]>
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> | |
10a24935 | 27 | #include "byte-order.h" |
064af421 BP |
28 | #include "classifier.h" |
29 | #include "coverage.h" | |
30 | #include "discovery.h" | |
31 | #include "dpif.h" | |
4f2cad2c | 32 | #include "dynamic-string.h" |
064af421 | 33 | #include "fail-open.h" |
ca0f572c BP |
34 | #include "hash.h" |
35 | #include "hmap.h" | |
064af421 BP |
36 | #include "in-band.h" |
37 | #include "mac-learning.h" | |
38 | #include "netdev.h" | |
39 | #include "netflow.h" | |
09246b99 | 40 | #include "nx-match.h" |
064af421 BP |
41 | #include "odp-util.h" |
42 | #include "ofp-print.h" | |
fa37b408 | 43 | #include "ofp-util.h" |
72b06300 | 44 | #include "ofproto-sflow.h" |
064af421 BP |
45 | #include "ofpbuf.h" |
46 | #include "openflow/nicira-ext.h" | |
47 | #include "openflow/openflow.h" | |
064af421 BP |
48 | #include "openvswitch/datapath-protocol.h" |
49 | #include "packets.h" | |
50 | #include "pinsched.h" | |
51 | #include "pktbuf.h" | |
52 | #include "poll-loop.h" | |
064af421 BP |
53 | #include "rconn.h" |
54 | #include "shash.h" | |
55 | #include "status.h" | |
fe55ad15 | 56 | #include "stream-ssl.h" |
064af421 BP |
57 | #include "svec.h" |
58 | #include "tag.h" | |
59 | #include "timeval.h" | |
4f2cad2c | 60 | #include "unixctl.h" |
064af421 | 61 | #include "vconn.h" |
5136ce49 | 62 | #include "vlog.h" |
064af421 | 63 | |
d98e6007 | 64 | VLOG_DEFINE_THIS_MODULE(ofproto); |
064af421 | 65 | |
72b06300 | 66 | #include "sflow_api.h" |
064af421 | 67 | |
064af421 | 68 | struct ofport { |
ca0f572c | 69 | struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */ |
064af421 BP |
70 | struct netdev *netdev; |
71 | struct ofp_phy_port opp; /* In host byte order. */ | |
ca0f572c | 72 | uint16_t odp_port; |
064af421 BP |
73 | }; |
74 | ||
75 | static void ofport_free(struct ofport *); | |
76 | static void hton_ofp_phy_port(struct ofp_phy_port *); | |
77 | ||
78 | static int xlate_actions(const union ofp_action *in, size_t n_in, | |
ae412e7d | 79 | const struct flow *, struct ofproto *, |
064af421 BP |
80 | const struct ofpbuf *packet, |
81 | struct odp_actions *out, tag_type *tags, | |
6a07af36 | 82 | bool *may_set_up_flow, uint16_t *nf_output_iface); |
064af421 | 83 | |
bcf84111 | 84 | /* An OpenFlow flow. */ |
064af421 | 85 | struct rule { |
0c43ad9a | 86 | long long int used; /* Time last used; time created if not used. */ |
064af421 | 87 | long long int created; /* Creation time. */ |
064af421 | 88 | |
bcf84111 | 89 | /* These statistics: |
064af421 | 90 | * |
bcf84111 BP |
91 | * - Do include packets and bytes from facets that have been deleted or |
92 | * whose own statistics have been folded into the rule. | |
79eee1eb | 93 | * |
bcf84111 BP |
94 | * - Do include packets and bytes sent "by hand" that were accounted to |
95 | * the rule without any facet being involved (this is a rare corner | |
96 | * case in rule_execute()). | |
064af421 | 97 | * |
bcf84111 BP |
98 | * - Do not include packet or bytes that can be obtained from any facet's |
99 | * packet_count or byte_count member or that can be obtained from the | |
100 | * datapath by, e.g., dpif_flow_get() for any facet. | |
101 | */ | |
102 | uint64_t packet_count; /* Number of packets received. */ | |
103 | uint64_t byte_count; /* Number of bytes received. */ | |
104 | ||
105 | ovs_be64 flow_cookie; /* Controller-issued identifier. */ | |
106 | ||
107 | struct cls_rule cr; /* In owning ofproto's classifier. */ | |
108 | uint16_t idle_timeout; /* In seconds from time of last use. */ | |
109 | uint16_t hard_timeout; /* In seconds from time of creation. */ | |
110 | bool send_flow_removed; /* Send a flow removed message? */ | |
111 | int n_actions; /* Number of elements in actions[]. */ | |
112 | union ofp_action *actions; /* OpenFlow actions. */ | |
113 | struct list facets; /* List of "struct facet"s. */ | |
064af421 BP |
114 | }; |
115 | ||
bcf84111 BP |
116 | static struct rule *rule_from_cls_rule(const struct cls_rule *); |
117 | static bool rule_is_hidden(const struct rule *); | |
064af421 | 118 | |
bcf84111 | 119 | static struct rule *rule_create(const struct cls_rule *, |
0193b2af | 120 | const union ofp_action *, size_t n_actions, |
ca069229 | 121 | uint16_t idle_timeout, uint16_t hard_timeout, |
8054fc48 | 122 | ovs_be64 flow_cookie, bool send_flow_removed); |
064af421 | 123 | static void rule_destroy(struct ofproto *, struct rule *); |
bcf84111 BP |
124 | static void rule_free(struct rule *); |
125 | ||
126 | static struct rule *rule_lookup(struct ofproto *, const struct flow *); | |
afe75089 | 127 | static void rule_insert(struct ofproto *, struct rule *); |
064af421 | 128 | static void rule_remove(struct ofproto *, struct rule *); |
bcf84111 BP |
129 | |
130 | static void rule_send_removed(struct ofproto *, struct rule *, uint8_t reason); | |
131 | ||
132 | /* An exact-match instantiation of an OpenFlow flow. */ | |
133 | struct facet { | |
134 | long long int used; /* Time last used; time created if not used. */ | |
135 | ||
136 | /* These statistics: | |
137 | * | |
138 | * - Do include packets and bytes sent "by hand", e.g. with | |
139 | * dpif_execute(). | |
140 | * | |
141 | * - Do include packets and bytes that were obtained from the datapath | |
142 | * when a flow was deleted (e.g. dpif_flow_del()) or when its | |
143 | * statistics were reset (e.g. dpif_flow_put() with ODPPF_ZERO_STATS). | |
144 | * | |
145 | * - Do not include any packets or bytes that can currently be obtained | |
146 | * from the datapath by, e.g., dpif_flow_get(). | |
147 | */ | |
148 | uint64_t packet_count; /* Number of packets received. */ | |
149 | uint64_t byte_count; /* Number of bytes received. */ | |
150 | ||
151 | /* Number of bytes passed to account_cb. This may include bytes that can | |
152 | * currently obtained from the datapath (thus, it can be greater than | |
153 | * byte_count). */ | |
154 | uint64_t accounted_bytes; | |
155 | ||
156 | struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */ | |
157 | struct list list_node; /* In owning rule's 'facets' list. */ | |
158 | struct rule *rule; /* Owning rule. */ | |
159 | struct flow flow; /* Exact-match flow. */ | |
160 | bool installed; /* Installed in datapath? */ | |
161 | bool may_install; /* True ordinarily; false if actions must | |
162 | * be reassessed for every packet. */ | |
163 | int n_actions; /* Number of elements in actions[]. */ | |
164 | union odp_action *actions; /* Datapath actions. */ | |
165 | tag_type tags; /* Tags (set only by hooks). */ | |
166 | struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ | |
167 | }; | |
168 | ||
169 | static struct facet *facet_create(struct ofproto *, struct rule *, | |
170 | const struct flow *, | |
171 | const struct ofpbuf *packet); | |
172 | static void facet_remove(struct ofproto *, struct facet *); | |
173 | static void facet_free(struct facet *); | |
174 | ||
175 | static struct facet *facet_lookup_valid(struct ofproto *, const struct flow *); | |
176 | static bool facet_revalidate(struct ofproto *, struct facet *); | |
177 | ||
178 | static void facet_install(struct ofproto *, struct facet *, bool zero_stats); | |
179 | static void facet_uninstall(struct ofproto *, struct facet *); | |
d530fcd2 | 180 | static void facet_flush_stats(struct ofproto *, struct facet *); |
bcf84111 | 181 | |
7f7ae89d BP |
182 | static void facet_make_actions(struct ofproto *, struct facet *, |
183 | const struct ofpbuf *packet); | |
bcf84111 BP |
184 | static void facet_update_stats(struct ofproto *, struct facet *, |
185 | const struct odp_flow_stats *); | |
064af421 | 186 | |
76ce9432 BP |
187 | /* ofproto supports two kinds of OpenFlow connections: |
188 | * | |
5899143f BP |
189 | * - "Primary" connections to ordinary OpenFlow controllers. ofproto |
190 | * maintains persistent connections to these controllers and by default | |
191 | * sends them asynchronous messages such as packet-ins. | |
76ce9432 | 192 | * |
5899143f | 193 | * - "Service" connections, e.g. from ovs-ofctl. When these connections |
76ce9432 BP |
194 | * drop, it is the other side's responsibility to reconnect them if |
195 | * necessary. ofproto does not send them asynchronous messages by default. | |
7d674866 BP |
196 | * |
197 | * Currently, active (tcp, ssl, unix) connections are always "primary" | |
198 | * connections and passive (ptcp, pssl, punix) connections are always "service" | |
199 | * connections. There is no inherent reason for this, but it reflects the | |
200 | * common case. | |
76ce9432 BP |
201 | */ |
202 | enum ofconn_type { | |
5899143f BP |
203 | OFCONN_PRIMARY, /* An ordinary OpenFlow controller. */ |
204 | OFCONN_SERVICE /* A service connection, e.g. "ovs-ofctl". */ | |
76ce9432 | 205 | }; |
064af421 | 206 | |
7d674866 BP |
207 | /* A listener for incoming OpenFlow "service" connections. */ |
208 | struct ofservice { | |
209 | struct hmap_node node; /* In struct ofproto's "services" hmap. */ | |
210 | struct pvconn *pvconn; /* OpenFlow connection listener. */ | |
211 | ||
212 | /* These are not used by ofservice directly. They are settings for | |
213 | * accepted "struct ofconn"s from the pvconn. */ | |
214 | int probe_interval; /* Max idle time before probing, in seconds. */ | |
215 | int rate_limit; /* Max packet-in rate in packets per second. */ | |
216 | int burst_limit; /* Limit on accumulating packet credits. */ | |
217 | }; | |
218 | ||
219 | static struct ofservice *ofservice_lookup(struct ofproto *, | |
220 | const char *target); | |
221 | static int ofservice_create(struct ofproto *, | |
222 | const struct ofproto_controller *); | |
223 | static void ofservice_reconfigure(struct ofservice *, | |
224 | const struct ofproto_controller *); | |
225 | static void ofservice_destroy(struct ofproto *, struct ofservice *); | |
226 | ||
76ce9432 BP |
227 | /* An OpenFlow connection. */ |
228 | struct ofconn { | |
229 | struct ofproto *ofproto; /* The ofproto that owns this connection. */ | |
230 | struct list node; /* In struct ofproto's "all_conns" list. */ | |
231 | struct rconn *rconn; /* OpenFlow connection. */ | |
232 | enum ofconn_type type; /* Type. */ | |
b70eac89 | 233 | int flow_format; /* One of NXFF_*. */ |
76ce9432 BP |
234 | |
235 | /* OFPT_PACKET_IN related data. */ | |
236 | struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */ | |
237 | struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */ | |
238 | struct pktbuf *pktbuf; /* OpenFlow packet buffers. */ | |
239 | int miss_send_len; /* Bytes to send of buffered packets. */ | |
240 | ||
241 | /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow | |
242 | * requests, and the maximum number before we stop reading OpenFlow | |
243 | * requests. */ | |
064af421 BP |
244 | #define OFCONN_REPLY_MAX 100 |
245 | struct rconn_packet_counter *reply_counter; | |
76ce9432 | 246 | |
5899143f | 247 | /* type == OFCONN_PRIMARY only. */ |
9deba63b | 248 | enum nx_role role; /* Role. */ |
76ce9432 BP |
249 | struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */ |
250 | struct discovery *discovery; /* Controller discovery object, if enabled. */ | |
251 | struct status_category *ss; /* Switch status category. */ | |
d2ede7bc | 252 | enum ofproto_band band; /* In-band or out-of-band? */ |
064af421 BP |
253 | }; |
254 | ||
76ce9432 BP |
255 | /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's |
256 | * "schedulers" array. Their values are 0 and 1, and their meanings and values | |
257 | * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In | |
258 | * case anything ever changes, check their values here. */ | |
259 | #define N_SCHEDULERS 2 | |
260 | BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0); | |
261 | BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR); | |
262 | BUILD_ASSERT_DECL(OFPR_ACTION == 1); | |
263 | BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR); | |
264 | ||
265 | static struct ofconn *ofconn_create(struct ofproto *, struct rconn *, | |
266 | enum ofconn_type); | |
c475ae67 | 267 | static void ofconn_destroy(struct ofconn *); |
3269c562 | 268 | static void ofconn_run(struct ofconn *); |
064af421 | 269 | static void ofconn_wait(struct ofconn *); |
c91248b3 | 270 | static bool ofconn_receives_async_msgs(const struct ofconn *); |
eb15cdbb | 271 | static char *ofconn_make_name(const struct ofproto *, const char *target); |
7d674866 | 272 | static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst); |
c91248b3 | 273 | |
064af421 BP |
274 | static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn, |
275 | struct rconn_packet_counter *counter); | |
276 | ||
76ce9432 BP |
277 | static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg); |
278 | static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn); | |
279 | ||
064af421 BP |
280 | struct ofproto { |
281 | /* Settings. */ | |
282 | uint64_t datapath_id; /* Datapath ID. */ | |
283 | uint64_t fallback_dpid; /* Datapath ID if no better choice found. */ | |
5a719c38 JP |
284 | char *mfr_desc; /* Manufacturer. */ |
285 | char *hw_desc; /* Hardware. */ | |
286 | char *sw_desc; /* Software version. */ | |
287 | char *serial_desc; /* Serial number. */ | |
8abc4ed7 | 288 | char *dp_desc; /* Datapath description. */ |
064af421 BP |
289 | |
290 | /* Datapath. */ | |
c228a364 | 291 | struct dpif *dpif; |
e9e28be3 | 292 | struct netdev_monitor *netdev_monitor; |
ca0f572c | 293 | struct hmap ports; /* Contains "struct ofport"s. */ |
064af421 BP |
294 | struct shash port_by_name; |
295 | uint32_t max_ports; | |
296 | ||
297 | /* Configuration. */ | |
298 | struct switch_status *switch_status; | |
064af421 | 299 | struct fail_open *fail_open; |
064af421 | 300 | struct netflow *netflow; |
72b06300 | 301 | struct ofproto_sflow *sflow; |
064af421 | 302 | |
d2ede7bc BP |
303 | /* In-band control. */ |
304 | struct in_band *in_band; | |
305 | long long int next_in_band_update; | |
917e50e1 BP |
306 | struct sockaddr_in *extra_in_band_remotes; |
307 | size_t n_extra_remotes; | |
308 | ||
064af421 BP |
309 | /* Flow table. */ |
310 | struct classifier cls; | |
064af421 | 311 | long long int next_expiration; |
bcf84111 BP |
312 | |
313 | /* Facets. */ | |
314 | struct hmap facets; | |
315 | bool need_revalidate; | |
064af421 BP |
316 | struct tag_set revalidate_set; |
317 | ||
318 | /* OpenFlow connections. */ | |
76ce9432 BP |
319 | struct hmap controllers; /* Controller "struct ofconn"s. */ |
320 | struct list all_conns; /* Contains "struct ofconn"s. */ | |
31681a5d | 321 | enum ofproto_fail_mode fail_mode; |
7d674866 BP |
322 | |
323 | /* OpenFlow listeners. */ | |
324 | struct hmap services; /* Contains "struct ofservice"s. */ | |
064af421 BP |
325 | struct pvconn **snoops; |
326 | size_t n_snoops; | |
327 | ||
328 | /* Hooks for ovs-vswitchd. */ | |
329 | const struct ofhooks *ofhooks; | |
330 | void *aux; | |
331 | ||
332 | /* Used by default ofhooks. */ | |
333 | struct mac_learning *ml; | |
334 | }; | |
335 | ||
336 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
337 | ||
338 | static const struct ofhooks default_ofhooks; | |
339 | ||
fa60c019 | 340 | static uint64_t pick_datapath_id(const struct ofproto *); |
064af421 | 341 | static uint64_t pick_fallback_dpid(void); |
76ce9432 | 342 | |
0de7a4b4 | 343 | static int ofproto_expire(struct ofproto *); |
4a4cdb3b | 344 | |
064af421 BP |
345 | static void handle_odp_msg(struct ofproto *, struct ofpbuf *); |
346 | ||
3269c562 | 347 | static void handle_openflow(struct ofconn *, struct ofpbuf *); |
064af421 | 348 | |
ca0f572c | 349 | static struct ofport *get_port(const struct ofproto *, uint16_t odp_port); |
064af421 BP |
350 | static void update_port(struct ofproto *, const char *devname); |
351 | static int init_ports(struct ofproto *); | |
352 | static void reinit_ports(struct ofproto *); | |
353 | ||
354 | int | |
1a6f1e2a JG |
355 | ofproto_create(const char *datapath, const char *datapath_type, |
356 | const struct ofhooks *ofhooks, void *aux, | |
064af421 BP |
357 | struct ofproto **ofprotop) |
358 | { | |
064af421 BP |
359 | struct odp_stats stats; |
360 | struct ofproto *p; | |
c228a364 | 361 | struct dpif *dpif; |
064af421 BP |
362 | int error; |
363 | ||
364 | *ofprotop = NULL; | |
365 | ||
366 | /* Connect to datapath and start listening for messages. */ | |
1a6f1e2a | 367 | error = dpif_open(datapath, datapath_type, &dpif); |
064af421 BP |
368 | if (error) { |
369 | VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error)); | |
370 | return error; | |
371 | } | |
c228a364 | 372 | error = dpif_get_dp_stats(dpif, &stats); |
064af421 BP |
373 | if (error) { |
374 | VLOG_ERR("failed to obtain stats for datapath %s: %s", | |
375 | datapath, strerror(error)); | |
c228a364 | 376 | dpif_close(dpif); |
064af421 BP |
377 | return error; |
378 | } | |
72b06300 | 379 | error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW); |
064af421 BP |
380 | if (error) { |
381 | VLOG_ERR("failed to listen on datapath %s: %s", | |
382 | datapath, strerror(error)); | |
c228a364 | 383 | dpif_close(dpif); |
064af421 BP |
384 | return error; |
385 | } | |
c228a364 | 386 | dpif_flow_flush(dpif); |
8f24562a | 387 | dpif_recv_purge(dpif); |
064af421 BP |
388 | |
389 | /* Initialize settings. */ | |
ec6fde61 | 390 | p = xzalloc(sizeof *p); |
064af421 | 391 | p->fallback_dpid = pick_fallback_dpid(); |
fa60c019 | 392 | p->datapath_id = p->fallback_dpid; |
5a719c38 JP |
393 | p->mfr_desc = xstrdup(DEFAULT_MFR_DESC); |
394 | p->hw_desc = xstrdup(DEFAULT_HW_DESC); | |
395 | p->sw_desc = xstrdup(DEFAULT_SW_DESC); | |
396 | p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC); | |
23ff2821 | 397 | p->dp_desc = xstrdup(DEFAULT_DP_DESC); |
064af421 BP |
398 | |
399 | /* Initialize datapath. */ | |
400 | p->dpif = dpif; | |
8b61709d | 401 | p->netdev_monitor = netdev_monitor_create(); |
ca0f572c | 402 | hmap_init(&p->ports); |
064af421 BP |
403 | shash_init(&p->port_by_name); |
404 | p->max_ports = stats.max_ports; | |
405 | ||
406 | /* Initialize submodules. */ | |
407 | p->switch_status = switch_status_create(p); | |
408 | p->in_band = NULL; | |
064af421 | 409 | p->fail_open = NULL; |
064af421 | 410 | p->netflow = NULL; |
72b06300 | 411 | p->sflow = NULL; |
064af421 BP |
412 | |
413 | /* Initialize flow table. */ | |
414 | classifier_init(&p->cls); | |
064af421 | 415 | p->next_expiration = time_msec() + 1000; |
bcf84111 BP |
416 | |
417 | /* Initialize facet table. */ | |
418 | hmap_init(&p->facets); | |
419 | p->need_revalidate = false; | |
064af421 BP |
420 | tag_set_init(&p->revalidate_set); |
421 | ||
422 | /* Initialize OpenFlow connections. */ | |
423 | list_init(&p->all_conns); | |
76ce9432 | 424 | hmap_init(&p->controllers); |
7d674866 | 425 | hmap_init(&p->services); |
064af421 BP |
426 | p->snoops = NULL; |
427 | p->n_snoops = 0; | |
428 | ||
429 | /* Initialize hooks. */ | |
430 | if (ofhooks) { | |
431 | p->ofhooks = ofhooks; | |
432 | p->aux = aux; | |
433 | p->ml = NULL; | |
434 | } else { | |
435 | p->ofhooks = &default_ofhooks; | |
436 | p->aux = p; | |
437 | p->ml = mac_learning_create(); | |
438 | } | |
439 | ||
fa60c019 BP |
440 | /* Pick final datapath ID. */ |
441 | p->datapath_id = pick_datapath_id(p); | |
b123cc3c | 442 | VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id); |
fa60c019 | 443 | |
064af421 BP |
444 | *ofprotop = p; |
445 | return 0; | |
446 | } | |
447 | ||
448 | void | |
449 | ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id) | |
450 | { | |
451 | uint64_t old_dpid = p->datapath_id; | |
fa60c019 | 452 | p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p); |
064af421 | 453 | if (p->datapath_id != old_dpid) { |
b123cc3c | 454 | VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id); |
76ce9432 BP |
455 | |
456 | /* Force all active connections to reconnect, since there is no way to | |
457 | * notify a controller that the datapath ID has changed. */ | |
fa05809b | 458 | ofproto_reconnect_controllers(p); |
064af421 BP |
459 | } |
460 | } | |
461 | ||
76ce9432 BP |
462 | static bool |
463 | is_discovery_controller(const struct ofproto_controller *c) | |
464 | { | |
465 | return !strcmp(c->target, "discover"); | |
466 | } | |
467 | ||
468 | static bool | |
469 | is_in_band_controller(const struct ofproto_controller *c) | |
470 | { | |
471 | return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND; | |
472 | } | |
473 | ||
474 | /* Creates a new controller in 'ofproto'. Some of the settings are initially | |
475 | * drawn from 'c', but update_controller() needs to be called later to finish | |
476 | * the new ofconn's configuration. */ | |
477 | static void | |
478 | add_controller(struct ofproto *ofproto, const struct ofproto_controller *c) | |
479 | { | |
480 | struct discovery *discovery; | |
481 | struct ofconn *ofconn; | |
482 | ||
483 | if (is_discovery_controller(c)) { | |
484 | int error = discovery_create(c->accept_re, c->update_resolv_conf, | |
485 | ofproto->dpif, ofproto->switch_status, | |
486 | &discovery); | |
487 | if (error) { | |
488 | return; | |
489 | } | |
490 | } else { | |
491 | discovery = NULL; | |
492 | } | |
493 | ||
5899143f | 494 | ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY); |
76ce9432 BP |
495 | ofconn->pktbuf = pktbuf_create(); |
496 | ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN; | |
497 | if (discovery) { | |
498 | ofconn->discovery = discovery; | |
499 | } else { | |
eb15cdbb BP |
500 | char *name = ofconn_make_name(ofproto, c->target); |
501 | rconn_connect(ofconn->rconn, c->target, name); | |
502 | free(name); | |
76ce9432 BP |
503 | } |
504 | hmap_insert(&ofproto->controllers, &ofconn->hmap_node, | |
505 | hash_string(c->target, 0)); | |
506 | } | |
507 | ||
508 | /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's | |
509 | * target or turn discovery on or off (these are done by creating new ofconns | |
510 | * and deleting old ones), but it can update the rest of an ofconn's | |
511 | * settings. */ | |
512 | static void | |
513 | update_controller(struct ofconn *ofconn, const struct ofproto_controller *c) | |
064af421 | 514 | { |
76ce9432 | 515 | int probe_interval; |
79c9f2ee | 516 | |
d2ede7bc BP |
517 | ofconn->band = (is_in_band_controller(c) |
518 | ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND); | |
519 | ||
76ce9432 | 520 | rconn_set_max_backoff(ofconn->rconn, c->max_backoff); |
79c9f2ee | 521 | |
76ce9432 BP |
522 | probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0; |
523 | rconn_set_probe_interval(ofconn->rconn, probe_interval); | |
79c9f2ee | 524 | |
76ce9432 BP |
525 | if (ofconn->discovery) { |
526 | discovery_set_update_resolv_conf(ofconn->discovery, | |
527 | c->update_resolv_conf); | |
528 | discovery_set_accept_controller_re(ofconn->discovery, c->accept_re); | |
529 | } | |
79c9f2ee | 530 | |
7d674866 | 531 | ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit); |
76ce9432 | 532 | } |
79c9f2ee | 533 | |
76ce9432 BP |
534 | static const char * |
535 | ofconn_get_target(const struct ofconn *ofconn) | |
536 | { | |
eb15cdbb | 537 | return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn); |
76ce9432 BP |
538 | } |
539 | ||
540 | static struct ofconn * | |
541 | find_controller_by_target(struct ofproto *ofproto, const char *target) | |
542 | { | |
543 | struct ofconn *ofconn; | |
544 | ||
4e8e4213 | 545 | HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node, |
76ce9432 BP |
546 | hash_string(target, 0), &ofproto->controllers) { |
547 | if (!strcmp(ofconn_get_target(ofconn), target)) { | |
548 | return ofconn; | |
79c9f2ee | 549 | } |
064af421 | 550 | } |
76ce9432 BP |
551 | return NULL; |
552 | } | |
064af421 | 553 | |
d2ede7bc BP |
554 | static void |
555 | update_in_band_remotes(struct ofproto *ofproto) | |
556 | { | |
557 | const struct ofconn *ofconn; | |
558 | struct sockaddr_in *addrs; | |
917e50e1 | 559 | size_t max_addrs, n_addrs; |
d2ede7bc | 560 | bool discovery; |
917e50e1 | 561 | size_t i; |
d2ede7bc | 562 | |
917e50e1 BP |
563 | /* Allocate enough memory for as many remotes as we could possibly have. */ |
564 | max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers); | |
565 | addrs = xmalloc(max_addrs * sizeof *addrs); | |
d2ede7bc BP |
566 | n_addrs = 0; |
567 | ||
568 | /* Add all the remotes. */ | |
569 | discovery = false; | |
4e8e4213 | 570 | HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) { |
d2ede7bc BP |
571 | struct sockaddr_in *sin = &addrs[n_addrs]; |
572 | ||
487ec65f BP |
573 | if (ofconn->band == OFPROTO_OUT_OF_BAND) { |
574 | continue; | |
575 | } | |
576 | ||
d2ede7bc BP |
577 | sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn); |
578 | if (sin->sin_addr.s_addr) { | |
579 | sin->sin_port = rconn_get_remote_port(ofconn->rconn); | |
580 | n_addrs++; | |
581 | } | |
582 | if (ofconn->discovery) { | |
583 | discovery = true; | |
584 | } | |
585 | } | |
917e50e1 BP |
586 | for (i = 0; i < ofproto->n_extra_remotes; i++) { |
587 | addrs[n_addrs++] = ofproto->extra_in_band_remotes[i]; | |
588 | } | |
d2ede7bc BP |
589 | |
590 | /* Create or update or destroy in-band. | |
591 | * | |
592 | * Ordinarily we only enable in-band if there's at least one remote | |
593 | * address, but discovery needs the in-band rules for DHCP to be installed | |
594 | * even before we know any remote addresses. */ | |
595 | if (n_addrs || discovery) { | |
596 | if (!ofproto->in_band) { | |
597 | in_band_create(ofproto, ofproto->dpif, ofproto->switch_status, | |
598 | &ofproto->in_band); | |
599 | } | |
40cae670 BP |
600 | if (ofproto->in_band) { |
601 | in_band_set_remotes(ofproto->in_band, addrs, n_addrs); | |
602 | } | |
d2ede7bc BP |
603 | ofproto->next_in_band_update = time_msec() + 1000; |
