]>
Commit | Line | Data |
---|---|---|
064af421 | 1 | /* |
db5ce514 | 2 | * Copyright (c) 2009, 2010, 2011 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" | |
53ddd40a | 38 | #include "multipath.h" |
064af421 BP |
39 | #include "netdev.h" |
40 | #include "netflow.h" | |
cdee00fd | 41 | #include "netlink.h" |
09246b99 | 42 | #include "nx-match.h" |
064af421 BP |
43 | #include "odp-util.h" |
44 | #include "ofp-print.h" | |
fa37b408 | 45 | #include "ofp-util.h" |
72b06300 | 46 | #include "ofproto-sflow.h" |
064af421 BP |
47 | #include "ofpbuf.h" |
48 | #include "openflow/nicira-ext.h" | |
49 | #include "openflow/openflow.h" | |
064af421 BP |
50 | #include "openvswitch/datapath-protocol.h" |
51 | #include "packets.h" | |
52 | #include "pinsched.h" | |
53 | #include "pktbuf.h" | |
54 | #include "poll-loop.h" | |
064af421 BP |
55 | #include "rconn.h" |
56 | #include "shash.h" | |
57 | #include "status.h" | |
fe55ad15 | 58 | #include "stream-ssl.h" |
064af421 BP |
59 | #include "svec.h" |
60 | #include "tag.h" | |
61 | #include "timeval.h" | |
c4617b3c | 62 | #include "unaligned.h" |
4f2cad2c | 63 | #include "unixctl.h" |
064af421 | 64 | #include "vconn.h" |
5136ce49 | 65 | #include "vlog.h" |
064af421 | 66 | |
d98e6007 | 67 | VLOG_DEFINE_THIS_MODULE(ofproto); |
064af421 | 68 | |
cc01d0bb BP |
69 | COVERAGE_DEFINE(facet_changed_rule); |
70 | COVERAGE_DEFINE(facet_revalidate); | |
d76f09ea | 71 | COVERAGE_DEFINE(odp_overflow); |
d76f09ea BP |
72 | COVERAGE_DEFINE(ofproto_agg_request); |
73 | COVERAGE_DEFINE(ofproto_costly_flags); | |
74 | COVERAGE_DEFINE(ofproto_ctlr_action); | |
cc01d0bb | 75 | COVERAGE_DEFINE(ofproto_del_rule); |
d76f09ea BP |
76 | COVERAGE_DEFINE(ofproto_error); |
77 | COVERAGE_DEFINE(ofproto_expiration); | |
78 | COVERAGE_DEFINE(ofproto_expired); | |
79 | COVERAGE_DEFINE(ofproto_flows_req); | |
80 | COVERAGE_DEFINE(ofproto_flush); | |
81 | COVERAGE_DEFINE(ofproto_invalidated); | |
d76f09ea | 82 | COVERAGE_DEFINE(ofproto_no_packet_in); |
d76f09ea BP |
83 | COVERAGE_DEFINE(ofproto_ofconn_stuck); |
84 | COVERAGE_DEFINE(ofproto_ofp2odp); | |
85 | COVERAGE_DEFINE(ofproto_packet_in); | |
86 | COVERAGE_DEFINE(ofproto_packet_out); | |
87 | COVERAGE_DEFINE(ofproto_queue_req); | |
88 | COVERAGE_DEFINE(ofproto_recv_openflow); | |
89 | COVERAGE_DEFINE(ofproto_reinit_ports); | |
d76f09ea BP |
90 | COVERAGE_DEFINE(ofproto_unexpected_rule); |
91 | COVERAGE_DEFINE(ofproto_uninstallable); | |
92 | COVERAGE_DEFINE(ofproto_update_port); | |
93 | ||
878ae780 EJ |
94 | /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a |
95 | * flow translation. */ | |
96 | #define MAX_RESUBMIT_RECURSION 16 | |
97 | ||
f29152ca BP |
98 | struct rule; |
99 | ||
064af421 | 100 | struct ofport { |
ca0f572c | 101 | struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */ |
064af421 BP |
102 | struct netdev *netdev; |
103 | struct ofp_phy_port opp; /* In host byte order. */ | |
ca0f572c | 104 | uint16_t odp_port; |
064af421 BP |
105 | }; |
106 | ||
107 | static void ofport_free(struct ofport *); | |
108 | static void hton_ofp_phy_port(struct ofp_phy_port *); | |
109 | ||
f29152ca BP |
110 | struct action_xlate_ctx { |
111 | /* action_xlate_ctx_init() initializes these members. */ | |
112 | ||
113 | /* The ofproto. */ | |
114 | struct ofproto *ofproto; | |
115 | ||
116 | /* Flow to which the OpenFlow actions apply. xlate_actions() will modify | |
117 | * this flow when actions change header fields. */ | |
118 | struct flow flow; | |
119 | ||
120 | /* The packet corresponding to 'flow', or a null pointer if we are | |
121 | * revalidating without a packet to refer to. */ | |
122 | const struct ofpbuf *packet; | |
123 | ||
7aa697dd BP |
124 | /* If nonnull, called just before executing a resubmit action. |
125 | * | |
126 | * This is normally null so the client has to set it manually after | |
127 | * calling action_xlate_ctx_init(). */ | |
878ae780 | 128 | void (*resubmit_hook)(struct action_xlate_ctx *, struct rule *); |
7aa697dd | 129 | |
ebe482fd EJ |
130 | /* If true, the speciality of 'flow' should be checked before executing |
131 | * its actions. If special_cb returns false on 'flow' rendered | |
132 | * uninstallable and no actions will be executed. */ | |
133 | bool check_special; | |
134 | ||
f29152ca BP |
135 | /* xlate_actions() initializes and uses these members. The client might want |
136 | * to look at them after it returns. */ | |
137 | ||
cdee00fd | 138 | struct ofpbuf *odp_actions; /* Datapath actions. */ |
f29152ca BP |
139 | tag_type tags; /* Tags associated with OFPP_NORMAL actions. */ |
140 | bool may_set_up_flow; /* True ordinarily; false if the actions must | |
141 | * be reassessed for every packet. */ | |
142 | uint16_t nf_output_iface; /* Output interface index for NetFlow. */ | |
143 | ||
144 | /* xlate_actions() initializes and uses these members, but the client has no | |
145 | * reason to look at them. */ | |
146 | ||
147 | int recurse; /* Recursion level, via xlate_table_action. */ | |
cdee00fd | 148 | int last_pop_priority; /* Offset in 'odp_actions' just past most |
7aec165d | 149 | * recent ODP_ACTION_ATTR_SET_PRIORITY. */ |
f29152ca BP |
150 | }; |
151 | ||
152 | static void action_xlate_ctx_init(struct action_xlate_ctx *, | |
153 | struct ofproto *, const struct flow *, | |
154 | const struct ofpbuf *); | |
cdee00fd BP |
155 | static struct ofpbuf *xlate_actions(struct action_xlate_ctx *, |
156 | const union ofp_action *in, size_t n_in); | |
064af421 | 157 | |
bcf84111 | 158 | /* An OpenFlow flow. */ |
064af421 | 159 | struct rule { |
0c43ad9a | 160 | long long int used; /* Time last used; time created if not used. */ |
064af421 | 161 | long long int created; /* Creation time. */ |
064af421 | 162 | |
bcf84111 | 163 | /* These statistics: |
064af421 | 164 | * |
bcf84111 BP |
165 | * - Do include packets and bytes from facets that have been deleted or |
166 | * whose own statistics have been folded into the rule. | |
79eee1eb | 167 | * |
bcf84111 BP |
168 | * - Do include packets and bytes sent "by hand" that were accounted to |
169 | * the rule without any facet being involved (this is a rare corner | |
170 | * case in rule_execute()). | |
064af421 | 171 | * |
bcf84111 BP |
172 | * - Do not include packet or bytes that can be obtained from any facet's |
173 | * packet_count or byte_count member or that can be obtained from the | |
174 | * datapath by, e.g., dpif_flow_get() for any facet. | |
175 | */ | |
176 | uint64_t packet_count; /* Number of packets received. */ | |
177 | uint64_t byte_count; /* Number of bytes received. */ | |
178 | ||
179 | ovs_be64 flow_cookie; /* Controller-issued identifier. */ | |
180 | ||
181 | struct cls_rule cr; /* In owning ofproto's classifier. */ | |
182 | uint16_t idle_timeout; /* In seconds from time of last use. */ | |
183 | uint16_t hard_timeout; /* In seconds from time of creation. */ | |
184 | bool send_flow_removed; /* Send a flow removed message? */ | |
185 | int n_actions; /* Number of elements in actions[]. */ | |
186 | union ofp_action *actions; /* OpenFlow actions. */ | |
187 | struct list facets; /* List of "struct facet"s. */ | |
064af421 BP |
188 | }; |
189 | ||
bcf84111 BP |
190 | static struct rule *rule_from_cls_rule(const struct cls_rule *); |
191 | static bool rule_is_hidden(const struct rule *); | |
064af421 | 192 | |
bcf84111 | 193 | static struct rule *rule_create(const struct cls_rule *, |
0193b2af | 194 | const union ofp_action *, size_t n_actions, |
ca069229 | 195 | uint16_t idle_timeout, uint16_t hard_timeout, |
8054fc48 | 196 | ovs_be64 flow_cookie, bool send_flow_removed); |
064af421 | 197 | static void rule_destroy(struct ofproto *, struct rule *); |
bcf84111 BP |
198 | static void rule_free(struct rule *); |
199 | ||
200 | static struct rule *rule_lookup(struct ofproto *, const struct flow *); | |
afe75089 | 201 | static void rule_insert(struct ofproto *, struct rule *); |
064af421 | 202 | static void rule_remove(struct ofproto *, struct rule *); |
bcf84111 BP |
203 | |
204 | static void rule_send_removed(struct ofproto *, struct rule *, uint8_t reason); | |
3394b5b6 EJ |
205 | static void rule_get_stats(const struct rule *, uint64_t *packets, |
206 | uint64_t *bytes); | |
bcf84111 BP |
207 | |
208 | /* An exact-match instantiation of an OpenFlow flow. */ | |
209 | struct facet { | |
210 | long long int used; /* Time last used; time created if not used. */ | |
211 | ||
212 | /* These statistics: | |
213 | * | |
214 | * - Do include packets and bytes sent "by hand", e.g. with | |
215 | * dpif_execute(). | |
216 | * | |
217 | * - Do include packets and bytes that were obtained from the datapath | |
218 | * when a flow was deleted (e.g. dpif_flow_del()) or when its | |
ba25b8f4 BP |
219 | * statistics were reset (e.g. dpif_flow_put() with |
220 | * DPIF_FP_ZERO_STATS). | |
bcf84111 BP |
221 | * |
222 | * - Do not include any packets or bytes that can currently be obtained | |
223 | * from the datapath by, e.g., dpif_flow_get(). | |
224 | */ | |
225 | uint64_t packet_count; /* Number of packets received. */ | |
226 | uint64_t byte_count; /* Number of bytes received. */ | |
227 | ||
3394b5b6 EJ |
228 | uint64_t dp_packet_count; /* Last known packet count in the datapath. */ |
229 | uint64_t dp_byte_count; /* Last known byte count in the datapath. */ | |
230 | ||
878ae780 EJ |
231 | uint64_t rs_packet_count; /* Packets pushed to resubmit children. */ |
232 | uint64_t rs_byte_count; /* Bytes pushed to resubmit children. */ | |
233 | long long int rs_used; /* Used time pushed to resubmit children. */ | |
234 | ||
bcf84111 BP |
235 | /* Number of bytes passed to account_cb. This may include bytes that can |
236 | * currently obtained from the datapath (thus, it can be greater than | |
237 | * byte_count). */ | |
238 | uint64_t accounted_bytes; | |
239 | ||
240 | struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */ | |
241 | struct list list_node; /* In owning rule's 'facets' list. */ | |
242 | struct rule *rule; /* Owning rule. */ | |
243 | struct flow flow; /* Exact-match flow. */ | |
244 | bool installed; /* Installed in datapath? */ | |
245 | bool may_install; /* True ordinarily; false if actions must | |
246 | * be reassessed for every packet. */ | |
cf22f8cb | 247 | size_t actions_len; /* Number of bytes in actions[]. */ |
cdee00fd | 248 | struct nlattr *actions; /* Datapath actions. */ |
bcf84111 BP |
249 | tag_type tags; /* Tags (set only by hooks). */ |
250 | struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ | |
251 | }; | |
252 | ||
253 | static struct facet *facet_create(struct ofproto *, struct rule *, | |
254 | const struct flow *, | |
255 | const struct ofpbuf *packet); | |
256 | static void facet_remove(struct ofproto *, struct facet *); | |
257 | static void facet_free(struct facet *); | |
258 | ||
259 | static struct facet *facet_lookup_valid(struct ofproto *, const struct flow *); | |
260 | static bool facet_revalidate(struct ofproto *, struct facet *); | |
261 | ||
262 | static void facet_install(struct ofproto *, struct facet *, bool zero_stats); | |
263 | static void facet_uninstall(struct ofproto *, struct facet *); | |
d530fcd2 | 264 | static void facet_flush_stats(struct ofproto *, struct facet *); |
bcf84111 | 265 | |
7f7ae89d BP |
266 | static void facet_make_actions(struct ofproto *, struct facet *, |
267 | const struct ofpbuf *packet); | |
bcf84111 | 268 | static void facet_update_stats(struct ofproto *, struct facet *, |
c97fb132 | 269 | const struct dpif_flow_stats *); |
878ae780 | 270 | static void facet_push_stats(struct ofproto *, struct facet *); |
064af421 | 271 | |
76ce9432 BP |
272 | /* ofproto supports two kinds of OpenFlow connections: |
273 | * | |
5899143f BP |
274 | * - "Primary" connections to ordinary OpenFlow controllers. ofproto |
275 | * maintains persistent connections to these controllers and by default | |
276 | * sends them asynchronous messages such as packet-ins. | |
76ce9432 | 277 | * |
5899143f | 278 | * - "Service" connections, e.g. from ovs-ofctl. When these connections |
76ce9432 BP |
279 | * drop, it is the other side's responsibility to reconnect them if |
280 | * necessary. ofproto does not send them asynchronous messages by default. | |
7d674866 BP |
281 | * |
