data; |
43253595 BP |
4729 | struct ofp_packet_in *opi; |
4730 | uint8_t reason; | |
4731 | uint16_t total_len; | |
4732 | uint16_t in_port; | |
30ea5d93 | 4733 | int max_len; |
43253595 BP |
4734 | |
4735 | /* Extract relevant header fields */ | |
30ea5d93 BP |
4736 | if (msg->type == _ODPL_ACTION_NR) { |
4737 | reason = OFPR_ACTION; | |
4738 | max_len = msg->arg; | |
4739 | } else { | |
4740 | reason = OFPR_NO_MATCH; | |
4741 | max_len = INT_MAX; | |
4742 | } | |
43253595 BP |
4743 | total_len = msg->length - sizeof *msg; |
4744 | in_port = odp_port_to_ofp_port(msg->port); | |
4745 | ||
4746 | /* Repurpose packet buffer by overwriting header. */ | |
4747 | ofpbuf_pull(packet, sizeof(struct odp_msg)); | |
4748 | opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data)); | |
4749 | opi->header.version = OFP_VERSION; | |
4750 | opi->header.type = OFPT_PACKET_IN; | |
4751 | opi->total_len = htons(total_len); | |
4752 | opi->in_port = htons(in_port); | |
4753 | opi->reason = reason; | |
30ea5d93 BP |
4754 | |
4755 | return max_len; | |
43253595 BP |
4756 | } |
4757 | ||
4758 | /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or | |
4759 | * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller | |
4760 | * as necessary according to their individual configurations. | |
4761 | * | |
4762 | * 'packet' must have sufficient headroom to convert it into a struct | |
4763 | * ofp_packet_in (e.g. as returned by dpif_recv()). | |
4764 | * | |
4765 | * Takes ownership of 'packet'. */ | |
4766 | static void | |
4767 | send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet) | |
4768 | { | |
76ce9432 | 4769 | struct ofconn *ofconn, *prev; |
30ea5d93 | 4770 | int max_len; |
064af421 | 4771 | |
30ea5d93 | 4772 | max_len = do_convert_to_packet_in(packet); |
76ce9432 BP |
4773 | |
4774 | prev = NULL; | |
4e8e4213 | 4775 | LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) { |
c91248b3 | 4776 | if (ofconn_receives_async_msgs(ofconn)) { |
9deba63b | 4777 | if (prev) { |
30ea5d93 | 4778 | schedule_packet_in(prev, packet, max_len, true); |
9deba63b BP |
4779 | } |
4780 | prev = ofconn; | |
064af421 | 4781 | } |
76ce9432 BP |
4782 | } |
4783 | if (prev) { | |
30ea5d93 | 4784 | schedule_packet_in(prev, packet, max_len, false); |
76ce9432 BP |
4785 | } else { |
4786 | ofpbuf_delete(packet); | |
064af421 | 4787 | } |
064af421 BP |
4788 | } |
4789 | ||
4790 | static uint64_t | |
fa60c019 | 4791 | pick_datapath_id(const struct ofproto *ofproto) |
064af421 | 4792 | { |
fa60c019 | 4793 | const struct ofport *port; |
064af421 | 4794 | |
ca0f572c | 4795 | port = get_port(ofproto, ODPP_LOCAL); |
fa60c019 BP |
4796 | if (port) { |
4797 | uint8_t ea[ETH_ADDR_LEN]; | |
4798 | int error; | |
4799 | ||
4800 | error = netdev_get_etheraddr(port->netdev, ea); | |
064af421 BP |
4801 | if (!error) { |
4802 | return eth_addr_to_uint64(ea); | |
4803 | } | |
4804 | VLOG_WARN("could not get MAC address for %s (%s)", | |
fa60c019 | 4805 | netdev_get_name(port->netdev), strerror(error)); |
064af421 | 4806 | } |
fa60c019 | 4807 | return ofproto->fallback_dpid; |
064af421 BP |
4808 | } |
4809 | ||
4810 | static uint64_t | |
4811 | pick_fallback_dpid(void) | |
4812 | { | |
4813 | uint8_t ea[ETH_ADDR_LEN]; | |
70150daf | 4814 | eth_addr_nicira_random(ea); |
064af421 BP |
4815 | return eth_addr_to_uint64(ea); |
4816 | } | |
4817 | \f | |
4818 | static bool | |
4819 | default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet, | |
4820 | struct odp_actions *actions, tag_type *tags, | |
6a07af36 | 4821 | uint16_t *nf_output_iface, void *ofproto_) |
064af421 BP |
4822 | { |
4823 | struct ofproto *ofproto = ofproto_; | |
4824 | int out_port; | |
4825 | ||
4826 | /* Drop frames for reserved multicast addresses. */ | |
4827 | if (eth_addr_is_reserved(flow->dl_dst)) { | |
4828 | return true; | |
4829 | } | |
4830 | ||
4831 | /* Learn source MAC (but don't try to learn from revalidation). */ | |
4832 | if (packet != NULL) { | |
4833 | tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src, | |
7febb910 JG |
4834 | 0, flow->in_port, |
4835 | GRAT_ARP_LOCK_NONE); | |
064af421 BP |
4836 | if (rev_tag) { |
4837 | /* The log messages here could actually be useful in debugging, | |
4838 | * so keep the rate limit relatively high. */ | |
4839 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); | |
4840 | VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16, | |
4841 | ETH_ADDR_ARGS(flow->dl_src), flow->in_port); | |
4842 | ofproto_revalidate(ofproto, rev_tag); | |
4843 | } | |
4844 | } | |
4845 | ||
4846 | /* Determine output port. */ | |
7febb910 JG |
4847 | out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags, |
4848 | NULL); | |
064af421 | 4849 | if (out_port < 0) { |
6a07af36 | 4850 | add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface); |
064af421 BP |
4851 | } else if (out_port != flow->in_port) { |
4852 | odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port; | |
6a07af36 | 4853 | *nf_output_iface = out_port; |
064af421 BP |
4854 | } else { |
4855 | /* Drop. */ | |
4856 | } | |
4857 | ||
4858 | return true; | |
4859 | } | |
4860 | ||
4861 | static const struct ofhooks default_ofhooks = { | |
064af421 BP |
4862 | default_normal_ofhook_cb, |
4863 | NULL, | |
4864 | NULL | |
4865 | }; |