604 | } else { | |
605 | in_band_destroy(ofproto->in_band); | |
606 | ofproto->in_band = NULL; | |
607 | } | |
608 | ||
609 | /* Clean up. */ | |
610 | free(addrs); | |
611 | } | |
612 | ||
31681a5d JP |
613 | static void |
614 | update_fail_open(struct ofproto *p) | |
615 | { | |
616 | struct ofconn *ofconn; | |
617 | ||
618 | if (!hmap_is_empty(&p->controllers) | |
619 | && p->fail_mode == OFPROTO_FAIL_STANDALONE) { | |
620 | struct rconn **rconns; | |
621 | size_t n; | |
622 | ||
623 | if (!p->fail_open) { | |
624 | p->fail_open = fail_open_create(p, p->switch_status); | |
625 | } | |
626 | ||
627 | n = 0; | |
628 | rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns); | |
4e8e4213 | 629 | HMAP_FOR_EACH (ofconn, hmap_node, &p->controllers) { |
31681a5d JP |
630 | rconns[n++] = ofconn->rconn; |
631 | } | |
632 | ||
633 | fail_open_set_controllers(p->fail_open, rconns, n); | |
634 | /* p->fail_open takes ownership of 'rconns'. */ | |
635 | } else { | |
636 | fail_open_destroy(p->fail_open); | |
637 | p->fail_open = NULL; | |
638 | } | |
639 | } | |
640 | ||
76ce9432 BP |
641 | void |
642 | ofproto_set_controllers(struct ofproto *p, | |
643 | const struct ofproto_controller *controllers, | |
644 | size_t n_controllers) | |
645 | { | |
646 | struct shash new_controllers; | |
7d674866 BP |
647 | struct ofconn *ofconn, *next_ofconn; |
648 | struct ofservice *ofservice, *next_ofservice; | |
76ce9432 | 649 | bool ss_exists; |
76ce9432 | 650 | size_t i; |
79c9f2ee | 651 | |
7d674866 BP |
652 | /* Create newly configured controllers and services. |
653 | * Create a name to ofproto_controller mapping in 'new_controllers'. */ | |
76ce9432 BP |
654 | shash_init(&new_controllers); |
655 | for (i = 0; i < n_controllers; i++) { | |
656 | const struct ofproto_controller *c = &controllers[i]; | |
657 | ||
7d674866 BP |
658 | if (!vconn_verify_name(c->target) || !strcmp(c->target, "discover")) { |
659 | if (!find_controller_by_target(p, c->target)) { | |
660 | add_controller(p, c); | |
661 | } | |
662 | } else if (!pvconn_verify_name(c->target)) { | |
663 | if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) { | |
664 | continue; | |
665 | } | |
666 | } else { | |
667 | VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"", | |
668 | dpif_name(p->dpif), c->target); | |
669 | continue; | |
76ce9432 | 670 | } |
7d674866 BP |
671 | |
672 | shash_add_once(&new_controllers, c->target, &controllers[i]); | |
76ce9432 BP |
673 | } |
674 | ||
7d674866 BP |
675 | /* Delete controllers that are no longer configured. |
676 | * Update configuration of all now-existing controllers. */ | |
76ce9432 | 677 | ss_exists = false; |
4e8e4213 | 678 | HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &p->controllers) { |
76ce9432 BP |
679 | struct ofproto_controller *c; |
680 | ||
681 | c = shash_find_data(&new_controllers, ofconn_get_target(ofconn)); | |
682 | if (!c) { | |
683 | ofconn_destroy(ofconn); | |
79c9f2ee | 684 | } else { |
76ce9432 | 685 | update_controller(ofconn, c); |
76ce9432 BP |
686 | if (ofconn->ss) { |
687 | ss_exists = true; | |
688 | } | |
76ce9432 BP |
689 | } |
690 | } | |
7d674866 BP |
691 | |
692 | /* Delete services that are no longer configured. | |
693 | * Update configuration of all now-existing services. */ | |
4e8e4213 | 694 | HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) { |
7d674866 BP |
695 | struct ofproto_controller *c; |
696 | ||
697 | c = shash_find_data(&new_controllers, | |
698 | pvconn_get_name(ofservice->pvconn)); | |
699 | if (!c) { | |
700 | ofservice_destroy(p, ofservice); | |
701 | } else { | |
702 | ofservice_reconfigure(ofservice, c); | |
703 | } | |
704 | } | |
705 | ||
76ce9432 BP |
706 | shash_destroy(&new_controllers); |
707 | ||
d2ede7bc | 708 | update_in_band_remotes(p); |
31681a5d | 709 | update_fail_open(p); |
79c9f2ee | 710 | |
76ce9432 BP |
711 | if (!hmap_is_empty(&p->controllers) && !ss_exists) { |
712 | ofconn = CONTAINER_OF(hmap_first(&p->controllers), | |
713 | struct ofconn, hmap_node); | |
714 | ofconn->ss = switch_status_register(p->switch_status, "remote", | |
715 | rconn_status_cb, ofconn->rconn); | |
79c9f2ee | 716 | } |
064af421 BP |
717 | } |
718 | ||
31681a5d JP |
719 | void |
720 | ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode) | |
721 | { | |
722 | p->fail_mode = fail_mode; | |
723 | update_fail_open(p); | |
724 | } | |
725 | ||
fa05809b BP |
726 | /* Drops the connections between 'ofproto' and all of its controllers, forcing |
727 | * them to reconnect. */ | |
728 | void | |
729 | ofproto_reconnect_controllers(struct ofproto *ofproto) | |
730 | { | |
731 | struct ofconn *ofconn; | |
732 | ||
4e8e4213 | 733 | LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) { |
fa05809b BP |
734 | rconn_reconnect(ofconn->rconn); |
735 | } | |
736 | } | |
737 | ||
917e50e1 BP |
738 | static bool |
739 | any_extras_changed(const struct ofproto *ofproto, | |
740 | const struct sockaddr_in *extras, size_t n) | |
741 | { | |
742 | size_t i; | |
743 | ||
744 | if (n != ofproto->n_extra_remotes) { | |
745 | return true; | |
746 | } | |
747 | ||
748 | for (i = 0; i < n; i++) { | |
749 | const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i]; | |
750 | const struct sockaddr_in *new = &extras[i]; | |
751 | ||
752 | if (old->sin_addr.s_addr != new->sin_addr.s_addr || | |
753 | old->sin_port != new->sin_port) { | |
754 | return true; | |
755 | } | |
756 | } | |
757 | ||
758 | return false; | |
759 | } | |
760 | ||
761 | /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s | |
762 | * in-band control should guarantee access, in the same way that in-band | |
763 | * control guarantees access to OpenFlow controllers. */ | |
764 | void | |
765 | ofproto_set_extra_in_band_remotes(struct ofproto *ofproto, | |
766 | const struct sockaddr_in *extras, size_t n) | |
767 | { | |
768 | if (!any_extras_changed(ofproto, extras, n)) { | |
769 | return; | |
770 | } | |
771 | ||
772 | free(ofproto->extra_in_band_remotes); | |
773 | ofproto->n_extra_remotes = n; | |
774 | ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras); | |
775 | ||
776 | update_in_band_remotes(ofproto); | |
777 | } | |
778 | ||
064af421 BP |
779 | void |
780 | ofproto_set_desc(struct ofproto *p, | |
5a719c38 JP |
781 | const char *mfr_desc, const char *hw_desc, |
782 | const char *sw_desc, const char *serial_desc, | |
8abc4ed7 | 783 | const char *dp_desc) |
064af421 | 784 | { |
5a719c38 JP |
785 | struct ofp_desc_stats *ods; |
786 | ||
787 | if (mfr_desc) { | |
788 | if (strlen(mfr_desc) >= sizeof ods->mfr_desc) { | |
789 | VLOG_WARN("truncating mfr_desc, must be less than %zu characters", | |
790 | sizeof ods->mfr_desc); | |
791 | } | |
792 | free(p->mfr_desc); | |
793 | p->mfr_desc = xstrdup(mfr_desc); | |
064af421 | 794 | } |
5a719c38 JP |
795 | if (hw_desc) { |
796 | if (strlen(hw_desc) >= sizeof ods->hw_desc) { | |
797 | VLOG_WARN("truncating hw_desc, must be less than %zu characters", | |
798 | sizeof ods->hw_desc); | |
799 | } | |
800 | free(p->hw_desc); | |
801 | p->hw_desc = xstrdup(hw_desc); | |
064af421 | 802 | } |
5a719c38 JP |
803 | if (sw_desc) { |
804 | if (strlen(sw_desc) >= sizeof ods->sw_desc) { | |
805 | VLOG_WARN("truncating sw_desc, must be less than %zu characters", | |
806 | sizeof ods->sw_desc); | |
807 | } | |
808 | free(p->sw_desc); | |
809 | p->sw_desc = xstrdup(sw_desc); | |
810 | } | |
811 | if (serial_desc) { | |
812 | if (strlen(serial_desc) >= sizeof ods->serial_num) { | |
813 | VLOG_WARN("truncating serial_desc, must be less than %zu " | |
814 | "characters", | |
815 | sizeof ods->serial_num); | |
816 | } | |
817 | free(p->serial_desc); | |
818 | p->serial_desc = xstrdup(serial_desc); | |
064af421 | 819 | } |
8abc4ed7 | 820 | if (dp_desc) { |
5a719c38 JP |
821 | if (strlen(dp_desc) >= sizeof ods->dp_desc) { |
822 | VLOG_WARN("truncating dp_desc, must be less than %zu characters", | |
823 | sizeof ods->dp_desc); | |
824 | } | |
8abc4ed7 JP |
825 | free(p->dp_desc); |
826 | p->dp_desc = xstrdup(dp_desc); | |
827 | } | |
064af421 BP |
828 | } |
829 | ||
064af421 BP |
830 | static int |
831 | set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp, | |
832 | const struct svec *svec) | |
833 | { | |
834 | struct pvconn **pvconns = *pvconnsp; | |
835 | size_t n_pvconns = *n_pvconnsp; | |
836 | int retval = 0; | |
837 | size_t i; | |
838 | ||
839 | for (i = 0; i < n_pvconns; i++) { | |
840 | pvconn_close(pvconns[i]); | |
841 | } | |
842 | free(pvconns); | |
843 | ||
844 | pvconns = xmalloc(svec->n * sizeof *pvconns); | |
845 | n_pvconns = 0; | |
846 | for (i = 0; i < svec->n; i++) { | |
847 | const char *name = svec->names[i]; | |
848 | struct pvconn *pvconn; | |
849 | int error; | |
850 | ||
851 | error = pvconn_open(name, &pvconn); | |
852 | if (!error) { | |
853 | pvconns[n_pvconns++] = pvconn; | |
854 | } else { | |
855 | VLOG_ERR("failed to listen on %s: %s", name, strerror(error)); | |
856 | if (!retval) { | |
857 | retval = error; | |
858 | } | |
859 | } | |
860 | } | |
861 | ||
862 | *pvconnsp = pvconns; | |
863 | *n_pvconnsp = n_pvconns; | |
864 | ||
865 | return retval; | |
866 | } | |
867 | ||
064af421 BP |
868 | int |
869 | ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops) | |
870 | { | |
871 | return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops); | |
872 | } | |
873 | ||
874 | int | |
0193b2af JG |
875 | ofproto_set_netflow(struct ofproto *ofproto, |
876 | const struct netflow_options *nf_options) | |
064af421 | 877 | { |
76343538 | 878 | if (nf_options && nf_options->collectors.n) { |
064af421 BP |
879 | if (!ofproto->netflow) { |
880 | ofproto->netflow = netflow_create(); | |
881 | } | |
0193b2af | 882 | return netflow_set_options(ofproto->netflow, nf_options); |
064af421 BP |
883 | } else { |
884 | netflow_destroy(ofproto->netflow); | |
885 | ofproto->netflow = NULL; | |
886 | return 0; | |
887 | } | |
888 | } | |
889 | ||
72b06300 BP |
890 | void |
891 | ofproto_set_sflow(struct ofproto *ofproto, | |
892 | const struct ofproto_sflow_options *oso) | |
893 | { | |
894 | struct ofproto_sflow *os = ofproto->sflow; | |
895 | if (oso) { | |
896 | if (!os) { | |
897 | struct ofport *ofport; | |
72b06300 BP |
898 | |
899 | os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif); | |
4e8e4213 | 900 | HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) { |
ca0f572c | 901 | ofproto_sflow_add_port(os, ofport->odp_port, |
72b06300 BP |
902 | netdev_get_name(ofport->netdev)); |
903 | } | |
904 | } | |
905 | ofproto_sflow_set_options(os, oso); | |
906 | } else { | |
907 | ofproto_sflow_destroy(os); | |
908 | ofproto->sflow = NULL; | |
909 | } | |
910 | } | |
911 | ||
064af421 BP |
912 | uint64_t |
913 | ofproto_get_datapath_id(const struct ofproto *ofproto) | |
914 | { | |
915 | return ofproto->datapath_id; | |
916 | } | |
917 | ||
76ce9432 | 918 | bool |
7d674866 | 919 | ofproto_has_primary_controller(const struct ofproto *ofproto) |
064af421 | 920 | { |
76ce9432 | 921 | return !hmap_is_empty(&ofproto->controllers); |
064af421 BP |
922 | } |
923 | ||
abdfe474 JP |
924 | enum ofproto_fail_mode |
925 | ofproto_get_fail_mode(const struct ofproto *p) | |
926 | { | |
927 | return p->fail_mode; | |
928 | } | |
929 | ||
064af421 BP |
930 | void |
931 | ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops) | |
932 | { | |
933 | size_t i; | |
934 | ||
935 | for (i = 0; i < ofproto->n_snoops; i++) { | |
936 | svec_add(snoops, pvconn_get_name(ofproto->snoops[i])); | |
937 | } | |
938 | } | |
939 | ||
940 | void | |
941 | ofproto_destroy(struct ofproto *p) | |
942 | { | |
7d674866 | 943 | struct ofservice *ofservice, *next_ofservice; |
064af421 | 944 | struct ofconn *ofconn, *next_ofconn; |
ca0f572c | 945 | struct ofport *ofport, *next_ofport; |
064af421 BP |
946 | size_t i; |
947 | ||
948 | if (!p) { | |
949 | return; | |
950 | } | |
951 | ||
f7de2cdf | 952 | /* Destroy fail-open and in-band early, since they touch the classifier. */ |
79c9f2ee BP |
953 | fail_open_destroy(p->fail_open); |
954 | p->fail_open = NULL; | |
955 | ||
956 | in_band_destroy(p->in_band); | |
957 | p->in_band = NULL; | |
917e50e1 | 958 | free(p->extra_in_band_remotes); |
2f6d3445 | 959 | |
064af421 BP |
960 | ofproto_flush_flows(p); |
961 | classifier_destroy(&p->cls); | |
bcf84111 | 962 | hmap_destroy(&p->facets); |
064af421 | 963 | |
4e8e4213 | 964 | LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) { |
c475ae67 | 965 | ofconn_destroy(ofconn); |
064af421 | 966 | } |
76ce9432 | 967 | hmap_destroy(&p->controllers); |
064af421 | 968 | |
c228a364 | 969 | dpif_close(p->dpif); |
e9e28be3 | 970 | netdev_monitor_destroy(p->netdev_monitor); |
4e8e4213 | 971 | HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) { |
ca0f572c | 972 | hmap_remove(&p->ports, &ofport->hmap_node); |
064af421 BP |
973 | ofport_free(ofport); |
974 | } | |
975 | shash_destroy(&p->port_by_name); | |
976 | ||
977 | switch_status_destroy(p->switch_status); | |
064af421 | 978 | netflow_destroy(p->netflow); |
72b06300 | 979 | ofproto_sflow_destroy(p->sflow); |
064af421 | 980 | |
4e8e4213 | 981 | HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) { |
7d674866 | 982 | ofservice_destroy(p, ofservice); |
064af421 | 983 | } |
7d674866 | 984 | hmap_destroy(&p->services); |
064af421 BP |
985 | |
986 | for (i = 0; i < p->n_snoops; i++) { | |
987 | pvconn_close(p->snoops[i]); | |
988 | } | |
989 | free(p->snoops); | |
990 | ||
991 | mac_learning_destroy(p->ml); | |
992 | ||
5a719c38 JP |
993 | free(p->mfr_desc); |
994 | free(p->hw_desc); | |
995 | free(p->sw_desc); | |
996 | free(p->serial_desc); | |
cb871ae0 JP |
997 | free(p->dp_desc); |
998 | ||
ca0f572c | 999 | hmap_destroy(&p->ports); |
3b917492 | 1000 | |
064af421 BP |
1001 | free(p); |
1002 | } | |
1003 | ||
1004 | int | |
1005 | ofproto_run(struct ofproto *p) | |
1006 | { | |
1007 | int error = ofproto_run1(p); | |
1008 | if (!error) { | |
1009 | error = ofproto_run2(p, false); | |
1010 | } | |
1011 | return error; | |
1012 | } | |
1013 | ||
e9e28be3 BP |
1014 | static void |
1015 | process_port_change(struct ofproto *ofproto, int error, char *devname) | |
1016 | { | |
1017 | if (error == ENOBUFS) { | |
1018 | reinit_ports(ofproto); | |
1019 | } else if (!error) { | |
1020 | update_port(ofproto, devname); | |
1021 | free(devname); | |
1022 | } | |
1023 | } | |
1024 | ||
e2bfacb6 BP |
1025 | /* Returns a "preference level" for snooping 'ofconn'. A higher return value |
1026 | * means that 'ofconn' is more interesting for monitoring than a lower return | |
1027 | * value. */ | |
1028 | static int | |
1029 | snoop_preference(const struct ofconn *ofconn) | |
1030 | { | |
1031 | switch (ofconn->role) { | |
1032 | case NX_ROLE_MASTER: | |
1033 | return 3; | |
1034 | case NX_ROLE_OTHER: | |
1035 | return 2; | |
1036 | case NX_ROLE_SLAVE: | |
1037 | return 1; | |
1038 | default: | |
1039 | /* Shouldn't happen. */ | |
1040 | return 0; | |
1041 | } | |
1042 | } | |
1043 | ||
76ce9432 BP |
1044 | /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'. |
1045 | * Connects this vconn to a controller. */ | |
1046 | static void | |
1047 | add_snooper(struct ofproto *ofproto, struct vconn *vconn) | |
1048 | { | |
e2bfacb6 | 1049 | struct ofconn *ofconn, *best; |
76ce9432 | 1050 | |
e2bfacb6 BP |
1051 | /* Pick a controller for monitoring. */ |
1052 | best = NULL; | |
4e8e4213 | 1053 | LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) { |
5899143f | 1054 | if (ofconn->type == OFCONN_PRIMARY |
e2bfacb6 BP |
1055 | && (!best || snoop_preference(ofconn) > snoop_preference(best))) { |
1056 | best = ofconn; | |
76ce9432 | 1057 | } |
e2bfacb6 | 1058 | } |
76ce9432 | 1059 | |
e2bfacb6 BP |
1060 | if (best) { |
1061 | rconn_add_monitor(best->rconn, vconn); | |
1062 | } else { | |
1063 | VLOG_INFO_RL(&rl, "no controller connection to snoop"); | |
1064 | vconn_close(vconn); | |
76ce9432 | 1065 | } |
76ce9432 BP |
1066 | } |
1067 | ||
064af421 BP |
1068 | int |
1069 | ofproto_run1(struct ofproto *p) | |
1070 | { | |
1071 | struct ofconn *ofconn, *next_ofconn; | |
7d674866 | 1072 | struct ofservice *ofservice; |
064af421 BP |
1073 | char *devname; |
1074 | int error; | |
1075 | int i; | |
1076 | ||
149f577a JG |
1077 | if (shash_is_empty(&p->port_by_name)) { |
1078 | init_ports(p); | |
1079 | } | |
1080 | ||
064af421 BP |
1081 | for (i = 0; i < 50; i++) { |
1082 | struct ofpbuf *buf; | |
064af421 | 1083 | |
c228a364 | 1084 | error = dpif_recv(p->dpif, &buf); |
064af421 BP |
1085 | if (error) { |
1086 | if (error == ENODEV) { | |
1087 | /* Someone destroyed the datapath behind our back. The caller | |
1088 | * better destroy us and give up, because we're just going to | |
1089 | * spin from here on out. */ | |
39a559f2 BP |
1090 | static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5); |
1091 | VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally", | |
c228a364 | 1092 | dpif_name(p->dpif)); |
064af421 BP |
1093 | return ENODEV; |
1094 | } | |
1095 | break; | |
1096 | } | |
1097 | ||
1098 | handle_odp_msg(p, buf); | |
1099 | } | |
1100 | ||
e9e28be3 BP |
1101 | while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) { |
1102 | process_port_change(p, error, devname); | |
1103 | } | |
1104 | while ((error = netdev_monitor_poll(p->netdev_monitor, | |
1105 | &devname)) != EAGAIN) { | |
1106 | process_port_change(p, error, devname); | |
064af421 BP |
1107 | } |
1108 | ||
1109 | if (p->in_band) { | |
d2ede7bc BP |
1110 | if (time_msec() >= p->next_in_band_update) { |
1111 | update_in_band_remotes(p); | |
1112 | } | |
064af421 BP |
1113 | in_band_run(p->in_band); |
1114 | } | |
064af421 | 1115 | |
4e8e4213 | 1116 | LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) { |
3269c562 | 1117 | ofconn_run(ofconn); |
064af421 BP |
1118 | } |
1119 | ||
7778bd15 BP |
1120 | /* Fail-open maintenance. Do this after processing the ofconns since |
1121 | * fail-open checks the status of the controller rconn. */ | |
1122 | if (p->fail_open) { | |
1123 | fail_open_run(p->fail_open); | |
1124 | } | |
1125 | ||
4e8e4213 | 1126 | HMAP_FOR_EACH (ofservice, node, &p->services) { |
064af421 BP |
1127 | struct vconn *vconn; |
1128 | int retval; | |
1129 | ||
7d674866 | 1130 | retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn); |
064af421 | 1131 | if (!retval) { |
9794e806 | 1132 | struct rconn *rconn; |
eb15cdbb | 1133 | char *name; |
9794e806 | 1134 | |
7d674866 | 1135 | rconn = rconn_create(ofservice->probe_interval, 0); |
eb15cdbb BP |
1136 | name = ofconn_make_name(p, vconn_get_name(vconn)); |
1137 | rconn_connect_unreliably(rconn, vconn, name); | |
1138 | free(name); | |
1139 | ||
7d674866 BP |
1140 | ofconn = ofconn_create(p, rconn, OFCONN_SERVICE); |
1141 | ofconn_set_rate_limit(ofconn, ofservice->rate_limit, | |
1142 | ofservice->burst_limit); | |
064af421 BP |
1143 | } else if (retval != EAGAIN) { |
1144 | VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval)); | |
1145 | } | |
1146 | } | |
1147 | ||
1148 | for (i = 0; i < p->n_snoops; i++) { | |
1149 | struct vconn *vconn; | |
1150 | int retval; | |
1151 | ||
1152 | retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn); | |
1153 | if (!retval) { | |
76ce9432 | 1154 | add_snooper(p, vconn); |
064af421 BP |
1155 | } else if (retval != EAGAIN) { |
1156 | VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval)); | |
1157 | } | |
1158 | } | |
1159 | ||
1160 | if (time_msec() >= p->next_expiration) { | |
0de7a4b4 BP |
1161 | int delay = ofproto_expire(p); |
1162 | p->next_expiration = time_msec() + delay; | |
064af421 | 1163 | COVERAGE_INC(ofproto_expiration); |
064af421 BP |
1164 | } |
1165 | ||
1166 | if (p->netflow) { | |
1167 | netflow_run(p->netflow); | |
1168 | } | |
72b06300 BP |
1169 | if (p->sflow) { |
1170 | ofproto_sflow_run(p->sflow); | |
1171 | } | |
064af421 BP |
1172 | |
1173 | return 0; | |
1174 | } | |
1175 | ||
064af421 BP |
1176 | int |
1177 | ofproto_run2(struct ofproto *p, bool revalidate_all) | |
1178 | { | |
bcf84111 BP |
1179 | /* Figure out what we need to revalidate now, if anything. */ |
1180 | struct tag_set revalidate_set = p->revalidate_set; | |
1181 | if (p->need_revalidate) { | |
1182 | revalidate_all = true; | |
1183 | } | |
1184 | ||
1185 | /* Clear the revalidation flags. */ | |
1186 | tag_set_init(&p->revalidate_set); | |
1187 | p->need_revalidate = false; | |
1188 | ||
1189 | /* Now revalidate if there's anything to do. */ | |
1190 | if (revalidate_all || !tag_set_is_empty(&revalidate_set)) { | |
1191 | struct facet *facet, *next; | |
1192 | ||
1193 | HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &p->facets) { | |
1194 | if (revalidate_all | |
1195 | || tag_set_intersects(&revalidate_set, facet->tags)) { | |
1196 | facet_revalidate(p, facet); | |
1197 | } | |
1198 | } | |
064af421 BP |
1199 | } |
1200 | ||
1201 | return 0; | |
1202 | } | |
1203 | ||
1204 | void | |
1205 | ofproto_wait(struct ofproto *p) | |
1206 | { | |
7d674866 | 1207 | struct ofservice *ofservice; |
064af421 BP |
1208 | struct ofconn *ofconn; |
1209 | size_t i; | |
1210 | ||
c228a364 | 1211 | dpif_recv_wait(p->dpif); |
e9e28be3 BP |
1212 | dpif_port_poll_wait(p->dpif); |
1213 | netdev_monitor_poll_wait(p->netdev_monitor); | |
4e8e4213 | 1214 | LIST_FOR_EACH (ofconn, node, &p->all_conns) { |
064af421 BP |
1215 | ofconn_wait(ofconn); |
1216 | } | |
1217 | if (p->in_band) { | |
7cf8b266 | 1218 | poll_timer_wait_until(p->next_in_band_update); |
064af421 BP |
1219 | in_band_wait(p->in_band); |
1220 | } | |
064af421 BP |
1221 | if (p->fail_open) { |
1222 | fail_open_wait(p->fail_open); | |
1223 | } | |
72b06300 BP |
1224 | if (p->sflow) { |
1225 | ofproto_sflow_wait(p->sflow); | |
1226 | } | |
064af421 BP |
1227 | if (!tag_set_is_empty(&p->revalidate_set)) { |
1228 | poll_immediate_wake(); | |
1229 | } | |
1230 | if (p->need_revalidate) { | |
1231 | /* Shouldn't happen, but if it does just go around again. */ | |
1232 | VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); | |
1233 | poll_immediate_wake(); | |
1234 | } else if (p->next_expiration != LLONG_MAX) { | |
7cf8b266 | 1235 | poll_timer_wait_until(p->next_expiration); |
064af421 | 1236 | } |
4e8e4213 | 1237 | HMAP_FOR_EACH (ofservice, node, &p->services) { |
7d674866 | 1238 | pvconn_wait(ofservice->pvconn); |
064af421 BP |
1239 | } |
1240 | for (i = 0; i < p->n_snoops; i++) { | |
1241 | pvconn_wait(p->snoops[i]); | |
1242 | } | |
1243 | } | |
1244 | ||
1245 | void | |
1246 | ofproto_revalidate(struct ofproto *ofproto, tag_type tag) | |
1247 | { | |
1248 | tag_set_add(&ofproto->revalidate_set, tag); | |
1249 | } | |
1250 | ||
1251 | struct tag_set * | |
1252 | ofproto_get_revalidate_set(struct ofproto *ofproto) | |
1253 | { | |
1254 | return &ofproto->revalidate_set; | |
1255 | } | |
1256 | ||
1257 | bool | |
1258 | ofproto_is_alive(const struct ofproto *p) | |
1259 | { | |
76ce9432 | 1260 | return !hmap_is_empty(&p->controllers); |
064af421 BP |
1261 | } |
1262 | ||
3a6ccc8c BP |
1263 | /* Deletes port number 'odp_port' from the datapath for 'ofproto'. |
1264 | * | |
1265 | * This is almost the same as calling dpif_port_del() directly on the | |
1266 | * datapath, but it also makes 'ofproto' close its open netdev for the port | |
1267 | * (if any). This makes it possible to create a new netdev of a different | |
1268 | * type under the same name, which otherwise the netdev library would refuse | |
1269 | * to do because of the conflict. (The netdev would eventually get closed on | |
1270 | * the next trip through ofproto_run(), but this interface is more direct.) | |
1271 | * | |
3a6ccc8c BP |
1272 | * Returns 0 if successful, otherwise a positive errno. */ |
1273 | int | |
1274 | ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port) | |
1275 | { | |
1276 | struct ofport *ofport = get_port(ofproto, odp_port); | |
1277 | const char *name = ofport ? (char *) ofport->opp.name : "<unknown>"; | |
1278 | int error; | |
1279 | ||
1280 | error = dpif_port_del(ofproto->dpif, odp_port); | |
1281 | if (error) { | |
1282 | VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)", | |
1283 | dpif_name(ofproto->dpif), odp_port, name, strerror(error)); | |
1284 | } else if (ofport) { | |
1285 | /* 'name' is ofport->opp.name and update_port() is going to destroy | |
1286 | * 'ofport'. Just in case update_port() refers to 'name' after it | |
1287 | * destroys 'ofport', make a copy of it around the update_port() | |
1288 | * call. */ | |
1289 | char *devname = xstrdup(name); | |
1290 | update_port(ofproto, devname); | |
1291 | free(devname); | |
1292 | } | |
1293 | return error; | |
1294 | } | |
1295 | ||
a4e2e1f2 EJ |
1296 | /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns |
1297 | * true if 'odp_port' exists and should be included, false otherwise. */ | |
1298 | bool | |
1299 | ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port) | |
1300 | { | |
1301 | struct ofport *ofport = get_port(ofproto, odp_port); | |
1302 | return ofport && !(ofport->opp.config & OFPPC_NO_FLOOD); | |
1303 | } | |
1304 | ||
064af421 | 1305 | int |
ae412e7d | 1306 | ofproto_send_packet(struct ofproto *p, const struct flow *flow, |
064af421 BP |
1307 | const union ofp_action *actions, size_t n_actions, |
1308 | const struct ofpbuf *packet) | |
1309 | { | |
1310 | struct odp_actions odp_actions; | |
1311 | int error; | |
1312 | ||
1313 | error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions, | |
6a07af36 | 1314 | NULL, NULL, NULL); |
064af421 BP |
1315 | if (error) { |
1316 | return error; | |
1317 | } | |
1318 | ||
1319 | /* XXX Should we translate the dpif_execute() errno value into an OpenFlow | |
1320 | * error code? */ | |