282 | * Currently, active (tcp, ssl, unix) connections are always "primary" | |
283 | * connections and passive (ptcp, pssl, punix) connections are always "service" | |
284 | * connections. There is no inherent reason for this, but it reflects the | |
285 | * common case. | |
76ce9432 BP |
286 | */ |
287 | enum ofconn_type { | |
5899143f BP |
288 | OFCONN_PRIMARY, /* An ordinary OpenFlow controller. */ |
289 | OFCONN_SERVICE /* A service connection, e.g. "ovs-ofctl". */ | |
76ce9432 | 290 | }; |
064af421 | 291 | |
7d674866 BP |
292 | /* A listener for incoming OpenFlow "service" connections. */ |
293 | struct ofservice { | |
294 | struct hmap_node node; /* In struct ofproto's "services" hmap. */ | |
295 | struct pvconn *pvconn; /* OpenFlow connection listener. */ | |
296 | ||
297 | /* These are not used by ofservice directly. They are settings for | |
298 | * accepted "struct ofconn"s from the pvconn. */ | |
299 | int probe_interval; /* Max idle time before probing, in seconds. */ | |
300 | int rate_limit; /* Max packet-in rate in packets per second. */ | |
301 | int burst_limit; /* Limit on accumulating packet credits. */ | |
302 | }; | |
303 | ||
304 | static struct ofservice *ofservice_lookup(struct ofproto *, | |
305 | const char *target); | |
306 | static int ofservice_create(struct ofproto *, | |
307 | const struct ofproto_controller *); | |
308 | static void ofservice_reconfigure(struct ofservice *, | |
309 | const struct ofproto_controller *); | |
310 | static void ofservice_destroy(struct ofproto *, struct ofservice *); | |
311 | ||
76ce9432 BP |
312 | /* An OpenFlow connection. */ |
313 | struct ofconn { | |
314 | struct ofproto *ofproto; /* The ofproto that owns this connection. */ | |
315 | struct list node; /* In struct ofproto's "all_conns" list. */ | |
316 | struct rconn *rconn; /* OpenFlow connection. */ | |
317 | enum ofconn_type type; /* Type. */ | |
492f7572 | 318 | enum nx_flow_format flow_format; /* Currently selected flow format. */ |
76ce9432 BP |
319 | |
320 | /* OFPT_PACKET_IN related data. */ | |
321 | struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */ | |
82272ede BP |
322 | #define N_SCHEDULERS 2 |
323 | struct pinsched *schedulers[N_SCHEDULERS]; | |
76ce9432 BP |
324 | struct pktbuf *pktbuf; /* OpenFlow packet buffers. */ |
325 | int miss_send_len; /* Bytes to send of buffered packets. */ | |
326 | ||
327 | /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow | |
328 | * requests, and the maximum number before we stop reading OpenFlow | |
329 | * requests. */ | |
064af421 BP |
330 | #define OFCONN_REPLY_MAX 100 |
331 | struct rconn_packet_counter *reply_counter; | |
76ce9432 | 332 | |
5899143f | 333 | /* type == OFCONN_PRIMARY only. */ |
9deba63b | 334 | enum nx_role role; /* Role. */ |
76ce9432 BP |
335 | struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */ |
336 | struct discovery *discovery; /* Controller discovery object, if enabled. */ | |
337 | struct status_category *ss; /* Switch status category. */ | |
d2ede7bc | 338 | enum ofproto_band band; /* In-band or out-of-band? */ |
064af421 BP |
339 | }; |
340 | ||
76ce9432 BP |
341 | |
342 | static struct ofconn *ofconn_create(struct ofproto *, struct rconn *, | |
343 | enum ofconn_type); | |
c475ae67 | 344 | static void ofconn_destroy(struct ofconn *); |
3269c562 | 345 | static void ofconn_run(struct ofconn *); |
064af421 | 346 | static void ofconn_wait(struct ofconn *); |
c91248b3 | 347 | static bool ofconn_receives_async_msgs(const struct ofconn *); |
eb15cdbb | 348 | static char *ofconn_make_name(const struct ofproto *, const char *target); |
7d674866 | 349 | static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst); |
c91248b3 | 350 | |
064af421 BP |
351 | static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn, |
352 | struct rconn_packet_counter *counter); | |
353 | ||
856081f6 BP |
354 | static void send_packet_in(struct ofproto *, struct dpif_upcall *, |
355 | const struct flow *, bool clone); | |
356 | static void do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn); | |
76ce9432 | 357 | |
064af421 BP |
358 | struct ofproto { |
359 | /* Settings. */ | |
360 | uint64_t datapath_id; /* Datapath ID. */ | |
361 | uint64_t fallback_dpid; /* Datapath ID if no better choice found. */ | |
5a719c38 JP |
362 | char *mfr_desc; /* Manufacturer. */ |
363 | char *hw_desc; /* Hardware. */ | |
364 | char *sw_desc; /* Software version. */ | |
365 | char *serial_desc; /* Serial number. */ | |
8abc4ed7 | 366 | char *dp_desc; /* Datapath description. */ |
064af421 BP |
367 | |
368 | /* Datapath. */ | |
c228a364 | 369 | struct dpif *dpif; |
e9e28be3 | 370 | struct netdev_monitor *netdev_monitor; |
ca0f572c | 371 | struct hmap ports; /* Contains "struct ofport"s. */ |
064af421 BP |
372 | struct shash port_by_name; |
373 | uint32_t max_ports; | |
374 | ||
375 | /* Configuration. */ | |
376 | struct switch_status *switch_status; | |
064af421 | 377 | struct fail_open *fail_open; |
064af421 | 378 | struct netflow *netflow; |
72b06300 | 379 | struct ofproto_sflow *sflow; |
064af421 | 380 | |
d2ede7bc BP |
381 | /* In-band control. */ |
382 | struct in_band *in_band; | |
383 | long long int next_in_band_update; | |
917e50e1 BP |
384 | struct sockaddr_in *extra_in_band_remotes; |
385 | size_t n_extra_remotes; | |
b1da6250 | 386 | int in_band_queue; |
917e50e1 | 387 | |
064af421 BP |
388 | /* Flow table. */ |
389 | struct classifier cls; | |
064af421 | 390 | long long int next_expiration; |
bcf84111 BP |
391 | |
392 | /* Facets. */ | |
393 | struct hmap facets; | |
394 | bool need_revalidate; | |
064af421 BP |
395 | struct tag_set revalidate_set; |
396 | ||
397 | /* OpenFlow connections. */ | |
76ce9432 BP |
398 | struct hmap controllers; /* Controller "struct ofconn"s. */ |
399 | struct list all_conns; /* Contains "struct ofconn"s. */ | |
31681a5d | 400 | enum ofproto_fail_mode fail_mode; |
7d674866 BP |
401 | |
402 | /* OpenFlow listeners. */ | |
403 | struct hmap services; /* Contains "struct ofservice"s. */ | |
064af421 BP |
404 | struct pvconn **snoops; |
405 | size_t n_snoops; | |
406 | ||
407 | /* Hooks for ovs-vswitchd. */ | |
408 | const struct ofhooks *ofhooks; | |
409 | void *aux; | |
410 | ||
411 | /* Used by default ofhooks. */ | |
412 | struct mac_learning *ml; | |
413 | }; | |
414 | ||
7aa697dd BP |
415 | /* Map from dpif name to struct ofproto, for use by unixctl commands. */ |
416 | static struct shash all_ofprotos = SHASH_INITIALIZER(&all_ofprotos); | |
417 | ||
064af421 BP |
418 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); |
419 | ||
420 | static const struct ofhooks default_ofhooks; | |
421 | ||
fa60c019 | 422 | static uint64_t pick_datapath_id(const struct ofproto *); |
064af421 | 423 | static uint64_t pick_fallback_dpid(void); |
76ce9432 | 424 | |
0de7a4b4 | 425 | static int ofproto_expire(struct ofproto *); |
878ae780 EJ |
426 | static void flow_push_stats(struct ofproto *, const struct rule *, |
427 | struct flow *, uint64_t packets, uint64_t bytes, | |
428 | long long int used); | |
4a4cdb3b | 429 | |
856081f6 | 430 | static void handle_upcall(struct ofproto *, struct dpif_upcall *); |
064af421 | 431 | |
3269c562 | 432 | static void handle_openflow(struct ofconn *, struct ofpbuf *); |
064af421 | 433 | |
ca0f572c | 434 | static struct ofport *get_port(const struct ofproto *, uint16_t odp_port); |
064af421 BP |
435 | static void update_port(struct ofproto *, const char *devname); |
436 | static int init_ports(struct ofproto *); | |
437 | static void reinit_ports(struct ofproto *); | |
438 | ||
7aa697dd BP |
439 | static void ofproto_unixctl_init(void); |
440 | ||
064af421 | 441 | int |
1a6f1e2a JG |
442 | ofproto_create(const char *datapath, const char *datapath_type, |
443 | const struct ofhooks *ofhooks, void *aux, | |
064af421 BP |
444 | struct ofproto **ofprotop) |
445 | { | |
064af421 | 446 | struct ofproto *p; |
c228a364 | 447 | struct dpif *dpif; |
064af421 BP |
448 | int error; |
449 | ||
450 | *ofprotop = NULL; | |
451 | ||
7aa697dd BP |
452 | ofproto_unixctl_init(); |
453 | ||
064af421 | 454 | /* Connect to datapath and start listening for messages. */ |
1a6f1e2a | 455 | error = dpif_open(datapath, datapath_type, &dpif); |
064af421 BP |
456 | if (error) { |
457 | VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error)); | |
458 | return error; | |
459 | } | |
82272ede BP |
460 | error = dpif_recv_set_mask(dpif, |
461 | ((1u << DPIF_UC_MISS) | | |
462 | (1u << DPIF_UC_ACTION) | | |
463 | (1u << DPIF_UC_SAMPLE))); | |
064af421 BP |
464 | if (error) { |
465 | VLOG_ERR("failed to listen on datapath %s: %s", | |
466 | datapath, strerror(error)); | |
c228a364 | 467 | dpif_close(dpif); |
064af421 BP |
468 | return error; |
469 | } | |
c228a364 | 470 | dpif_flow_flush(dpif); |
8f24562a | 471 | dpif_recv_purge(dpif); |
064af421 BP |
472 | |
473 | /* Initialize settings. */ | |
ec6fde61 | 474 | p = xzalloc(sizeof *p); |
064af421 | 475 | p->fallback_dpid = pick_fallback_dpid(); |
fa60c019 | 476 | p->datapath_id = p->fallback_dpid; |
5a719c38 JP |
477 | p->mfr_desc = xstrdup(DEFAULT_MFR_DESC); |
478 | p->hw_desc = xstrdup(DEFAULT_HW_DESC); | |
479 | p->sw_desc = xstrdup(DEFAULT_SW_DESC); | |
480 | p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC); | |
23ff2821 | 481 | p->dp_desc = xstrdup(DEFAULT_DP_DESC); |
064af421 BP |
482 | |
483 | /* Initialize datapath. */ | |
484 | p->dpif = dpif; | |
8b61709d | 485 | p->netdev_monitor = netdev_monitor_create(); |
ca0f572c | 486 | hmap_init(&p->ports); |
064af421 | 487 | shash_init(&p->port_by_name); |
996c1b3d | 488 | p->max_ports = dpif_get_max_ports(dpif); |
064af421 BP |
489 | |
490 | /* Initialize submodules. */ | |
491 | p->switch_status = switch_status_create(p); | |
064af421 | 492 | p->fail_open = NULL; |
064af421 | 493 | p->netflow = NULL; |
72b06300 | 494 | p->sflow = NULL; |
064af421 | 495 | |
b1da6250 BP |
496 | /* Initialize in-band control. */ |
497 | p->in_band = NULL; | |
498 | p->in_band_queue = -1; | |
499 | ||
064af421 BP |
500 | /* Initialize flow table. */ |
501 | classifier_init(&p->cls); | |
064af421 | 502 | p->next_expiration = time_msec() + 1000; |
bcf84111 BP |
503 | |
504 | /* Initialize facet table. */ | |
505 | hmap_init(&p->facets); | |
506 | p->need_revalidate = false; | |
064af421 BP |
507 | tag_set_init(&p->revalidate_set); |
508 | ||
509 | /* Initialize OpenFlow connections. */ | |
510 | list_init(&p->all_conns); | |
76ce9432 | 511 | hmap_init(&p->controllers); |
7d674866 | 512 | hmap_init(&p->services); |
064af421 BP |
513 | p->snoops = NULL; |
514 | p->n_snoops = 0; | |
515 | ||
516 | /* Initialize hooks. */ | |
517 | if (ofhooks) { | |
518 | p->ofhooks = ofhooks; | |
519 | p->aux = aux; | |
520 | p->ml = NULL; | |
521 | } else { | |
522 | p->ofhooks = &default_ofhooks; | |
523 | p->aux = p; | |
524 | p->ml = mac_learning_create(); | |
525 | } | |
526 | ||
fa60c019 BP |
527 | /* Pick final datapath ID. */ |
528 | p->datapath_id = pick_datapath_id(p); | |
b123cc3c | 529 | VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id); |
fa60c019 | 530 | |
7aa697dd BP |
531 | shash_add_once(&all_ofprotos, dpif_name(p->dpif), p); |
532 | ||
064af421 BP |
533 | *ofprotop = p; |
534 | return 0; | |
535 | } | |
536 | ||
537 | void | |
538 | ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id) | |
539 | { | |
540 | uint64_t old_dpid = p->datapath_id; | |
fa60c019 | 541 | p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p); |
064af421 | 542 | if (p->datapath_id != old_dpid) { |
b123cc3c | 543 | VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id); |
76ce9432 BP |
544 | |
545 | /* Force all active connections to reconnect, since there is no way to | |
546 | * notify a controller that the datapath ID has changed. */ | |
fa05809b | 547 | ofproto_reconnect_controllers(p); |
064af421 BP |
548 | } |
549 | } | |
550 | ||
76ce9432 BP |
551 | static bool |
552 | is_discovery_controller(const struct ofproto_controller *c) | |
553 | { | |
554 | return !strcmp(c->target, "discover"); | |
555 | } | |
556 | ||
557 | static bool | |
558 | is_in_band_controller(const struct ofproto_controller *c) | |
559 | { | |
560 | return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND; | |
561 | } | |
562 | ||
563 | /* Creates a new controller in 'ofproto'. Some of the settings are initially | |
564 | * drawn from 'c', but update_controller() needs to be called later to finish | |
565 | * the new ofconn's configuration. */ | |
566 | static void | |
567 | add_controller(struct ofproto *ofproto, const struct ofproto_controller *c) | |
568 | { | |
569 | struct discovery *discovery; | |
570 | struct ofconn *ofconn; | |
571 | ||
572 | if (is_discovery_controller(c)) { | |
573 | int error = discovery_create(c->accept_re, c->update_resolv_conf, | |
574 | ofproto->dpif, ofproto->switch_status, | |
575 | &discovery); | |
576 | if (error) { | |
577 | return; | |
578 | } | |
579 | } else { | |
580 | discovery = NULL; | |
581 | } | |
582 | ||
5899143f | 583 | ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY); |
76ce9432 BP |
584 | ofconn->pktbuf = pktbuf_create(); |
585 | ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN; | |
586 | if (discovery) { | |
587 | ofconn->discovery = discovery; | |
588 | } else { | |
eb15cdbb BP |
589 | char *name = ofconn_make_name(ofproto, c->target); |
590 | rconn_connect(ofconn->rconn, c->target, name); | |
591 | free(name); | |
76ce9432 BP |
592 | } |
593 | hmap_insert(&ofproto->controllers, &ofconn->hmap_node, | |
594 | hash_string(c->target, 0)); | |
595 | } | |
596 | ||
597 | /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's | |
598 | * target or turn discovery on or off (these are done by creating new ofconns | |
599 | * and deleting old ones), but it can update the rest of an ofconn's | |
600 | * settings. */ | |
601 | static void | |
602 | update_controller(struct ofconn *ofconn, const struct ofproto_controller *c) | |
064af421 | 603 | { |
76ce9432 | 604 | int probe_interval; |
79c9f2ee | 605 | |
d2ede7bc BP |
606 | ofconn->band = (is_in_band_controller(c) |
607 | ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND); | |
608 | ||
76ce9432 | 609 | rconn_set_max_backoff(ofconn->rconn, c->max_backoff); |
79c9f2ee | 610 | |
76ce9432 BP |
611 | probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0; |
612 | rconn_set_probe_interval(ofconn->rconn, probe_interval); | |
79c9f2ee | 613 | |
76ce9432 BP |
614 | if (ofconn->discovery) { |
615 | discovery_set_update_resolv_conf(ofconn->discovery, | |
616 | c->update_resolv_conf); | |
617 | discovery_set_accept_controller_re(ofconn->discovery, c->accept_re); | |
618 | } | |
79c9f2ee | 619 | |
7d674866 | 620 | ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit); |
76ce9432 | 621 | } |
79c9f2ee | 622 | |
76ce9432 BP |
623 | static const char * |
624 | ofconn_get_target(const struct ofconn *ofconn) | |
625 | { | |
eb15cdbb | 626 | return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn); |
76ce9432 BP |
627 | } |
628 | ||
629 | static struct ofconn * | |
630 | find_controller_by_target(struct ofproto *ofproto, const char *target) | |
631 | { | |
632 | struct ofconn *ofconn; | |
633 | ||
4e8e4213 | 634 | HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node, |
76ce9432 BP |
635 | hash_string(target, 0), &ofproto->controllers) { |
636 | if (!strcmp(ofconn_get_target(ofconn), target)) { | |
637 | return ofconn; | |
79c9f2ee | 638 | } |
064af421 | 639 | } |
76ce9432 BP |
640 | return NULL; |
641 | } | |
064af421 | 642 | |
d2ede7bc BP |
643 | static void |
644 | update_in_band_remotes(struct ofproto *ofproto) | |
645 | { | |
646 | const struct ofconn *ofconn; | |
647 | struct sockaddr_in *addrs; | |
917e50e1 | 648 | size_t max_addrs, n_addrs; |
d2ede7bc | 649 | bool discovery; |
917e50e1 | 650 | size_t i; |
d2ede7bc | 651 | |
917e50e1 BP |
652 | /* Allocate enough memory for as many remotes as we could possibly have. */ |
653 | max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers); | |
654 | addrs = xmalloc(max_addrs * sizeof *addrs); | |
d2ede7bc BP |
655 | n_addrs = 0; |
656 | ||
657 | /* Add all the remotes. */ | |
658 | discovery = false; | |
4e8e4213 | 659 | HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) { |
d2ede7bc BP |
660 | struct sockaddr_in *sin = &addrs[n_addrs]; |
661 | ||
487ec65f BP |
662 | if (ofconn->band == OFPROTO_OUT_OF_BAND) { |
663 | continue; | |
664 | } | |
665 | ||
d2ede7bc BP |
666 | sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn); |
667 | if (sin->sin_addr.s_addr) { | |
668 | sin->sin_port = rconn_get_remote_port(ofconn->rconn); | |
669 | n_addrs++; | |
670 | } | |
671 | if (ofconn->discovery) { | |
672 | discovery = true; | |
673 | } | |
674 | } | |
917e50e1 BP |
675 | for (i = 0; i < ofproto->n_extra_remotes; i++) { |
676 | addrs[n_addrs++] = ofproto->extra_in_band_remotes[i]; | |
677 | } | |
d2ede7bc BP |
678 | |
679 | /* Create or update or destroy in-band. | |
680 | * | |
681 | * Ordinarily we only enable in-band if there's at least one remote | |
682 | * address, but discovery needs the in-band rules for DHCP to be installed | |
683 | * even before we know any remote addresses. */ | |
684 | if (n_addrs || discovery) { | |
685 | if (!ofproto->in_band) { | |
686 | in_band_create(ofproto, ofproto->dpif, ofproto->switch_status, | |
687 | &ofproto->in_band); | |
688 | } | |
40cae670 BP |
689 | if (ofproto->in_band) { |
690 | in_band_set_remotes(ofproto->in_band, addrs, n_addrs); | |
691 | } | |
b1da6250 | 692 | in_band_set_queue(ofproto->in_band, ofproto->in_band_queue); |
d2ede7bc BP |
693 | ofproto->next_in_band_update = time_msec() + 1000; |
694 | } else { | |
695 | in_band_destroy(ofproto->in_band); | |
696 | ofproto->in_band = NULL; | |
697 | } | |
698 | ||
699 | /* Clean up. */ | |
700 | free(addrs); | |
701 | } | |
702 | ||
31681a5d JP |
703 | static void |
704 | update_fail_open(struct ofproto *p) | |
705 | { | |
706 | struct ofconn *ofconn; | |
707 | ||
708 | if (!hmap_is_empty(&p->controllers) | |
709 | && p->fail_mode == OFPROTO_FAIL_STANDALONE) { | |
710 | struct rconn **rconns; | |
711 | size_t n; | |
712 | ||
713 | if (!p->fail_open) { | |
714 | p->fail_open = fail_open_create(p, p->switch_status); | |
715 | } | |
716 | ||
717 | n = 0; | |
718 | rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns); | |
4e8e4213 | 719 | HMAP_FOR_EACH (ofconn, hmap_node, &p->controllers) { |
31681a5d JP |
720 | rconns[n++] = ofconn->rconn; |
721 | } | |
722 | ||
723 | fail_open_set_controllers(p->fail_open, rconns, n); | |
724 | /* p->fail_open takes ownership of 'rconns'. */ | |
725 | } else { | |
726 | fail_open_destroy(p->fail_open); | |
727 | p->fail_open = NULL; | |
728 | } | |
729 | } | |
730 | ||
76ce9432 BP |
731 | void |
732 | ofproto_set_controllers(struct ofproto *p, | |
733 | const struct ofproto_controller *controllers, | |
734 | size_t n_controllers) | |
735 | { | |
736 | struct shash new_controllers; | |
7d674866 BP |
737 | struct ofconn *ofconn, *next_ofconn; |
738 | struct ofservice *ofservice, *next_ofservice; | |
76ce9432 | 739 | bool ss_exists; |
76ce9432 | 740 | size_t i; |
79c9f2ee | 741 | |
7d674866 BP |
742 | /* Create newly configured controllers and services. |
743 | * Create a name to ofproto_controller mapping in 'new_controllers'. */ | |
76ce9432 BP |
744 | shash_init(&new_controllers); |
745 | for (i = 0; i < n_controllers; i++) { | |
746 | const struct ofproto_controller *c = &controllers[i]; | |
747 | ||
7d674866 BP |
748 | if (!vconn_verify_name(c->target) || !strcmp(c->target, "discover")) { |
749 | if (!find_controller_by_target(p, c->target)) { | |
750 | add_controller(p, c); | |
751 | } | |
752 | } else if (!pvconn_verify_name(c->target)) { | |
753 | if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) { | |
754 | continue; | |
755 | } | |
756 | } else { | |
757 | VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"", | |
758 | dpif_name(p->dpif), c->target); | |
759 | continue; | |
76ce9432 | 760 | } |
7d674866 BP |
761 | |
762 | shash_add_once(&new_controllers, c->target, &controllers[i]); | |
76ce9432 BP |
763 | } |
764 | ||
7d674866 BP |
765 | /* Delete controllers that are no longer configured. |
766 | * Update configuration of all now-existing controllers. */ | |
76ce9432 | 767 | ss_exists = false; |
4e8e4213 | 768 | HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &p->controllers) { |
76ce9432 BP |
769 | struct ofproto_controller *c; |
770 | ||
771 | c = shash_find_data(&new_controllers, ofconn_get_target(ofconn)); | |
772 | if (!c) { | |
773 | ofconn_destroy(ofconn); | |
79c9f2ee | 774 | } else { |
76ce9432 | 775 | update_controller(ofconn, c); |
76ce9432 BP |
776 | if (ofconn->ss) { |
777 | ss_exists = true; | |
778 | } | |
76ce9432 BP |
779 | } |
780 | } | |
7d674866 BP |
781 | |
782 | /* Delete services that are no longer configured. | |
783 | * Update configuration of all now-existing services. */ | |
4e8e4213 | 784 | HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) { |
7d674866 BP |
785 | struct ofproto_controller *c; |
786 | ||
787 | c = shash_find_data(&new_controllers, | |
788 | pvconn_get_name(ofservice->pvconn)); | |
789 | if (!c) { | |
790 | ofservice_destroy(p, ofservice); | |
791 | } else { | |
792 | ofservice_reconfigure(ofservice, c); | |
793 | } | |
794 | } | |
795 | ||
76ce9432 BP |
796 | shash_destroy(&new_controllers); |
797 | ||
d2ede7bc | 798 | update_in_band_remotes(p); |
31681a5d | 799 | update_fail_open(p); |
79c9f2ee | 800 | |
76ce9432 BP |
801 | if (!hmap_is_empty(&p->controllers) && !ss_exists) { |
802 | ofconn = CONTAINER_OF(hmap_first(&p->controllers), | |
803 | struct ofconn, hmap_node); | |
804 | ofconn->ss = switch_status_register(p->switch_status, "remote", | |
805 | rconn_status_cb, ofconn->rconn); | |
79c9f2ee | 806 | } |
064af421 BP |
807 | } |
808 | ||
31681a5d JP |
809 | void |
810 | ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode) | |
811 | { | |
812 | p->fail_mode = fail_mode; | |
813 | update_fail_open(p); | |
814 | } | |
815 | ||
fa05809b BP |
816 | /* Drops the connections between 'ofproto' and all of its controllers, forcing |
817 | * them to reconnect. */ | |
818 | void | |
819 | ofproto_reconnect_controllers(struct ofproto *ofproto) | |
820 | { | |
821 | struct ofconn *ofconn; | |
822 | ||
4e8e4213 | 823 | LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) { |
fa05809b BP |
824 | rconn_reconnect(ofconn->rconn); |
825 | } | |
826 | } | |
827 | ||
917e50e1 BP |
828 | static bool |
829 | any_extras_changed(const struct ofproto *ofproto, | |
830 | const struct sockaddr_in *extras, size_t n) | |
831 | { | |
832 | size_t i; | |
833 | ||
834 | if (n != ofproto->n_extra_remotes) { | |
835 | return true; | |
836 | } | |
837 | ||
838 | for (i = 0; i < n; i++) { | |
839 | const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i]; | |
840 | const struct sockaddr_in *new = &extras[i]; | |
841 | ||
842 | if (old->sin_addr.s_addr != new->sin_addr.s_addr || | |
843 | old->sin_port != new->sin_port) { | |
844 | return true; | |
845 | } | |
846 | } | |
847 | ||
848 | return false; | |
849 | } | |
850 | ||
851 | /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s | |
852 | * in-band control should guarantee access, in the same way that in-band | |
853 | * control guarantees access to OpenFlow controllers. */ | |
854 | void | |
855 | ofproto_set_extra_in_band_remotes(struct ofproto *ofproto, | |
856 | const struct sockaddr_in *extras, size_t n) | |
857 | { | |
858 | if (!any_extras_changed(ofproto, extras, n)) { | |
859 | return; | |
860 | } | |
861 | ||
862 | free(ofproto->extra_in_band_remotes); | |
863 | ofproto->n_extra_remotes = n; | |
864 | ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras); | |
865 | ||
866 | update_in_band_remotes(ofproto); | |
867 | } | |
868 | ||
b1da6250 BP |
869 | /* Sets the OpenFlow queue used by flows set up by in-band control on |
870 | * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control | |
871 | * flows will use the default queue. */ | |
872 | void | |
873 | ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id) | |
874 | { | |
875 | if (queue_id != ofproto->in_band_queue) { | |
876 | ofproto->in_band_queue = queue_id; | |
877 | update_in_band_remotes(ofproto); | |
878 | } | |
879 | } | |
880 | ||
064af421 BP |
881 | void |
882 | ofproto_set_desc(struct ofproto *p, | |
5a719c38 JP |
883 | const char *mfr_desc, const char *hw_desc, |
884 | const char *sw_desc, const char *serial_desc, | |
8abc4ed7 | 885 | const char *dp_desc) |
064af421 | 886 | { |
5a719c38 JP |
887 | struct ofp_desc_stats *ods; |
888 | ||