f1588b1f | 1321 | dpif_execute(p->dpif, odp_actions.actions, odp_actions.n_actions, packet); |
064af421 BP |
1322 | return 0; |
1323 | } | |
1324 | ||
fa8b054f BP |
1325 | /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and |
1326 | * performs the 'n_actions' actions in 'actions'. The new flow will not | |
1327 | * timeout. | |
1328 | * | |
1329 | * If cls_rule->priority is in the range of priorities supported by OpenFlow | |
1330 | * (0...65535, inclusive) then the flow will be visible to OpenFlow | |
1331 | * controllers; otherwise, it will be hidden. | |
1332 | * | |
1333 | * The caller retains ownership of 'cls_rule' and 'actions'. */ | |
064af421 | 1334 | void |
cf3fad8a | 1335 | ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule, |
fa8b054f | 1336 | const union ofp_action *actions, size_t n_actions) |
064af421 BP |
1337 | { |
1338 | struct rule *rule; | |
bcf84111 | 1339 | rule = rule_create(cls_rule, actions, n_actions, 0, 0, 0, false); |
afe75089 | 1340 | rule_insert(p, rule); |
064af421 BP |
1341 | } |
1342 | ||
1343 | void | |
cf3fad8a | 1344 | ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target) |
064af421 BP |
1345 | { |
1346 | struct rule *rule; | |
1347 | ||
1348 | rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls, | |
cf3fad8a | 1349 | target)); |
064af421 BP |
1350 | if (rule) { |
1351 | rule_remove(ofproto, rule); | |
1352 | } | |
1353 | } | |
1354 | ||
1355 | static void | |
1356 | destroy_rule(struct cls_rule *rule_, void *ofproto_) | |
1357 | { | |
1358 | struct rule *rule = rule_from_cls_rule(rule_); | |
1359 | struct ofproto *ofproto = ofproto_; | |
1360 | ||
064af421 BP |
1361 | rule_remove(ofproto, rule); |
1362 | } | |
1363 | ||
1364 | void | |
1365 | ofproto_flush_flows(struct ofproto *ofproto) | |
1366 | { | |
bcf84111 BP |
1367 | struct facet *facet, *next_facet; |
1368 | ||
064af421 | 1369 | COVERAGE_INC(ofproto_flush); |
bcf84111 BP |
1370 | |
1371 | HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) { | |
1372 | /* Mark the facet as not installed so that facet_remove() doesn't | |
1373 | * bother trying to uninstall it. There is no point in uninstalling it | |
1374 | * individually since we are about to blow away all the facets with | |
1375 | * dpif_flow_flush(). */ | |
1376 | facet->installed = false; | |
1377 | facet_remove(ofproto, facet); | |
1378 | } | |
3c4486a5 | 1379 | classifier_for_each(&ofproto->cls, destroy_rule, ofproto); |
c228a364 | 1380 | dpif_flow_flush(ofproto->dpif); |
064af421 BP |
1381 | if (ofproto->in_band) { |
1382 | in_band_flushed(ofproto->in_band); | |
1383 | } | |
1384 | if (ofproto->fail_open) { | |
1385 | fail_open_flushed(ofproto->fail_open); | |
1386 | } | |
1387 | } | |
1388 | \f | |
1389 | static void | |
1390 | reinit_ports(struct ofproto *p) | |
1391 | { | |
1392 | struct svec devnames; | |
1393 | struct ofport *ofport; | |
064af421 BP |
1394 | struct odp_port *odp_ports; |
1395 | size_t n_odp_ports; | |
1396 | size_t i; | |
1397 | ||
898bf89d JP |
1398 | COVERAGE_INC(ofproto_reinit_ports); |
1399 | ||
064af421 | 1400 | svec_init(&devnames); |
4e8e4213 | 1401 | HMAP_FOR_EACH (ofport, hmap_node, &p->ports) { |
064af421 BP |
1402 | svec_add (&devnames, (char *) ofport->opp.name); |
1403 | } | |
c228a364 | 1404 | dpif_port_list(p->dpif, &odp_ports, &n_odp_ports); |
064af421 BP |
1405 | for (i = 0; i < n_odp_ports; i++) { |
1406 | svec_add (&devnames, odp_ports[i].devname); | |
1407 | } | |
1408 | free(odp_ports); | |
1409 | ||
1410 | svec_sort_unique(&devnames); | |
1411 | for (i = 0; i < devnames.n; i++) { | |
1412 | update_port(p, devnames.names[i]); | |
1413 | } | |
1414 | svec_destroy(&devnames); | |
1415 | } | |
1416 | ||
064af421 BP |
1417 | static struct ofport * |
1418 | make_ofport(const struct odp_port *odp_port) | |
1419 | { | |
149f577a | 1420 | struct netdev_options netdev_options; |
064af421 BP |
1421 | enum netdev_flags flags; |
1422 | struct ofport *ofport; | |
1423 | struct netdev *netdev; | |
064af421 BP |
1424 | int error; |
1425 | ||
149f577a JG |
1426 | memset(&netdev_options, 0, sizeof netdev_options); |
1427 | netdev_options.name = odp_port->devname; | |
1428 | netdev_options.ethertype = NETDEV_ETH_TYPE_NONE; | |
149f577a JG |
1429 | |
1430 | error = netdev_open(&netdev_options, &netdev); | |
064af421 BP |
1431 | if (error) { |
1432 | VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s " | |
1433 | "cannot be opened (%s)", | |
1434 | odp_port->devname, odp_port->port, | |
1435 | odp_port->devname, strerror(error)); | |
1436 | return NULL; | |
1437 | } | |
1438 | ||
1439 | ofport = xmalloc(sizeof *ofport); | |
1440 | ofport->netdev = netdev; | |
ca0f572c | 1441 | ofport->odp_port = odp_port->port; |
064af421 | 1442 | ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port); |
80992a35 | 1443 | netdev_get_etheraddr(netdev, ofport->opp.hw_addr); |
064af421 BP |
1444 | memcpy(ofport->opp.name, odp_port->devname, |
1445 | MIN(sizeof ofport->opp.name, sizeof odp_port->devname)); | |
1446 | ofport->opp.name[sizeof ofport->opp.name - 1] = '\0'; | |
1447 | ||
1448 | netdev_get_flags(netdev, &flags); | |
1449 | ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN; | |
1450 | ||
85da620e | 1451 | ofport->opp.state = netdev_get_carrier(netdev) ? 0 : OFPPS_LINK_DOWN; |
064af421 BP |
1452 | |
1453 | netdev_get_features(netdev, | |
1454 | &ofport->opp.curr, &ofport->opp.advertised, | |
1455 | &ofport->opp.supported, &ofport->opp.peer); | |
1456 | return ofport; | |
1457 | } | |
1458 | ||
1459 | static bool | |
1460 | ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port) | |
1461 | { | |
ca0f572c | 1462 | if (get_port(p, odp_port->port)) { |
064af421 BP |
1463 | VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath", |
1464 | odp_port->port); | |
1465 | return true; | |
1466 | } else if (shash_find(&p->port_by_name, odp_port->devname)) { | |
1467 | VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath", | |
1468 | odp_port->devname); | |
1469 | return true; | |
1470 | } else { | |
1471 | return false; | |
1472 | } | |
1473 | } | |
1474 | ||
1475 | static int | |
1476 | ofport_equal(const struct ofport *a_, const struct ofport *b_) | |
1477 | { | |
1478 | const struct ofp_phy_port *a = &a_->opp; | |
1479 | const struct ofp_phy_port *b = &b_->opp; | |
1480 | ||
1481 | BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */ | |
1482 | return (a->port_no == b->port_no | |
1483 | && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr) | |
1484 | && !strcmp((char *) a->name, (char *) b->name) | |
1485 | && a->state == b->state | |
1486 | && a->config == b->config | |
1487 | && a->curr == b->curr | |
1488 | && a->advertised == b->advertised | |
1489 | && a->supported == b->supported | |
1490 | && a->peer == b->peer); | |
1491 | } | |
1492 | ||
1493 | static void | |
1494 | send_port_status(struct ofproto *p, const struct ofport *ofport, | |
1495 | uint8_t reason) | |
1496 | { | |
1497 | /* XXX Should limit the number of queued port status change messages. */ | |
1498 | struct ofconn *ofconn; | |
4e8e4213 | 1499 | LIST_FOR_EACH (ofconn, node, &p->all_conns) { |
064af421 BP |
1500 | struct ofp_port_status *ops; |
1501 | struct ofpbuf *b; | |
1502 | ||
c91248b3 | 1503 | if (!ofconn_receives_async_msgs(ofconn)) { |
9deba63b BP |
1504 | continue; |
1505 | } | |
1506 | ||
064af421 BP |
1507 | ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b); |
1508 | ops->reason = reason; | |
1509 | ops->desc = ofport->opp; | |
1510 | hton_ofp_phy_port(&ops->desc); | |
1511 | queue_tx(b, ofconn, NULL); | |
1512 | } | |
064af421 BP |
1513 | } |
1514 | ||
1515 | static void | |
1516 | ofport_install(struct ofproto *p, struct ofport *ofport) | |
1517 | { | |
72b06300 BP |
1518 | const char *netdev_name = (const char *) ofport->opp.name; |
1519 | ||
e9e28be3 | 1520 | netdev_monitor_add(p->netdev_monitor, ofport->netdev); |
ca0f572c | 1521 | hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0)); |
72b06300 BP |
1522 | shash_add(&p->port_by_name, netdev_name, ofport); |
1523 | if (p->sflow) { | |
ca0f572c | 1524 | ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name); |
72b06300 | 1525 | } |
064af421 BP |
1526 | } |
1527 | ||
1528 | static void | |
1529 | ofport_remove(struct ofproto *p, struct ofport *ofport) | |
1530 | { | |
e9e28be3 | 1531 | netdev_monitor_remove(p->netdev_monitor, ofport->netdev); |
ca0f572c | 1532 | hmap_remove(&p->ports, &ofport->hmap_node); |
064af421 BP |
1533 | shash_delete(&p->port_by_name, |
1534 | shash_find(&p->port_by_name, (char *) ofport->opp.name)); | |
72b06300 | 1535 | if (p->sflow) { |
ca0f572c | 1536 | ofproto_sflow_del_port(p->sflow, ofport->odp_port); |
72b06300 | 1537 | } |
064af421 BP |
1538 | } |
1539 | ||
1540 | static void | |
1541 | ofport_free(struct ofport *ofport) | |
1542 | { | |
1543 | if (ofport) { | |
1544 | netdev_close(ofport->netdev); | |
1545 | free(ofport); | |
1546 | } | |
1547 | } | |
1548 | ||
ca0f572c BP |
1549 | static struct ofport * |
1550 | get_port(const struct ofproto *ofproto, uint16_t odp_port) | |
1551 | { | |
1552 | struct ofport *port; | |
1553 | ||
4e8e4213 | 1554 | HMAP_FOR_EACH_IN_BUCKET (port, hmap_node, |
ca0f572c BP |
1555 | hash_int(odp_port, 0), &ofproto->ports) { |
1556 | if (port->odp_port == odp_port) { | |
1557 | return port; | |
1558 | } | |
1559 | } | |
1560 | return NULL; | |
1561 | } | |
1562 | ||
064af421 BP |
1563 | static void |
1564 | update_port(struct ofproto *p, const char *devname) | |
1565 | { | |
1566 | struct odp_port odp_port; | |
c874dc6d BP |
1567 | struct ofport *old_ofport; |
1568 | struct ofport *new_ofport; | |
064af421 BP |
1569 | int error; |
1570 | ||
1571 | COVERAGE_INC(ofproto_update_port); | |
c874dc6d BP |
1572 | |
1573 | /* Query the datapath for port information. */ | |
c228a364 | 1574 | error = dpif_port_query_by_name(p->dpif, devname, &odp_port); |
064af421 | 1575 | |
c874dc6d BP |
1576 | /* Find the old ofport. */ |
1577 | old_ofport = shash_find_data(&p->port_by_name, devname); | |
1578 | if (!error) { | |
1579 | if (!old_ofport) { | |
1580 | /* There's no port named 'devname' but there might be a port with | |
1581 | * the same port number. This could happen if a port is deleted | |
1582 | * and then a new one added in its place very quickly, or if a port | |
1583 | * is renamed. In the former case we want to send an OFPPR_DELETE | |
1584 | * and an OFPPR_ADD, and in the latter case we want to send a | |
1585 | * single OFPPR_MODIFY. We can distinguish the cases by comparing | |
1586 | * the old port's ifindex against the new port, or perhaps less | |
1587 | * reliably but more portably by comparing the old port's MAC | |
1588 | * against the new port's MAC. However, this code isn't that smart | |
1589 | * and always sends an OFPPR_MODIFY (XXX). */ | |
ca0f572c | 1590 | old_ofport = get_port(p, odp_port.port); |
064af421 | 1591 | } |
c874dc6d | 1592 | } else if (error != ENOENT && error != ENODEV) { |
064af421 BP |
1593 | VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error " |
1594 | "%s", strerror(error)); | |
1595 | return; | |
1596 | } | |
c874dc6d BP |
1597 | |
1598 | /* Create a new ofport. */ | |
1599 | new_ofport = !error ? make_ofport(&odp_port) : NULL; | |
1600 | ||
1601 | /* Eliminate a few pathological cases. */ | |
1602 | if (!old_ofport && !new_ofport) { | |
1603 | return; | |
1604 | } else if (old_ofport && new_ofport) { | |
1605 | /* Most of the 'config' bits are OpenFlow soft state, but | |
1606 | * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow | |
1607 | * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and | |
1608 | * leaves the other bits 0.) */ | |
1609 | new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN; | |
1610 | ||
1611 | if (ofport_equal(old_ofport, new_ofport)) { | |
1612 | /* False alarm--no change. */ | |
1613 | ofport_free(new_ofport); | |
1614 | return; | |
1615 | } | |
1616 | } | |
1617 | ||
1618 | /* Now deal with the normal cases. */ | |
1619 | if (old_ofport) { | |
1620 | ofport_remove(p, old_ofport); | |
1621 | } | |
1622 | if (new_ofport) { | |
1623 | ofport_install(p, new_ofport); | |
1624 | } | |
1625 | send_port_status(p, new_ofport ? new_ofport : old_ofport, | |
1626 | (!old_ofport ? OFPPR_ADD | |
1627 | : !new_ofport ? OFPPR_DELETE | |
1628 | : OFPPR_MODIFY)); | |
1629 | ofport_free(old_ofport); | |
064af421 BP |
1630 | } |
1631 | ||
1632 | static int | |
1633 | init_ports(struct ofproto *p) | |
1634 | { | |
1635 | struct odp_port *ports; | |
1636 | size_t n_ports; | |
1637 | size_t i; | |
1638 | int error; | |
1639 | ||
c228a364 | 1640 | error = dpif_port_list(p->dpif, &ports, &n_ports); |
064af421 BP |
1641 | if (error) { |
1642 | return error; | |
1643 | } | |
1644 | ||
1645 | for (i = 0; i < n_ports; i++) { | |
1646 | const struct odp_port *odp_port = &ports[i]; | |
1647 | if (!ofport_conflicts(p, odp_port)) { | |
1648 | struct ofport *ofport = make_ofport(odp_port); | |
1649 | if (ofport) { | |
1650 | ofport_install(p, ofport); | |
1651 | } | |
1652 | } | |
1653 | } | |
1654 | free(ports); | |
064af421 BP |
1655 | return 0; |
1656 | } | |
1657 | \f | |
1658 | static struct ofconn * | |
76ce9432 | 1659 | ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type) |
064af421 | 1660 | { |
76ce9432 BP |
1661 | struct ofconn *ofconn = xzalloc(sizeof *ofconn); |
1662 | ofconn->ofproto = p; | |
064af421 BP |
1663 | list_push_back(&p->all_conns, &ofconn->node); |
1664 | ofconn->rconn = rconn; | |
76ce9432 | 1665 | ofconn->type = type; |
b70eac89 | 1666 | ofconn->flow_format = NXFF_OPENFLOW10; |
9deba63b | 1667 | ofconn->role = NX_ROLE_OTHER; |
76ce9432 | 1668 | ofconn->packet_in_counter = rconn_packet_counter_create (); |
064af421 | 1669 | ofconn->pktbuf = NULL; |
064af421 | 1670 | ofconn->miss_send_len = 0; |
064af421 BP |
1671 | ofconn->reply_counter = rconn_packet_counter_create (); |
1672 | return ofconn; | |
1673 | } | |
1674 | ||
1675 | static void | |
c475ae67 | 1676 | ofconn_destroy(struct ofconn *ofconn) |
064af421 | 1677 | { |
5899143f | 1678 | if (ofconn->type == OFCONN_PRIMARY) { |
76ce9432 BP |
1679 | hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node); |
1680 | } | |
1681 | discovery_destroy(ofconn->discovery); | |
1682 | ||
064af421 | 1683 | list_remove(&ofconn->node); |
76ce9432 | 1684 | switch_status_unregister(ofconn->ss); |
064af421 BP |
1685 | rconn_destroy(ofconn->rconn); |
1686 | rconn_packet_counter_destroy(ofconn->packet_in_counter); | |
1687 | rconn_packet_counter_destroy(ofconn->reply_counter); | |
1688 | pktbuf_destroy(ofconn->pktbuf); | |
1689 | free(ofconn); | |
1690 | } | |
1691 | ||
1692 | static void | |
3269c562 | 1693 | ofconn_run(struct ofconn *ofconn) |
064af421 | 1694 | { |
3269c562 | 1695 | struct ofproto *p = ofconn->ofproto; |
064af421 | 1696 | int iteration; |
76ce9432 BP |
1697 | size_t i; |
1698 | ||
1699 | if (ofconn->discovery) { | |
1700 | char *controller_name; | |
1701 | if (rconn_is_connectivity_questionable(ofconn->rconn)) { | |
1702 | discovery_question_connectivity(ofconn->discovery); | |
1703 | } | |
1704 | if (discovery_run(ofconn->discovery, &controller_name)) { | |
1705 | if (controller_name) { | |
eb15cdbb BP |
1706 | char *ofconn_name = ofconn_make_name(p, controller_name); |
1707 | rconn_connect(ofconn->rconn, controller_name, ofconn_name); | |
1708 | free(ofconn_name); | |
76ce9432 BP |
1709 | } else { |
1710 | rconn_disconnect(ofconn->rconn); | |
1711 | } | |
1712 | } | |
1713 | } | |
1714 | ||
1715 | for (i = 0; i < N_SCHEDULERS; i++) { | |
1716 | pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn); | |
1717 | } | |
064af421 BP |
1718 | |
1719 | rconn_run(ofconn->rconn); | |
1720 | ||
1721 | if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) { | |
1722 | /* Limit the number of iterations to prevent other tasks from | |
1723 | * starving. */ | |
1724 | for (iteration = 0; iteration < 50; iteration++) { | |
1725 | struct ofpbuf *of_msg = rconn_recv(ofconn->rconn); | |
1726 | if (!of_msg) { | |
1727 | break; | |
1728 | } | |
7778bd15 BP |
1729 | if (p->fail_open) { |
1730 | fail_open_maybe_recover(p->fail_open); | |
1731 | } | |
3269c562 | 1732 | handle_openflow(ofconn, of_msg); |
064af421 BP |
1733 | ofpbuf_delete(of_msg); |
1734 | } | |
1735 | } | |
1736 | ||
76ce9432 | 1737 | if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) { |
c475ae67 | 1738 | ofconn_destroy(ofconn); |
064af421 BP |
1739 | } |
1740 | } | |
1741 | ||
1742 | static void | |
1743 | ofconn_wait(struct ofconn *ofconn) | |
1744 | { | |
76ce9432 BP |
1745 | int i; |
1746 | ||
1747 | if (ofconn->discovery) { | |
1748 | discovery_wait(ofconn->discovery); | |
1749 | } | |
1750 | for (i = 0; i < N_SCHEDULERS; i++) { | |
1751 | pinsched_wait(ofconn->schedulers[i]); | |
1752 | } | |
064af421 BP |
1753 | rconn_run_wait(ofconn->rconn); |
1754 | if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) { | |
1755 | rconn_recv_wait(ofconn->rconn); | |
1756 | } else { | |
1757 | COVERAGE_INC(ofproto_ofconn_stuck); | |
1758 | } | |
1759 | } | |
c91248b3 BP |
1760 | |
1761 | /* Returns true if 'ofconn' should receive asynchronous messages. */ | |
1762 | static bool | |
1763 | ofconn_receives_async_msgs(const struct ofconn *ofconn) | |
1764 | { | |
5899143f BP |
1765 | if (ofconn->type == OFCONN_PRIMARY) { |
1766 | /* Primary controllers always get asynchronous messages unless they | |
c91248b3 BP |
1767 | * have configured themselves as "slaves". */ |
1768 | return ofconn->role != NX_ROLE_SLAVE; | |
1769 | } else { | |
5899143f BP |
1770 | /* Service connections don't get asynchronous messages unless they have |
1771 | * explicitly asked for them by setting a nonzero miss send length. */ | |
c91248b3 BP |
1772 | return ofconn->miss_send_len > 0; |
1773 | } | |
1774 | } | |
eb15cdbb BP |
1775 | |
1776 | /* Returns a human-readable name for an OpenFlow connection between 'ofproto' | |
1777 | * and 'target', suitable for use in log messages for identifying the | |
1778 | * connection. | |
1779 | * | |
1780 | * The name is dynamically allocated. The caller should free it (with free()) | |
1781 | * when it is no longer needed. */ | |
1782 | static char * | |
1783 | ofconn_make_name(const struct ofproto *ofproto, const char *target) | |
1784 | { | |
1785 | return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target); | |
1786 | } | |
7d674866 BP |
1787 | |
1788 | static void | |
1789 | ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst) | |
1790 | { | |
1791 | int i; | |
1792 | ||
1793 | for (i = 0; i < N_SCHEDULERS; i++) { | |
1794 | struct pinsched **s = &ofconn->schedulers[i]; | |
1795 | ||
1796 | if (rate > 0) { | |
1797 | if (!*s) { | |
1798 | *s = pinsched_create(rate, burst, | |
1799 | ofconn->ofproto->switch_status); | |
1800 | } else { | |
1801 | pinsched_set_limits(*s, rate, burst); | |
1802 | } | |
1803 | } else { | |
1804 | pinsched_destroy(*s); | |
1805 | *s = NULL; | |
1806 | } | |
1807 | } | |
1808 | } | |
1809 | \f | |
1810 | static void | |
1811 | ofservice_reconfigure(struct ofservice *ofservice, | |
1812 | const struct ofproto_controller *c) | |
1813 | { | |
1814 | ofservice->probe_interval = c->probe_interval; | |
1815 | ofservice->rate_limit = c->rate_limit; | |
1816 | ofservice->burst_limit = c->burst_limit; | |
1817 | } | |
1818 | ||
1819 | /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a | |
1820 | * positive errno value. */ | |
1821 | static int | |
1822 | ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c) | |
1823 | { | |
1824 | struct ofservice *ofservice; | |
1825 | struct pvconn *pvconn; | |
1826 | int error; | |
1827 | ||
1828 | error = pvconn_open(c->target, &pvconn); | |
1829 | if (error) { | |
1830 | return error; | |
1831 | } | |
1832 | ||
1833 | ofservice = xzalloc(sizeof *ofservice); | |
1834 | hmap_insert(&ofproto->services, &ofservice->node, | |
1835 | hash_string(c->target, 0)); | |
1836 | ofservice->pvconn = pvconn; | |
1837 | ||
1838 | ofservice_reconfigure(ofservice, c); | |
1839 | ||
1840 | return 0; | |
1841 | } | |
1842 | ||
1843 | static void | |
1844 | ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice) | |
1845 | { | |
1846 | hmap_remove(&ofproto->services, &ofservice->node); | |
1847 | pvconn_close(ofservice->pvconn); | |
1848 | free(ofservice); | |
1849 | } | |
1850 | ||
1851 | /* Finds and returns the ofservice within 'ofproto' that has the given | |
1852 | * 'target', or a null pointer if none exists. */ | |
1853 | static struct ofservice * | |
1854 | ofservice_lookup(struct ofproto *ofproto, const char *target) | |
1855 | { | |
1856 | struct ofservice *ofservice; | |
1857 | ||
4e8e4213 BP |
1858 | HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0), |
1859 | &ofproto->services) { | |
7d674866 BP |
1860 | if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) { |
1861 | return ofservice; | |
1862 | } | |
1863 | } | |
1864 | return NULL; | |
1865 | } | |
064af421 | 1866 | \f |
bcf84111 BP |
1867 | /* Returns true if 'rule' should be hidden from the controller. |
1868 | * | |
1869 | * Rules with priority higher than UINT16_MAX are set up by ofproto itself | |
1870 | * (e.g. by in-band control) and are intentionally hidden from the | |
1871 | * controller. */ | |
1872 | static bool | |
1873 | rule_is_hidden(const struct rule *rule) | |
1874 | { | |
1875 | return rule->cr.priority > UINT16_MAX; | |
1876 | } | |
1877 | ||
1878 | /* Creates and returns a new rule initialized as specified. | |
1879 | * | |
1880 | * The caller is responsible for inserting the rule into the classifier (with | |
1881 | * rule_insert()). */ | |
064af421 | 1882 | static struct rule * |
bcf84111 | 1883 | rule_create(const struct cls_rule *cls_rule, |
064af421 | 1884 | const union ofp_action *actions, size_t n_actions, |
ca069229 | 1885 | uint16_t idle_timeout, uint16_t hard_timeout, |
8054fc48 | 1886 | ovs_be64 flow_cookie, bool send_flow_removed) |
064af421 | 1887 | { |
ec6fde61 | 1888 | struct rule *rule = xzalloc(sizeof *rule); |
bcf84111 | 1889 | rule->cr = *cls_rule; |
064af421 BP |
1890 | rule->idle_timeout = idle_timeout; |
1891 | rule->hard_timeout = hard_timeout; | |
39997502 | 1892 | rule->flow_cookie = flow_cookie; |
064af421 | 1893 | rule->used = rule->created = time_msec(); |
ca069229 | 1894 | rule->send_flow_removed = send_flow_removed; |
bcf84111 | 1895 | list_init(&rule->facets); |
3dffcf07 BP |
1896 | if (n_actions > 0) { |
1897 | rule->n_actions = n_actions; | |
1898 | rule->actions = xmemdup(actions, n_actions * sizeof *actions); | |
1899 | } | |
0193b2af | 1900 | |
064af421 BP |
1901 | return rule; |
1902 | } | |
1903 | ||
1904 | static struct rule * | |
1905 | rule_from_cls_rule(const struct cls_rule *cls_rule) | |
1906 | { | |
1907 | return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL; | |
1908 | } | |
1909 | ||
1910 | static void | |
1911 | rule_free(struct rule *rule) | |
1912 | { | |
1913 | free(rule->actions); | |
064af421 BP |
1914 | free(rule); |
1915 | } | |
1916 | ||
bcf84111 BP |
1917 | /* Destroys 'rule' and iterates through all of its facets and revalidates them, |
1918 | * destroying any that no longer has a rule (which is probably all of them). | |
064af421 | 1919 | * |
bcf84111 | 1920 | * The caller must have already removed 'rule' from the classifier. */ |
064af421 BP |
1921 | static void |
1922 | rule_destroy(struct ofproto *ofproto, struct rule *rule) | |
1923 | { | |
bcf84111 BP |
1924 | struct facet *facet, *next_facet; |
1925 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { | |
1926 | facet_revalidate(ofproto, facet); | |
064af421 BP |
1927 | } |
1928 | rule_free(rule); | |
1929 | } | |
1930 | ||
bcf84111 BP |
1931 | /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action |
1932 | * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't | |
1933 | * count). */ | |
064af421 | 1934 | static bool |
8054fc48 | 1935 | rule_has_out_port(const struct rule *rule, ovs_be16 out_port) |
064af421 BP |
1936 | { |
1937 | const union ofp_action *oa; | |
1938 | struct actions_iterator i; | |
1939 | ||
1940 | if (out_port == htons(OFPP_NONE)) { | |
1941 | return true; | |
1942 | } | |
1943 | for (oa = actions_first(&i, rule->actions, rule->n_actions); oa; | |
1944 | oa = actions_next(&i)) { | |
c1c9c9c4 | 1945 | if (action_outputs_to_port(oa, out_port)) { |
064af421 BP |
1946 | return true; |
1947 | } | |
1948 | } | |
1949 | return false; | |
1950 | } | |
1951 | ||
750638bb BP |
1952 | /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on |
1953 | * 'packet', which arrived on 'in_port'. | |
1954 | * | |
1955 | * Takes ownership of 'packet'. */ | |
9dbb9d5e BP |
1956 | static bool |
1957 | execute_odp_actions(struct ofproto *ofproto, uint16_t in_port, | |
1958 | const union odp_action *actions, size_t n_actions, | |
750638bb | 1959 | struct ofpbuf *packet) |
9dbb9d5e | 1960 | { |
750638bb | 1961 | if (n_actions == 1 && actions[0].type == ODPAT_CONTROLLER) { |
9dbb9d5e BP |
1962 | /* As an optimization, avoid a round-trip from userspace to kernel to |
1963 | * userspace. This also avoids possibly filling up kernel packet | |
1964 | * buffers along the way. */ | |
9dbb9d5e BP |
1965 | struct odp_msg *msg; |
1966 | ||
750638bb | 1967 | msg = ofpbuf_push_uninit(packet, sizeof *msg); |
9dbb9d5e BP |
1968 | msg->type = _ODPL_ACTION_NR; |
1969 | msg->length = sizeof(struct odp_msg) + packet->size; | |
1970 | msg->port = in_port; | |
1971 | msg->reserved = 0; | |
1972 | msg->arg = actions[0].controller.arg; | |
9dbb9d5e | 1973 | |
750638bb | 1974 | send_packet_in(ofproto, packet); |
9dbb9d5e | 1975 | |
750638bb BP |
1976 | return true; |
1977 | } else { | |
1978 | int error; | |
9dbb9d5e | 1979 | |
f1588b1f | 1980 | error = dpif_execute(ofproto->dpif, actions, n_actions, packet); |