889 | if (mfr_desc) { | |
890 | if (strlen(mfr_desc) >= sizeof ods->mfr_desc) { | |
891 | VLOG_WARN("truncating mfr_desc, must be less than %zu characters", | |
892 | sizeof ods->mfr_desc); | |
893 | } | |
894 | free(p->mfr_desc); | |
895 | p->mfr_desc = xstrdup(mfr_desc); | |
064af421 | 896 | } |
5a719c38 JP |
897 | if (hw_desc) { |
898 | if (strlen(hw_desc) >= sizeof ods->hw_desc) { | |
899 | VLOG_WARN("truncating hw_desc, must be less than %zu characters", | |
900 | sizeof ods->hw_desc); | |
901 | } | |
902 | free(p->hw_desc); | |
903 | p->hw_desc = xstrdup(hw_desc); | |
064af421 | 904 | } |
5a719c38 JP |
905 | if (sw_desc) { |
906 | if (strlen(sw_desc) >= sizeof ods->sw_desc) { | |
907 | VLOG_WARN("truncating sw_desc, must be less than %zu characters", | |
908 | sizeof ods->sw_desc); | |
909 | } | |
910 | free(p->sw_desc); | |
911 | p->sw_desc = xstrdup(sw_desc); | |
912 | } | |
913 | if (serial_desc) { | |
914 | if (strlen(serial_desc) >= sizeof ods->serial_num) { | |
915 | VLOG_WARN("truncating serial_desc, must be less than %zu " | |
916 | "characters", | |
917 | sizeof ods->serial_num); | |
918 | } | |
919 | free(p->serial_desc); | |
920 | p->serial_desc = xstrdup(serial_desc); | |
064af421 | 921 | } |
8abc4ed7 | 922 | if (dp_desc) { |
5a719c38 JP |
923 | if (strlen(dp_desc) >= sizeof ods->dp_desc) { |
924 | VLOG_WARN("truncating dp_desc, must be less than %zu characters", | |
925 | sizeof ods->dp_desc); | |
926 | } | |
8abc4ed7 JP |
927 | free(p->dp_desc); |
928 | p->dp_desc = xstrdup(dp_desc); | |
929 | } | |
064af421 BP |
930 | } |
931 | ||
064af421 BP |
932 | static int |
933 | set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp, | |
934 | const struct svec *svec) | |
935 | { | |
936 | struct pvconn **pvconns = *pvconnsp; | |
937 | size_t n_pvconns = *n_pvconnsp; | |
938 | int retval = 0; | |
939 | size_t i; | |
940 | ||
941 | for (i = 0; i < n_pvconns; i++) { | |
942 | pvconn_close(pvconns[i]); | |
943 | } | |
944 | free(pvconns); | |
945 | ||
946 | pvconns = xmalloc(svec->n * sizeof *pvconns); | |
947 | n_pvconns = 0; | |
948 | for (i = 0; i < svec->n; i++) { | |
949 | const char *name = svec->names[i]; | |
950 | struct pvconn *pvconn; | |
951 | int error; | |
952 | ||
953 | error = pvconn_open(name, &pvconn); | |
954 | if (!error) { | |
955 | pvconns[n_pvconns++] = pvconn; | |
956 | } else { | |
957 | VLOG_ERR("failed to listen on %s: %s", name, strerror(error)); | |
958 | if (!retval) { | |
959 | retval = error; | |
960 | } | |
961 | } | |
962 | } | |
963 | ||
964 | *pvconnsp = pvconns; | |
965 | *n_pvconnsp = n_pvconns; | |
966 | ||
967 | return retval; | |
968 | } | |
969 | ||
064af421 BP |
970 | int |
971 | ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops) | |
972 | { | |
973 | return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops); | |
974 | } | |
975 | ||
976 | int | |
0193b2af JG |
977 | ofproto_set_netflow(struct ofproto *ofproto, |
978 | const struct netflow_options *nf_options) | |
064af421 | 979 | { |
76343538 | 980 | if (nf_options && nf_options->collectors.n) { |
064af421 BP |
981 | if (!ofproto->netflow) { |
982 | ofproto->netflow = netflow_create(); | |
983 | } | |
0193b2af | 984 | return netflow_set_options(ofproto->netflow, nf_options); |
064af421 BP |
985 | } else { |
986 | netflow_destroy(ofproto->netflow); | |
987 | ofproto->netflow = NULL; | |
988 | return 0; | |
989 | } | |
990 | } | |
991 | ||
72b06300 BP |
992 | void |
993 | ofproto_set_sflow(struct ofproto *ofproto, | |
994 | const struct ofproto_sflow_options *oso) | |
995 | { | |
996 | struct ofproto_sflow *os = ofproto->sflow; | |
997 | if (oso) { | |
998 | if (!os) { | |
999 | struct ofport *ofport; | |
72b06300 BP |
1000 | |
1001 | os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif); | |
4e8e4213 | 1002 | HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) { |
ca0f572c | 1003 | ofproto_sflow_add_port(os, ofport->odp_port, |
72b06300 BP |
1004 | netdev_get_name(ofport->netdev)); |
1005 | } | |
1006 | } | |
1007 | ofproto_sflow_set_options(os, oso); | |
1008 | } else { | |
1009 | ofproto_sflow_destroy(os); | |
1010 | ofproto->sflow = NULL; | |
1011 | } | |
1012 | } | |
1013 | ||
064af421 BP |
1014 | uint64_t |
1015 | ofproto_get_datapath_id(const struct ofproto *ofproto) | |
1016 | { | |
1017 | return ofproto->datapath_id; | |
1018 | } | |
1019 | ||
76ce9432 | 1020 | bool |
7d674866 | 1021 | ofproto_has_primary_controller(const struct ofproto *ofproto) |
064af421 | 1022 | { |
76ce9432 | 1023 | return !hmap_is_empty(&ofproto->controllers); |
064af421 BP |
1024 | } |
1025 | ||
abdfe474 JP |
1026 | enum ofproto_fail_mode |
1027 | ofproto_get_fail_mode(const struct ofproto *p) | |
1028 | { | |
1029 | return p->fail_mode; | |
1030 | } | |
1031 | ||
064af421 BP |
1032 | void |
1033 | ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops) | |
1034 | { | |
1035 | size_t i; | |
1036 | ||
1037 | for (i = 0; i < ofproto->n_snoops; i++) { | |
1038 | svec_add(snoops, pvconn_get_name(ofproto->snoops[i])); | |
1039 | } | |
1040 | } | |
1041 | ||
1042 | void | |
1043 | ofproto_destroy(struct ofproto *p) | |
1044 | { | |
7d674866 | 1045 | struct ofservice *ofservice, *next_ofservice; |
064af421 | 1046 | struct ofconn *ofconn, *next_ofconn; |
ca0f572c | 1047 | struct ofport *ofport, *next_ofport; |
064af421 BP |
1048 | size_t i; |
1049 | ||
1050 | if (!p) { | |
1051 | return; | |
1052 | } | |
1053 | ||
7aa697dd BP |
1054 | shash_find_and_delete(&all_ofprotos, dpif_name(p->dpif)); |
1055 | ||
f7de2cdf | 1056 | /* Destroy fail-open and in-band early, since they touch the classifier. */ |
79c9f2ee BP |
1057 | fail_open_destroy(p->fail_open); |
1058 | p->fail_open = NULL; | |
1059 | ||
1060 | in_band_destroy(p->in_band); | |
1061 | p->in_band = NULL; | |
917e50e1 | 1062 | free(p->extra_in_band_remotes); |
2f6d3445 | 1063 | |
064af421 BP |
1064 | ofproto_flush_flows(p); |
1065 | classifier_destroy(&p->cls); | |
bcf84111 | 1066 | hmap_destroy(&p->facets); |
064af421 | 1067 | |
4e8e4213 | 1068 | LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) { |
c475ae67 | 1069 | ofconn_destroy(ofconn); |
064af421 | 1070 | } |
76ce9432 | 1071 | hmap_destroy(&p->controllers); |
064af421 | 1072 | |
c228a364 | 1073 | dpif_close(p->dpif); |
e9e28be3 | 1074 | netdev_monitor_destroy(p->netdev_monitor); |
4e8e4213 | 1075 | HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) { |
ca0f572c | 1076 | hmap_remove(&p->ports, &ofport->hmap_node); |
064af421 BP |
1077 | ofport_free(ofport); |
1078 | } | |
1079 | shash_destroy(&p->port_by_name); | |
1080 | ||
1081 | switch_status_destroy(p->switch_status); | |
064af421 | 1082 | netflow_destroy(p->netflow); |
72b06300 | 1083 | ofproto_sflow_destroy(p->sflow); |
064af421 | 1084 | |
4e8e4213 | 1085 | HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) { |
7d674866 | 1086 | ofservice_destroy(p, ofservice); |
064af421 | 1087 | } |
7d674866 | 1088 | hmap_destroy(&p->services); |
064af421 BP |
1089 | |
1090 | for (i = 0; i < p->n_snoops; i++) { | |
1091 | pvconn_close(p->snoops[i]); | |
1092 | } | |
1093 | free(p->snoops); | |
1094 | ||
1095 | mac_learning_destroy(p->ml); | |
1096 | ||
5a719c38 JP |
1097 | free(p->mfr_desc); |
1098 | free(p->hw_desc); | |
1099 | free(p->sw_desc); | |
1100 | free(p->serial_desc); | |
cb871ae0 JP |
1101 | free(p->dp_desc); |
1102 | ||
ca0f572c | 1103 | hmap_destroy(&p->ports); |
3b917492 | 1104 | |
064af421 BP |
1105 | free(p); |
1106 | } | |
1107 | ||
1108 | int | |
1109 | ofproto_run(struct ofproto *p) | |
1110 | { | |
1111 | int error = ofproto_run1(p); | |
1112 | if (!error) { | |
1113 | error = ofproto_run2(p, false); | |
1114 | } | |
1115 | return error; | |
1116 | } | |
1117 | ||
e9e28be3 BP |
1118 | static void |
1119 | process_port_change(struct ofproto *ofproto, int error, char *devname) | |
1120 | { | |
1121 | if (error == ENOBUFS) { | |
1122 | reinit_ports(ofproto); | |
1123 | } else if (!error) { | |
1124 | update_port(ofproto, devname); | |
1125 | free(devname); | |
1126 | } | |
1127 | } | |
1128 | ||
e2bfacb6 BP |
1129 | /* Returns a "preference level" for snooping 'ofconn'. A higher return value |
1130 | * means that 'ofconn' is more interesting for monitoring than a lower return | |
1131 | * value. */ | |
1132 | static int | |
1133 | snoop_preference(const struct ofconn *ofconn) | |
1134 | { | |
1135 | switch (ofconn->role) { | |
1136 | case NX_ROLE_MASTER: | |
1137 | return 3; | |
1138 | case NX_ROLE_OTHER: | |
1139 | return 2; | |
1140 | case NX_ROLE_SLAVE: | |
1141 | return 1; | |
1142 | default: | |
1143 | /* Shouldn't happen. */ | |
1144 | return 0; | |
1145 | } | |
1146 | } | |
1147 | ||
76ce9432 BP |
1148 | /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'. |
1149 | * Connects this vconn to a controller. */ | |
1150 | static void | |
1151 | add_snooper(struct ofproto *ofproto, struct vconn *vconn) | |
1152 | { | |
e2bfacb6 | 1153 | struct ofconn *ofconn, *best; |
76ce9432 | 1154 | |
e2bfacb6 BP |
1155 | /* Pick a controller for monitoring. */ |
1156 | best = NULL; | |
4e8e4213 | 1157 | LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) { |
5899143f | 1158 | if (ofconn->type == OFCONN_PRIMARY |
e2bfacb6 BP |
1159 | && (!best || snoop_preference(ofconn) > snoop_preference(best))) { |
1160 | best = ofconn; | |
76ce9432 | 1161 | } |
e2bfacb6 | 1162 | } |
76ce9432 | 1163 | |
e2bfacb6 BP |
1164 | if (best) { |
1165 | rconn_add_monitor(best->rconn, vconn); | |
1166 | } else { | |
1167 | VLOG_INFO_RL(&rl, "no controller connection to snoop"); | |
1168 | vconn_close(vconn); | |
76ce9432 | 1169 | } |
76ce9432 BP |
1170 | } |
1171 | ||
064af421 BP |
1172 | int |
1173 | ofproto_run1(struct ofproto *p) | |
1174 | { | |
1175 | struct ofconn *ofconn, *next_ofconn; | |
7d674866 | 1176 | struct ofservice *ofservice; |
064af421 BP |
1177 | char *devname; |
1178 | int error; | |
1179 | int i; | |
1180 | ||
149f577a JG |
1181 | if (shash_is_empty(&p->port_by_name)) { |
1182 | init_ports(p); | |
1183 | } | |
1184 | ||
064af421 | 1185 | for (i = 0; i < 50; i++) { |
856081f6 | 1186 | struct dpif_upcall packet; |
064af421 | 1187 | |
856081f6 | 1188 | error = dpif_recv(p->dpif, &packet); |
064af421 BP |
1189 | if (error) { |
1190 | if (error == ENODEV) { | |
1191 | /* Someone destroyed the datapath behind our back. The caller | |
1192 | * better destroy us and give up, because we're just going to | |
1193 | * spin from here on out. */ | |
39a559f2 BP |
1194 | static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5); |
1195 | VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally", | |
c228a364 | 1196 | dpif_name(p->dpif)); |
064af421 BP |
1197 | return ENODEV; |
1198 | } | |
1199 | break; | |
1200 | } | |
1201 | ||
856081f6 | 1202 | handle_upcall(p, &packet); |
064af421 BP |
1203 | } |
1204 | ||
e9e28be3 BP |
1205 | while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) { |
1206 | process_port_change(p, error, devname); | |
1207 | } | |
1208 | while ((error = netdev_monitor_poll(p->netdev_monitor, | |
1209 | &devname)) != EAGAIN) { | |
1210 | process_port_change(p, error, devname); | |
064af421 BP |
1211 | } |
1212 | ||
1213 | if (p->in_band) { | |
d2ede7bc BP |
1214 | if (time_msec() >= p->next_in_band_update) { |
1215 | update_in_band_remotes(p); | |
1216 | } | |
064af421 BP |
1217 | in_band_run(p->in_band); |
1218 | } | |
064af421 | 1219 | |
4e8e4213 | 1220 | LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) { |
3269c562 | 1221 | ofconn_run(ofconn); |
064af421 BP |
1222 | } |
1223 | ||
7778bd15 BP |
1224 | /* Fail-open maintenance. Do this after processing the ofconns since |
1225 | * fail-open checks the status of the controller rconn. */ | |
1226 | if (p->fail_open) { | |
1227 | fail_open_run(p->fail_open); | |
1228 | } | |
1229 | ||
4e8e4213 | 1230 | HMAP_FOR_EACH (ofservice, node, &p->services) { |
064af421 BP |
1231 | struct vconn *vconn; |
1232 | int retval; | |
1233 | ||
7d674866 | 1234 | retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn); |
064af421 | 1235 | if (!retval) { |
9794e806 | 1236 | struct rconn *rconn; |
eb15cdbb | 1237 | char *name; |
9794e806 | 1238 | |
7d674866 | 1239 | rconn = rconn_create(ofservice->probe_interval, 0); |
eb15cdbb BP |
1240 | name = ofconn_make_name(p, vconn_get_name(vconn)); |