750638bb BP |
1981 | ofpbuf_delete(packet); |
1982 | return !error; | |
1983 | } | |
9dbb9d5e BP |
1984 | } |
1985 | ||
bcf84111 BP |
1986 | /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s |
1987 | * statistics appropriately. 'packet' must have at least sizeof(struct | |
1988 | * ofp_packet_in) bytes of headroom. | |
064af421 | 1989 | * |
bcf84111 BP |
1990 | * For correct results, 'packet' must actually be in 'facet''s flow; that is, |
1991 | * applying flow_extract() to 'packet' would yield the same flow as | |
1992 | * 'facet->flow'. | |
064af421 | 1993 | * |
bcf84111 BP |
1994 | * 'facet' must have accurately composed ODP actions; that is, it must not be |
1995 | * in need of revalidation. | |
750638bb BP |
1996 | * |
1997 | * Takes ownership of 'packet'. */ | |
064af421 | 1998 | static void |
bcf84111 BP |
1999 | facet_execute(struct ofproto *ofproto, struct facet *facet, |
2000 | struct ofpbuf *packet) | |
064af421 | 2001 | { |
750638bb | 2002 | struct odp_flow_stats stats; |
bcf84111 BP |
2003 | |
2004 | assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in)); | |
2005 | ||
2006 | flow_extract_stats(&facet->flow, packet, &stats); | |
2007 | if (execute_odp_actions(ofproto, facet->flow.in_port, | |
2008 | facet->actions, facet->n_actions, packet)) { | |
2009 | facet_update_stats(ofproto, facet, &stats); | |
2010 | facet->used = time_msec(); | |
2011 | netflow_flow_update_time(ofproto->netflow, | |
2012 | &facet->nf_flow, facet->used); | |
2013 | } | |
2014 | } | |
2015 | ||
2016 | /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s | |
2017 | * statistics (or the statistics for one of its facets) appropriately. | |
2018 | * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom. | |
2019 | * | |
2020 | * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited | |
2021 | * with statistics for 'packet' either way. | |
2022 | * | |
2023 | * Takes ownership of 'packet'. */ | |
2024 | static void | |
2025 | rule_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port, | |
2026 | struct ofpbuf *packet) | |
2027 | { | |
2028 | struct facet *facet; | |
064af421 | 2029 | struct odp_actions a; |
bcf84111 BP |
2030 | struct flow flow; |
2031 | size_t size; | |
064af421 | 2032 | |
750638bb BP |
2033 | assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in)); |
2034 | ||
bcf84111 BP |
2035 | flow_extract(packet, 0, in_port, &flow); |
2036 | ||
2037 | /* First look for a related facet. If we find one, account it to that. */ | |
2038 | facet = facet_lookup_valid(ofproto, &flow); | |
2039 | if (facet && facet->rule == rule) { | |
2040 | facet_execute(ofproto, facet, packet); | |
2041 | return; | |
064af421 BP |
2042 | } |
2043 | ||
bcf84111 BP |
2044 | /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then |
2045 | * create a new facet for it and use that. */ | |
2046 | if (rule_lookup(ofproto, &flow) == rule) { | |
2047 | facet = facet_create(ofproto, rule, &flow, packet); | |
2048 | facet_execute(ofproto, facet, packet); | |
2049 | facet_install(ofproto, facet, true); | |
2050 | return; | |
2051 | } | |
2052 | ||
2053 | /* We can't account anything to a facet. If we were to try, then that | |
2054 | * facet would have a non-matching rule, busting our invariants. */ | |
2055 | if (xlate_actions(rule->actions, rule->n_actions, &flow, ofproto, | |
2056 | packet, &a, NULL, 0, NULL)) { | |
2057 | ofpbuf_delete(packet); | |
2058 | return; | |
2059 | } | |
2060 | size = packet->size; | |
2061 | if (execute_odp_actions(ofproto, in_port, | |
2062 | a.actions, a.n_actions, packet)) { | |
064af421 | 2063 | rule->used = time_msec(); |
bcf84111 BP |
2064 | rule->packet_count++; |
2065 | rule->byte_count += size; | |
064af421 BP |
2066 | } |
2067 | } | |
2068 | ||
afe75089 | 2069 | /* Inserts 'rule' into 'p''s flow table. */ |
064af421 | 2070 | static void |
afe75089 | 2071 | rule_insert(struct ofproto *p, struct rule *rule) |
064af421 BP |
2072 | { |
2073 | struct rule *displaced_rule; | |
2074 | ||
064af421 | 2075 | displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr)); |
bcf84111 BP |
2076 | if (displaced_rule) { |
2077 | rule_destroy(p, displaced_rule); | |
064af421 | 2078 | } |
bcf84111 | 2079 | p->need_revalidate = true; |
064af421 BP |
2080 | } |
2081 | ||
bcf84111 BP |
2082 | /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a |
2083 | * 'flow' and an example 'packet' within that flow. | |
2084 | * | |
2085 | * The caller must already have determined that no facet with an identical | |
2086 | * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in | |
2087 | * 'ofproto''s classifier table. */ | |
2088 | static struct facet * | |
2089 | facet_create(struct ofproto *ofproto, struct rule *rule, | |
2090 | const struct flow *flow, const struct ofpbuf *packet) | |
064af421 | 2091 | { |
bcf84111 | 2092 | struct facet *facet; |
fbb2ea0b | 2093 | |
bcf84111 BP |
2094 | facet = xzalloc(sizeof *facet); |
2095 | facet->used = time_msec(); | |
2096 | hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0)); | |
2097 | list_push_back(&rule->facets, &facet->list_node); | |
2098 | facet->rule = rule; | |
2099 | facet->flow = *flow; | |
2100 | netflow_flow_init(&facet->nf_flow); | |
2101 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used); | |
2102 | ||
2103 | facet_make_actions(ofproto, facet, packet); | |
064af421 | 2104 | |
bcf84111 BP |
2105 | return facet; |
2106 | } | |
2107 | ||
2108 | static void | |
2109 | facet_free(struct facet *facet) | |
2110 | { | |
2111 | free(facet->actions); | |
2112 | free(facet); | |
064af421 BP |
2113 | } |
2114 | ||
431d4707 | 2115 | /* Remove 'rule' from 'ofproto' and free up the associated memory: |
431d4707 BP |
2116 | * |
2117 | * - Removes 'rule' from the classifier. | |
2118 | * | |
bcf84111 BP |
2119 | * - If 'rule' has facets, revalidates them (and possibly uninstalls and |
2120 | * destroys them), via rule_destroy(). | |
431d4707 | 2121 | */ |
064af421 BP |
2122 | static void |
2123 | rule_remove(struct ofproto *ofproto, struct rule *rule) | |
2124 | { | |
bcf84111 BP |
2125 | COVERAGE_INC(ofproto_del_rule); |
2126 | ofproto->need_revalidate = true; | |
064af421 BP |
2127 | classifier_remove(&ofproto->cls, &rule->cr); |
2128 | rule_destroy(ofproto, rule); | |
2129 | } | |
2130 | ||
bcf84111 BP |
2131 | /* Remove 'facet' from 'ofproto' and free up the associated memory: |
2132 | * | |
2133 | * - If 'facet' was installed in the datapath, uninstalls it and updates its | |
2134 | * rule's statistics, via facet_uninstall(). | |
2135 | * | |
2136 | * - Removes 'facet' from its rule and from ofproto->facets. | |
2137 | */ | |
2138 | static void | |
2139 | facet_remove(struct ofproto *ofproto, struct facet *facet) | |
2140 | { | |
2141 | facet_uninstall(ofproto, facet); | |
d530fcd2 | 2142 | facet_flush_stats(ofproto, facet); |
bcf84111 BP |
2143 | hmap_remove(&ofproto->facets, &facet->hmap_node); |
2144 | list_remove(&facet->list_node); | |
2145 | facet_free(facet); | |
2146 | } | |
2147 | ||
7f7ae89d BP |
2148 | /* Composes the ODP actions for 'facet' based on its rule's actions. */ |
2149 | static void | |
bcf84111 BP |
2150 | facet_make_actions(struct ofproto *p, struct facet *facet, |
2151 | const struct ofpbuf *packet) | |
064af421 | 2152 | { |
bcf84111 | 2153 | const struct rule *rule = facet->rule; |
064af421 BP |
2154 | struct odp_actions a; |
2155 | size_t actions_len; | |
2156 | ||
bcf84111 BP |
2157 | xlate_actions(rule->actions, rule->n_actions, &facet->flow, p, |
2158 | packet, &a, &facet->tags, &facet->may_install, | |
2159 | &facet->nf_flow.output_iface); | |
064af421 BP |
2160 | |
2161 | actions_len = a.n_actions * sizeof *a.actions; | |
7f7ae89d BP |
2162 | if (facet->n_actions != a.n_actions |
2163 | || memcmp(facet->actions, a.actions, actions_len)) { | |
2164 | free(facet->actions); | |
2165 | facet->n_actions = a.n_actions; | |
2166 | facet->actions = xmemdup(a.actions, actions_len); | |
064af421 BP |
2167 | } |
2168 | } | |
2169 | ||
2170 | static int | |
bcf84111 | 2171 | facet_put__(struct ofproto *ofproto, struct facet *facet, int flags, |
064af421 BP |
2172 | struct odp_flow_put *put) |
2173 | { | |
2174 | memset(&put->flow.stats, 0, sizeof put->flow.stats); | |
bcf84111 BP |
2175 | odp_flow_key_from_flow(&put->flow.key, &facet->flow); |
2176 | put->flow.actions = facet->actions; | |
2177 | put->flow.n_actions = facet->n_actions; | |
ab48643b | 2178 | put->flow.flags = 0; |
064af421 | 2179 | put->flags = flags; |
c228a364 | 2180 | return dpif_flow_put(ofproto->dpif, put); |
064af421 BP |
2181 | } |
2182 | ||
bcf84111 BP |
2183 | /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If |
2184 | * 'zero_stats' is true, clears any existing statistics from the datapath for | |
2185 | * 'facet'. */ | |
064af421 | 2186 | static void |
bcf84111 | 2187 | facet_install(struct ofproto *p, struct facet *facet, bool zero_stats) |
064af421 | 2188 | { |
bcf84111 | 2189 | if (facet->may_install) { |
064af421 | 2190 | struct odp_flow_put put; |
bcf84111 | 2191 | int flags; |
064af421 | 2192 | |
bcf84111 BP |
2193 | flags = ODPPF_CREATE | ODPPF_MODIFY; |
2194 | if (zero_stats) { | |
2195 | flags |= ODPPF_ZERO_STATS; | |
2196 | } | |
2197 | if (!facet_put__(p, facet, flags, &put)) { | |
2198 | facet->installed = true; | |
2199 | } | |
064af421 BP |
2200 | } |
2201 | } | |
2202 | ||
bcf84111 BP |
2203 | /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up |
2204 | * to the accounting hook function in the ofhooks structure. */ | |
064af421 | 2205 | static void |
bcf84111 BP |
2206 | facet_account(struct ofproto *ofproto, |
2207 | struct facet *facet, uint64_t extra_bytes) | |
064af421 | 2208 | { |
bcf84111 | 2209 | uint64_t total_bytes = facet->byte_count + extra_bytes; |
064af421 BP |
2210 | |
2211 | if (ofproto->ofhooks->account_flow_cb | |
bcf84111 | 2212 | && total_bytes > facet->accounted_bytes) |
064af421 BP |
2213 | { |
2214 | ofproto->ofhooks->account_flow_cb( | |
bcf84111 BP |
2215 | &facet->flow, facet->tags, facet->actions, facet->n_actions, |
2216 | total_bytes - facet->accounted_bytes, ofproto->aux); | |
2217 | facet->accounted_bytes = total_bytes; | |
064af421 BP |
2218 | } |
2219 | } | |
2220 | ||
d530fcd2 | 2221 | /* If 'rule' is installed in the datapath, uninstalls it. */ |
064af421 | 2222 | static void |
bcf84111 | 2223 | facet_uninstall(struct ofproto *p, struct facet *facet) |
064af421 | 2224 | { |
bcf84111 | 2225 | if (facet->installed) { |
064af421 BP |
2226 | struct odp_flow odp_flow; |
2227 | ||
bcf84111 | 2228 | odp_flow_key_from_flow(&odp_flow.key, &facet->flow); |
064af421 BP |
2229 | odp_flow.actions = NULL; |
2230 | odp_flow.n_actions = 0; | |
ab48643b | 2231 | odp_flow.flags = 0; |
c228a364 | 2232 | if (!dpif_flow_del(p->dpif, &odp_flow)) { |
bcf84111 | 2233 | facet_update_stats(p, facet, &odp_flow.stats); |
064af421 | 2234 | } |
bcf84111 | 2235 | facet->installed = false; |
064af421 BP |
2236 | } |
2237 | } | |
2238 | ||
bcf84111 BP |
2239 | /* Returns true if the only action for 'facet' is to send to the controller. |
2240 | * (We don't report NetFlow expiration messages for such facets because they | |
2241 | * are just part of the control logic for the network, not real traffic). */ | |
0193b2af | 2242 | static bool |
bcf84111 | 2243 | facet_is_controller_flow(struct facet *facet) |
0193b2af | 2244 | { |
bcf84111 BP |
2245 | return (facet |
2246 | && facet->rule->n_actions == 1 | |
2247 | && action_outputs_to_port(&facet->rule->actions[0], | |
c1c9c9c4 | 2248 | htons(OFPP_CONTROLLER))); |
0193b2af JG |
2249 | } |
2250 | ||
bcf84111 BP |
2251 | /* Folds all of 'facet''s statistics into its rule. Also updates the |
2252 | * accounting ofhook and emits a NetFlow expiration if appropriate. */ | |
064af421 | 2253 | static void |
d530fcd2 | 2254 | facet_flush_stats(struct ofproto *ofproto, struct facet *facet) |
064af421 | 2255 | { |
bcf84111 | 2256 | facet_account(ofproto, facet, 0); |
064af421 | 2257 | |
bcf84111 | 2258 | if (ofproto->netflow && !facet_is_controller_flow(facet)) { |
064af421 | 2259 | struct ofexpired expired; |
bcf84111 BP |
2260 | expired.flow = facet->flow; |
2261 | expired.packet_count = facet->packet_count; | |
2262 | expired.byte_count = facet->byte_count; | |
2263 | expired.used = facet->used; | |
2264 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
2265 | } | |
2266 | ||
2267 | facet->rule->packet_count += facet->packet_count; | |
2268 | facet->rule->byte_count += facet->byte_count; | |
2269 | ||
2270 | /* Reset counters to prevent double counting if 'facet' ever gets | |
2271 | * reinstalled. */ | |
2272 | facet->packet_count = 0; | |
2273 | facet->byte_count = 0; | |
2274 | facet->accounted_bytes = 0; | |
2275 | ||
2276 | netflow_flow_clear(&facet->nf_flow); | |
2277 | } | |
2278 | ||
2279 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
2280 | * Returns it if found, otherwise a null pointer. | |
2281 | * | |
2282 | * The returned facet might need revalidation; use facet_lookup_valid() | |
2283 | * instead if that is important. */ | |
2284 | static struct facet * | |
2285 | facet_find(struct ofproto *ofproto, const struct flow *flow) | |
2286 | { | |
2287 | struct facet *facet; | |
2288 | ||
2289 | HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0), | |
2290 | &ofproto->facets) { | |
2291 | if (flow_equal(flow, &facet->flow)) { | |
2292 | return facet; | |
2293 | } | |
2294 | } | |
2295 | ||
2296 | return NULL; | |
2297 | } | |
2298 | ||
2299 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
2300 | * Returns it if found, otherwise a null pointer. | |
2301 | * | |
2302 | * The returned facet is guaranteed to be valid. */ | |
2303 | static struct facet * | |
2304 | facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow) | |
2305 | { | |
2306 | struct facet *facet = facet_find(ofproto, flow); | |
2307 | ||
2308 | /* The facet we found might not be valid, since we could be in need of | |
2309 | * revalidation. If it is not valid, don't return it. */ | |
2310 | if (facet | |
2311 | && ofproto->need_revalidate | |
2312 | && !facet_revalidate(ofproto, facet)) { | |
2313 | COVERAGE_INC(ofproto_invalidated); | |
2314 | return NULL; | |
064af421 | 2315 | } |
064af421 | 2316 | |
bcf84111 BP |
2317 | return facet; |
2318 | } | |
0193b2af | 2319 | |
bcf84111 BP |
2320 | /* Re-searches 'ofproto''s classifier for a rule matching 'facet': |
2321 | * | |
2322 | * - If the rule found is different from 'facet''s current rule, moves | |
2323 | * 'facet' to the new rule and recompiles its actions. | |
2324 | * | |
2325 | * - If the rule found is the same as 'facet''s current rule, leaves 'facet' | |
2326 | * where it is and recompiles its actions anyway. | |
2327 | * | |
2328 | * - If there is none, destroys 'facet'. | |
2329 | * | |
d530fcd2 | 2330 | * Returns true if 'facet' still exists, false if it has been destroyed. */ |
bcf84111 BP |
2331 | static bool |
2332 | facet_revalidate(struct ofproto *ofproto, struct facet *facet) | |
2333 | { | |
d530fcd2 BP |
2334 | struct rule *new_rule; |
2335 | struct odp_actions a; | |
2336 | size_t actions_len; | |
2337 | uint16_t new_nf_output_iface; | |
2338 | bool actions_changed; | |
bcf84111 BP |
2339 | |
2340 | COVERAGE_INC(facet_revalidate); | |
d530fcd2 BP |
2341 | |
2342 | /* Determine the new rule. */ | |
2343 | new_rule = rule_lookup(ofproto, &facet->flow); | |
2344 | if (!new_rule) { | |
2345 | /* No new rule, so delete the facet. */ | |
bcf84111 BP |
2346 | facet_remove(ofproto, facet); |
2347 | return false; | |
2348 | } | |
2349 | ||
d530fcd2 BP |
2350 | /* Calculate new ODP actions. |
2351 | * | |
2352 | * We are very cautious about actually modifying 'facet' state at this | |
2353 | * point, because we might need to, e.g., emit a NetFlow expiration and, if | |
2354 | * so, we need to have the old state around to properly compose it. */ | |
2355 | xlate_actions(new_rule->actions, new_rule->n_actions, &facet->flow, | |
2356 | ofproto, NULL, &a, &facet->tags, &facet->may_install, | |
2357 | &new_nf_output_iface); | |
2358 | actions_len = a.n_actions * sizeof *a.actions; | |
2359 | actions_changed = (facet->n_actions != a.n_actions | |
2360 | || memcmp(facet->actions, a.actions, actions_len)); | |
2361 | ||
2362 | /* If the ODP actions changed or the installability changed, then we need | |
2363 | * to talk to the datapath. */ | |
2364 | if (actions_changed || facet->may_install != facet->installed) { | |
2365 | if (facet->may_install) { | |
2366 | struct odp_flow_put put; | |
2367 | ||
2368 | memset(&put.flow.stats, 0, sizeof put.flow.stats); | |
2369 | odp_flow_key_from_flow(&put.flow.key, &facet->flow); | |
2370 | put.flow.actions = a.actions; | |
2371 | put.flow.n_actions = a.n_actions; | |
2372 | put.flow.flags = 0; | |
2373 | put.flags = ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS; | |
2374 | dpif_flow_put(ofproto->dpif, &put); | |
2375 | ||
2376 | facet_update_stats(ofproto, facet, &put.flow.stats); | |
2377 | } else { | |
2378 | facet_uninstall(ofproto, facet); | |
2379 | } | |
2380 | ||
2381 | /* The datapath flow is gone or has zeroed stats, so push stats out of | |
2382 | * 'facet' into 'rule'. */ | |
2383 | facet_flush_stats(ofproto, facet); | |
2384 | } | |
2385 | ||
2386 | /* Update 'facet' now that we've taken care of all the old state. */ | |
2387 | facet->nf_flow.output_iface = new_nf_output_iface; | |
2388 | if (actions_changed) { | |
2389 | free(facet->actions); | |
2390 | facet->n_actions = a.n_actions; | |
2391 | facet->actions = xmemdup(a.actions, actions_len); | |
2392 | } | |
2393 | if (facet->rule != new_rule) { | |
bcf84111 BP |
2394 | COVERAGE_INC(facet_changed_rule); |
2395 | list_remove(&facet->list_node); | |
d530fcd2 BP |
2396 | list_push_back(&new_rule->facets, &facet->list_node); |
2397 | facet->rule = new_rule; | |
2398 | facet->used = new_rule->created; | |
0c0afbec | 2399 | } |
bcf84111 | 2400 | |
bcf84111 | 2401 | return true; |
064af421 BP |
2402 | } |
2403 | \f | |
2404 | static void | |
2405 | queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn, | |
2406 | struct rconn_packet_counter *counter) | |
2407 | { | |
2408 | update_openflow_length(msg); | |
2409 | if (rconn_send(ofconn->rconn, msg, counter)) { | |
2410 | ofpbuf_delete(msg); | |
2411 | } | |
2412 | } | |
2413 | ||
064af421 BP |
2414 | static void |
2415 | send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh, | |
2416 | int error) | |
2417 | { | |
26c112c2 BP |
2418 | struct ofpbuf *buf = make_ofp_error_msg(error, oh); |
2419 | if (buf) { | |
2420 | COVERAGE_INC(ofproto_error); | |
2421 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
2422 | } | |
064af421 BP |
2423 | } |
2424 | ||
2425 | static void | |
2426 | hton_ofp_phy_port(struct ofp_phy_port *opp) | |
2427 | { | |
2428 | opp->port_no = htons(opp->port_no); | |
2429 | opp->config = htonl(opp->config); | |
2430 | opp->state = htonl(opp->state); | |
2431 | opp->curr = htonl(opp->curr); | |
2432 | opp->advertised = htonl(opp->advertised); | |
2433 | opp->supported = htonl(opp->supported); | |
2434 | opp->peer = htonl(opp->peer); | |
2435 | } | |
2436 | ||
2437 | static int | |
2438 | handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh) | |
2439 | { | |
2440 | struct ofp_header *rq = oh; | |
2441 | queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter); | |
2442 | return 0; | |
2443 | } | |
2444 | ||
2445 | static int | |
3269c562 | 2446 | handle_features_request(struct ofconn *ofconn, struct ofp_header *oh) |
064af421 BP |
2447 | { |
2448 | struct ofp_switch_features *osf; | |
2449 | struct ofpbuf *buf; | |
064af421 BP |
2450 | struct ofport *port; |
2451 | ||
2452 | osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf); | |
3269c562 | 2453 | osf->datapath_id = htonll(ofconn->ofproto->datapath_id); |
064af421 BP |
2454 | osf->n_buffers = htonl(pktbuf_capacity()); |
2455 | osf->n_tables = 2; | |
2456 | osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS | | |
0254ae23 | 2457 | OFPC_PORT_STATS | OFPC_ARP_MATCH_IP); |
064af421 BP |
2458 | osf->actions = htonl((1u << OFPAT_OUTPUT) | |
2459 | (1u << OFPAT_SET_VLAN_VID) | | |
2460 | (1u << OFPAT_SET_VLAN_PCP) | | |
2461 | (1u << OFPAT_STRIP_VLAN) | | |
2462 | (1u << OFPAT_SET_DL_SRC) | | |
2463 | (1u << OFPAT_SET_DL_DST) | | |
2464 | (1u << OFPAT_SET_NW_SRC) | | |
2465 | (1u << OFPAT_SET_NW_DST) | | |
959a2ecd | 2466 | (1u << OFPAT_SET_NW_TOS) | |
064af421 | 2467 | (1u << OFPAT_SET_TP_SRC) | |
c1c9c9c4 BP |
2468 | (1u << OFPAT_SET_TP_DST) | |
2469 | (1u << OFPAT_ENQUEUE)); | |
064af421 | 2470 | |
3269c562 | 2471 | HMAP_FOR_EACH (port, hmap_node, &ofconn->ofproto->ports) { |
064af421 BP |
2472 | hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp)); |
2473 | } | |
2474 | ||
2475 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
2476 | return 0; | |
2477 | } | |
2478 | ||
2479 | static int | |
3269c562 | 2480 | handle_get_config_request(struct ofconn *ofconn, struct ofp_header *oh) |
064af421 BP |
2481 | { |
2482 | struct ofpbuf *buf; | |
2483 | struct ofp_switch_config *osc; | |
2484 | uint16_t flags; | |
2485 | bool drop_frags; | |
2486 | ||
2487 | /* Figure out flags. */ | |
3269c562 | 2488 | dpif_get_drop_frags(ofconn->ofproto->dpif, &drop_frags); |
064af421 | 2489 | flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL; |
064af421 BP |
2490 | |
2491 | /* Send reply. */ | |
2492 | osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf); | |
2493 | osc->flags = htons(flags); | |
2494 | osc->miss_send_len = htons(ofconn->miss_send_len); | |
2495 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
2496 | ||
2497 | return 0; | |
2498 | } | |
2499 | ||
2500 | static int | |
3269c562 | 2501 | handle_set_config(struct ofconn *ofconn, struct ofp_switch_config *osc) |
064af421 BP |
2502 | { |
2503 | uint16_t flags; | |
2504 | int error; | |
2505 | ||
2506 | error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc); | |
2507 | if (error) { | |
2508 | return error; | |
2509 | } | |
2510 | flags = ntohs(osc->flags); | |
2511 | ||
5899143f | 2512 | if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) { |
064af421 BP |
2513 | switch (flags & OFPC_FRAG_MASK) { |
2514 | case OFPC_FRAG_NORMAL: | |
3269c562 | 2515 | dpif_set_drop_frags(ofconn->ofproto->dpif, false); |
064af421 BP |
2516 | break; |
2517 | case OFPC_FRAG_DROP: | |
3269c562 | 2518 | dpif_set_drop_frags(ofconn->ofproto->dpif, true); |
064af421 BP |
2519 | break; |
2520 | default: | |
2521 | VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")", | |
2522 | osc->flags); | |
2523 | break; | |
2524 | } | |
2525 | } | |
2526 | ||
064af421 BP |
2527 | ofconn->miss_send_len = ntohs(osc->miss_send_len); |
2528 | ||
2529 | return 0; | |
2530 | } | |
2531 | ||
064af421 | 2532 | static void |
aae51f53 | 2533 | add_controller_action(struct odp_actions *actions, uint16_t max_len) |
064af421 BP |
2534 | { |
2535 | union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER); | |
aae51f53 | 2536 | a->controller.arg = max_len; |
064af421 BP |
2537 | } |
2538 | ||
2539 | struct action_xlate_ctx { | |
2540 | /* Input. */ | |
ae412e7d | 2541 | struct flow flow; /* Flow to which these actions correspond. */ |
064af421 BP |
2542 | int recurse; /* Recursion level, via xlate_table_action. */ |
2543 | struct ofproto *ofproto; | |
2544 | const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a | |
2545 | * null pointer if we are revalidating | |
2546 | * without a packet to refer to. */ | |
2547 | ||
2548 | /* Output. */ | |
2549 | struct odp_actions *out; /* Datapath actions. */ | |
1eb0942d | 2550 | tag_type tags; /* Tags associated with OFPP_NORMAL actions. */ |
d6fbec6d | 2551 | bool may_set_up_flow; /* True ordinarily; false if the actions must |
064af421 | 2552 | * be reassessed for every packet. */ |
6a07af36 | 2553 | uint16_t nf_output_iface; /* Output interface index for NetFlow. */ |
064af421 BP |
2554 | }; |
2555 | ||
5f8bfd69 BP |
2556 | /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a |
2557 | * flow translation. */ | |
2558 | #define MAX_RESUBMIT_RECURSION 8 | |
2559 | ||
064af421 BP |
2560 | static void do_xlate_actions(const union ofp_action *in, size_t n_in, |
2561 | struct action_xlate_ctx *ctx); | |
2562 | ||
2563 | static void | |
2564 | add_output_action(struct action_xlate_ctx *ctx, uint16_t port) | |
2565 | { | |
ca0f572c | 2566 | const struct ofport *ofport = get_port(ctx->ofproto, port); |
6cfaf517 BP |
2567 | |
2568 | if (ofport) { | |
2569 | if (ofport->opp.config & OFPPC_NO_FWD) { | |
2570 | /* Forwarding disabled on port. */ | |
2571 | return; | |
2572 | } | |
2573 | } else { | |
2574 | /* | |
2575 | * We don't have an ofport record for this port, but it doesn't hurt to | |
2576 | * allow forwarding to it anyhow. Maybe such a port will appear later | |
2577 | * and we're pre-populating the flow table. | |
2578 | */ | |
064af421 | 2579 | } |
6cfaf517 BP |
2580 | |
2581 | odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port; | |
6a07af36 | 2582 | ctx->nf_output_iface = port; |
064af421 BP |
2583 | } |
2584 | ||
2585 | static struct rule * | |
bcf84111 | 2586 | rule_lookup(struct ofproto *ofproto, const struct flow *flow) |
064af421 | 2587 | { |
3c4486a5 | 2588 | return rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow)); |
064af421 BP |
2589 | } |
2590 | ||
2591 | static void | |
2592 | xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port) | |
2593 | { | |
5f8bfd69 | 2594 | if (ctx->recurse < MAX_RESUBMIT_RECURSION) { |
2c5d1389 | 2595 | uint16_t old_in_port; |
064af421 | 2596 | struct rule *rule; |
064af421 | 2597 | |
2c5d1389 BP |
2598 | /* Look up a flow with 'in_port' as the input port. Then restore the |
2599 | * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will | |
2600 | * have surprising behavior). */ | |
2601 | old_in_port = ctx->flow.in_port; | |
e18fe8a2 | 2602 | ctx->flow.in_port = in_port; |
bcf84111 | 2603 | rule = rule_lookup(ctx->ofproto, &ctx->flow); |
2c5d1389 BP |
2604 | ctx->flow.in_port = old_in_port; |
2605 | ||
064af421 | 2606 | if (rule) { |
064af421 BP |
2607 | ctx->recurse++; |
2608 | do_xlate_actions(rule->actions, rule->n_actions, ctx); | |
2609 | ctx->recurse--; | |
2610 | } | |
5f8bfd69 BP |
2611 | } else { |
2612 | struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
2613 | ||
2614 | VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times", | |
2615 | MAX_RESUBMIT_RECURSION); | |
064af421 BP |
2616 | } |
2617 | } | |
2618 | ||
f1588b1f BP |
2619 | static void |
2620 | flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, uint32_t mask, | |
2621 | uint16_t *nf_output_iface, struct odp_actions *actions) | |
2622 | { | |
2623 | struct ofport *ofport; | |
2624 | ||
2625 | HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) { | |
2626 | uint16_t odp_port = ofport->odp_port; | |
2627 | if (odp_port != odp_in_port && !(ofport->opp.config & mask)) { | |
2628 | odp_actions_add(actions, ODPAT_OUTPUT)->output.port = odp_port; | |
2629 | } | |
2630 | } | |
2631 | *nf_output_iface = NF_OUT_FLOOD; | |
2632 | } | |
2633 | ||
064af421 | 2634 | static void |
aae51f53 BP |
2635 | xlate_output_action__(struct action_xlate_ctx *ctx, |
2636 | uint16_t port, uint16_t max_len) | |
064af421 BP |
2637 | { |
2638 | uint16_t odp_port; | |
6a07af36 JG |
2639 | uint16_t prev_nf_output_iface = ctx->nf_output_iface; |
2640 | ||
2641 | ctx->nf_output_iface = NF_OUT_DROP; | |
064af421 | 2642 | |
aae51f53 | 2643 | switch (port) { |
064af421 | 2644 | case OFPP_IN_PORT: |
e18fe8a2 | 2645 | add_output_action(ctx, ctx->flow.in_port); |
064af421 BP |
2646 | break; |
2647 | case OFPP_TABLE: | |
e18fe8a2 | 2648 | xlate_table_action(ctx, ctx->flow.in_port); |
064af421 BP |
2649 | break; |
2650 | case OFPP_NORMAL: | |
e18fe8a2 | 2651 | if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet, |
1eb0942d | 2652 | ctx->out, &ctx->tags, |
6a07af36 | 2653 | &ctx->nf_output_iface, |
064af421 BP |
2654 | ctx->ofproto->aux)) { |
2655 | COVERAGE_INC(ofproto_uninstallable); | |
d6fbec6d | 2656 | ctx->may_set_up_flow = false; |
064af421 BP |
2657 | } |
2658 | break; | |
2659 | case OFPP_FLOOD: | |
f1588b1f BP |
2660 | flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD, |
2661 | &ctx->nf_output_iface, ctx->out); | |
9628cd42 | 2662 | break; |
064af421 | 2663 | case OFPP_ALL: |
f1588b1f BP |
2664 | flood_packets(ctx->ofproto, ctx->flow.in_port, 0, |
2665 | &ctx->nf_output_iface, ctx->out); | |
064af421 BP |
2666 | break; |
2667 | case OFPP_CONTROLLER: | |
aae51f53 | 2668 | add_controller_action(ctx->out, max_len); |
064af421 BP |
2669 | break; |
2670 | case OFPP_LOCAL: | |
2671 | add_output_action(ctx, ODPP_LOCAL); | |
2672 | break; | |
2673 | default: | |
aae51f53 | 2674 | odp_port = ofp_port_to_odp_port(port); |
e18fe8a2 | 2675 | if (odp_port != ctx->flow.in_port) { |
064af421 BP |
2676 | add_output_action(ctx, odp_port); |
2677 | } | |
2678 | break; | |
2679 | } | |
6a07af36 JG |
2680 | |
2681 | if (prev_nf_output_iface == NF_OUT_FLOOD) { | |
2682 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
2683 | } else if (ctx->nf_output_iface == NF_OUT_DROP) { | |
2684 | ctx->nf_output_iface = prev_nf_output_iface; | |
2685 | } else if (prev_nf_output_iface != NF_OUT_DROP && | |
2686 | ctx->nf_output_iface != NF_OUT_FLOOD) { | |
2687 | ctx->nf_output_iface = NF_OUT_MULTI; | |
2688 | } | |
064af421 BP |
2689 | } |
2690 | ||
aae51f53 BP |
2691 | static void |
2692 | xlate_output_action(struct action_xlate_ctx *ctx, | |
2693 | const struct ofp_action_output *oao) | |
2694 | { | |
2695 | xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len)); | |
2696 | } | |
2697 | ||
c1c9c9c4 BP |
2698 | /* If the final ODP action in 'ctx' is "pop priority", drop it, as an |
2699 | * optimization, because we're going to add another action that sets the | |
2700 | * priority immediately after, or because there are no actions following the | |
2701 | * pop. */ | |
2702 | static void | |
2703 | remove_pop_action(struct action_xlate_ctx *ctx) | |
2704 | { | |
2705 | size_t n = ctx->out->n_actions; | |
2706 | if (n > 0 && ctx->out->actions[n - 1].type == ODPAT_POP_PRIORITY) { | |
2707 | ctx->out->n_actions--; | |
2708 | } | |
2709 | } | |
2710 | ||
2711 | static void | |
2712 | xlate_enqueue_action(struct action_xlate_ctx *ctx, | |
2713 | const struct ofp_action_enqueue *oae) | |
2714 | { | |
2715 | uint16_t ofp_port, odp_port; | |
aae51f53 BP |
2716 | uint32_t priority; |
2717 | int error; | |
2718 | ||
2719 | error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id), | |
2720 | &priority); | |
2721 | if (error) { | |
2722 | /* Fall back to ordinary output action. */ | |
2723 | xlate_output_action__(ctx, ntohs(oae->port), 0); | |
2724 | return; | |
2725 | } | |
c1c9c9c4 BP |
2726 | |
2727 | /* Figure out ODP output port. */ | |
2728 | ofp_port = ntohs(oae->port); | |
2729 | if (ofp_port != OFPP_IN_PORT) { | |
2730 | odp_port = ofp_port_to_odp_port(ofp_port); | |
2731 | } else { | |
2732 | odp_port = ctx->flow.in_port; | |
2733 | } | |
2734 | ||
2735 | /* Add ODP actions. */ | |
2736 | remove_pop_action(ctx); | |
2737 | odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority | |
aae51f53 | 2738 | = priority; |
c1c9c9c4 BP |
2739 | add_output_action(ctx, odp_port); |
2740 | odp_actions_add(ctx->out, ODPAT_POP_PRIORITY); | |
2741 | ||
2742 | /* Update NetFlow output port. */ | |
2743 | if (ctx->nf_output_iface == NF_OUT_DROP) { | |
2744 | ctx->nf_output_iface = odp_port; | |
2745 | } else if (ctx->nf_output_iface != NF_OUT_FLOOD) { | |
2746 | ctx->nf_output_iface = NF_OUT_MULTI; | |
2747 | } | |
2748 | } | |
2749 | ||
eedc0097 JP |
2750 | static void |
2751 | xlate_set_queue_action(struct action_xlate_ctx *ctx, | |
2752 | const struct nx_action_set_queue *nasq) | |
2753 | { | |
2754 | uint32_t priority; | |
2755 | int error; | |
2756 | ||
2757 | error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id), | |
2758 | &priority); | |
2759 | if (error) { | |
2760 | /* Couldn't translate queue to a priority, so ignore. A warning | |
2761 | * has already been logged. */ | |
2762 | return; | |
2763 | } | |
2764 | ||
2765 | remove_pop_action(ctx); | |
2766 | odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority | |
2767 | = priority; | |
2768 | } | |
2769 | ||
350a665f BP |
2770 | static void |
2771 | xlate_set_dl_tci(struct action_xlate_ctx *ctx) | |
2772 | { | |
2773 | ovs_be16 dl_vlan = ctx->flow.dl_vlan; | |
2774 | uint8_t dl_vlan_pcp = ctx->flow.dl_vlan_pcp; | |
2775 | ||
2776 | if (dl_vlan == htons(OFP_VLAN_NONE)) { | |
2777 | odp_actions_add(ctx->out, ODPAT_STRIP_VLAN); | |
2778 | } else { | |
2779 | union odp_action *oa = odp_actions_add(ctx->out, ODPAT_SET_DL_TCI); | |
2780 | oa->dl_tci.tci = htons(ntohs(dl_vlan & htons(VLAN_VID_MASK)) | |
2781 | | (dl_vlan_pcp << VLAN_PCP_SHIFT) | |
2782 | | VLAN_CFI); | |
2783 | } | |
2784 | } | |
2785 | ||
b6c9e612 BP |
2786 | static void |
2787 | xlate_reg_move_action(struct action_xlate_ctx *ctx, | |
2788 | const struct nx_action_reg_move *narm) | |
2789 | { | |
2790 | ovs_be16 old_vlan = ctx->flow.dl_vlan; | |
2791 | uint8_t old_pcp = ctx->flow.dl_vlan_pcp; | |
2792 | ||
2793 | nxm_execute_reg_move(narm, &ctx->flow); | |
2794 | ||
2795 | if (ctx->flow.dl_vlan != old_vlan || ctx->flow.dl_vlan_pcp != old_pcp) { | |
2796 | xlate_set_dl_tci(ctx); | |
2797 | } | |
2798 | } | |
2799 | ||
064af421 BP |
2800 | static void |
2801 | xlate_nicira_action(struct action_xlate_ctx *ctx, | |
2802 | const struct nx_action_header *nah) | |
2803 | { | |
2804 | const struct nx_action_resubmit *nar; | |
659586ef | 2805 | const struct nx_action_set_tunnel *nast; |
eedc0097 | 2806 | const struct nx_action_set_queue *nasq; |
659586ef | 2807 | union odp_action *oa; |
064af421 BP |
2808 | int subtype = ntohs(nah->subtype); |
2809 | ||
2810 | assert(nah->vendor == htonl(NX_VENDOR_ID)); | |
2811 | switch (subtype) { | |
2812 | case NXAST_RESUBMIT: | |
2813 | nar = (const struct nx_action_resubmit *) nah; | |
2814 | xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port))); | |
2815 | break; | |
2816 | ||
659586ef JG |
2817 | case NXAST_SET_TUNNEL: |
2818 | nast = (const struct nx_action_set_tunnel *) nah; | |
2819 | oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL); | |
2820 | ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id; | |
2821 | break; | |
2822 | ||
401eeb92 BP |
2823 | case NXAST_DROP_SPOOFED_ARP: |
2824 | if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) { | |
2825 | odp_actions_add(ctx->out, ODPAT_DROP_SPOOFED_ARP); | |
2826 | } | |
2827 | break; | |
2828 | ||
eedc0097 JP |
2829 | case NXAST_SET_QUEUE: |
2830 | nasq = (const struct nx_action_set_queue *) nah; | |
2831 | xlate_set_queue_action(ctx, nasq); | |
2832 | break; | |
2833 | ||
2834 | case NXAST_POP_QUEUE: | |
2835 | odp_actions_add(ctx->out, ODPAT_POP_PRIORITY); | |
2836 | break; | |
2837 | ||
b6c9e612 BP |
2838 | case NXAST_REG_MOVE: |
2839 | xlate_reg_move_action(ctx, (const struct nx_action_reg_move *) nah); | |
2840 | break; | |
2841 | ||
2842 | case NXAST_REG_LOAD: | |
2843 | nxm_execute_reg_load((const struct nx_action_reg_load *) nah, | |
2844 | &ctx->flow); | |
2845 | break; | |
2846 | ||
999f0d45 | 2847 | /* If you add a new action here that modifies flow data, don't forget to |
c1c9c9c4 | 2848 | * update the flow key in ctx->flow at the same time. */ |
999f0d45 | 2849 | |
064af421 BP |
2850 | default: |
2851 | VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype); | |
2852 | break; | |
2853 | } | |
2854 | } | |
2855 | ||
2856 | static void | |
2857 | do_xlate_actions(const union ofp_action *in, size_t n_in, | |
2858 | struct action_xlate_ctx *ctx) | |
2859 | { | |
2860 | struct actions_iterator iter; | |
2861 | const union ofp_action *ia; | |
2862 | const struct ofport *port; | |
2863 | ||
ca0f572c | 2864 | port = get_port(ctx->ofproto, ctx->flow.in_port); |
064af421 | 2865 | if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) && |
ba186119 | 2866 | port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp) |
064af421 BP |
2867 | ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) { |
2868 | /* Drop this flow. */ | |
2869 | return; | |
2870 | } | |
2871 | ||
2872 | for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) { | |
2873 | uint16_t type = ntohs(ia->type); | |
2874 | union odp_action *oa; | |
2875 | ||
2876 | switch (type) { | |
2877 | case OFPAT_OUTPUT: | |
2878 | xlate_output_action(ctx, &ia->output); | |
2879 | break; | |
2880 | ||
2881 | case OFPAT_SET_VLAN_VID: | |
f1670a48 | 2882 | ctx->flow.dl_vlan = ia->vlan_vid.vlan_vid; |
350a665f | 2883 | xlate_set_dl_tci(ctx); |
064af421 BP |
2884 | break; |
2885 | ||
2886 | case OFPAT_SET_VLAN_PCP: | |
f1670a48 | 2887 | ctx->flow.dl_vlan_pcp = ia->vlan_pcp.vlan_pcp; |
350a665f | 2888 | xlate_set_dl_tci(ctx); |
064af421 BP |
2889 | break; |
2890 | ||
2891 | case OFPAT_STRIP_VLAN: | |
7ac0f8db | 2892 | ctx->flow.dl_vlan = htons(OFP_VLAN_NONE); |
999f0d45 | 2893 | ctx->flow.dl_vlan_pcp = 0; |
350a665f | 2894 | xlate_set_dl_tci(ctx); |
064af421 BP |
2895 | break; |
2896 | ||
2897 | case OFPAT_SET_DL_SRC: | |
2898 | oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC); | |
2899 | memcpy(oa->dl_addr.dl_addr, | |
2900 | ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); | |
999f0d45 BP |
2901 | memcpy(ctx->flow.dl_src, |
2902 | ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); | |
064af421 BP |
2903 | break; |
2904 | ||
2905 | case OFPAT_SET_DL_DST: | |
2906 | oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST); | |
2907 | memcpy(oa->dl_addr.dl_addr, | |
2908 | ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); | |
999f0d45 BP |
2909 | memcpy(ctx->flow.dl_dst, |
2910 | ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); | |
064af421 BP |
2911 | break; |
2912 | ||
2913 | case OFPAT_SET_NW_SRC: | |
2914 | oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC); | |
999f0d45 | 2915 | ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr; |
064af421 BP |
2916 | break; |
2917 | ||
2d70a31a JP |
2918 | case OFPAT_SET_NW_DST: |
2919 | oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST); | |
999f0d45 | 2920 | ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr; |
2d38e234 | 2921 | break; |
959a2ecd JP |
2922 | |
2923 | case OFPAT_SET_NW_TOS: | |
2924 | oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS); | |
999f0d45 | 2925 | ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos; |
2d70a31a JP |
2926 | break; |
2927 | ||
064af421 BP |
2928 | case OFPAT_SET_TP_SRC: |
2929 | oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC); | |
999f0d45 | 2930 | ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port; |
064af421 BP |
2931 | break; |
2932 | ||
2d70a31a JP |
2933 | case OFPAT_SET_TP_DST: |
2934 | oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST); | |
999f0d45 | 2935 | ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port; |
2d70a31a JP |
2936 | break; |
2937 | ||
064af421 BP |
2938 | case OFPAT_VENDOR: |
2939 | xlate_nicira_action(ctx, (const struct nx_action_header *) ia); | |
2940 | break; | |
2941 | ||
c1c9c9c4 BP |
2942 | case OFPAT_ENQUEUE: |
2943 | xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia); | |
2944 | break; | |
2945 | ||
064af421 BP |
2946 | default: |
2947 | VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type); | |
2948 | break; | |
2949 | } | |
2950 | } | |
2951 | } | |
2952 | ||
2953 | static int | |
2954 | xlate_actions(const union ofp_action *in, size_t n_in, | |
ae412e7d | 2955 | const struct flow *flow, struct ofproto *ofproto, |
064af421 | 2956 | const struct ofpbuf *packet, |
6a07af36 JG |
2957 | struct odp_actions *out, tag_type *tags, bool *may_set_up_flow, |
2958 | uint16_t *nf_output_iface) | |
064af421 | 2959 | { |
064af421 | 2960 | struct action_xlate_ctx ctx; |
1eb0942d | 2961 | |
064af421 BP |
2962 | COVERAGE_INC(ofproto_ofp2odp); |
2963 | odp_actions_init(out); | |
e18fe8a2 | 2964 | ctx.flow = *flow; |
064af421 BP |
2965 | ctx.recurse = 0; |
2966 | ctx.ofproto = ofproto; | |
2967 | ctx.packet = packet; | |
2968 | ctx.out = out; | |
1eb0942d | 2969 | ctx.tags = 0; |
d6fbec6d | 2970 | ctx.may_set_up_flow = true; |
6a07af36 | 2971 | ctx.nf_output_iface = NF_OUT_DROP; |
064af421 | 2972 | do_xlate_actions(in, n_in, &ctx); |
c1c9c9c4 | 2973 | remove_pop_action(&ctx); |
0ad9b732 | 2974 | |
d6fbec6d | 2975 | /* Check with in-band control to see if we're allowed to set up this |
0ad9b732 JP |
2976 | * flow. */ |
2977 | if (!in_band_rule_check(ofproto->in_band, flow, out)) { | |
d6fbec6d | 2978 | ctx.may_set_up_flow = false; |
0ad9b732 JP |
2979 | } |
2980 | ||
1eb0942d BP |
2981 | if (tags) { |
2982 | *tags = ctx.tags; | |
2983 | } | |
d6fbec6d BP |
2984 | if (may_set_up_flow) { |
2985 | *may_set_up_flow = ctx.may_set_up_flow; | |
064af421 | 2986 | } |
6a07af36 JG |
2987 | if (nf_output_iface) { |
2988 | *nf_output_iface = ctx.nf_output_iface; | |
064af421 BP |
2989 | } |
2990 | if (odp_actions_overflow(out)) { | |
9026bf34 | 2991 | COVERAGE_INC(odp_overflow); |
064af421 BP |
2992 | odp_actions_init(out); |
2993 | return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY); | |
2994 | } | |
2995 | return 0; | |
2996 | } | |
2997 | ||
9deba63b BP |
2998 | /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow |
2999 | * error message code (composed with ofp_mkerr()) for the caller to propagate | |
3000 | * upward. Otherwise, returns 0. | |
3001 | * | |
2228b50d | 3002 | * The log message mentions 'msg_type'. */ |
9deba63b | 3003 | static int |
2228b50d | 3004 | reject_slave_controller(struct ofconn *ofconn, const const char *msg_type) |
9deba63b | 3005 | { |
5899143f | 3006 | if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) { |
9deba63b | 3007 | static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5); |
9deba63b | 3008 | VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller", |
2228b50d | 3009 | msg_type); |
9deba63b BP |
3010 | |
3011 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM); | |
3012 | } else { | |
3013 | return 0; | |
3014 | } | |
3015 | } | |
3016 | ||
064af421 | 3017 | static int |
3269c562 | 3018 | handle_packet_out(struct ofconn *ofconn, struct ofp_header *oh) |
064af421 | 3019 | { |
3269c562 | 3020 | struct ofproto *p = ofconn->ofproto; |
064af421 BP |
3021 | struct ofp_packet_out *opo; |
3022 | struct ofpbuf payload, *buffer; | |
ac51afaf BP |
3023 | union ofp_action *ofp_actions; |
3024 | struct odp_actions odp_actions; | |
3025 | struct ofpbuf request; | |
ae412e7d | 3026 | struct flow flow; |
ac51afaf | 3027 | size_t n_ofp_actions; |
064af421 | 3028 | uint16_t in_port; |
064af421 BP |
3029 | int error; |
3030 | ||
ac51afaf BP |
3031 | COVERAGE_INC(ofproto_packet_out); |
3032 | ||
2228b50d | 3033 | error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT"); |
9deba63b BP |
3034 | if (error) { |
3035 | return error; | |
3036 | } | |
3037 | ||
ac51afaf BP |
3038 | /* Get ofp_packet_out. */ |
3039 | request.data = oh; | |
3040 | request.size = ntohs(oh->length); | |
3041 | opo = ofpbuf_try_pull(&request, offsetof(struct ofp_packet_out, actions)); | |
3042 | if (!opo) { | |
3043 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
3044 | } | |
3045 | ||
3046 | /* Get actions. */ | |
3047 | error = ofputil_pull_actions(&request, ntohs(opo->actions_len), | |
3048 | &ofp_actions, &n_ofp_actions); | |
064af421 BP |
3049 | if (error) { |
3050 | return error; | |
3051 | } | |
064af421 | 3052 | |
ac51afaf | 3053 | /* Get payload. */ |
064af421 BP |
3054 | if (opo->buffer_id != htonl(UINT32_MAX)) { |
3055 | error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id), | |
3056 | &buffer, &in_port); | |
7778bd15 | 3057 | if (error || !buffer) { |
064af421 BP |
3058 | return error; |
3059 | } | |
3060 | payload = *buffer; | |
3061 | } else { | |
ac51afaf | 3062 | payload = request; |
064af421 BP |
3063 | buffer = NULL; |
3064 | } | |
3065 | ||
ac51afaf BP |
3066 | /* Extract flow, check actions. */ |
3067 | flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), | |
3068 | &flow); | |
f1defbf9 | 3069 | error = validate_actions(ofp_actions, n_ofp_actions, &flow, p->max_ports); |
ac51afaf BP |
3070 | if (error) { |
3071 | goto exit; | |
3072 | } | |
3073 | ||
3074 | /* Send. */ | |
3075 | error = xlate_actions(ofp_actions, n_ofp_actions, &flow, p, &payload, | |
3076 | &odp_actions, NULL, NULL, NULL); | |
700712e2 | 3077 | if (!error) { |
ac51afaf BP |
3078 | dpif_execute(p->dpif, odp_actions.actions, odp_actions.n_actions, |
3079 | &payload); | |
064af421 | 3080 | } |
064af421 | 3081 | |
ac51afaf BP |
3082 | exit: |
3083 | ofpbuf_delete(buffer); | |
3084 | return 0; | |
064af421 BP |
3085 | } |
3086 | ||
3087 | static void | |
3088 | update_port_config(struct ofproto *p, struct ofport *port, | |
3089 | uint32_t config, uint32_t mask) | |
3090 | { | |
3091 | mask &= config ^ port->opp.config; | |
3092 | if (mask & OFPPC_PORT_DOWN) { | |
3093 | if (config & OFPPC_PORT_DOWN) { | |
3094 | netdev_turn_flags_off(port->netdev, NETDEV_UP, true); | |
3095 | } else { | |
3096 | netdev_turn_flags_on(port->netdev, NETDEV_UP, true); | |
3097 | } | |
3098 | } | |
f1588b1f BP |
3099 | #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \ |
3100 | OFPPC_NO_FWD | OFPPC_NO_FLOOD) | |
064af421 BP |
3101 | if (mask & REVALIDATE_BITS) { |
3102 | COVERAGE_INC(ofproto_costly_flags); | |
3103 | port->opp.config ^= mask & REVALIDATE_BITS; | |
3104 | p->need_revalidate = true; | |
3105 | } | |
3106 | #undef REVALIDATE_BITS | |
064af421 BP |
3107 | if (mask & OFPPC_NO_PACKET_IN) { |
3108 | port->opp.config ^= OFPPC_NO_PACKET_IN; | |
3109 | } | |
3110 | } | |
3111 | ||
3112 | static int | |
3269c562 | 3113 | handle_port_mod(struct ofconn *ofconn, struct ofp_header *oh) |
064af421 | 3114 | { |
3269c562 | 3115 | struct ofproto *p = ofconn->ofproto; |
064af421 BP |
3116 | const struct ofp_port_mod *opm; |
3117 | struct ofport *port; | |
3118 | int error; | |
3119 | ||
2228b50d | 3120 | error = reject_slave_controller(ofconn, "OFPT_PORT_MOD"); |
9deba63b BP |
3121 | if (error) { |
3122 | return error; | |
3123 | } | |
064af421 BP |
3124 | error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm); |
3125 | if (error) { | |
3126 | return error; | |
3127 | } | |
3128 | opm = (struct ofp_port_mod *) oh; | |
3129 | ||
ca0f572c | 3130 | port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no))); |
064af421 BP |
3131 | if (!port) { |
3132 | return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT); | |
3133 | } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) { | |
3134 | return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR); | |
3135 | } else { | |
3136 | update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask)); | |
3137 | if (opm->advertise) { | |
3138 | netdev_set_advertisements(port->netdev, ntohl(opm->advertise)); | |
3139 | } | |
3140 | } | |
3141 | return 0; | |
3142 | } | |
3143 | ||
3144 | static struct ofpbuf * | |
06a5e131 | 3145 | make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len) |
064af421 BP |
3146 | { |
3147 | struct ofp_stats_reply *osr; | |
3148 | struct ofpbuf *msg; | |
3149 | ||
3150 | msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX)); | |
3151 | osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg); | |
3152 | osr->type = type; | |
3153 | osr->flags = htons(0); | |
3154 | return msg; | |
3155 | } | |
3156 | ||
3157 | static struct ofpbuf * | |
06a5e131 | 3158 | start_ofp_stats_reply(const struct ofp_stats_request *request, size_t body_len) |
064af421 | 3159 | { |
06a5e131 | 3160 | return make_ofp_stats_reply(request->header.xid, request->type, body_len); |
064af421 BP |
3161 | } |
3162 | ||
3163 | static void * | |
06a5e131 BP |
3164 | append_ofp_stats_reply(size_t nbytes, struct ofconn *ofconn, |
3165 | struct ofpbuf **msgp) | |
064af421 BP |
3166 | { |
3167 | struct ofpbuf *msg = *msgp; | |
3168 | assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply)); | |
3169 | if (nbytes + msg->size > UINT16_MAX) { | |
3170 | struct ofp_stats_reply *reply = msg->data; | |
3171 | reply->flags = htons(OFPSF_REPLY_MORE); | |
06a5e131 | 3172 | *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes); |
064af421 BP |
3173 | queue_tx(msg, ofconn, ofconn->reply_counter); |
3174 | } | |
3175 | return ofpbuf_put_uninit(*msgp, nbytes); | |
3176 | } | |
3177 | ||
09246b99 BP |
3178 | static struct ofpbuf * |
3179 | make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len) | |
3180 | { | |
3181 | struct nicira_stats_msg *nsm; | |
3182 | struct ofpbuf *msg; | |
3183 | ||
3184 | msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX)); | |
3185 | nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg); | |
3186 | nsm->type = htons(OFPST_VENDOR); | |
3187 | nsm->flags = htons(0); | |
3188 | nsm->vendor = htonl(NX_VENDOR_ID); | |
3189 | nsm->subtype = htonl(subtype); | |
3190 | return msg; | |
3191 | } | |
3192 | ||
3193 | static struct ofpbuf * | |
3194 | start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len) | |
3195 | { | |
3196 | return make_nxstats_reply(request->header.xid, request->subtype, body_len); | |
3197 | } | |
3198 | ||
3199 | static void | |
3200 | append_nxstats_reply(size_t nbytes, struct ofconn *ofconn, | |
3201 | struct ofpbuf **msgp) | |
3202 | { | |
3203 | struct ofpbuf *msg = *msgp; | |
3204 | assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg)); | |
3205 | if (nbytes + msg->size > UINT16_MAX) { | |
3206 | struct nicira_stats_msg *reply = msg->data; | |
3207 | reply->flags = htons(OFPSF_REPLY_MORE); | |
3208 | *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes); | |
3209 | queue_tx(msg, ofconn, ofconn->reply_counter); | |
3210 | } | |
3211 | ofpbuf_prealloc_tailroom(*msgp, nbytes); | |
3212 | } | |
3213 | ||
064af421 | 3214 | static int |
3269c562 BP |
3215 | handle_desc_stats_request(struct ofconn *ofconn, |
3216 | struct ofp_stats_request *request) | |
064af421 | 3217 | { |
3269c562 | 3218 | struct ofproto *p = ofconn->ofproto; |
064af421 BP |
3219 | struct ofp_desc_stats *ods; |
3220 | struct ofpbuf *msg; | |
3221 | ||
06a5e131 BP |
3222 | msg = start_ofp_stats_reply(request, sizeof *ods); |
3223 | ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg); | |
5a719c38 JP |
3224 | memset(ods, 0, sizeof *ods); |
3225 | ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc); | |
3226 | ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc); | |
3227 | ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc); | |
3228 | ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num); | |
3229 | ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc); | |
064af421 BP |
3230 | queue_tx(msg, ofconn, ofconn->reply_counter); |
3231 | ||
3232 | return 0; | |
3233 | } | |
3234 | ||
064af421 | 3235 | static int |
3269c562 | 3236 | handle_table_stats_request(struct ofconn *ofconn, |
064af421 BP |
3237 | struct ofp_stats_request *request) |
3238 | { | |
3269c562 | 3239 | struct ofproto *p = ofconn->ofproto; |
064af421 BP |
3240 | struct ofp_table_stats *ots; |
3241 | struct ofpbuf *msg; | |
064af421 | 3242 | |
06a5e131 | 3243 | msg = start_ofp_stats_reply(request, sizeof *ots * 2); |
064af421 | 3244 | |
064af421 | 3245 | /* Classifier table. */ |
06a5e131 | 3246 | ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg); |
064af421 | 3247 | memset(ots, 0, sizeof *ots); |
064af421 | 3248 | strcpy(ots->name, "classifier"); |