1241 | rconn_connect_unreliably(rconn, vconn, name); | |
1242 | free(name); | |
1243 | ||
7d674866 BP |
1244 | ofconn = ofconn_create(p, rconn, OFCONN_SERVICE); |
1245 | ofconn_set_rate_limit(ofconn, ofservice->rate_limit, | |
1246 | ofservice->burst_limit); | |
064af421 BP |
1247 | } else if (retval != EAGAIN) { |
1248 | VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval)); | |
1249 | } | |
1250 | } | |
1251 | ||
1252 | for (i = 0; i < p->n_snoops; i++) { | |
1253 | struct vconn *vconn; | |
1254 | int retval; | |
1255 | ||
1256 | retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn); | |
1257 | if (!retval) { | |
76ce9432 | 1258 | add_snooper(p, vconn); |
064af421 BP |
1259 | } else if (retval != EAGAIN) { |
1260 | VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval)); | |
1261 | } | |
1262 | } | |
1263 | ||
1264 | if (time_msec() >= p->next_expiration) { | |
0de7a4b4 BP |
1265 | int delay = ofproto_expire(p); |
1266 | p->next_expiration = time_msec() + delay; | |
064af421 | 1267 | COVERAGE_INC(ofproto_expiration); |
064af421 BP |
1268 | } |
1269 | ||
1270 | if (p->netflow) { | |
1271 | netflow_run(p->netflow); | |
1272 | } | |
72b06300 BP |
1273 | if (p->sflow) { |
1274 | ofproto_sflow_run(p->sflow); | |
1275 | } | |
064af421 BP |
1276 | |
1277 | return 0; | |
1278 | } | |
1279 | ||
064af421 BP |
1280 | int |
1281 | ofproto_run2(struct ofproto *p, bool revalidate_all) | |
1282 | { | |
bcf84111 BP |
1283 | /* Figure out what we need to revalidate now, if anything. */ |
1284 | struct tag_set revalidate_set = p->revalidate_set; | |
1285 | if (p->need_revalidate) { | |
1286 | revalidate_all = true; | |
1287 | } | |
1288 | ||
1289 | /* Clear the revalidation flags. */ | |
1290 | tag_set_init(&p->revalidate_set); | |
1291 | p->need_revalidate = false; | |
1292 | ||
1293 | /* Now revalidate if there's anything to do. */ | |
1294 | if (revalidate_all || !tag_set_is_empty(&revalidate_set)) { | |
1295 | struct facet *facet, *next; | |
1296 | ||
1297 | HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &p->facets) { | |
1298 | if (revalidate_all | |
1299 | || tag_set_intersects(&revalidate_set, facet->tags)) { | |
1300 | facet_revalidate(p, facet); | |
1301 | } | |
1302 | } | |
064af421 BP |
1303 | } |
1304 | ||
1305 | return 0; | |
1306 | } | |
1307 | ||
1308 | void | |
1309 | ofproto_wait(struct ofproto *p) | |
1310 | { | |
7d674866 | 1311 | struct ofservice *ofservice; |
064af421 BP |
1312 | struct ofconn *ofconn; |
1313 | size_t i; | |
1314 | ||
c228a364 | 1315 | dpif_recv_wait(p->dpif); |
e9e28be3 BP |
1316 | dpif_port_poll_wait(p->dpif); |
1317 | netdev_monitor_poll_wait(p->netdev_monitor); | |
4e8e4213 | 1318 | LIST_FOR_EACH (ofconn, node, &p->all_conns) { |
064af421 BP |
1319 | ofconn_wait(ofconn); |
1320 | } | |
1321 | if (p->in_band) { | |
7cf8b266 | 1322 | poll_timer_wait_until(p->next_in_band_update); |
064af421 BP |
1323 | in_band_wait(p->in_band); |
1324 | } | |
064af421 BP |
1325 | if (p->fail_open) { |
1326 | fail_open_wait(p->fail_open); | |
1327 | } | |
72b06300 BP |
1328 | if (p->sflow) { |
1329 | ofproto_sflow_wait(p->sflow); | |
1330 | } | |
064af421 BP |
1331 | if (!tag_set_is_empty(&p->revalidate_set)) { |
1332 | poll_immediate_wake(); | |
1333 | } | |
1334 | if (p->need_revalidate) { | |
1335 | /* Shouldn't happen, but if it does just go around again. */ | |
1336 | VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); | |
1337 | poll_immediate_wake(); | |
1338 | } else if (p->next_expiration != LLONG_MAX) { | |
7cf8b266 | 1339 | poll_timer_wait_until(p->next_expiration); |
064af421 | 1340 | } |
4e8e4213 | 1341 | HMAP_FOR_EACH (ofservice, node, &p->services) { |
7d674866 | 1342 | pvconn_wait(ofservice->pvconn); |
064af421 BP |
1343 | } |
1344 | for (i = 0; i < p->n_snoops; i++) { | |
1345 | pvconn_wait(p->snoops[i]); | |
1346 | } | |
1347 | } | |
1348 | ||
1349 | void | |
1350 | ofproto_revalidate(struct ofproto *ofproto, tag_type tag) | |
1351 | { | |
1352 | tag_set_add(&ofproto->revalidate_set, tag); | |
1353 | } | |
1354 | ||
1355 | struct tag_set * | |
1356 | ofproto_get_revalidate_set(struct ofproto *ofproto) | |
1357 | { | |
1358 | return &ofproto->revalidate_set; | |
1359 | } | |
1360 | ||
1361 | bool | |
1362 | ofproto_is_alive(const struct ofproto *p) | |
1363 | { | |
76ce9432 | 1364 | return !hmap_is_empty(&p->controllers); |
064af421 BP |
1365 | } |
1366 | ||
bffc0589 | 1367 | void |
2cdcb898 | 1368 | ofproto_get_ofproto_controller_info(const struct ofproto *ofproto, |
bffc0589 AE |
1369 | struct shash *info) |
1370 | { | |
1371 | const struct ofconn *ofconn; | |
1372 | ||
1373 | shash_init(info); | |
1374 | ||
1375 | HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) { | |
1376 | const struct rconn *rconn = ofconn->rconn; | |
2cdcb898 AE |
1377 | time_t now = time_now(); |
1378 | time_t last_connection = rconn_get_last_connection(rconn); | |
1379 | time_t last_disconnect = rconn_get_last_disconnect(rconn); | |
bffc0589 AE |
1380 | const int last_error = rconn_get_last_error(rconn); |
1381 | struct ofproto_controller_info *cinfo = xmalloc(sizeof *cinfo); | |
1382 | ||
1383 | shash_add(info, rconn_get_target(rconn), cinfo); | |
1384 | ||
1385 | cinfo->is_connected = rconn_is_connected(rconn); | |
1386 | cinfo->role = ofconn->role; | |
1387 | ||
1388 | cinfo->pairs.n = 0; | |
1389 | ||
c18ea70d | 1390 | if (last_error) { |
bffc0589 AE |
1391 | cinfo->pairs.keys[cinfo->pairs.n] = "last_error"; |
1392 | cinfo->pairs.values[cinfo->pairs.n++] = | |
c18ea70d | 1393 | xstrdup(ovs_retval_to_string(last_error)); |
bffc0589 AE |
1394 | } |
1395 | ||
1396 | cinfo->pairs.keys[cinfo->pairs.n] = "state"; | |
1397 | cinfo->pairs.values[cinfo->pairs.n++] = | |
1398 | xstrdup(rconn_get_state(rconn)); | |
1399 | ||
2cdcb898 AE |
1400 | if (last_connection != TIME_MIN) { |
1401 | cinfo->pairs.keys[cinfo->pairs.n] = "sec_since_connect"; | |
1402 | cinfo->pairs.values[cinfo->pairs.n++] | |
1403 | = xasprintf("%ld", (long int) (now - last_connection)); | |
1404 | } | |
1405 | ||
1406 | if (last_disconnect != TIME_MIN) { | |
1407 | cinfo->pairs.keys[cinfo->pairs.n] = "sec_since_disconnect"; | |
1408 | cinfo->pairs.values[cinfo->pairs.n++] | |
1409 | = xasprintf("%ld", (long int) (now - last_disconnect)); | |
24190e86 | 1410 | } |
bffc0589 AE |
1411 | } |
1412 | } | |
1413 | ||
1414 | void | |
1415 | ofproto_free_ofproto_controller_info(struct shash *info) | |
1416 | { | |
1417 | struct shash_node *node; | |
1418 | ||
1419 | SHASH_FOR_EACH (node, info) { | |
1420 | struct ofproto_controller_info *cinfo = node->data; | |
1421 | while (cinfo->pairs.n) { | |
1422 | free((char *) cinfo->pairs.values[--cinfo->pairs.n]); | |
1423 | } | |
1424 | free(cinfo); | |
1425 | } | |
1426 | shash_destroy(info); | |
1427 | } | |
1428 | ||
3a6ccc8c BP |
1429 | /* Deletes port number 'odp_port' from the datapath for 'ofproto'. |
1430 | * | |
1431 | * This is almost the same as calling dpif_port_del() directly on the | |
1432 | * datapath, but it also makes 'ofproto' close its open netdev for the port | |
1433 | * (if any). This makes it possible to create a new netdev of a different | |
1434 | * type under the same name, which otherwise the netdev library would refuse | |
1435 | * to do because of the conflict. (The netdev would eventually get closed on | |
1436 | * the next trip through ofproto_run(), but this interface is more direct.) | |
1437 | * | |
3a6ccc8c BP |
1438 | * Returns 0 if successful, otherwise a positive errno. */ |
1439 | int | |
1440 | ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port) | |
1441 | { | |
1442 | struct ofport *ofport = get_port(ofproto, odp_port); | |
0b61210e | 1443 | const char *name = ofport ? ofport->opp.name : "<unknown>"; |
3a6ccc8c BP |
1444 | int error; |
1445 | ||
1446 | error = dpif_port_del(ofproto->dpif, odp_port); | |
1447 | if (error) { | |
1448 | VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)", | |
1449 | dpif_name(ofproto->dpif), odp_port, name, strerror(error)); | |
1450 | } else if (ofport) { | |
1451 | /* 'name' is ofport->opp.name and update_port() is going to destroy | |
1452 | * 'ofport'. Just in case update_port() refers to 'name' after it | |
1453 | * destroys 'ofport', make a copy of it around the update_port() | |
1454 | * call. */ | |
1455 | char *devname = xstrdup(name); | |
1456 | update_port(ofproto, devname); | |
1457 | free(devname); | |
1458 | } | |
1459 | return error; | |
1460 | } | |
1461 | ||
a4e2e1f2 EJ |
1462 | /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns |
1463 | * true if 'odp_port' exists and should be included, false otherwise. */ | |
1464 | bool | |
1465 | ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port) | |
1466 | { | |
1467 | struct ofport *ofport = get_port(ofproto, odp_port); | |
1468 | return ofport && !(ofport->opp.config & OFPPC_NO_FLOOD); | |
1469 | } | |
1470 | ||
064af421 | 1471 | int |
ae412e7d | 1472 | ofproto_send_packet(struct ofproto *p, const struct flow *flow, |
064af421 BP |
1473 | const union ofp_action *actions, size_t n_actions, |
1474 | const struct ofpbuf *packet) | |
1475 | { | |
f29152ca | 1476 | struct action_xlate_ctx ctx; |
cdee00fd | 1477 | struct ofpbuf *odp_actions; |
064af421 | 1478 | |
f29152ca | 1479 | action_xlate_ctx_init(&ctx, p, flow, packet); |
ebe482fd EJ |
1480 | /* Always xlate packets originated in this function. */ |
1481 | ctx.check_special = false; | |
cdee00fd | 1482 | odp_actions = xlate_actions(&ctx, actions, n_actions); |
064af421 BP |
1483 | |
1484 | /* XXX Should we translate the dpif_execute() errno value into an OpenFlow | |
1485 | * error code? */ | |
cdee00fd BP |
1486 | dpif_execute(p->dpif, odp_actions->data, odp_actions->size, packet); |
1487 | ||
1488 | ofpbuf_delete(odp_actions); | |
1489 | ||
064af421 BP |
1490 | return 0; |
1491 | } | |
1492 | ||
fa8b054f BP |
1493 | /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and |
1494 | * performs the 'n_actions' actions in 'actions'. The new flow will not | |
1495 | * timeout. | |
1496 | * | |
1497 | * If cls_rule->priority is in the range of priorities supported by OpenFlow | |
1498 | * (0...65535, inclusive) then the flow will be visible to OpenFlow | |
1499 | * controllers; otherwise, it will be hidden. | |
1500 | * | |
1501 | * The caller retains ownership of 'cls_rule' and 'actions'. */ | |
064af421 | 1502 | void |
cf3fad8a | 1503 | ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule, |
fa8b054f | 1504 | const union ofp_action *actions, size_t n_actions) |
064af421 BP |
1505 | { |
1506 | struct rule *rule; | |
bcf84111 | 1507 | rule = rule_create(cls_rule, actions, n_actions, 0, 0, 0, false); |
afe75089 | 1508 | rule_insert(p, rule); |
064af421 BP |
1509 | } |
1510 | ||
1511 | void | |
cf3fad8a | 1512 | ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target) |
064af421 BP |
1513 | { |
1514 | struct rule *rule; | |
1515 | ||
1516 | rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls, | |
cf3fad8a | 1517 | target)); |
064af421 BP |
1518 | if (rule) { |
1519 | rule_remove(ofproto, rule); | |
1520 | } | |
1521 | } | |
1522 | ||
064af421 BP |
1523 | void |
1524 | ofproto_flush_flows(struct ofproto *ofproto) | |
1525 | { | |
bcf84111 | 1526 | struct facet *facet, *next_facet; |
5ecc9d81 BP |
1527 | struct rule *rule, *next_rule; |
1528 | struct cls_cursor cursor; | |
bcf84111 | 1529 | |
064af421 | 1530 | COVERAGE_INC(ofproto_flush); |
bcf84111 BP |
1531 | |
1532 | HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) { | |
1533 | /* Mark the facet as not installed so that facet_remove() doesn't | |
1534 | * bother trying to uninstall it. There is no point in uninstalling it | |
1535 | * individually since we are about to blow away all the facets with | |
1536 | * dpif_flow_flush(). */ | |
1537 | facet->installed = false; | |
a2c6a63c EJ |
1538 | facet->dp_packet_count = 0; |
1539 | facet->dp_byte_count = 0; | |
bcf84111 BP |
1540 | facet_remove(ofproto, facet); |
1541 | } | |
5ecc9d81 BP |
1542 | |
1543 | cls_cursor_init(&cursor, &ofproto->cls, NULL); | |
1544 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) { | |
1545 | rule_remove(ofproto, rule); | |
1546 | } | |
1547 | ||
c228a364 | 1548 | dpif_flow_flush(ofproto->dpif); |
064af421 BP |
1549 | if (ofproto->in_band) { |
1550 | in_band_flushed(ofproto->in_band); | |
1551 | } | |
1552 | if (ofproto->fail_open) { | |
1553 | fail_open_flushed(ofproto->fail_open); | |
1554 | } | |
1555 | } | |
1556 | \f | |
1557 | static void | |
1558 | reinit_ports(struct ofproto *p) | |
1559 | { | |
b0ec0f27 | 1560 | struct dpif_port_dump dump; |
c77d9d13 BP |
1561 | struct shash_node *node; |
1562 | struct shash devnames; | |
064af421 | 1563 | struct ofport *ofport; |
4c738a8d | 1564 | struct dpif_port dpif_port; |
064af421 | 1565 | |
898bf89d JP |
1566 | COVERAGE_INC(ofproto_reinit_ports); |
1567 | ||
c77d9d13 | 1568 | shash_init(&devnames); |
4e8e4213 | 1569 | HMAP_FOR_EACH (ofport, hmap_node, &p->ports) { |
c77d9d13 | 1570 | shash_add_once (&devnames, ofport->opp.name, NULL); |
064af421 | 1571 | } |
4c738a8d BP |
1572 | DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) { |
1573 | shash_add_once (&devnames, dpif_port.name, NULL); | |
064af421 | 1574 | } |
064af421 | 1575 | |
c77d9d13 BP |
1576 | SHASH_FOR_EACH (node, &devnames) { |
1577 | update_port(p, node->name); | |
064af421 | 1578 | } |
c77d9d13 | 1579 | shash_destroy(&devnames); |
064af421 BP |
1580 | } |
1581 | ||
064af421 | 1582 | static struct ofport * |
4c738a8d | 1583 | make_ofport(const struct dpif_port *dpif_port) |
064af421 | 1584 | { |
149f577a | 1585 | struct netdev_options netdev_options; |
064af421 BP |
1586 | enum netdev_flags flags; |
1587 | struct ofport *ofport; | |
1588 | struct netdev *netdev; | |
064af421 BP |
1589 | int error; |
1590 | ||
149f577a | 1591 | memset(&netdev_options, 0, sizeof netdev_options); |
4c738a8d BP |
1592 | netdev_options.name = dpif_port->name; |
1593 | netdev_options.type = dpif_port->type; | |
149f577a | 1594 | netdev_options.ethertype = NETDEV_ETH_TYPE_NONE; |
149f577a JG |
1595 | |
1596 | error = netdev_open(&netdev_options, &netdev); | |
064af421 BP |
1597 | if (error) { |
1598 | VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s " | |
1599 | "cannot be opened (%s)", | |
4c738a8d BP |
1600 | dpif_port->name, dpif_port->port_no, |
1601 | dpif_port->name, strerror(error)); | |
064af421 BP |
1602 | return NULL; |
1603 | } | |
1604 | ||
0a6f5542 | 1605 | ofport = xzalloc(sizeof *ofport); |
064af421 | 1606 | ofport->netdev = netdev; |
4c738a8d BP |
1607 | ofport->odp_port = dpif_port->port_no; |
1608 | ofport->opp.port_no = odp_port_to_ofp_port(dpif_port->port_no); | |
80992a35 | 1609 | netdev_get_etheraddr(netdev, ofport->opp.hw_addr); |
4c738a8d | 1610 | ovs_strlcpy(ofport->opp.name, dpif_port->name, sizeof ofport->opp.name); |
064af421 BP |
1611 | |
1612 | netdev_get_flags(netdev, &flags); | |
1613 | ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN; | |
1614 | ||
85da620e | 1615 | ofport->opp.state = netdev_get_carrier(netdev) ? 0 : OFPPS_LINK_DOWN; |
064af421 BP |
1616 | |
1617 | netdev_get_features(netdev, | |
1618 | &ofport->opp.curr, &ofport->opp.advertised, | |
1619 | &ofport->opp.supported, &ofport->opp.peer); | |
1620 | return ofport; | |
1621 | } | |
1622 | ||
1623 | static bool | |
4c738a8d | 1624 | ofport_conflicts(const struct ofproto *p, const struct dpif_port *dpif_port) |
064af421 | 1625 | { |
4c738a8d | 1626 | if (get_port(p, dpif_port->port_no)) { |
064af421 | 1627 | VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath", |
4c738a8d | 1628 | dpif_port->port_no); |
064af421 | 1629 | return true; |
4c738a8d | 1630 | } else if (shash_find(&p->port_by_name, dpif_port->name)) { |
064af421 | 1631 | VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath", |
4c738a8d | 1632 | dpif_port->name); |
064af421 BP |
1633 | return true; |
1634 | } else { | |
1635 | return false; | |
1636 | } | |
1637 | } | |
1638 | ||
1639 | static int | |
1640 | ofport_equal(const struct ofport *a_, const struct ofport *b_) | |
1641 | { | |
1642 | const struct ofp_phy_port *a = &a_->opp; | |
1643 | const struct ofp_phy_port *b = &b_->opp; | |
1644 | ||
1645 | BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */ | |
1646 | return (a->port_no == b->port_no | |
1647 | && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr) | |
0b61210e | 1648 | && !strcmp(a->name, b->name) |
064af421 BP |
1649 | && a->state == b->state |
1650 | && a->config == b->config | |
1651 | && a->curr == b->curr | |
1652 | && a->advertised == b->advertised | |
1653 | && a->supported == b->supported | |
1654 | && a->peer == b->peer); | |
1655 | } | |
1656 | ||
1657 | static void | |
1658 | send_port_status(struct ofproto *p, const struct ofport *ofport, | |
1659 | uint8_t reason) | |
1660 | { | |
1661 | /* XXX Should limit the number of queued port status change messages. */ | |
1662 | struct ofconn *ofconn; | |
4e8e4213 | 1663 | LIST_FOR_EACH (ofconn, node, &p->all_conns) { |
064af421 BP |
1664 | struct ofp_port_status *ops; |
1665 | struct ofpbuf *b; | |
1666 | ||
197a992f BP |
1667 | /* Primary controllers, even slaves, should always get port status |
1668 | updates. Otherwise obey ofconn_receives_async_msgs(). */ | |
1669 | if (ofconn->type != OFCONN_PRIMARY | |
1670 | && !ofconn_receives_async_msgs(ofconn)) { | |
9deba63b BP |
1671 | continue; |
1672 | } | |
1673 | ||
064af421 BP |
1674 | ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b); |
1675 | ops->reason = reason; | |
1676 | ops->desc = ofport->opp; | |
1677 | hton_ofp_phy_port(&ops->desc); | |
1678 | queue_tx(b, ofconn, NULL); | |
1679 | } | |
064af421 BP |
1680 | } |
1681 | ||
1682 | static void | |
1683 | ofport_install(struct ofproto *p, struct ofport *ofport) | |
1684 | { | |
0b61210e | 1685 | const char *netdev_name = ofport->opp.name; |
72b06300 | 1686 | |
e9e28be3 | 1687 | netdev_monitor_add(p->netdev_monitor, ofport->netdev); |
ca0f572c | 1688 | hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0)); |
72b06300 BP |
1689 | shash_add(&p->port_by_name, netdev_name, ofport); |
1690 | if (p->sflow) { | |
ca0f572c | 1691 | ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name); |
72b06300 | 1692 | } |
064af421 BP |
1693 | } |
1694 | ||
1695 | static void | |
1696 | ofport_remove(struct ofproto *p, struct ofport *ofport) | |
1697 | { | |
e9e28be3 | 1698 | netdev_monitor_remove(p->netdev_monitor, ofport->netdev); |
ca0f572c | 1699 | hmap_remove(&p->ports, &ofport->hmap_node); |
064af421 | 1700 | shash_delete(&p->port_by_name, |
0b61210e | 1701 | shash_find(&p->port_by_name, ofport->opp.name)); |
72b06300 | 1702 | if (p->sflow) { |
ca0f572c | 1703 | ofproto_sflow_del_port(p->sflow, ofport->odp_port); |
72b06300 | 1704 | } |
064af421 BP |
1705 | } |
1706 | ||
1707 | static void | |
1708 | ofport_free(struct ofport *ofport) | |
1709 | { | |
1710 | if (ofport) { | |
1711 | netdev_close(ofport->netdev); | |
1712 | free(ofport); | |
1713 | } | |
1714 | } | |
1715 | ||
ca0f572c BP |
1716 | static struct ofport * |
1717 | get_port(const struct ofproto *ofproto, uint16_t odp_port) | |
1718 | { | |
1719 | struct ofport *port; | |
1720 | ||
4e8e4213 | 1721 | HMAP_FOR_EACH_IN_BUCKET (port, hmap_node, |
ca0f572c BP |
1722 | hash_int(odp_port, 0), &ofproto->ports) { |
1723 | if (port->odp_port == odp_port) { | |
1724 | return port; | |
1725 | } | |
1726 | } | |
1727 | return NULL; | |
1728 | } | |
1729 | ||
064af421 BP |
1730 | static void |
1731 | update_port(struct ofproto *p, const char *devname) | |
1732 | { | |
4c738a8d | 1733 | struct dpif_port dpif_port; |
c874dc6d BP |
1734 | struct ofport *old_ofport; |
1735 | struct ofport *new_ofport; | |
064af421 BP |
1736 | int error; |
1737 | ||
1738 | COVERAGE_INC(ofproto_update_port); | |
c874dc6d BP |
1739 | |
1740 | /* Query the datapath for port information. */ | |
4c738a8d | 1741 | error = dpif_port_query_by_name(p->dpif, devname, &dpif_port); |
064af421 | 1742 | |
c874dc6d BP |
1743 | /* Find the old ofport. */ |
1744 | old_ofport = shash_find_data(&p->port_by_name, devname); | |
1745 | if (!error) { | |
1746 | if (!old_ofport) { | |
1747 | /* There's no port named 'devname' but there might be a port with | |
1748 | * the same port number. This could happen if a port is deleted | |
1749 | * and then a new one added in its place very quickly, or if a port | |
1750 | * is renamed. In the former case we want to send an OFPPR_DELETE | |
1751 | * and an OFPPR_ADD, and in the latter case we want to send a | |
1752 | * single OFPPR_MODIFY. We can distinguish the cases by comparing | |
1753 | * the old port's ifindex against the new port, or perhaps less | |
1754 | * reliably but more portably by comparing the old port's MAC | |
1755 | * against the new port's MAC. However, this code isn't that smart | |
1756 | * and always sends an OFPPR_MODIFY (XXX). */ | |
4c738a8d | 1757 | old_ofport = get_port(p, dpif_port.port_no); |
064af421 | 1758 | } |
c874dc6d | 1759 | } else if (error != ENOENT && error != ENODEV) { |
064af421 BP |
1760 | VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error " |
1761 | "%s", strerror(error)); | |
4c738a8d | 1762 | goto exit; |
064af421 | 1763 | } |
c874dc6d BP |
1764 | |
1765 | /* Create a new ofport. */ | |
4c738a8d | 1766 | new_ofport = !error ? make_ofport(&dpif_port) : NULL; |
c874dc6d BP |
1767 | |
1768 | /* Eliminate a few pathological cases. */ | |
1769 | if (!old_ofport && !new_ofport) { | |
4c738a8d | 1770 | goto exit; |
c874dc6d BP |
1771 | } else if (old_ofport && new_ofport) { |
1772 | /* Most of the 'config' bits are OpenFlow soft state, but | |
bc4a55cd BP |
1773 | * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the |
1774 | * OpenFlow bits from old_ofport. (make_ofport() only sets | |
1775 | * OFPPC_PORT_DOWN and leaves the other bits 0.) */ | |
c874dc6d BP |
1776 | new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN; |
1777 | ||
1778 | if (ofport_equal(old_ofport, new_ofport)) { | |
1779 | /* False alarm--no change. */ | |
1780 | ofport_free(new_ofport); | |
4c738a8d | 1781 | goto exit; |
c874dc6d BP |
1782 | } |
1783 | } | |
1784 | ||
1785 | /* Now deal with the normal cases. */ | |
1786 | if (old_ofport) { | |
1787 | ofport_remove(p, old_ofport); | |
1788 | } | |
1789 | if (new_ofport) { | |
1790 | ofport_install(p, new_ofport); | |
1791 | } | |
1792 | send_port_status(p, new_ofport ? new_ofport : old_ofport, | |
1793 | (!old_ofport ? OFPPR_ADD | |
1794 | : !new_ofport ? OFPPR_DELETE | |
1795 | : OFPPR_MODIFY)); | |
1796 | ofport_free(old_ofport); | |
4c738a8d BP |
1797 | |
1798 | exit: | |
1799 | dpif_port_destroy(&dpif_port); | |
064af421 BP |
1800 | } |
1801 | ||
1802 | static int | |
1803 | init_ports(struct ofproto *p) | |
1804 | { | |
b0ec0f27 | 1805 | struct dpif_port_dump dump; |
4c738a8d | 1806 | struct dpif_port dpif_port; |
064af421 | 1807 | |
4c738a8d BP |
1808 | DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) { |
1809 | if (!ofport_conflicts(p, &dpif_port)) { | |
1810 | struct ofport *ofport = make_ofport(&dpif_port); | |
064af421 BP |
1811 | if (ofport) { |
1812 | ofport_install(p, ofport); | |
1813 | } | |
1814 | } | |
1815 | } | |
b0ec0f27 | 1816 | |
064af421 BP |
1817 | return 0; |
1818 | } | |
1819 | \f | |
1820 | static struct ofconn * | |
76ce9432 | 1821 | ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type) |
064af421 | 1822 | { |
76ce9432 BP |
1823 | struct ofconn *ofconn = xzalloc(sizeof *ofconn); |
1824 | ofconn->ofproto = p; | |
064af421 BP |
1825 | list_push_back(&p->all_conns, &ofconn->node); |
1826 | ofconn->rconn = rconn; | |
76ce9432 | 1827 | ofconn->type = type; |
b70eac89 | 1828 | ofconn->flow_format = NXFF_OPENFLOW10; |
9deba63b | 1829 | ofconn->role = NX_ROLE_OTHER; |
76ce9432 | 1830 | ofconn->packet_in_counter = rconn_packet_counter_create (); |
064af421 | 1831 | ofconn->pktbuf = NULL; |
064af421 | 1832 | ofconn->miss_send_len = 0; |
064af421 BP |
1833 | ofconn->reply_counter = rconn_packet_counter_create (); |
1834 | return ofconn; | |
1835 | } | |
1836 | ||
1837 | static void | |
c475ae67 | 1838 | ofconn_destroy(struct ofconn *ofconn) |
064af421 | 1839 | { |
5899143f | 1840 | if (ofconn->type == OFCONN_PRIMARY) { |
76ce9432 BP |
1841 | hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node); |
1842 | } | |
1843 | discovery_destroy(ofconn->discovery); | |
1844 | ||
064af421 | 1845 | list_remove(&ofconn->node); |
76ce9432 | 1846 | switch_status_unregister(ofconn->ss); |
064af421 BP |
1847 | rconn_destroy(ofconn->rconn); |
1848 | rconn_packet_counter_destroy(ofconn->packet_in_counter); | |