b70eac89 | 3249 | ots->wildcards = (ofconn->flow_format == NXFF_OPENFLOW10 |
f9bfea14 | 3250 | ? htonl(OFPFW_ALL) : htonl(OVSFW_ALL)); |
ad828225 | 3251 | ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */ |
bcf84111 | 3252 | ots->active_count = htonl(classifier_count(&p->cls)); |
064af421 BP |
3253 | ots->lookup_count = htonll(0); /* XXX */ |
3254 | ots->matched_count = htonll(0); /* XXX */ | |
3255 | ||
3256 | queue_tx(msg, ofconn, ofconn->reply_counter); | |
3257 | return 0; | |
3258 | } | |
3259 | ||
abaad8cf | 3260 | static void |
ca0f572c | 3261 | append_port_stat(struct ofport *port, struct ofconn *ofconn, |
a4948b95 | 3262 | struct ofpbuf **msgp) |
abaad8cf JP |
3263 | { |
3264 | struct netdev_stats stats; | |
3265 | struct ofp_port_stats *ops; | |
3266 | ||
d295e8e9 JP |
3267 | /* Intentionally ignore return value, since errors will set |
3268 | * 'stats' to all-1s, which is correct for OpenFlow, and | |
abaad8cf JP |
3269 | * netdev_get_stats() will log errors. */ |
3270 | netdev_get_stats(port->netdev, &stats); | |
3271 | ||
06a5e131 | 3272 | ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp); |
ca0f572c | 3273 | ops->port_no = htons(port->opp.port_no); |
abaad8cf JP |
3274 | memset(ops->pad, 0, sizeof ops->pad); |
3275 | ops->rx_packets = htonll(stats.rx_packets); | |
3276 | ops->tx_packets = htonll(stats.tx_packets); | |
3277 | ops->rx_bytes = htonll(stats.rx_bytes); | |
3278 | ops->tx_bytes = htonll(stats.tx_bytes); | |
3279 | ops->rx_dropped = htonll(stats.rx_dropped); | |
3280 | ops->tx_dropped = htonll(stats.tx_dropped); | |
3281 | ops->rx_errors = htonll(stats.rx_errors); | |
3282 | ops->tx_errors = htonll(stats.tx_errors); | |
3283 | ops->rx_frame_err = htonll(stats.rx_frame_errors); | |
3284 | ops->rx_over_err = htonll(stats.rx_over_errors); | |
3285 | ops->rx_crc_err = htonll(stats.rx_crc_errors); | |
3286 | ops->collisions = htonll(stats.collisions); | |
3287 | } | |
3288 | ||
064af421 | 3289 | static int |
3269c562 | 3290 | handle_port_stats_request(struct ofconn *ofconn, struct ofp_stats_request *osr, |
abaad8cf | 3291 | size_t arg_size) |
064af421 | 3292 | { |
3269c562 | 3293 | struct ofproto *p = ofconn->ofproto; |
abaad8cf | 3294 | struct ofp_port_stats_request *psr; |
064af421 BP |
3295 | struct ofp_port_stats *ops; |
3296 | struct ofpbuf *msg; | |
3297 | struct ofport *port; | |
064af421 | 3298 | |
abaad8cf JP |
3299 | if (arg_size != sizeof *psr) { |
3300 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
3301 | } | |
3302 | psr = (struct ofp_port_stats_request *) osr->body; | |
3303 | ||
06a5e131 | 3304 | msg = start_ofp_stats_reply(osr, sizeof *ops * 16); |
abaad8cf | 3305 | if (psr->port_no != htons(OFPP_NONE)) { |
ca0f572c | 3306 | port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no))); |
abaad8cf | 3307 | if (port) { |
ca0f572c | 3308 | append_port_stat(port, ofconn, &msg); |
abaad8cf JP |
3309 | } |
3310 | } else { | |
4e8e4213 | 3311 | HMAP_FOR_EACH (port, hmap_node, &p->ports) { |
ca0f572c | 3312 | append_port_stat(port, ofconn, &msg); |
abaad8cf | 3313 | } |
064af421 BP |
3314 | } |
3315 | ||
3316 | queue_tx(msg, ofconn, ofconn->reply_counter); | |
3317 | return 0; | |
3318 | } | |
3319 | ||
3320 | struct flow_stats_cbdata { | |
064af421 | 3321 | struct ofconn *ofconn; |
8054fc48 | 3322 | ovs_be16 out_port; |
064af421 BP |
3323 | struct ofpbuf *msg; |
3324 | }; | |
3325 | ||
01149cfd | 3326 | /* Obtains statistic counters for 'rule' within 'p' and stores them into |
bcf84111 BP |
3327 | * '*packet_countp' and '*byte_countp'. The returned statistics include |
3328 | * statistics for all of 'rule''s facets. */ | |
064af421 BP |
3329 | static void |
3330 | query_stats(struct ofproto *p, struct rule *rule, | |
3331 | uint64_t *packet_countp, uint64_t *byte_countp) | |
3332 | { | |
3333 | uint64_t packet_count, byte_count; | |
bcf84111 | 3334 | struct facet *facet; |
064af421 BP |
3335 | struct odp_flow *odp_flows; |
3336 | size_t n_odp_flows; | |
3337 | ||
01149cfd | 3338 | /* Start from historical data for 'rule' itself that are no longer tracked |
bcf84111 | 3339 | * by the datapath. This counts, for example, facets that have expired. */ |
b3137fe8 JG |
3340 | packet_count = rule->packet_count; |
3341 | byte_count = rule->byte_count; | |
3342 | ||
bcf84111 | 3343 | /* Prepare to ask the datapath for statistics on all of the rule's facets. |
01149cfd BP |
3344 | * |
3345 | * Also, add any statistics that are not tracked by the datapath for each | |
bcf84111 | 3346 | * facet. This includes, for example, statistics for packets that were |
01149cfd | 3347 | * executed "by hand" by ofproto via dpif_execute() but must be accounted |
bcf84111 BP |
3348 | * to a rule. */ |
3349 | odp_flows = xzalloc(list_size(&rule->facets) * sizeof *odp_flows); | |
3350 | n_odp_flows = 0; | |
3351 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
3352 | struct odp_flow *odp_flow = &odp_flows[n_odp_flows++]; | |
3353 | odp_flow_key_from_flow(&odp_flow->key, &facet->flow); | |
3354 | packet_count += facet->packet_count; | |
3355 | byte_count += facet->byte_count; | |
064af421 BP |
3356 | } |
3357 | ||
28998b22 | 3358 | /* Fetch up-to-date statistics from the datapath and add them in. */ |
c228a364 | 3359 | if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) { |
064af421 | 3360 | size_t i; |
bcf84111 | 3361 | |
064af421 BP |
3362 | for (i = 0; i < n_odp_flows; i++) { |
3363 | struct odp_flow *odp_flow = &odp_flows[i]; | |
3364 | packet_count += odp_flow->stats.n_packets; | |
3365 | byte_count += odp_flow->stats.n_bytes; | |
3366 | } | |
3367 | } | |
3368 | free(odp_flows); | |
3369 | ||
01149cfd | 3370 | /* Return the stats to the caller. */ |
064af421 BP |
3371 | *packet_countp = packet_count; |
3372 | *byte_countp = byte_count; | |
3373 | } | |
3374 | ||
c6ebb8fb BP |
3375 | static void |
3376 | calc_flow_duration(long long int start, ovs_be32 *sec, ovs_be32 *nsec) | |
3377 | { | |
3378 | long long int msecs = time_msec() - start; | |
3379 | *sec = htonl(msecs / 1000); | |
3380 | *nsec = htonl((msecs % 1000) * (1000 * 1000)); | |
3381 | } | |
3382 | ||
064af421 BP |
3383 | static void |
3384 | flow_stats_cb(struct cls_rule *rule_, void *cbdata_) | |
3385 | { | |
3386 | struct rule *rule = rule_from_cls_rule(rule_); | |
3387 | struct flow_stats_cbdata *cbdata = cbdata_; | |
3388 | struct ofp_flow_stats *ofs; | |
3389 | uint64_t packet_count, byte_count; | |
3390 | size_t act_len, len; | |
3391 | ||
3392 | if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) { | |
3393 | return; | |
3394 | } | |
3395 | ||
3396 | act_len = sizeof *rule->actions * rule->n_actions; | |
3397 | len = offsetof(struct ofp_flow_stats, actions) + act_len; | |
3398 | ||
3269c562 | 3399 | query_stats(cbdata->ofconn->ofproto, rule, &packet_count, &byte_count); |
064af421 | 3400 | |
06a5e131 | 3401 | ofs = append_ofp_stats_reply(len, cbdata->ofconn, &cbdata->msg); |
064af421 | 3402 | ofs->length = htons(len); |
ad828225 | 3403 | ofs->table_id = 0; |
064af421 | 3404 | ofs->pad = 0; |
659586ef | 3405 | flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, |
b70eac89 | 3406 | cbdata->ofconn->flow_format, &ofs->match); |
c6ebb8fb | 3407 | calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec); |
39997502 | 3408 | ofs->cookie = rule->flow_cookie; |
064af421 BP |
3409 | ofs->priority = htons(rule->cr.priority); |
3410 | ofs->idle_timeout = htons(rule->idle_timeout); | |
3411 | ofs->hard_timeout = htons(rule->hard_timeout); | |
39997502 | 3412 | memset(ofs->pad2, 0, sizeof ofs->pad2); |
064af421 BP |
3413 | ofs->packet_count = htonll(packet_count); |
3414 | ofs->byte_count = htonll(byte_count); | |
3dffcf07 BP |
3415 | if (rule->n_actions > 0) { |
3416 | memcpy(ofs->actions, rule->actions, act_len); | |
3417 | } | |
064af421 BP |
3418 | } |
3419 | ||
3c4486a5 BP |
3420 | static bool |
3421 | is_valid_table(uint8_t table_id) | |
064af421 | 3422 | { |
3c4486a5 | 3423 | return table_id == 0 || table_id == 0xff; |
064af421 BP |
3424 | } |
3425 | ||
3426 | static int | |
3269c562 BP |
3427 | handle_flow_stats_request(struct ofconn *ofconn, |
3428 | const struct ofp_stats_request *osr, size_t arg_size) | |
064af421 BP |
3429 | { |
3430 | struct ofp_flow_stats_request *fsr; | |
3431 | struct flow_stats_cbdata cbdata; | |
064af421 BP |
3432 | |
3433 | if (arg_size != sizeof *fsr) { | |
49bdc010 | 3434 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); |
064af421 BP |
3435 | } |
3436 | fsr = (struct ofp_flow_stats_request *) osr->body; | |
3437 | ||
3438 | COVERAGE_INC(ofproto_flows_req); | |
06a5e131 | 3439 | cbdata.msg = start_ofp_stats_reply(osr, 1024); |
3c4486a5 BP |
3440 | if (is_valid_table(fsr->table_id)) { |
3441 | struct cls_rule target; | |
3442 | ||
3443 | cbdata.ofconn = ofconn; | |
3444 | cbdata.out_port = fsr->out_port; | |
3445 | cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0, &target); | |
3446 | classifier_for_each_match(&ofconn->ofproto->cls, &target, | |
3447 | flow_stats_cb, &cbdata); | |
3448 | } | |
3449 | ||
064af421 BP |
3450 | queue_tx(cbdata.msg, ofconn, ofconn->reply_counter); |
3451 | return 0; | |
3452 | } | |
3453 | ||
09246b99 BP |
3454 | static void |
3455 | nx_flow_stats_cb(struct cls_rule *rule_, void *cbdata_) | |
3456 | { | |
3457 | struct rule *rule = rule_from_cls_rule(rule_); | |
3458 | struct flow_stats_cbdata *cbdata = cbdata_; | |
3459 | struct nx_flow_stats *nfs; | |
3460 | uint64_t packet_count, byte_count; | |
3461 | size_t act_len, start_len; | |
3462 | ||
3463 | if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) { | |
3464 | return; | |
3465 | } | |
3466 | ||
3467 | query_stats(cbdata->ofconn->ofproto, rule, &packet_count, &byte_count); | |
3468 | ||
3469 | act_len = sizeof *rule->actions * rule->n_actions; | |
3470 | ||
3471 | start_len = cbdata->msg->size; | |
3472 | append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, | |
3473 | cbdata->ofconn, &cbdata->msg); | |
3474 | nfs = ofpbuf_put_uninit(cbdata->msg, sizeof *nfs); | |
3475 | nfs->table_id = 0; | |
3476 | nfs->pad = 0; | |
3477 | calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec); | |
3478 | nfs->cookie = rule->flow_cookie; | |
3479 | nfs->priority = htons(rule->cr.priority); | |
3480 | nfs->idle_timeout = htons(rule->idle_timeout); | |
3481 | nfs->hard_timeout = htons(rule->hard_timeout); | |
3482 | nfs->match_len = htons(nx_put_match(cbdata->msg, &rule->cr)); | |
3483 | memset(nfs->pad2, 0, sizeof nfs->pad2); | |
3484 | nfs->packet_count = htonll(packet_count); | |
3485 | nfs->byte_count = htonll(byte_count); | |
3486 | if (rule->n_actions > 0) { | |
3487 | ofpbuf_put(cbdata->msg, rule->actions, act_len); | |
3488 | } | |
3489 | nfs->length = htons(cbdata->msg->size - start_len); | |
3490 | } | |
3491 | ||
3492 | static int | |
3493 | handle_nxst_flow(struct ofconn *ofconn, struct ofpbuf *b) | |
3494 | { | |
3495 | struct nx_flow_stats_request *nfsr; | |
3496 | struct flow_stats_cbdata cbdata; | |
3497 | struct cls_rule target; | |
3498 | int error; | |
3499 | ||
3500 | /* Dissect the message. */ | |
3501 | nfsr = ofpbuf_try_pull(b, sizeof *nfsr); | |
3502 | if (!nfsr) { | |
3503 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
3504 | } | |
3505 | error = nx_pull_match(b, ntohs(nfsr->match_len), 0, &target); | |
3506 | if (error) { | |
3507 | return error; | |
3508 | } | |
3509 | ||
3510 | COVERAGE_INC(ofproto_flows_req); | |
09246b99 | 3511 | cbdata.msg = start_nxstats_reply(&nfsr->nsm, 1024); |
3c4486a5 BP |
3512 | if (is_valid_table(nfsr->table_id)) { |
3513 | cbdata.ofconn = ofconn; | |
3514 | cbdata.out_port = nfsr->out_port; | |
3515 | classifier_for_each_match(&ofconn->ofproto->cls, &target, | |
3516 | nx_flow_stats_cb, &cbdata); | |
3517 | } | |
09246b99 BP |
3518 | queue_tx(cbdata.msg, ofconn, ofconn->reply_counter); |
3519 | return 0; | |
3520 | } | |
3521 | ||
4f2cad2c JP |
3522 | struct flow_stats_ds_cbdata { |
3523 | struct ofproto *ofproto; | |
3524 | struct ds *results; | |
3525 | }; | |
3526 | ||
3527 | static void | |
3528 | flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_) | |
3529 | { | |
3530 | struct rule *rule = rule_from_cls_rule(rule_); | |
3531 | struct flow_stats_ds_cbdata *cbdata = cbdata_; | |
3532 | struct ds *results = cbdata->results; | |
3533 | struct ofp_match match; | |
3534 | uint64_t packet_count, byte_count; | |
3535 | size_t act_len = sizeof *rule->actions * rule->n_actions; | |
3536 | ||
4f2cad2c | 3537 | query_stats(cbdata->ofproto, rule, &packet_count, &byte_count); |
659586ef | 3538 | flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, |
b70eac89 | 3539 | NXFF_OPENFLOW10, &match); |
4f2cad2c JP |
3540 | |
3541 | ds_put_format(results, "duration=%llds, ", | |
3542 | (time_msec() - rule->created) / 1000); | |
52ae00b3 | 3543 | ds_put_format(results, "priority=%u, ", rule->cr.priority); |
4f2cad2c JP |
3544 | ds_put_format(results, "n_packets=%"PRIu64", ", packet_count); |
3545 | ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count); | |
3546 | ofp_print_match(results, &match, true); | |
3dffcf07 BP |
3547 | if (act_len > 0) { |
3548 | ofp_print_actions(results, &rule->actions->header, act_len); | |
3c8552c1 JP |
3549 | } else { |
3550 | ds_put_cstr(results, "drop"); | |
3dffcf07 | 3551 | } |
4f2cad2c JP |
3552 | ds_put_cstr(results, "\n"); |
3553 | } | |
3554 | ||
d295e8e9 | 3555 | /* Adds a pretty-printed description of all flows to 'results', including |
4f2cad2c JP |
3556 | * those marked hidden by secchan (e.g., by in-band control). */ |
3557 | void | |
3558 | ofproto_get_all_flows(struct ofproto *p, struct ds *results) | |
3559 | { | |
3560 | struct ofp_match match; | |
3561 | struct cls_rule target; | |
3562 | struct flow_stats_ds_cbdata cbdata; | |
3563 | ||
3564 | memset(&match, 0, sizeof match); | |
659586ef | 3565 | match.wildcards = htonl(OVSFW_ALL); |
4f2cad2c JP |
3566 | |
3567 | cbdata.ofproto = p; | |
3568 | cbdata.results = results; | |
3569 | ||
f9bfea14 | 3570 | cls_rule_from_match(&match, 0, NXFF_OPENFLOW10, 0, &target); |
3c4486a5 | 3571 | classifier_for_each_match(&p->cls, &target, flow_stats_ds_cb, &cbdata); |
4f2cad2c JP |
3572 | } |
3573 | ||
064af421 BP |
3574 | struct aggregate_stats_cbdata { |
3575 | struct ofproto *ofproto; | |
8054fc48 | 3576 | ovs_be16 out_port; |
064af421 BP |
3577 | uint64_t packet_count; |
3578 | uint64_t byte_count; | |
3579 | uint32_t n_flows; | |
3580 | }; | |
3581 | ||
3582 | static void | |
3583 | aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_) | |
3584 | { | |
3585 | struct rule *rule = rule_from_cls_rule(rule_); | |
3586 | struct aggregate_stats_cbdata *cbdata = cbdata_; | |
3587 | uint64_t packet_count, byte_count; | |
3588 | ||
3589 | if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) { | |
3590 | return; | |
3591 | } | |
3592 | ||
3593 | query_stats(cbdata->ofproto, rule, &packet_count, &byte_count); | |
3594 | ||
3595 | cbdata->packet_count += packet_count; | |
3596 | cbdata->byte_count += byte_count; | |
3597 | cbdata->n_flows++; | |
3598 | } | |
3599 | ||
27d34fce BP |
3600 | static void |
3601 | query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target, | |
734bbeb4 | 3602 | ovs_be16 out_port, uint8_t table_id, |
27d34fce BP |
3603 | struct ofp_aggregate_stats_reply *oasr) |
3604 | { | |
3605 | struct aggregate_stats_cbdata cbdata; | |
3606 | ||
3607 | COVERAGE_INC(ofproto_agg_request); | |
27d34fce BP |
3608 | cbdata.packet_count = 0; |
3609 | cbdata.byte_count = 0; | |
3610 | cbdata.n_flows = 0; | |
3c4486a5 BP |
3611 | if (is_valid_table(table_id)) { |
3612 | cbdata.ofproto = ofproto; | |
3613 | cbdata.out_port = out_port; | |
3614 | ||
3615 | classifier_for_each_match(&ofproto->cls, target, | |
3616 | aggregate_stats_cb, &cbdata); | |
3617 | } | |
27d34fce BP |
3618 | |
3619 | oasr->flow_count = htonl(cbdata.n_flows); | |
3620 | oasr->packet_count = htonll(cbdata.packet_count); | |
3621 | oasr->byte_count = htonll(cbdata.byte_count); | |
3622 | memset(oasr->pad, 0, sizeof oasr->pad); | |
3623 | } | |
3624 | ||
064af421 | 3625 | static int |
3269c562 | 3626 | handle_aggregate_stats_request(struct ofconn *ofconn, |
064af421 BP |
3627 | const struct ofp_stats_request *osr, |
3628 | size_t arg_size) | |
3629 | { | |
27d34fce | 3630 | struct ofp_aggregate_stats_request *request; |
064af421 | 3631 | struct ofp_aggregate_stats_reply *reply; |
064af421 BP |
3632 | struct cls_rule target; |
3633 | struct ofpbuf *msg; | |
3634 | ||
27d34fce | 3635 | if (arg_size != sizeof *request) { |
49bdc010 | 3636 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); |
064af421 | 3637 | } |
27d34fce | 3638 | request = (struct ofp_aggregate_stats_request *) osr->body; |
064af421 | 3639 | |
27d34fce | 3640 | cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0, &target); |
064af421 | 3641 | |
06a5e131 BP |
3642 | msg = start_ofp_stats_reply(osr, sizeof *reply); |
3643 | reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg); | |
27d34fce BP |
3644 | query_aggregate_stats(ofconn->ofproto, &target, request->out_port, |
3645 | request->table_id, reply); | |
064af421 BP |
3646 | queue_tx(msg, ofconn, ofconn->reply_counter); |
3647 | return 0; | |
3648 | } | |
3649 | ||
09246b99 BP |
3650 | static int |
3651 | handle_nxst_aggregate(struct ofconn *ofconn, struct ofpbuf *b) | |
3652 | { | |
3653 | struct nx_aggregate_stats_request *request; | |
3654 | struct ofp_aggregate_stats_reply *reply; | |
3655 | struct cls_rule target; | |
3656 | struct ofpbuf *buf; | |
3657 | int error; | |
3658 | ||
3659 | /* Dissect the message. */ | |
3660 | request = ofpbuf_try_pull(b, sizeof *request); | |
3661 | if (!request) { | |
3662 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
3663 | } | |
3664 | error = nx_pull_match(b, ntohs(request->match_len), 0, &target); | |
3665 | if (error) { | |
3666 | return error; | |
3667 | } | |
3668 | ||
3669 | /* Reply. */ | |
3670 | COVERAGE_INC(ofproto_flows_req); | |
3671 | buf = start_nxstats_reply(&request->nsm, sizeof *reply); | |
3672 | reply = ofpbuf_put_uninit(buf, sizeof *reply); | |
3673 | query_aggregate_stats(ofconn->ofproto, &target, request->out_port, | |
3674 | request->table_id, reply); | |
3675 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
3676 | ||
3677 | return 0; | |
3678 | } | |
3679 | ||
c1c9c9c4 BP |
3680 | struct queue_stats_cbdata { |
3681 | struct ofconn *ofconn; | |
ca0f572c | 3682 | struct ofport *ofport; |
c1c9c9c4 | 3683 | struct ofpbuf *msg; |
c1c9c9c4 BP |
3684 | }; |
3685 | ||
3686 | static void | |
db9220c3 | 3687 | put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id, |
c1c9c9c4 BP |
3688 | const struct netdev_queue_stats *stats) |
3689 | { | |
3690 | struct ofp_queue_stats *reply; | |
3691 | ||
06a5e131 | 3692 | reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg); |
ca0f572c | 3693 | reply->port_no = htons(cbdata->ofport->opp.port_no); |
c1c9c9c4 BP |
3694 | memset(reply->pad, 0, sizeof reply->pad); |
3695 | reply->queue_id = htonl(queue_id); | |
3696 | reply->tx_bytes = htonll(stats->tx_bytes); | |
3697 | reply->tx_packets = htonll(stats->tx_packets); | |
3698 | reply->tx_errors = htonll(stats->tx_errors); | |
3699 | } | |
3700 | ||
3701 | static void | |
db9220c3 | 3702 | handle_queue_stats_dump_cb(uint32_t queue_id, |
c1c9c9c4 BP |
3703 | struct netdev_queue_stats *stats, |
3704 | void *cbdata_) | |
3705 | { | |
3706 | struct queue_stats_cbdata *cbdata = cbdata_; | |
3707 | ||
3708 | put_queue_stats(cbdata, queue_id, stats); | |
3709 | } | |
3710 | ||
3711 | static void | |
ca0f572c | 3712 | handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id, |
c1c9c9c4 BP |
3713 | struct queue_stats_cbdata *cbdata) |
3714 | { | |
ca0f572c | 3715 | cbdata->ofport = port; |
c1c9c9c4 BP |
3716 | if (queue_id == OFPQ_ALL) { |
3717 | netdev_dump_queue_stats(port->netdev, | |
3718 | handle_queue_stats_dump_cb, cbdata); | |
3719 | } else { | |
3720 | struct netdev_queue_stats stats; | |
3721 | ||
1ac788f6 BP |
3722 | if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) { |
3723 | put_queue_stats(cbdata, queue_id, &stats); | |
3724 | } | |
c1c9c9c4 BP |
3725 | } |
3726 | } | |
3727 | ||
3728 | static int | |
3269c562 | 3729 | handle_queue_stats_request(struct ofconn *ofconn, |
c1c9c9c4 BP |
3730 | const struct ofp_stats_request *osr, |
3731 | size_t arg_size) | |
3732 | { | |
3269c562 | 3733 | struct ofproto *ofproto = ofconn->ofproto; |
c1c9c9c4 BP |
3734 | struct ofp_queue_stats_request *qsr; |
3735 | struct queue_stats_cbdata cbdata; | |
3736 | struct ofport *port; | |
3737 | unsigned int port_no; | |
3738 | uint32_t queue_id; | |
3739 | ||
3740 | if (arg_size != sizeof *qsr) { | |
3741 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
3742 | } | |
3743 | qsr = (struct ofp_queue_stats_request *) osr->body; | |
3744 | ||
3745 | COVERAGE_INC(ofproto_queue_req); | |
3746 | ||
3747 | cbdata.ofconn = ofconn; | |
06a5e131 | 3748 | cbdata.msg = start_ofp_stats_reply(osr, 128); |
c1c9c9c4 BP |
3749 | |
3750 | port_no = ntohs(qsr->port_no); | |
3751 | queue_id = ntohl(qsr->queue_id); | |
3752 | if (port_no == OFPP_ALL) { | |
4e8e4213 | 3753 | HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) { |
ca0f572c | 3754 | handle_queue_stats_for_port(port, queue_id, &cbdata); |
c1c9c9c4 BP |
3755 | } |
3756 | } else if (port_no < ofproto->max_ports) { | |
ca0f572c | 3757 | port = get_port(ofproto, ofp_port_to_odp_port(port_no)); |
c1c9c9c4 | 3758 | if (port) { |
ca0f572c | 3759 | handle_queue_stats_for_port(port, queue_id, &cbdata); |
c1c9c9c4 BP |
3760 | } |
3761 | } else { | |
3762 | ofpbuf_delete(cbdata.msg); | |
3763 | return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT); | |
3764 | } | |
3765 | queue_tx(cbdata.msg, ofconn, ofconn->reply_counter); | |
3766 | ||
3767 | return 0; | |
3768 | } | |
3769 | ||
09246b99 BP |
3770 | static int |
3771 | handle_vendor_stats_request(struct ofconn *ofconn, | |
3772 | struct ofp_stats_request *osr, size_t arg_size) | |
3773 | { | |
3774 | struct nicira_stats_msg *nsm; | |
3775 | struct ofpbuf b; | |
3776 | ovs_be32 vendor; | |
3777 | ||
3778 | if (arg_size < 4) { | |
3779 | VLOG_WARN_RL(&rl, "truncated vendor stats request body"); | |
3780 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
3781 | } | |
3782 | ||
3783 | memcpy(&vendor, osr->body, sizeof vendor); | |
3784 | if (vendor != htonl(NX_VENDOR_ID)) { | |
3785 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR); | |
3786 | } | |
3787 | ||
487cedcf | 3788 | if (ntohs(osr->header.length) < sizeof(struct nicira_stats_msg)) { |
09246b99 BP |
3789 | VLOG_WARN_RL(&rl, "truncated Nicira stats request"); |
3790 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
3791 | } | |
3792 | ||
3793 | nsm = (struct nicira_stats_msg *) osr; | |
3794 | b.data = nsm; | |
3795 | b.size = ntohs(nsm->header.length); | |
3796 | switch (ntohl(nsm->subtype)) { | |
3797 | case NXST_FLOW: | |
3798 | return handle_nxst_flow(ofconn, &b); | |
3799 | ||
3800 | case NXST_AGGREGATE: | |
3801 | return handle_nxst_aggregate(ofconn, &b); | |
3802 | ||
3803 | default: | |
3804 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); | |
3805 | } | |
3806 | } | |
3807 | ||
064af421 | 3808 | static int |
3269c562 | 3809 | handle_stats_request(struct ofconn *ofconn, struct ofp_header *oh) |
064af421 BP |
3810 | { |
3811 | struct ofp_stats_request *osr; | |
3812 | size_t arg_size; | |
3813 | int error; | |
3814 | ||
3815 | error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr, | |
3816 | 1, &arg_size); | |
3817 | if (error) { | |
3818 | return error; | |
3819 | } | |
3820 | osr = (struct ofp_stats_request *) oh; | |
3821 | ||
3822 | switch (ntohs(osr->type)) { | |
3823 | case OFPST_DESC: | |
3269c562 | 3824 | return handle_desc_stats_request(ofconn, osr); |
064af421 BP |
3825 | |
3826 | case OFPST_FLOW: | |
3269c562 | 3827 | return handle_flow_stats_request(ofconn, osr, arg_size); |
064af421 BP |
3828 | |
3829 | case OFPST_AGGREGATE: | |
3269c562 | 3830 | return handle_aggregate_stats_request(ofconn, osr, arg_size); |
064af421 BP |
3831 | |
3832 | case OFPST_TABLE: | |
3269c562 | 3833 | return handle_table_stats_request(ofconn, osr); |
064af421 BP |
3834 | |
3835 | case OFPST_PORT: | |
3269c562 | 3836 | return handle_port_stats_request(ofconn, osr, arg_size); |
064af421 | 3837 | |
c1c9c9c4 | 3838 | case OFPST_QUEUE: |
3269c562 | 3839 | return handle_queue_stats_request(ofconn, osr, arg_size); |
c1c9c9c4 | 3840 | |
064af421 | 3841 | case OFPST_VENDOR: |
09246b99 | 3842 | return handle_vendor_stats_request(ofconn, osr, arg_size); |
064af421 BP |
3843 | |
3844 | default: | |
3845 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT); | |
3846 | } | |
3847 | } | |
3848 | ||
3849 | static long long int | |
3850 | msec_from_nsec(uint64_t sec, uint32_t nsec) | |
3851 | { | |
3852 | return !sec ? 0 : sec * 1000 + nsec / 1000000; | |
3853 | } | |
3854 | ||
3855 | static void | |
bcf84111 BP |
3856 | facet_update_time(struct ofproto *ofproto, struct facet *facet, |
3857 | const struct odp_flow_stats *stats) | |
064af421 BP |
3858 | { |
3859 | long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec); | |
bcf84111 BP |
3860 | if (used > facet->used) { |
3861 | facet->used = used; | |
3862 | if (used > facet->rule->used) { | |
3863 | facet->rule->used = used; | |
4836f9f2 | 3864 | } |
bcf84111 | 3865 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used); |
064af421 BP |
3866 | } |
3867 | } | |
3868 | ||
bcf84111 BP |
3869 | /* Folds the statistics from 'stats' into the counters in 'facet'. |
3870 | * | |
3871 | * Because of the meaning of a facet's counters, it only makes sense to do this | |
3872 | * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a | |