1849 | rconn_packet_counter_destroy(ofconn->reply_counter); | |
1850 | pktbuf_destroy(ofconn->pktbuf); | |
1851 | free(ofconn); | |
1852 | } | |
1853 | ||
1854 | static void | |
3269c562 | 1855 | ofconn_run(struct ofconn *ofconn) |
064af421 | 1856 | { |
3269c562 | 1857 | struct ofproto *p = ofconn->ofproto; |
064af421 | 1858 | int iteration; |
76ce9432 BP |
1859 | size_t i; |
1860 | ||
1861 | if (ofconn->discovery) { | |
1862 | char *controller_name; | |
1863 | if (rconn_is_connectivity_questionable(ofconn->rconn)) { | |
1864 | discovery_question_connectivity(ofconn->discovery); | |
1865 | } | |
1866 | if (discovery_run(ofconn->discovery, &controller_name)) { | |
1867 | if (controller_name) { | |
eb15cdbb BP |
1868 | char *ofconn_name = ofconn_make_name(p, controller_name); |
1869 | rconn_connect(ofconn->rconn, controller_name, ofconn_name); | |
1870 | free(ofconn_name); | |
eaf9d084 | 1871 | free(controller_name); |
76ce9432 BP |
1872 | } else { |
1873 | rconn_disconnect(ofconn->rconn); | |
1874 | } | |
1875 | } | |
1876 | } | |
1877 | ||
1878 | for (i = 0; i < N_SCHEDULERS; i++) { | |
1879 | pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn); | |
1880 | } | |
064af421 BP |
1881 | |
1882 | rconn_run(ofconn->rconn); | |
1883 | ||
1884 | if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) { | |
1885 | /* Limit the number of iterations to prevent other tasks from | |
1886 | * starving. */ | |
1887 | for (iteration = 0; iteration < 50; iteration++) { | |
1888 | struct ofpbuf *of_msg = rconn_recv(ofconn->rconn); | |
1889 | if (!of_msg) { | |
1890 | break; | |
1891 | } | |
7778bd15 BP |
1892 | if (p->fail_open) { |
1893 | fail_open_maybe_recover(p->fail_open); | |
1894 | } | |
3269c562 | 1895 | handle_openflow(ofconn, of_msg); |
064af421 BP |
1896 | ofpbuf_delete(of_msg); |
1897 | } | |
1898 | } | |
1899 | ||
76ce9432 | 1900 | if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) { |
c475ae67 | 1901 | ofconn_destroy(ofconn); |
064af421 BP |
1902 | } |
1903 | } | |
1904 | ||
1905 | static void | |
1906 | ofconn_wait(struct ofconn *ofconn) | |
1907 | { | |
76ce9432 BP |
1908 | int i; |
1909 | ||
1910 | if (ofconn->discovery) { | |
1911 | discovery_wait(ofconn->discovery); | |
1912 | } | |
1913 | for (i = 0; i < N_SCHEDULERS; i++) { | |
1914 | pinsched_wait(ofconn->schedulers[i]); | |
1915 | } | |
064af421 BP |
1916 | rconn_run_wait(ofconn->rconn); |
1917 | if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) { | |
1918 | rconn_recv_wait(ofconn->rconn); | |
1919 | } else { | |
1920 | COVERAGE_INC(ofproto_ofconn_stuck); | |
1921 | } | |
1922 | } | |
c91248b3 BP |
1923 | |
1924 | /* Returns true if 'ofconn' should receive asynchronous messages. */ | |
1925 | static bool | |
1926 | ofconn_receives_async_msgs(const struct ofconn *ofconn) | |
1927 | { | |
5899143f BP |
1928 | if (ofconn->type == OFCONN_PRIMARY) { |
1929 | /* Primary controllers always get asynchronous messages unless they | |
c91248b3 BP |
1930 | * have configured themselves as "slaves". */ |
1931 | return ofconn->role != NX_ROLE_SLAVE; | |
1932 | } else { | |
5899143f BP |
1933 | /* Service connections don't get asynchronous messages unless they have |
1934 | * explicitly asked for them by setting a nonzero miss send length. */ | |
c91248b3 BP |
1935 | return ofconn->miss_send_len > 0; |
1936 | } | |
1937 | } | |
eb15cdbb BP |
1938 | |
1939 | /* Returns a human-readable name for an OpenFlow connection between 'ofproto' | |
1940 | * and 'target', suitable for use in log messages for identifying the | |
1941 | * connection. | |
1942 | * | |
1943 | * The name is dynamically allocated. The caller should free it (with free()) | |
1944 | * when it is no longer needed. */ | |
1945 | static char * | |
1946 | ofconn_make_name(const struct ofproto *ofproto, const char *target) | |
1947 | { | |
1948 | return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target); | |
1949 | } | |
7d674866 BP |
1950 | |
1951 | static void | |
1952 | ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst) | |
1953 | { | |
1954 | int i; | |
1955 | ||
1956 | for (i = 0; i < N_SCHEDULERS; i++) { | |
1957 | struct pinsched **s = &ofconn->schedulers[i]; | |
1958 | ||
1959 | if (rate > 0) { | |
1960 | if (!*s) { | |
1961 | *s = pinsched_create(rate, burst, | |
1962 | ofconn->ofproto->switch_status); | |
1963 | } else { | |
1964 | pinsched_set_limits(*s, rate, burst); | |
1965 | } | |
1966 | } else { | |
1967 | pinsched_destroy(*s); | |
1968 | *s = NULL; | |
1969 | } | |
1970 | } | |
1971 | } | |
1972 | \f | |
1973 | static void | |
1974 | ofservice_reconfigure(struct ofservice *ofservice, | |
1975 | const struct ofproto_controller *c) | |
1976 | { | |
1977 | ofservice->probe_interval = c->probe_interval; | |
1978 | ofservice->rate_limit = c->rate_limit; | |
1979 | ofservice->burst_limit = c->burst_limit; | |
1980 | } | |
1981 | ||
1982 | /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a | |
1983 | * positive errno value. */ | |
1984 | static int | |
1985 | ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c) | |
1986 | { | |
1987 | struct ofservice *ofservice; | |
1988 | struct pvconn *pvconn; | |
1989 | int error; | |
1990 | ||
1991 | error = pvconn_open(c->target, &pvconn); | |
1992 | if (error) { | |
1993 | return error; | |
1994 | } | |
1995 | ||
1996 | ofservice = xzalloc(sizeof *ofservice); | |
1997 | hmap_insert(&ofproto->services, &ofservice->node, | |
1998 | hash_string(c->target, 0)); | |
1999 | ofservice->pvconn = pvconn; | |
2000 | ||
2001 | ofservice_reconfigure(ofservice, c); | |
2002 | ||
2003 | return 0; | |
2004 | } | |
2005 | ||
2006 | static void | |
2007 | ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice) | |
2008 | { | |
2009 | hmap_remove(&ofproto->services, &ofservice->node); | |
2010 | pvconn_close(ofservice->pvconn); | |
2011 | free(ofservice); | |
2012 | } | |
2013 | ||
2014 | /* Finds and returns the ofservice within 'ofproto' that has the given | |
2015 | * 'target', or a null pointer if none exists. */ | |
2016 | static struct ofservice * | |
2017 | ofservice_lookup(struct ofproto *ofproto, const char *target) | |
2018 | { | |
2019 | struct ofservice *ofservice; | |
2020 | ||
4e8e4213 BP |
2021 | HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0), |
2022 | &ofproto->services) { | |
7d674866 BP |
2023 | if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) { |
2024 | return ofservice; | |
2025 | } | |
2026 | } | |
2027 | return NULL; | |
2028 | } | |
064af421 | 2029 | \f |
bcf84111 BP |
2030 | /* Returns true if 'rule' should be hidden from the controller. |
2031 | * | |
2032 | * Rules with priority higher than UINT16_MAX are set up by ofproto itself | |
2033 | * (e.g. by in-band control) and are intentionally hidden from the | |
2034 | * controller. */ | |
2035 | static bool | |
2036 | rule_is_hidden(const struct rule *rule) | |
2037 | { | |
2038 | return rule->cr.priority > UINT16_MAX; | |
2039 | } | |
2040 | ||
2041 | /* Creates and returns a new rule initialized as specified. | |
2042 | * | |
2043 | * The caller is responsible for inserting the rule into the classifier (with | |
2044 | * rule_insert()). */ | |
064af421 | 2045 | static struct rule * |
bcf84111 | 2046 | rule_create(const struct cls_rule *cls_rule, |
064af421 | 2047 | const union ofp_action *actions, size_t n_actions, |
ca069229 | 2048 | uint16_t idle_timeout, uint16_t hard_timeout, |
8054fc48 | 2049 | ovs_be64 flow_cookie, bool send_flow_removed) |
064af421 | 2050 | { |
ec6fde61 | 2051 | struct rule *rule = xzalloc(sizeof *rule); |
bcf84111 | 2052 | rule->cr = *cls_rule; |
064af421 BP |
2053 | rule->idle_timeout = idle_timeout; |
2054 | rule->hard_timeout = hard_timeout; | |
39997502 | 2055 | rule->flow_cookie = flow_cookie; |
064af421 | 2056 | rule->used = rule->created = time_msec(); |
ca069229 | 2057 | rule->send_flow_removed = send_flow_removed; |
bcf84111 | 2058 | list_init(&rule->facets); |
3dffcf07 BP |
2059 | if (n_actions > 0) { |
2060 | rule->n_actions = n_actions; | |
2061 | rule->actions = xmemdup(actions, n_actions * sizeof *actions); | |
2062 | } | |
0193b2af | 2063 | |
064af421 BP |
2064 | return rule; |
2065 | } | |
2066 | ||
2067 | static struct rule * | |
2068 | rule_from_cls_rule(const struct cls_rule *cls_rule) | |
2069 | { | |
2070 | return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL; | |
2071 | } | |
2072 | ||
2073 | static void | |
2074 | rule_free(struct rule *rule) | |
2075 | { | |
2076 | free(rule->actions); | |
064af421 BP |
2077 | free(rule); |
2078 | } | |
2079 | ||
bcf84111 BP |
2080 | /* Destroys 'rule' and iterates through all of its facets and revalidates them, |
2081 | * destroying any that no longer has a rule (which is probably all of them). | |
064af421 | 2082 | * |
bcf84111 | 2083 | * The caller must have already removed 'rule' from the classifier. */ |
064af421 BP |
2084 | static void |
2085 | rule_destroy(struct ofproto *ofproto, struct rule *rule) | |
2086 | { | |
bcf84111 BP |
2087 | struct facet *facet, *next_facet; |
2088 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { | |
2089 | facet_revalidate(ofproto, facet); | |
064af421 BP |
2090 | } |
2091 | rule_free(rule); | |
2092 | } | |
2093 | ||
bcf84111 BP |
2094 | /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action |
2095 | * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't | |
2096 | * count). */ | |
064af421 | 2097 | static bool |
8054fc48 | 2098 | rule_has_out_port(const struct rule *rule, ovs_be16 out_port) |
064af421 BP |
2099 | { |
2100 | const union ofp_action *oa; | |
2101 | struct actions_iterator i; | |
2102 | ||
2103 | if (out_port == htons(OFPP_NONE)) { | |
2104 | return true; | |
2105 | } | |
2106 | for (oa = actions_first(&i, rule->actions, rule->n_actions); oa; | |
2107 | oa = actions_next(&i)) { | |
c1c9c9c4 | 2108 | if (action_outputs_to_port(oa, out_port)) { |
064af421 BP |
2109 | return true; |
2110 | } | |
2111 | } | |
2112 | return false; | |
2113 | } | |
2114 | ||
750638bb BP |
2115 | /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on |
2116 | * 'packet', which arrived on 'in_port'. | |
2117 | * | |
2118 | * Takes ownership of 'packet'. */ | |
9dbb9d5e | 2119 | static bool |
856081f6 | 2120 | execute_odp_actions(struct ofproto *ofproto, const struct flow *flow, |
cf22f8cb | 2121 | const struct nlattr *odp_actions, size_t actions_len, |
750638bb | 2122 | struct ofpbuf *packet) |
9dbb9d5e | 2123 | { |
b9298d3f | 2124 | if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t)) |
7aec165d | 2125 | && odp_actions->nla_type == ODP_ACTION_ATTR_CONTROLLER) { |
9dbb9d5e BP |
2126 | /* As an optimization, avoid a round-trip from userspace to kernel to |
2127 | * userspace. This also avoids possibly filling up kernel packet | |
2128 | * buffers along the way. */ | |
856081f6 | 2129 | struct dpif_upcall upcall; |
9dbb9d5e | 2130 | |
82272ede | 2131 | upcall.type = DPIF_UC_ACTION; |
856081f6 BP |
2132 | upcall.packet = packet; |
2133 | upcall.key = NULL; | |
2134 | upcall.key_len = 0; | |
2135 | upcall.userdata = nl_attr_get_u64(odp_actions); | |
2136 | upcall.sample_pool = 0; | |
2137 | upcall.actions = NULL; | |
2138 | upcall.actions_len = 0; | |
9dbb9d5e | 2139 | |
856081f6 | 2140 | send_packet_in(ofproto, &upcall, flow, false); |
9dbb9d5e | 2141 | |
750638bb BP |
2142 | return true; |
2143 | } else { | |
2144 | int error; | |
9dbb9d5e | 2145 | |
cdee00fd | 2146 | error = dpif_execute(ofproto->dpif, odp_actions, actions_len, packet); |
750638bb BP |
2147 | ofpbuf_delete(packet); |
2148 | return !error; | |
2149 | } | |
9dbb9d5e BP |
2150 | } |
2151 | ||
bcf84111 BP |
2152 | /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s |
2153 | * statistics appropriately. 'packet' must have at least sizeof(struct | |
2154 | * ofp_packet_in) bytes of headroom. | |
064af421 | 2155 | * |
bcf84111 BP |
2156 | * For correct results, 'packet' must actually be in 'facet''s flow; that is, |
2157 | * applying flow_extract() to 'packet' would yield the same flow as | |
2158 | * 'facet->flow'. | |
064af421 | 2159 | * |
bcf84111 BP |
2160 | * 'facet' must have accurately composed ODP actions; that is, it must not be |
2161 | * in need of revalidation. | |
750638bb BP |
2162 | * |
2163 | * Takes ownership of 'packet'. */ | |
064af421 | 2164 | static void |
bcf84111 BP |