3873 | * packet that was sent by hand or if it represents statistics that have been | |
3874 | * cleared out of the datapath. */ | |
064af421 | 3875 | static void |
bcf84111 BP |
3876 | facet_update_stats(struct ofproto *ofproto, struct facet *facet, |
3877 | const struct odp_flow_stats *stats) | |
064af421 | 3878 | { |
064af421 | 3879 | if (stats->n_packets) { |
bcf84111 BP |
3880 | facet_update_time(ofproto, facet, stats); |
3881 | facet->packet_count += stats->n_packets; | |
3882 | facet->byte_count += stats->n_bytes; | |
3883 | netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags); | |
064af421 BP |
3884 | } |
3885 | } | |
3886 | ||
3052b0c5 BP |
3887 | struct flow_mod { |
3888 | struct cls_rule cr; | |
3889 | ovs_be64 cookie; | |
3890 | uint16_t command; | |
3891 | uint16_t idle_timeout; | |
3892 | uint16_t hard_timeout; | |
3893 | uint32_t buffer_id; | |
3894 | uint16_t out_port; | |
3895 | uint16_t flags; | |
3896 | union ofp_action *actions; | |
3897 | size_t n_actions; | |
3898 | }; | |
3899 | ||
79eee1eb BP |
3900 | /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT |
3901 | * in which no matching flow already exists in the flow table. | |
3902 | * | |
3903 | * Adds the flow specified by 'ofm', which is followed by 'n_actions' | |
3269c562 BP |
3904 | * ofp_actions, to ofconn->ofproto's flow table. Returns 0 on success or an |
3905 | * OpenFlow error code as encoded by ofp_mkerr() on failure. | |
79eee1eb BP |
3906 | * |
3907 | * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, | |
3908 | * if any. */ | |
064af421 | 3909 | static int |
3052b0c5 | 3910 | add_flow(struct ofconn *ofconn, struct flow_mod *fm) |
064af421 | 3911 | { |
3269c562 | 3912 | struct ofproto *p = ofconn->ofproto; |
064af421 BP |
3913 | struct ofpbuf *packet; |
3914 | struct rule *rule; | |
3915 | uint16_t in_port; | |
3916 | int error; | |
3917 | ||
3052b0c5 BP |
3918 | if (fm->flags & OFPFF_CHECK_OVERLAP |
3919 | && classifier_rule_overlaps(&p->cls, &fm->cr)) { | |
3920 | return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP); | |
49bdc010 JP |
3921 | } |
3922 | ||
064af421 | 3923 | error = 0; |
3052b0c5 BP |
3924 | if (fm->buffer_id != UINT32_MAX) { |
3925 | error = pktbuf_retrieve(ofconn->pktbuf, fm->buffer_id, | |
064af421 | 3926 | &packet, &in_port); |
212fe71c BP |
3927 | } else { |
3928 | packet = NULL; | |
165cd8a3 | 3929 | in_port = UINT16_MAX; |
064af421 BP |
3930 | } |
3931 | ||
bcf84111 BP |
3932 | rule = rule_create(&fm->cr, fm->actions, fm->n_actions, |
3933 | fm->idle_timeout, fm->hard_timeout, fm->cookie, | |
3934 | fm->flags & OFPFF_SEND_FLOW_REM); | |
afe75089 BP |
3935 | rule_insert(p, rule); |
3936 | if (packet) { | |
3937 | rule_execute(p, rule, in_port, packet); | |
3938 | } | |
064af421 BP |
3939 | return error; |
3940 | } | |
3941 | ||
79eee1eb | 3942 | static struct rule * |
3052b0c5 | 3943 | find_flow_strict(struct ofproto *p, const struct flow_mod *fm) |
064af421 | 3944 | { |
3052b0c5 | 3945 | return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr)); |
79eee1eb | 3946 | } |
064af421 | 3947 | |
79eee1eb | 3948 | static int |
3269c562 | 3949 | send_buffered_packet(struct ofconn *ofconn, |
3052b0c5 | 3950 | struct rule *rule, uint32_t buffer_id) |
79eee1eb BP |
3951 | { |
3952 | struct ofpbuf *packet; | |
3953 | uint16_t in_port; | |
79eee1eb | 3954 | int error; |
064af421 | 3955 | |
3052b0c5 | 3956 | if (buffer_id == UINT32_MAX) { |
79eee1eb | 3957 | return 0; |
064af421 | 3958 | } |
79eee1eb | 3959 | |
3052b0c5 | 3960 | error = pktbuf_retrieve(ofconn->pktbuf, buffer_id, &packet, &in_port); |
79eee1eb BP |
3961 | if (error) { |
3962 | return error; | |
3963 | } | |
3964 | ||
bcf84111 | 3965 | rule_execute(ofconn->ofproto, rule, in_port, packet); |
79eee1eb | 3966 | |
064af421 BP |
3967 | return 0; |
3968 | } | |
79eee1eb BP |
3969 | \f |
3970 | /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */ | |
064af421 BP |
3971 | |
3972 | struct modify_flows_cbdata { | |
3973 | struct ofproto *ofproto; | |
3052b0c5 | 3974 | const struct flow_mod *fm; |
79eee1eb | 3975 | struct rule *match; |
064af421 BP |
3976 | }; |
3977 | ||
3052b0c5 BP |
3978 | static int modify_flow(struct ofproto *, const struct flow_mod *, |
3979 | struct rule *); | |
79eee1eb BP |
3980 | static void modify_flows_cb(struct cls_rule *, void *cbdata_); |
3981 | ||
3982 | /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as | |
3983 | * encoded by ofp_mkerr() on failure. | |
3984 | * | |
3985 | * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, | |
3986 | * if any. */ | |
3987 | static int | |
3052b0c5 | 3988 | modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm) |
79eee1eb BP |
3989 | { |
3990 | struct modify_flows_cbdata cbdata; | |
79eee1eb | 3991 | |
3269c562 | 3992 | cbdata.ofproto = ofconn->ofproto; |
3052b0c5 | 3993 | cbdata.fm = fm; |
79eee1eb BP |
3994 | cbdata.match = NULL; |
3995 | ||
3c4486a5 | 3996 | classifier_for_each_match(&ofconn->ofproto->cls, &fm->cr, |
79eee1eb BP |
3997 | modify_flows_cb, &cbdata); |
3998 | if (cbdata.match) { | |
3999 | /* This credits the packet to whichever flow happened to happened to | |
4000 | * match last. That's weird. Maybe we should do a lookup for the | |
4001 | * flow that actually matches the packet? Who knows. */ | |
3052b0c5 | 4002 | send_buffered_packet(ofconn, cbdata.match, fm->buffer_id); |
79eee1eb BP |
4003 | return 0; |
4004 | } else { | |
3052b0c5 | 4005 | return add_flow(ofconn, fm); |
79eee1eb BP |
4006 | } |
4007 | } | |
4008 | ||
4009 | /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error | |
4010 | * code as encoded by ofp_mkerr() on failure. | |
4011 | * | |
4012 | * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, | |
4013 | * if any. */ | |
4014 | static int | |
3052b0c5 | 4015 | modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm) |
79eee1eb | 4016 | { |
3052b0c5 BP |
4017 | struct ofproto *p = ofconn->ofproto; |
4018 | struct rule *rule = find_flow_strict(p, fm); | |
79eee1eb | 4019 | if (rule && !rule_is_hidden(rule)) { |
3052b0c5 BP |
4020 | modify_flow(p, fm, rule); |
4021 | return send_buffered_packet(ofconn, rule, fm->buffer_id); | |
79eee1eb | 4022 | } else { |
3052b0c5 | 4023 | return add_flow(ofconn, fm); |
79eee1eb BP |
4024 | } |
4025 | } | |
4026 | ||
4027 | /* Callback for modify_flows_loose(). */ | |
064af421 BP |
4028 | static void |
4029 | modify_flows_cb(struct cls_rule *rule_, void *cbdata_) | |
4030 | { | |
4031 | struct rule *rule = rule_from_cls_rule(rule_); | |
4032 | struct modify_flows_cbdata *cbdata = cbdata_; | |
4033 | ||
79eee1eb BP |
4034 | if (!rule_is_hidden(rule)) { |
4035 | cbdata->match = rule; | |
3052b0c5 | 4036 | modify_flow(cbdata->ofproto, cbdata->fm, rule); |
064af421 | 4037 | } |
064af421 BP |
4038 | } |
4039 | ||
79eee1eb BP |
4040 | /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has |
4041 | * been identified as a flow in 'p''s flow table to be modified, by changing | |
4042 | * the rule's actions to match those in 'ofm' (which is followed by 'n_actions' | |
4043 | * ofp_action[] structures). */ | |
064af421 | 4044 | static int |
3052b0c5 | 4045 | modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule) |
064af421 | 4046 | { |
3052b0c5 | 4047 | size_t actions_len = fm->n_actions * sizeof *rule->actions; |
79eee1eb | 4048 | |
3052b0c5 | 4049 | rule->flow_cookie = fm->cookie; |
79eee1eb BP |
4050 | |
4051 | /* If the actions are the same, do nothing. */ | |
3052b0c5 BP |
4052 | if (fm->n_actions == rule->n_actions |
4053 | && (!fm->n_actions | |
4054 | || !memcmp(fm->actions, rule->actions, actions_len))) { | |
79eee1eb BP |
4055 | return 0; |
4056 | } | |
4057 | ||
4058 | /* Replace actions. */ | |
4059 | free(rule->actions); | |
3052b0c5 BP |
4060 | rule->actions = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL; |
4061 | rule->n_actions = fm->n_actions; | |
79eee1eb | 4062 | |
bcf84111 | 4063 | p->need_revalidate = true; |
79eee1eb BP |
4064 | |
4065 | return 0; | |
4066 | } | |
4067 | \f | |
4068 | /* OFPFC_DELETE implementation. */ | |
4069 | ||
4070 | struct delete_flows_cbdata { | |
4071 | struct ofproto *ofproto; | |
8054fc48 | 4072 | ovs_be16 out_port; |
79eee1eb BP |
4073 | }; |
4074 | ||
4075 | static void delete_flows_cb(struct cls_rule *, void *cbdata_); | |
8054fc48 | 4076 | static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port); |
79eee1eb BP |
4077 | |
4078 | /* Implements OFPFC_DELETE. */ | |
4079 | static void | |
3052b0c5 | 4080 | delete_flows_loose(struct ofproto *p, const struct flow_mod *fm) |
79eee1eb BP |
4081 | { |
4082 | struct delete_flows_cbdata cbdata; | |
064af421 | 4083 | |
3052b0c5 BP |
4084 | cbdata.ofproto = p; |
4085 | cbdata.out_port = htons(fm->out_port); | |
064af421 | 4086 | |
3c4486a5 | 4087 | classifier_for_each_match(&p->cls, &fm->cr, delete_flows_cb, &cbdata); |
064af421 BP |
4088 | } |
4089 | ||
79eee1eb BP |
4090 | /* Implements OFPFC_DELETE_STRICT. */ |
4091 | static void | |
3052b0c5 | 4092 | delete_flow_strict(struct ofproto *p, struct flow_mod *fm) |
79eee1eb | 4093 | { |
3052b0c5 | 4094 | struct rule *rule = find_flow_strict(p, fm); |
79eee1eb | 4095 | if (rule) { |
3052b0c5 | 4096 | delete_flow(p, rule, htons(fm->out_port)); |
79eee1eb BP |
4097 | } |
4098 | } | |
4099 | ||
4100 | /* Callback for delete_flows_loose(). */ | |
4101 | static void | |
4102 | delete_flows_cb(struct cls_rule *rule_, void *cbdata_) | |
4103 | { | |
4104 | struct rule *rule = rule_from_cls_rule(rule_); | |
4105 | struct delete_flows_cbdata *cbdata = cbdata_; | |
4106 | ||
4107 | delete_flow(cbdata->ofproto, rule, cbdata->out_port); | |
4108 | } | |
4109 | ||
4110 | /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has | |
4111 | * been identified as a flow to delete from 'p''s flow table, by deleting the | |
4112 | * flow and sending out a OFPT_FLOW_REMOVED message to any interested | |
4113 | * controller. | |
4114 | * | |
4115 | * Will not delete 'rule' if it is hidden. Will delete 'rule' only if | |
4116 | * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the | |
4117 | * specified 'out_port'. */ | |
4118 | static void | |
8054fc48 | 4119 | delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port) |
79eee1eb BP |
4120 | { |
4121 | if (rule_is_hidden(rule)) { | |
4122 | return; | |
4123 | } | |
4124 | ||
4125 | if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) { | |
4126 | return; | |
4127 | } | |
4128 | ||
bcf84111 | 4129 | rule_send_removed(p, rule, OFPRR_DELETE); |
79eee1eb BP |
4130 | rule_remove(p, rule); |
4131 | } | |
4132 | \f | |
064af421 | 4133 | static int |
3052b0c5 | 4134 | flow_mod_core(struct ofconn *ofconn, struct flow_mod *fm) |
064af421 | 4135 | { |
3052b0c5 | 4136 | struct ofproto *p = ofconn->ofproto; |
064af421 BP |
4137 | int error; |
4138 | ||
3052b0c5 | 4139 | error = reject_slave_controller(ofconn, "flow_mod"); |
9deba63b BP |
4140 | if (error) { |
4141 | return error; | |
4142 | } | |
3052b0c5 | 4143 | |
f1defbf9 BP |
4144 | error = validate_actions(fm->actions, fm->n_actions, |
4145 | &fm->cr.flow, p->max_ports); | |
064af421 BP |
4146 | if (error) { |
4147 | return error; | |
4148 | } | |
4149 | ||
49bdc010 JP |
4150 | /* We do not support the emergency flow cache. It will hopefully |
4151 | * get dropped from OpenFlow in the near future. */ | |
3052b0c5 | 4152 | if (fm->flags & OFPFF_EMERG) { |
49bdc010 JP |
4153 | /* There isn't a good fit for an error code, so just state that the |
4154 | * flow table is full. */ | |
4155 | return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL); | |
4156 | } | |
4157 | ||
3052b0c5 BP |
4158 | switch (fm->command) { |
4159 | case OFPFC_ADD: | |
4160 | return add_flow(ofconn, fm); | |
4161 | ||
4162 | case OFPFC_MODIFY: | |
4163 | return modify_flows_loose(ofconn, fm); | |
4164 | ||
4165 | case OFPFC_MODIFY_STRICT: | |
4166 | return modify_flow_strict(ofconn, fm); | |
4167 | ||
4168 | case OFPFC_DELETE: | |
4169 | delete_flows_loose(p, fm); | |
4170 | return 0; | |
4171 | ||
4172 | case OFPFC_DELETE_STRICT: | |
4173 | delete_flow_strict(p, fm); | |
4174 | return 0; | |
4175 | ||
4176 | default: | |
4177 | return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND); | |
4178 | } | |
4179 | } | |
4180 | ||
4181 | static int | |
09246b99 | 4182 | handle_ofpt_flow_mod(struct ofconn *ofconn, struct ofp_header *oh) |
3052b0c5 BP |
4183 | { |
4184 | struct ofp_match orig_match; | |
4185 | struct ofp_flow_mod *ofm; | |
4186 | struct flow_mod fm; | |
4187 | struct ofpbuf b; | |
4188 | int error; | |
4189 | ||
4190 | b.data = oh; | |
4191 | b.size = ntohs(oh->length); | |
4192 | ||
4193 | /* Dissect the message. */ | |
4194 | ofm = ofpbuf_try_pull(&b, sizeof *ofm); | |
4195 | if (!ofm) { | |
4196 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
4197 | } | |
4198 | error = ofputil_pull_actions(&b, b.size, &fm.actions, &fm.n_actions); | |
4199 | if (error) { | |
4200 | return error; | |
4201 | } | |
4202 | ||
4203 | /* Normalize ofm->match. If normalization actually changes anything, then | |
3f09c339 BP |
4204 | * log the differences. */ |
4205 | ofm->match.pad1[0] = ofm->match.pad2[0] = 0; | |
4206 | orig_match = ofm->match; | |
064af421 | 4207 | normalize_match(&ofm->match); |
3f09c339 BP |
4208 | if (memcmp(&ofm->match, &orig_match, sizeof orig_match)) { |
4209 | static struct vlog_rate_limit normal_rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
4210 | if (!VLOG_DROP_INFO(&normal_rl)) { | |
4211 | char *old = ofp_match_to_literal_string(&orig_match); | |
4212 | char *new = ofp_match_to_literal_string(&ofm->match); | |
4213 | VLOG_INFO("%s: normalization changed ofp_match, details:", | |
4214 | rconn_get_name(ofconn->rconn)); | |
4215 | VLOG_INFO(" pre: %s", old); | |
4216 | VLOG_INFO("post: %s", new); | |
4217 | free(old); | |
4218 | free(new); | |
4219 | } | |
4220 | } | |
4221 | ||
3052b0c5 BP |
4222 | /* Translate the message. */ |
4223 | cls_rule_from_match(&ofm->match, ntohs(ofm->priority), ofconn->flow_format, | |
4224 | ofm->cookie, &fm.cr); | |
4225 | fm.cookie = ofm->cookie; | |
4226 | fm.command = ntohs(ofm->command); | |
4227 | fm.idle_timeout = ntohs(ofm->idle_timeout); | |
4228 | fm.hard_timeout = ntohs(ofm->hard_timeout); | |
4229 | fm.buffer_id = ntohl(ofm->buffer_id); | |
4230 | fm.out_port = ntohs(ofm->out_port); | |
4231 | fm.flags = ntohs(ofm->flags); | |
064af421 | 4232 | |
3052b0c5 BP |
4233 | /* Execute the command. */ |
4234 | return flow_mod_core(ofconn, &fm); | |
064af421 BP |
4235 | } |
4236 | ||
09246b99 BP |
4237 | static int |
4238 | handle_nxt_flow_mod(struct ofconn *ofconn, struct ofp_header *oh) | |
4239 | { | |
4240 | struct nx_flow_mod *nfm; | |
4241 | struct flow_mod fm; | |
4242 | struct ofpbuf b; | |
4243 | int error; | |
4244 | ||
4245 | b.data = oh; | |
4246 | b.size = ntohs(oh->length); | |
4247 | ||
4248 | /* Dissect the message. */ | |
4249 | nfm = ofpbuf_try_pull(&b, sizeof *nfm); | |
4250 | if (!nfm) { | |
4251 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
4252 | } | |
4253 | error = nx_pull_match(&b, ntohs(nfm->match_len), ntohs(nfm->priority), | |
4254 | &fm.cr); | |
4255 | if (error) { | |
4256 | return error; | |
4257 | } | |
4258 | error = ofputil_pull_actions(&b, b.size, &fm.actions, &fm.n_actions); | |
4259 | if (error) { | |
4260 | return error; | |
4261 | } | |
4262 | ||
4263 | /* Translate the message. */ | |
4264 | fm.cookie = nfm->cookie; | |
4265 | fm.command = ntohs(nfm->command); | |
4266 | fm.idle_timeout = ntohs(nfm->idle_timeout); | |
4267 | fm.hard_timeout = ntohs(nfm->hard_timeout); | |
4268 | fm.buffer_id = ntohl(nfm->buffer_id); | |
4269 | fm.out_port = ntohs(nfm->out_port); | |
4270 | fm.flags = ntohs(nfm->flags); | |
4271 | ||
4272 | /* Execute the command. */ | |
4273 | return flow_mod_core(ofconn, &fm); | |
4274 | } | |
4275 | ||
659586ef | 4276 | static int |
b70eac89 | 4277 | handle_tun_id_from_cookie(struct ofconn *ofconn, struct nxt_tun_id_cookie *msg) |
659586ef JG |
4278 | { |
4279 | int error; | |
4280 | ||
4281 | error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg); | |
4282 | if (error) { | |
4283 | return error; | |
4284 | } | |
4285 | ||
b70eac89 | 4286 | ofconn->flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10; |
659586ef JG |
4287 | return 0; |
4288 | } | |
4289 | ||
9deba63b | 4290 | static int |
3269c562 | 4291 | handle_role_request(struct ofconn *ofconn, struct nicira_header *msg) |
9deba63b BP |
4292 | { |
4293 | struct nx_role_request *nrr; | |
4294 | struct nx_role_request *reply; | |
4295 | struct ofpbuf *buf; | |
4296 | uint32_t role; | |
4297 | ||
4298 | if (ntohs(msg->header.length) != sizeof *nrr) { | |
100e95db | 4299 | VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)", |
9deba63b BP |
4300 | ntohs(msg->header.length), sizeof *nrr); |
4301 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
4302 | } | |
4303 | nrr = (struct nx_role_request *) msg; | |
4304 | ||
5899143f | 4305 | if (ofconn->type != OFCONN_PRIMARY) { |
9deba63b BP |
4306 | VLOG_WARN_RL(&rl, "ignoring role request on non-controller " |
4307 | "connection"); | |
4308 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM); | |
4309 | } | |
4310 | ||
4311 | role = ntohl(nrr->role); | |
4312 | if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER | |
4313 | && role != NX_ROLE_SLAVE) { | |
4314 | VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role); | |
4315 | ||
4316 | /* There's no good error code for this. */ | |
4317 | return ofp_mkerr(OFPET_BAD_REQUEST, -1); | |
4318 | } | |
4319 | ||
4320 | if (role == NX_ROLE_MASTER) { | |
4321 | struct ofconn *other; | |
4322 | ||
3269c562 | 4323 | HMAP_FOR_EACH (other, hmap_node, &ofconn->ofproto->controllers) { |
9deba63b BP |
4324 | if (other->role == NX_ROLE_MASTER) { |
4325 | other->role = NX_ROLE_SLAVE; | |
4326 | } | |
4327 | } | |
4328 | } | |
4329 | ofconn->role = role; | |
4330 | ||
0bd0c660 BP |
4331 | reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, msg->header.xid, |
4332 | &buf); | |
9deba63b BP |
4333 | reply->role = htonl(role); |
4334 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
4335 | ||
4336 | return 0; | |
4337 | } | |
4338 | ||
09246b99 BP |
4339 | static int |
4340 | handle_nxt_set_flow_format(struct ofconn *ofconn, | |
4341 | struct nxt_set_flow_format *msg) | |
4342 | { | |
4343 | uint32_t format; | |
4344 | int error; | |
4345 | ||
4346 | error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg); | |
4347 | if (error) { | |
4348 | return error; | |
4349 | } | |
4350 | ||
4351 | format = ntohl(msg->format); | |
4352 | if (format == NXFF_OPENFLOW10 | |
4353 | || format == NXFF_TUN_ID_FROM_COOKIE | |
4354 | || format == NXFF_NXM) { | |
4355 | ofconn->flow_format = format; | |
4356 | return 0; | |
4357 | } else { | |
4358 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM); | |
4359 | } | |
4360 | } | |
4361 | ||
064af421 | 4362 | static int |
3269c562 | 4363 | handle_vendor(struct ofconn *ofconn, void *msg) |
064af421 | 4364 | { |
3269c562 | 4365 | struct ofproto *p = ofconn->ofproto; |
064af421 BP |
4366 | struct ofp_vendor_header *ovh = msg; |
4367 | struct nicira_header *nh; | |
4368 | ||
4369 | if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) { | |
100e95db | 4370 | VLOG_WARN_RL(&rl, "received vendor message of length %u " |
659586ef JG |
4371 | "(expected at least %zu)", |
4372 | ntohs(ovh->header.length), sizeof(struct ofp_vendor_header)); | |
49bdc010 | 4373 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); |
064af421 BP |
4374 | } |
4375 | if (ovh->vendor != htonl(NX_VENDOR_ID)) { | |
4376 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR); | |
4377 | } | |
4378 | if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) { | |
100e95db | 4379 | VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u " |
659586ef JG |
4380 | "(expected at least %zu)", |
4381 | ntohs(ovh->header.length), sizeof(struct nicira_header)); | |
49bdc010 | 4382 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); |
064af421 BP |
4383 | } |
4384 | ||
4385 | nh = msg; | |
4386 | switch (ntohl(nh->subtype)) { | |
4387 | case NXT_STATUS_REQUEST: | |
4388 | return switch_status_handle_request(p->switch_status, ofconn->rconn, | |
4389 | msg); | |
659586ef JG |
4390 | |
4391 | case NXT_TUN_ID_FROM_COOKIE: | |
b70eac89 | 4392 | return handle_tun_id_from_cookie(ofconn, msg); |
9deba63b BP |
4393 | |
4394 | case NXT_ROLE_REQUEST: | |
3269c562 | 4395 | return handle_role_request(ofconn, msg); |
09246b99 BP |
4396 | |
4397 | case NXT_SET_FLOW_FORMAT: | |
4398 | return handle_nxt_set_flow_format(ofconn, msg); | |
4399 | ||
4400 | case NXT_FLOW_MOD: | |
4401 | return handle_nxt_flow_mod(ofconn, &ovh->header); | |
064af421 BP |
4402 | } |
4403 | ||
4404 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); | |
4405 | } | |
4406 | ||
246e61ea JP |
4407 | static int |
4408 | handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh) | |
4409 | { | |
4410 | struct ofp_header *ob; | |
4411 | struct ofpbuf *buf; | |
4412 | ||
4413 | /* Currently, everything executes synchronously, so we can just | |
4414 | * immediately send the barrier reply. */ | |
4415 | ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf); | |
4416 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
4417 | return 0; | |
4418 | } | |
4419 | ||
064af421 | 4420 | static void |
3269c562 | 4421 | handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg) |
064af421 BP |
4422 | { |
4423 | struct ofp_header *oh = ofp_msg->data; | |
4424 | int error; | |
4425 | ||
4426 | COVERAGE_INC(ofproto_recv_openflow); | |
4427 | switch (oh->type) { | |
4428 | case OFPT_ECHO_REQUEST: | |
4429 | error = handle_echo_request(ofconn, oh); | |
4430 | break; | |
4431 | ||
4432 | case OFPT_ECHO_REPLY: | |
4433 | error = 0; | |
4434 | break; | |
4435 | ||
4436 | case OFPT_FEATURES_REQUEST: | |
3269c562 | 4437 | error = handle_features_request(ofconn, oh); |
064af421 BP |
4438 | break; |
4439 | ||
4440 | case OFPT_GET_CONFIG_REQUEST: | |
3269c562 | 4441 | error = handle_get_config_request(ofconn, oh); |
064af421 BP |
4442 | break; |
4443 | ||
4444 | case OFPT_SET_CONFIG: | |
3269c562 | 4445 | error = handle_set_config(ofconn, ofp_msg->data); |
064af421 BP |
4446 | break; |
4447 | ||
4448 | case OFPT_PACKET_OUT: | |
3269c562 | 4449 | error = handle_packet_out(ofconn, ofp_msg->data); |
064af421 BP |
4450 | break; |
4451 | ||
4452 | case OFPT_PORT_MOD: | |
3269c562 | 4453 | error = handle_port_mod(ofconn, oh); |
064af421 BP |
4454 | break; |
4455 | ||
4456 | case OFPT_FLOW_MOD: | |
09246b99 | 4457 | error = handle_ofpt_flow_mod(ofconn, ofp_msg->data); |
064af421 BP |
4458 | break; |
4459 | ||
4460 | case OFPT_STATS_REQUEST: | |
3269c562 | 4461 | error = handle_stats_request(ofconn, oh); |
064af421 BP |
4462 | break; |
4463 | ||
4464 | case OFPT_VENDOR: | |
3269c562 | 4465 | error = handle_vendor(ofconn, ofp_msg->data); |
064af421 BP |
4466 | break; |
4467 | ||
246e61ea JP |
4468 | case OFPT_BARRIER_REQUEST: |
4469 | error = handle_barrier_request(ofconn, oh); | |
4470 | break; | |
4471 | ||
064af421 BP |
4472 | default: |
4473 | if (VLOG_IS_WARN_ENABLED()) { | |
4474 | char *s = ofp_to_string(oh, ntohs(oh->length), 2); | |
4475 | VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s); | |
4476 | free(s); | |
4477 | } | |
4478 | error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE); | |
4479 | break; | |
4480 | } | |
4481 | ||
4482 | if (error) { | |
4483 | send_error_oh(ofconn, ofp_msg->data, error); | |
4484 | } | |
4485 | } | |
4486 | \f | |
4487 | static void | |
72b06300 | 4488 | handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet) |
064af421 BP |
4489 | { |
4490 | struct odp_msg *msg = packet->data; | |
064af421 | 4491 | struct ofpbuf payload; |
bcf84111 | 4492 | struct facet *facet; |
ae412e7d | 4493 | struct flow flow; |
064af421 | 4494 | |
064af421 BP |
4495 | payload.data = msg + 1; |
4496 | payload.size = msg->length - sizeof *msg; | |
659586ef | 4497 | flow_extract(&payload, msg->arg, msg->port, &flow); |
064af421 | 4498 | |
0ad9b732 JP |
4499 | /* Check with in-band control to see if this packet should be sent |
4500 | * to the local port regardless of the flow table. */ | |
4501 | if (in_band_msg_in_hook(p->in_band, &flow, &payload)) { | |
4502 | union odp_action action; | |
4503 | ||
4504 | memset(&action, 0, sizeof(action)); | |
4505 | action.output.type = ODPAT_OUTPUT; | |
4506 | action.output.port = ODPP_LOCAL; | |
f1588b1f | 4507 | dpif_execute(p->dpif, &action, 1, &payload); |
0ad9b732 JP |
4508 | } |
4509 | ||
bcf84111 BP |
4510 | facet = facet_lookup_valid(p, &flow); |
4511 | if (!facet) { | |
4512 | struct rule *rule = rule_lookup(p, &flow); | |
4513 | if (!rule) { | |
4514 | /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */ | |
4515 | struct ofport *port = get_port(p, msg->port); | |
4516 | if (port) { | |
4517 | if (port->opp.config & OFPPC_NO_PACKET_IN) { | |
4518 | COVERAGE_INC(ofproto_no_packet_in); | |
4519 | /* XXX install 'drop' flow entry */ | |
4520 | ofpbuf_delete(packet); | |
4521 | return; | |
4522 | } | |
4523 | } else { | |
4524 | VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, | |
4525 | msg->port); | |
064af421 | 4526 | } |
064af421 | 4527 | |
bcf84111 BP |
4528 | COVERAGE_INC(ofproto_packet_in); |
4529 | send_packet_in(p, packet); | |
4530 | return; | |
064af421 | 4531 | } |
bcf84111 BP |
4532 | |
4533 | facet = facet_create(p, rule, &flow, packet); | |
4534 | } else if (!facet->may_install) { | |
4535 | /* The facet is not installable, that is, we need to process every | |
4536 | * packet, so process the current packet's actions into 'facet'. */ | |
4537 | facet_make_actions(p, facet, packet); | |
064af421 BP |
4538 | } |
4539 | ||
bcf84111 | 4540 | if (facet->rule->cr.priority == FAIL_OPEN_PRIORITY) { |
7778bd15 BP |
4541 | /* |
4542 | * Extra-special case for fail-open mode. | |
4543 | * | |
4544 | * We are in fail-open mode and the packet matched the fail-open rule, | |
4545 | * but we are connected to a controller too. We should send the packet | |
4546 | * up to the controller in the hope that it will try to set up a flow | |
4547 | * and thereby allow us to exit fail-open. | |
4548 | * | |
4549 | * See the top-level comment in fail-open.c for more information. | |
4550 | */ | |
c9b5816c BP |
4551 | send_packet_in(p, ofpbuf_clone_with_headroom(packet, |
4552 | DPIF_RECV_MSG_PADDING)); | |
7778bd15 | 4553 | } |
750638bb BP |
4554 | |
4555 | ofpbuf_pull(packet, sizeof *msg); | |
bcf84111 BP |
4556 | facet_execute(p, facet, packet); |
4557 | facet_install(p, facet, false); | |
064af421 | 4558 | } |
72b06300 BP |
4559 | |
4560 | static void | |
4561 | handle_odp_msg(struct ofproto *p, struct ofpbuf *packet) | |
4562 | { | |
4563 | struct odp_msg *msg = packet->data; | |
4564 | ||
4565 | switch (msg->type) { | |
4566 | case _ODPL_ACTION_NR: | |
4567 | COVERAGE_INC(ofproto_ctlr_action); | |
76ce9432 | 4568 | send_packet_in(p, packet); |
72b06300 BP |
4569 | break; |
4570 | ||
4571 | case _ODPL_SFLOW_NR: | |
4572 | if (p->sflow) { | |
4573 | ofproto_sflow_received(p->sflow, msg); | |
4574 | } | |
4575 | ofpbuf_delete(packet); | |
4576 | break; | |
4577 | ||
4578 | case _ODPL_MISS_NR: | |
4579 | handle_odp_miss_msg(p, packet); | |
4580 | break; | |
4581 | ||
4582 | default: | |
4583 | VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32, | |
4584 | msg->type); | |
4585 | break; | |
4586 | } | |
4587 | } | |
064af421 | 4588 | \f |
4a4cdb3b BP |
4589 | /* Flow expiration. */ |
4590 | ||
4591 | struct expire_cbdata { | |
4592 | struct ofproto *ofproto; | |
0de7a4b4 | 4593 | int dp_max_idle; |
4a4cdb3b BP |
4594 | }; |
4595 | ||
0de7a4b4 | 4596 | static int ofproto_dp_max_idle(const struct ofproto *); |
4a4cdb3b BP |
4597 | static void ofproto_update_used(struct ofproto *); |
4598 | static void rule_expire(struct cls_rule *, void *cbdata); | |
bcf84111 | 4599 | static void ofproto_expire_facets(struct ofproto *, int dp_max_idle); |
4a4cdb3b BP |
4600 | |
4601 | /* This function is called periodically by ofproto_run(). Its job is to | |
4602 | * collect updates for the flows that have been installed into the datapath, | |
4603 | * most importantly when they last were used, and then use that information to | |
0de7a4b4 BP |
4604 | * expire flows that have not been used recently. |
4605 | * | |
4606 | * Returns the number of milliseconds after which it should be called again. */ | |
4607 | static int | |
4a4cdb3b BP |
4608 | ofproto_expire(struct ofproto *ofproto) |
4609 | { | |
4610 | struct expire_cbdata cbdata; | |
4611 | ||
4612 | /* Update 'used' for each flow in the datapath. */ | |
4613 | ofproto_update_used(ofproto); | |
4614 | ||
bcf84111 | 4615 | /* Expire facets that have been idle too long. */ |
0de7a4b4 | 4616 | cbdata.dp_max_idle = ofproto_dp_max_idle(ofproto); |
bcf84111 BP |
4617 | ofproto_expire_facets(ofproto, cbdata.dp_max_idle); |
4618 | ||
4619 | /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */ | |
4620 | cbdata.ofproto = ofproto; | |
3c4486a5 | 4621 | classifier_for_each(&ofproto->cls, rule_expire, &cbdata); |
4a4cdb3b BP |
4622 | |
4623 | /* Let the hook know that we're at a stable point: all outstanding data | |
4624 | * in existing flows has been accounted to the account_cb. Thus, the | |
4625 | * hook can now reasonably do operations that depend on having accurate | |
4626 | * flow volume accounting (currently, that's just bond rebalancing). */ | |
4627 | if (ofproto->ofhooks->account_checkpoint_cb) { | |
4628 | ofproto->ofhooks->account_checkpoint_cb(ofproto->aux); | |
4629 | } | |
0de7a4b4 BP |
4630 | |
4631 | return MIN(cbdata.dp_max_idle, 1000); | |
4a4cdb3b BP |
4632 | } |
4633 | ||
bcf84111 | 4634 | /* Update 'used' member of installed facets. */ |
4a4cdb3b BP |
4635 | static void |
4636 | ofproto_update_used(struct ofproto *p) | |
4637 | { | |
4638 | struct odp_flow *flows; | |
4639 | size_t n_flows; | |
4640 | size_t i; | |
4641 | int error; | |
4642 | ||
4643 | error = dpif_flow_list_all(p->dpif, &flows, &n_flows); | |
4644 | if (error) { | |
4645 | return; | |
4646 | } | |
4647 | ||
4648 | for (i = 0; i < n_flows; i++) { | |
4649 | struct odp_flow *f = &flows[i]; | |
bcf84111 | 4650 | struct facet *facet; |
ae412e7d | 4651 | struct flow flow; |
14608a15 BP |
4652 | |
4653 | odp_flow_key_to_flow(&f->key, &flow); | |
bcf84111 | 4654 | facet = facet_find(p, &flow); |
4a4cdb3b | 4655 | |
bcf84111 BP |
4656 | if (facet && facet->installed) { |
4657 | facet_update_time(p, facet, &f->stats); | |
4658 | facet_account(p, facet, f->stats.n_bytes); | |
4a4cdb3b BP |
4659 | } else { |
4660 | /* There's a flow in the datapath that we know nothing about. | |
4661 | * Delete it. */ | |
4662 | COVERAGE_INC(ofproto_unexpected_rule); | |
4663 | dpif_flow_del(p->dpif, f); | |
4664 | } | |
4665 | ||
4666 | } | |
4667 | free(flows); | |
4668 | } | |
4669 | ||
0de7a4b4 | 4670 | /* Calculates and returns the number of milliseconds of idle time after which |
bcf84111 | 4671 | * facets should expire from the datapath and we should fold their statistics |
0de7a4b4 BP |
4672 | * into their parent rules in userspace. */ |
4673 | static int | |
4674 | ofproto_dp_max_idle(const struct ofproto *ofproto) | |
4675 | { | |
4676 | /* | |
4677 | * Idle time histogram. | |
4678 | * | |
bcf84111 | 4679 | * Most of the time a switch has a relatively small number of facets. When |
0de7a4b4 BP |
4680 | * this is the case we might as well keep statistics for all of them in |
4681 | * userspace and to cache them in the kernel datapath for performance as | |
4682 | * well. | |
4683 | * | |
bcf84111 | 4684 | * As the number of facets increases, the memory required to maintain |
0de7a4b4 | 4685 | * statistics about them in userspace and in the kernel becomes |
bcf84111 | 4686 | * significant. However, with a large number of facets it is likely that |
0de7a4b4 BP |
4687 | * only a few of them are "heavy hitters" that consume a large amount of |
4688 | * bandwidth. At this point, only heavy hitters are worth caching in the | |
bcf84111 | 4689 | * kernel and maintaining in userspaces; other facets we can discard. |
0de7a4b4 BP |
4690 | * |
4691 | * The technique used to compute the idle time is to build a histogram with | |
bcf84111 BP |
4692 | * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet |
4693 | * that is installed in the kernel gets dropped in the appropriate bucket. | |
0de7a4b4 | 4694 | * After the histogram has been built, we compute the cutoff so that only |
bcf84111 BP |
4695 | * the most-recently-used 1% of facets (but at least 1000 flows) are kept |
4696 | * cached. At least the most-recently-used bucket of facets is kept, so | |
4697 | * actually an arbitrary number of facets can be kept in any given | |
0de7a4b4 BP |
4698 | * expiration run (though the next run will delete most of those unless |
4699 | * they receive additional data). | |
4700 | * | |
bcf84111 BP |
4701 | * This requires a second pass through the facets, in addition to the pass |
4702 | * made by ofproto_update_used(), because the former function never looks | |
4703 | * at uninstallable facets. | |
0de7a4b4 BP |
4704 | */ |
4705 | enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) }; | |
4706 | enum { N_BUCKETS = 5000 / BUCKET_WIDTH }; | |
4707 | int buckets[N_BUCKETS] = { 0 }; | |
bcf84111 | 4708 | struct facet *facet; |
0de7a4b4 | 4709 | int total, bucket; |
0de7a4b4 BP |
4710 | long long int now; |
4711 | int i; | |
4712 | ||
bcf84111 | 4713 | total = hmap_count(&ofproto->facets); |
0de7a4b4 BP |
4714 | if (total <= 1000) { |
4715 | return N_BUCKETS * BUCKET_WIDTH; | |
4716 | } | |
4717 | ||
4718 | /* Build histogram. */ | |
4719 | now = time_msec(); | |
bcf84111 BP |
4720 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { |
4721 | long long int idle = now - facet->used; | |
0de7a4b4 BP |
4722 | int bucket = (idle <= 0 ? 0 |
4723 | : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1 | |
4724 | : (unsigned int) idle / BUCKET_WIDTH); | |
4725 | buckets[bucket]++; | |
4726 | } | |
4727 | ||
4728 | /* Find the first bucket whose flows should be expired. */ | |
4729 | for (bucket = 0; bucket < N_BUCKETS; bucket++) { | |
4730 | if (buckets[bucket]) { | |
4731 | int subtotal = 0; | |
4732 | do { | |
4733 | subtotal += buckets[bucket++]; | |
4734 | } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100)); | |
4735 | break; | |
4736 | } | |
4737 | } | |
4738 | ||
4739 | if (VLOG_IS_DBG_ENABLED()) { | |
4740 | struct ds s; | |
4741 | ||
4742 | ds_init(&s); | |
4743 | ds_put_cstr(&s, "keep"); | |
4744 | for (i = 0; i < N_BUCKETS; i++) { | |
4745 | if (i == bucket) { | |
4746 | ds_put_cstr(&s, ", drop"); | |
4747 | } | |
4748 | if (buckets[i]) { | |
4749 | ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]); | |
4750 | } | |
4751 | } | |
4752 | VLOG_INFO("%s: %s (msec:count)", | |
4753 | dpif_name(ofproto->dpif), ds_cstr(&s)); | |
4754 | ds_destroy(&s); | |
4755 | } | |
4756 | ||
4757 | return bucket * BUCKET_WIDTH; | |
4758 | } | |
4759 | ||
4a4cdb3b | 4760 | static void |
bcf84111 | 4761 | facet_active_timeout(struct ofproto *ofproto, struct facet *facet) |
4a4cdb3b | 4762 | { |
bcf84111 BP |
4763 | if (ofproto->netflow && !facet_is_controller_flow(facet) && |
4764 | netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) { | |
4a4cdb3b BP |
4765 | struct ofexpired expired; |
4766 | struct odp_flow odp_flow; | |
4767 | ||
4768 | /* Get updated flow stats. | |
4769 | * | |
4770 | * XXX We could avoid this call entirely if (1) ofproto_update_used() | |
4771 | * updated TCP flags and (2) the dpif_flow_list_all() in | |
4772 | * ofproto_update_used() zeroed TCP flags. */ | |
4773 | memset(&odp_flow, 0, sizeof odp_flow); | |
bcf84111 BP |
4774 | if (facet->installed) { |
4775 | odp_flow_key_from_flow(&odp_flow.key, &facet->flow); | |
4a4cdb3b BP |
4776 | odp_flow.flags = ODPFF_ZERO_TCP_FLAGS; |
4777 | dpif_flow_get(ofproto->dpif, &odp_flow); | |
4778 | ||
4779 | if (odp_flow.stats.n_packets) { | |
bcf84111 BP |
4780 | facet_update_time(ofproto, facet, &odp_flow.stats); |
4781 | netflow_flow_update_flags(&facet->nf_flow, | |
4a4cdb3b BP |
4782 | odp_flow.stats.tcp_flags); |
4783 | } | |
4784 | } | |
4785 | ||
bcf84111 BP |
4786 | expired.flow = facet->flow; |
4787 | expired.packet_count = facet->packet_count + | |
4a4cdb3b | 4788 | odp_flow.stats.n_packets; |
bcf84111 BP |
4789 | expired.byte_count = facet->byte_count + odp_flow.stats.n_bytes; |
4790 | expired.used = facet->used; | |
4791 | ||
4792 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
4793 | } | |
4794 | } | |
4a4cdb3b | 4795 | |
bcf84111 BP |
4796 | static void |
4797 | ofproto_expire_facets(struct ofproto *ofproto, int dp_max_idle) | |
4798 | { | |
4799 | long long int cutoff = time_msec() - dp_max_idle; | |
4800 | struct facet *facet, *next_facet; | |
4801 | ||
4802 | HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) { | |
4803 | facet_active_timeout(ofproto, facet); | |
4804 | if (facet->used < cutoff) { | |
4805 | facet_remove(ofproto, facet); | |
4806 | } | |
4a4cdb3b BP |
4807 | } |
4808 | } | |
4809 | ||
4810 | /* If 'cls_rule' is an OpenFlow rule, that has expired according to OpenFlow | |
4811 | * rules, then delete it entirely. | |
4812 | * | |
4a4cdb3b BP |
4813 | * (This is a callback function for classifier_for_each().) */ |
4814 | static void | |
4815 | rule_expire(struct cls_rule *cls_rule, void *cbdata_) | |
4816 | { | |
4817 | struct expire_cbdata *cbdata = cbdata_; | |
4a4cdb3b | 4818 | struct rule *rule = rule_from_cls_rule(cls_rule); |
bcf84111 BP |
4819 | struct facet *facet, *next_facet; |
4820 | long long int now; | |
4821 | uint8_t reason; | |
4a4cdb3b | 4822 | |
bcf84111 | 4823 | /* Has 'rule' expired? */ |
4a4cdb3b | 4824 | now = time_msec(); |
bcf84111 BP |
4825 | if (rule->hard_timeout |
4826 | && now > rule->created + rule->hard_timeout * 1000) { | |
4827 | reason = OFPRR_HARD_TIMEOUT; | |
4828 | } else if (rule->idle_timeout && list_is_empty(&rule->facets) | |
4829 | && now >rule->used + rule->idle_timeout * 1000) { | |
4830 | reason = OFPRR_IDLE_TIMEOUT; | |
4a4cdb3b | 4831 | } else { |
bcf84111 | 4832 | return; |
4a4cdb3b | 4833 | } |
064af421 | 4834 | |
bcf84111 | 4835 | COVERAGE_INC(ofproto_expired); |
064af421 | 4836 | |
bcf84111 BP |
4837 | /* Update stats. (This is a no-op if the rule expired due to an idle |
4838 | * timeout, because that only happens when the rule has no facets left.) */ | |
4839 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { | |
4840 | facet_remove(cbdata->ofproto, facet); | |
064af421 BP |
4841 | } |
4842 | ||
bcf84111 BP |
4843 | /* Get rid of the rule. */ |
4844 | if (!rule_is_hidden(rule)) { | |
4845 | rule_send_removed(cbdata->ofproto, rule, reason); | |
4846 | } | |
4847 | rule_remove(cbdata->ofproto, rule); | |
064af421 | 4848 | } |
bcf84111 | 4849 | \f |
064af421 | 4850 | static struct ofpbuf * |
09246b99 BP |
4851 | compose_ofp_flow_removed(struct ofconn *ofconn, const struct rule *rule, |
4852 | uint8_t reason) | |
064af421 | 4853 | { |
ca069229 | 4854 | struct ofp_flow_removed *ofr; |
064af421 BP |
4855 | struct ofpbuf *buf; |
4856 | ||
ca069229 | 4857 | ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf); |
b70eac89 | 4858 | flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, ofconn->flow_format, |
659586ef | 4859 | &ofr->match); |
39997502 | 4860 | ofr->cookie = rule->flow_cookie; |
ca069229 JP |
4861 | ofr->priority = htons(rule->cr.priority); |
4862 | ofr->reason = reason; | |
c6ebb8fb | 4863 | calc_flow_duration(rule->created, &ofr->duration_sec, &ofr->duration_nsec); |
ca069229 JP |
4864 | ofr->idle_timeout = htons(rule->idle_timeout); |
4865 | ofr->packet_count = htonll(rule->packet_count); | |
4866 | ofr->byte_count = htonll(rule->byte_count); | |
064af421 BP |
4867 | |
4868 | return buf; | |
4869 | } | |
4870 | ||
09246b99 BP |
4871 | static struct ofpbuf * |
4872 | compose_nx_flow_removed(const struct rule *rule, uint8_t reason) | |
4873 | { | |
4874 | struct nx_flow_removed *nfr; | |
4875 | struct ofpbuf *buf; | |
4876 | int match_len; | |
4877 | ||
4878 | nfr = make_nxmsg(sizeof *nfr, NXT_FLOW_REMOVED, &buf); | |
4879 | ||
4880 | match_len = nx_put_match(buf, &rule->cr); | |
4881 | ||
4882 | nfr->cookie = rule->flow_cookie; | |
4883 | nfr->priority = htons(rule->cr.priority); | |
4884 | nfr->reason = reason; | |
4885 | calc_flow_duration(rule->created, &nfr->duration_sec, &nfr->duration_nsec); | |
4886 | nfr->idle_timeout = htons(rule->idle_timeout); | |
4887 | nfr->match_len = htons(match_len); | |
4888 | nfr->packet_count = htonll(rule->packet_count); | |
4889 | nfr->byte_count = htonll(rule->byte_count); | |
4890 | ||
4891 | return buf; | |
4892 | } | |
4893 | ||
ca069229 | 4894 | static void |
bcf84111 | 4895 | rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason) |
064af421 BP |
4896 | { |
4897 | struct ofconn *ofconn; | |
064af421 | 4898 | |
3b587616 BP |
4899 | if (!rule->send_flow_removed) { |
4900 | return; | |
4901 | } | |
4902 | ||
4e8e4213 | 4903 | LIST_FOR_EACH (ofconn, node, &p->all_conns) { |
7a0efeb5 BP |
4904 | struct ofpbuf *msg; |
4905 | ||
4906 | if (!rconn_is_connected(ofconn->rconn) | |
4907 | || !ofconn_receives_async_msgs(ofconn)) { | |
4908 | continue; | |
064af421 | 4909 | } |
7a0efeb5 | 4910 | |
09246b99 BP |
4911 | msg = (ofconn->flow_format == NXFF_NXM |
4912 | ? compose_nx_flow_removed(rule, reason) | |
4913 | : compose_ofp_flow_removed(ofconn, rule, reason)); | |
6d6c7259 BP |
4914 | |
4915 | /* Account flow expirations under ofconn->reply_counter, the counter | |
4916 | * for replies to OpenFlow requests. That works because preventing | |
4917 | * OpenFlow requests from being processed also prevents new flows from | |
4918 | * being added (and expiring). (It also prevents processing OpenFlow | |
4919 | * requests that would not add new flows, so it is imperfect.) */ | |
7a0efeb5 | 4920 | queue_tx(msg, ofconn, ofconn->reply_counter); |
064af421 BP |
4921 | } |
4922 | } | |
4923 | ||
43253595 | 4924 | /* pinsched callback for sending 'packet' on 'ofconn'. */ |
064af421 | 4925 | static void |
76ce9432 | 4926 | do_send_packet_in(struct ofpbuf *packet, void *ofconn_) |
064af421 | 4927 | { |
76ce9432 | 4928 | struct ofconn *ofconn = ofconn_; |
43253595 BP |
4929 | |
4930 | rconn_send_with_limit(ofconn->rconn, packet, | |
4931 | ofconn->packet_in_counter, 100); | |
4932 | } | |
4933 | ||
4934 | /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and | |
4935 | * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s | |
4936 | * packet scheduler for sending. | |
4937 | * | |
30ea5d93 BP |
4938 | * 'max_len' specifies the maximum number of bytes of the packet to send on |
4939 | * 'ofconn' (INT_MAX specifies no limit). | |
4940 | * | |
43253595 BP |
4941 | * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise, |
4942 | * ownership is transferred to this function. */ | |
4943 | static void | |
30ea5d93 BP |
4944 | schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len, |
4945 | bool clone) | |
43253595 | 4946 | { |
76ce9432 | 4947 | struct ofproto *ofproto = ofconn->ofproto; |
43253595 BP |
4948 | struct ofp_packet_in *opi = packet->data; |
4949 | uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port)); | |
4950 | int send_len, trim_size; | |
76ce9432 | 4951 | uint32_t buffer_id; |
064af421 | 4952 | |
43253595 BP |
4953 | /* Get buffer. */ |
4954 | if (opi->reason == OFPR_ACTION) { | |
76ce9432 | 4955 | buffer_id = UINT32_MAX; |
43253595 BP |
4956 | } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) { |
4957 | buffer_id = pktbuf_get_null(); | |
89b9612d BP |
4958 | } else if (!ofconn->pktbuf) { |
4959 | buffer_id = UINT32_MAX; | |
76ce9432 | 4960 | } else { |
43253595 BP |
4961 | struct ofpbuf payload; |
4962 | payload.data = opi->data; | |
4963 | payload.size = packet->size - offsetof(struct ofp_packet_in, data); | |
4964 | buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port); | |
76ce9432 | 4965 | } |
372179d4 | 4966 | |
43253595 BP |
4967 | /* Figure out how much of the packet to send. */ |
4968 | send_len = ntohs(opi->total_len); | |
4969 | if (buffer_id != UINT32_MAX) { | |
4970 | send_len = MIN(send_len, ofconn->miss_send_len); | |
4971 | } | |
30ea5d93 | 4972 | send_len = MIN(send_len, max_len); |
064af421 | 4973 | |
43253595 BP |
4974 | /* Adjust packet length and clone if necessary. */ |
4975 | trim_size = offsetof(struct ofp_packet_in, data) + send_len; | |
4976 | if (clone) { | |
4977 | packet = ofpbuf_clone_data(packet->data, trim_size); | |
4978 | opi = packet->data; | |
4979 | } else { | |
4980 | packet->size = trim_size; | |
4981 | } | |
4982 | ||
4983 | /* Update packet headers. */ | |
4984 | opi->buffer_id = htonl(buffer_id); | |
4985 | update_openflow_length(packet); | |
4986 | ||
4987 | /* Hand over to packet scheduler. It might immediately call into | |
4988 | * do_send_packet_in() or it might buffer it for a while (until a later | |
4989 | * call to pinsched_run()). */ | |
4990 | pinsched_send(ofconn->schedulers[opi->reason], in_port, | |
4991 | packet, do_send_packet_in, ofconn); | |
064af421 BP |
4992 | } |
4993 | ||
43253595 BP |
4994 | /* Replace struct odp_msg header in 'packet' by equivalent struct |
4995 | * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as | |
4996 | * returned by dpif_recv()). | |
4997 | * | |
4998 | * The conversion is not complete: the caller still needs to trim any unneeded | |
4999 | * payload off the end of the buffer, set the length in the OpenFlow header, | |
5000 | * and set buffer_id. Those require us to know the controller settings and so | |
30ea5d93 BP |
5001 | * must be done on a per-controller basis. |
5002 | * | |
5003 | * Returns the maximum number of bytes of the packet that should be sent to | |
5004 | * the controller (INT_MAX if no limit). */ | |
5005 | static int | |
43253595 | 5006 | do_convert_to_packet_in(struct ofpbuf *packet) |
064af421 | 5007 | { |
76ce9432 | 5008 | struct odp_msg *msg = packet->data; |
43253595 BP |
5009 | struct ofp_packet_in *opi; |
5010 | uint8_t reason; | |
5011 | uint16_t total_len; | |
5012 | uint16_t in_port; | |
30ea5d93 | 5013 | int max_len; |
43253595 BP |
5014 | |
5015 | /* Extract relevant header fields */ | |
30ea5d93 BP |
5016 | if (msg->type == _ODPL_ACTION_NR) { |
5017 | reason = OFPR_ACTION; | |
5018 | max_len = msg->arg; | |
5019 | } else { | |
5020 | reason = OFPR_NO_MATCH; | |
5021 | max_len = INT_MAX; | |
5022 | } | |
43253595 BP |
5023 | total_len = msg->length - sizeof *msg; |
5024 | in_port = odp_port_to_ofp_port(msg->port); | |
5025 | ||
5026 | /* Repurpose packet buffer by overwriting header. */ | |
5027 | ofpbuf_pull(packet, sizeof(struct odp_msg)); | |
5028 | opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data)); | |
5029 | opi->header.version = OFP_VERSION; | |
5030 | opi->header.type = OFPT_PACKET_IN; | |
5031 | opi->total_len = htons(total_len); | |
5032 | opi->in_port = htons(in_port); | |
5033 | opi->reason = reason; | |
30ea5d93 BP |
5034 | |
5035 | return max_len; | |
43253595 BP |
5036 | } |
5037 | ||
5038 | /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or | |
5039 | * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller | |
5040 | * as necessary according to their individual configurations. | |
5041 | * | |
5042 | * 'packet' must have sufficient headroom to convert it into a struct | |
5043 | * ofp_packet_in (e.g. as returned by dpif_recv()). | |
5044 | * | |
5045 | * Takes ownership of 'packet'. */ | |
5046 | static void | |
5047 | send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet) | |
5048 | { | |
76ce9432 | 5049 | struct ofconn *ofconn, *prev; |
30ea5d93 | 5050 | int max_len; |
064af421 | 5051 | |
30ea5d93 | 5052 | max_len = do_convert_to_packet_in(packet); |
76ce9432 BP |
5053 | |
5054 | prev = NULL; | |
4e8e4213 | 5055 | LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) { |
c91248b3 | 5056 | if (ofconn_receives_async_msgs(ofconn)) { |
9deba63b | 5057 | if (prev) { |
30ea5d93 | 5058 | schedule_packet_in(prev, packet, max_len, true); |
9deba63b BP |
5059 | } |
5060 | prev = ofconn; | |
064af421 | 5061 | } |
76ce9432 BP |
5062 | } |
5063 | if (prev) { | |
30ea5d93 | 5064 | schedule_packet_in(prev, packet, max_len, false); |
76ce9432 BP |
5065 | } else { |
5066 | ofpbuf_delete(packet); | |
064af421 | 5067 | } |
064af421 BP |
5068 | } |
5069 | ||
5070 | static uint64_t | |
fa60c019 | 5071 | pick_datapath_id(const struct ofproto *ofproto) |
064af421 | 5072 | { |
fa60c019 | 5073 | const struct ofport *port; |
064af421 | 5074 | |
ca0f572c | 5075 | port = get_port(ofproto, ODPP_LOCAL); |
fa60c019 BP |
5076 | if (port) { |
5077 | uint8_t ea[ETH_ADDR_LEN]; | |
5078 | int error; | |
5079 | ||
5080 | error = netdev_get_etheraddr(port->netdev, ea); | |
064af421 BP |
5081 | if (!error) { |
5082 | return eth_addr_to_uint64(ea); | |
5083 | } | |
5084 | VLOG_WARN("could not get MAC address for %s (%s)", | |
fa60c019 | 5085 | netdev_get_name(port->netdev), strerror(error)); |
064af421 | 5086 | } |
fa60c019 | 5087 | return ofproto->fallback_dpid; |
064af421 BP |
5088 | } |
5089 | ||
5090 | static uint64_t | |
5091 | pick_fallback_dpid(void) | |
5092 | { | |
5093 | uint8_t ea[ETH_ADDR_LEN]; | |
70150daf | 5094 | eth_addr_nicira_random(ea); |
064af421 BP |
5095 | return eth_addr_to_uint64(ea); |
5096 | } | |
5097 | \f | |
5098 | static bool | |
ae412e7d | 5099 | default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet, |
064af421 | 5100 | struct odp_actions *actions, tag_type *tags, |
6a07af36 | 5101 | uint16_t *nf_output_iface, void *ofproto_) |
064af421 BP |
5102 | { |
5103 | struct ofproto *ofproto = ofproto_; | |
5104 | int out_port; | |
5105 | ||
5106 | /* Drop frames for reserved multicast addresses. */ | |
5107 | if (eth_addr_is_reserved(flow->dl_dst)) { | |
5108 | return true; | |
5109 | } | |
5110 | ||
5111 | /* Learn source MAC (but don't try to learn from revalidation). */ | |
5112 | if (packet != NULL) { | |
5113 | tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src, | |
7febb910 JG |
5114 | 0, flow->in_port, |
5115 | GRAT_ARP_LOCK_NONE); | |
064af421 BP |
5116 | if (rev_tag) { |
5117 | /* The log messages here could actually be useful in debugging, | |
5118 | * so keep the rate limit relatively high. */ | |
5119 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); | |
5120 | VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16, | |
5121 | ETH_ADDR_ARGS(flow->dl_src), flow->in_port); | |
5122 | ofproto_revalidate(ofproto, rev_tag); | |
5123 | } | |
5124 | } | |
5125 | ||
5126 | /* Determine output port. */ | |
7febb910 JG |
5127 | out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags, |
5128 | NULL); | |
064af421 | 5129 | if (out_port < 0) { |
f1588b1f BP |
5130 | flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD, |
5131 | nf_output_iface, actions); | |
064af421 BP |
5132 | } else if (out_port != flow->in_port) { |
5133 | odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port; | |
6a07af36 | 5134 | *nf_output_iface = out_port; |
064af421 BP |
5135 | } else { |
5136 | /* Drop. */ | |
5137 | } | |
5138 | ||
5139 | return true; | |
5140 | } | |
5141 | ||
5142 | static const struct ofhooks default_ofhooks = { | |
064af421 BP |
5143 | default_normal_ofhook_cb, |
5144 | NULL, | |
5145 | NULL | |
5146 | }; |