2165 | facet_execute(struct ofproto *ofproto, struct facet *facet, |
2166 | struct ofpbuf *packet) | |
064af421 | 2167 | { |
c97fb132 | 2168 | struct dpif_flow_stats stats; |
bcf84111 BP |
2169 | |
2170 | assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in)); | |
2171 | ||
2172 | flow_extract_stats(&facet->flow, packet, &stats); | |
9fe215f0 | 2173 | stats.used = time_msec(); |
856081f6 | 2174 | if (execute_odp_actions(ofproto, &facet->flow, |
cdee00fd | 2175 | facet->actions, facet->actions_len, packet)) { |
878ae780 | 2176 | facet_update_stats(ofproto, facet, &stats); |
bcf84111 BP |
2177 | } |
2178 | } | |
2179 | ||
2180 | /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s | |
2181 | * statistics (or the statistics for one of its facets) appropriately. | |
2182 | * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom. | |
2183 | * | |
2184 | * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited | |
2185 | * with statistics for 'packet' either way. | |
2186 | * | |
2187 | * Takes ownership of 'packet'. */ | |
2188 | static void | |
2189 | rule_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port, | |
2190 | struct ofpbuf *packet) | |
2191 | { | |
f29152ca | 2192 | struct action_xlate_ctx ctx; |
cdee00fd | 2193 | struct ofpbuf *odp_actions; |
bcf84111 | 2194 | struct facet *facet; |
bcf84111 BP |
2195 | struct flow flow; |
2196 | size_t size; | |
064af421 | 2197 | |
750638bb BP |
2198 | assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in)); |
2199 | ||
bcf84111 BP |
2200 | flow_extract(packet, 0, in_port, &flow); |
2201 | ||
2202 | /* First look for a related facet. If we find one, account it to that. */ | |
2203 | facet = facet_lookup_valid(ofproto, &flow); | |
2204 | if (facet && facet->rule == rule) { | |
2205 | facet_execute(ofproto, facet, packet); | |
2206 | return; | |
064af421 BP |
2207 | } |
2208 | ||
bcf84111 BP |
2209 | /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then |
2210 | * create a new facet for it and use that. */ | |
2211 | if (rule_lookup(ofproto, &flow) == rule) { | |
2212 | facet = facet_create(ofproto, rule, &flow, packet); | |
2213 | facet_execute(ofproto, facet, packet); | |
2214 | facet_install(ofproto, facet, true); | |
2215 | return; | |
2216 | } | |
2217 | ||
2218 | /* We can't account anything to a facet. If we were to try, then that | |
2219 | * facet would have a non-matching rule, busting our invariants. */ | |
f29152ca | 2220 | action_xlate_ctx_init(&ctx, ofproto, &flow, packet); |
cdee00fd | 2221 | odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions); |
bcf84111 | 2222 | size = packet->size; |
856081f6 | 2223 | if (execute_odp_actions(ofproto, &flow, odp_actions->data, |
cdee00fd | 2224 | odp_actions->size, packet)) { |
064af421 | 2225 | rule->used = time_msec(); |
bcf84111 BP |
2226 | rule->packet_count++; |
2227 | rule->byte_count += size; | |
878ae780 | 2228 | flow_push_stats(ofproto, rule, &flow, 1, size, rule->used); |
064af421 | 2229 | } |
cdee00fd | 2230 | ofpbuf_delete(odp_actions); |
064af421 BP |
2231 | } |
2232 | ||
afe75089 | 2233 | /* Inserts 'rule' into 'p''s flow table. */ |
064af421 | 2234 | static void |
afe75089 | 2235 | rule_insert(struct ofproto *p, struct rule *rule) |
064af421 BP |
2236 | { |
2237 | struct rule *displaced_rule; | |
2238 | ||
064af421 | 2239 | displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr)); |
bcf84111 BP |
2240 | if (displaced_rule) { |
2241 | rule_destroy(p, displaced_rule); | |
064af421 | 2242 | } |
bcf84111 | 2243 | p->need_revalidate = true; |
064af421 BP |
2244 | } |
2245 | ||
bcf84111 BP |
2246 | /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a |
2247 | * 'flow' and an example 'packet' within that flow. | |
2248 | * | |
2249 | * The caller must already have determined that no facet with an identical | |
2250 | * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in | |
2251 | * 'ofproto''s classifier table. */ | |
2252 | static struct facet * | |
2253 | facet_create(struct ofproto *ofproto, struct rule *rule, | |
2254 | const struct flow *flow, const struct ofpbuf *packet) | |
064af421 | 2255 | { |
bcf84111 | 2256 | struct facet *facet; |
fbb2ea0b | 2257 | |
bcf84111 BP |
2258 | facet = xzalloc(sizeof *facet); |
2259 | facet->used = time_msec(); | |
2260 | hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0)); | |
2261 | list_push_back(&rule->facets, &facet->list_node); | |
2262 | facet->rule = rule; | |
2263 | facet->flow = *flow; | |
2264 | netflow_flow_init(&facet->nf_flow); | |
2265 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used); | |
2266 | ||
2267 | facet_make_actions(ofproto, facet, packet); | |
064af421 | 2268 | |
bcf84111 BP |
2269 | return facet; |
2270 | } | |
2271 | ||
2272 | static void | |
2273 | facet_free(struct facet *facet) | |
2274 | { | |
2275 | free(facet->actions); | |
2276 | free(facet); | |
064af421 BP |
2277 | } |
2278 | ||
431d4707 | 2279 | /* Remove 'rule' from 'ofproto' and free up the associated memory: |
431d4707 BP |
2280 | * |
2281 | * - Removes 'rule' from the classifier. | |
2282 | * | |
bcf84111 BP |
2283 | * - If 'rule' has facets, revalidates them (and possibly uninstalls and |
2284 | * destroys them), via rule_destroy(). | |
431d4707 | 2285 | */ |
064af421 BP |
2286 | static void |
2287 | rule_remove(struct ofproto *ofproto, struct rule *rule) | |
2288 | { | |
bcf84111 BP |
2289 | COVERAGE_INC(ofproto_del_rule); |
2290 | ofproto->need_revalidate = true; | |
064af421 BP |
2291 | classifier_remove(&ofproto->cls, &rule->cr); |
2292 | rule_destroy(ofproto, rule); | |
2293 | } | |
2294 | ||
bcf84111 BP |
2295 | /* Remove 'facet' from 'ofproto' and free up the associated memory: |
2296 | * | |
2297 | * - If 'facet' was installed in the datapath, uninstalls it and updates its | |
2298 | * rule's statistics, via facet_uninstall(). | |
2299 | * | |
2300 | * - Removes 'facet' from its rule and from ofproto->facets. | |
2301 | */ | |
2302 | static void | |
2303 | facet_remove(struct ofproto *ofproto, struct facet *facet) | |
2304 | { | |
2305 | facet_uninstall(ofproto, facet); | |
d530fcd2 | 2306 | facet_flush_stats(ofproto, facet); |
bcf84111 BP |
2307 | hmap_remove(&ofproto->facets, &facet->hmap_node); |
2308 | list_remove(&facet->list_node); | |
2309 | facet_free(facet); | |
2310 | } | |
2311 | ||
7f7ae89d BP |
2312 | /* Composes the ODP actions for 'facet' based on its rule's actions. */ |
2313 | static void | |
bcf84111 BP |
2314 | facet_make_actions(struct ofproto *p, struct facet *facet, |
2315 | const struct ofpbuf *packet) | |
064af421 | 2316 | { |
bcf84111 | 2317 | const struct rule *rule = facet->rule; |
cdee00fd | 2318 | struct ofpbuf *odp_actions; |
f29152ca | 2319 | struct action_xlate_ctx ctx; |
064af421 | 2320 | |
f29152ca | 2321 | action_xlate_ctx_init(&ctx, p, &facet->flow, packet); |
cdee00fd | 2322 | odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions); |
19739aee JP |
2323 | facet->tags = ctx.tags; |
2324 | facet->may_install = ctx.may_set_up_flow; | |
2325 | facet->nf_flow.output_iface = ctx.nf_output_iface; | |
064af421 | 2326 | |
cdee00fd BP |
2327 | if (facet->actions_len != odp_actions->size |
2328 | || memcmp(facet->actions, odp_actions->data, odp_actions->size)) { | |
7f7ae89d | 2329 | free(facet->actions); |
cdee00fd BP |
2330 | facet->actions_len = odp_actions->size; |
2331 | facet->actions = xmemdup(odp_actions->data, odp_actions->size); | |
064af421 | 2332 | } |
cdee00fd BP |
2333 | |
2334 | ofpbuf_delete(odp_actions); | |
064af421 BP |
2335 | } |
2336 | ||
2337 | static int | |
ba25b8f4 | 2338 | facet_put__(struct ofproto *ofproto, struct facet *facet, |
c62b0064 BP |
2339 | const struct nlattr *actions, size_t actions_len, |
2340 | struct dpif_flow_stats *stats) | |
064af421 | 2341 | { |
36956a7d | 2342 | uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S]; |
c62b0064 | 2343 | enum dpif_flow_put_flags flags; |
36956a7d BP |
2344 | struct ofpbuf key; |
2345 | ||
c62b0064 BP |
2346 | flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; |
2347 | if (stats) { | |
2348 | flags |= DPIF_FP_ZERO_STATS; | |
3394b5b6 EJ |
2349 | facet->dp_packet_count = 0; |
2350 | facet->dp_byte_count = 0; | |
c62b0064 BP |
2351 | } |
2352 | ||
36956a7d BP |
2353 | ofpbuf_use_stack(&key, keybuf, sizeof keybuf); |
2354 | odp_flow_key_from_flow(&key, &facet->flow); | |
2355 | assert(key.base == keybuf); | |
2356 | ||
feebdea2 | 2357 | return dpif_flow_put(ofproto->dpif, flags, key.data, key.size, |
c62b0064 | 2358 | actions, actions_len, stats); |
064af421 BP |
2359 | } |
2360 | ||
bcf84111 BP |
2361 | /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If |
2362 | * 'zero_stats' is true, clears any existing statistics from the datapath for | |
2363 | * 'facet'. */ | |
064af421 | 2364 | static void |
bcf84111 | 2365 | facet_install(struct ofproto *p, struct facet *facet, bool zero_stats) |
064af421 | 2366 | { |
c62b0064 BP |
2367 | struct dpif_flow_stats stats; |
2368 | ||
2369 | if (facet->may_install | |
2370 | && !facet_put__(p, facet, facet->actions, facet->actions_len, | |
2371 | zero_stats ? &stats : NULL)) { | |
2372 | facet->installed = true; | |
064af421 BP |
2373 | } |
2374 | } | |
2375 | ||
bcf84111 BP |
2376 | /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up |
2377 | * to the accounting hook function in the ofhooks structure. */ | |
064af421 | 2378 | static void |
bcf84111 BP |
2379 | facet_account(struct ofproto *ofproto, |
2380 | struct facet *facet, uint64_t extra_bytes) | |
064af421 | 2381 | { |
bcf84111 | 2382 | uint64_t total_bytes = facet->byte_count + extra_bytes; |
064af421 BP |
2383 | |
2384 | if (ofproto->ofhooks->account_flow_cb | |
bcf84111 | 2385 | && total_bytes > facet->accounted_bytes) |
064af421 BP |
2386 | { |
2387 | ofproto->ofhooks->account_flow_cb( | |
cdee00fd | 2388 | &facet->flow, facet->tags, facet->actions, facet->actions_len, |
bcf84111 BP |
2389 | total_bytes - facet->accounted_bytes, ofproto->aux); |
2390 | facet->accounted_bytes = total_bytes; | |
064af421 BP |
2391 | } |
2392 | } | |
2393 | ||
d530fcd2 | 2394 | /* If 'rule' is installed in the datapath, uninstalls it. */ |
064af421 | 2395 | static void |
bcf84111 | 2396 | facet_uninstall(struct ofproto *p, struct facet *facet) |
064af421 | 2397 | { |
bcf84111 | 2398 | if (facet->installed) { |
36956a7d | 2399 | uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S]; |
c97fb132 | 2400 | struct dpif_flow_stats stats; |
36956a7d | 2401 | struct ofpbuf key; |
064af421 | 2402 | |
36956a7d BP |
2403 | ofpbuf_use_stack(&key, keybuf, sizeof keybuf); |
2404 | odp_flow_key_from_flow(&key, &facet->flow); | |
2405 | assert(key.base == keybuf); | |
2406 | ||
feebdea2 BP |
2407 | if (!dpif_flow_del(p->dpif, key.data, key.size, &stats)) { |
2408 | facet_update_stats(p, facet, &stats); | |
064af421 | 2409 | } |
bcf84111 | 2410 | facet->installed = false; |
3394b5b6 EJ |
2411 | facet->dp_packet_count = 0; |
2412 | facet->dp_byte_count = 0; | |
a2c6a63c EJ |
2413 | } else { |
2414 | assert(facet->dp_packet_count == 0); | |
2415 | assert(facet->dp_byte_count == 0); | |
064af421 BP |
2416 | } |
2417 | } | |
2418 | ||
bcf84111 BP |
2419 | /* Returns true if the only action for 'facet' is to send to the controller. |
2420 | * (We don't report NetFlow expiration messages for such facets because they | |
2421 | * are just part of the control logic for the network, not real traffic). */ | |
0193b2af | 2422 | static bool |
bcf84111 | 2423 | facet_is_controller_flow(struct facet *facet) |
0193b2af | 2424 | { |
bcf84111 BP |
2425 | return (facet |
2426 | && facet->rule->n_actions == 1 | |
2427 | && action_outputs_to_port(&facet->rule->actions[0], | |
c1c9c9c4 | 2428 | htons(OFPP_CONTROLLER))); |
0193b2af JG |
2429 | } |
2430 | ||
bcf84111 | 2431 | /* Folds all of 'facet''s statistics into its rule. Also updates the |
3394b5b6 EJ |
2432 | * accounting ofhook and emits a NetFlow expiration if appropriate. All of |
2433 | * 'facet''s statistics in the datapath should have been zeroed and folded into | |
2434 | * its packet and byte counts before this function is called. */ | |
064af421 | 2435 | static void |
d530fcd2 | 2436 | facet_flush_stats(struct ofproto *ofproto, struct facet *facet) |
064af421 | 2437 | { |
3394b5b6 EJ |
2438 | assert(!facet->dp_byte_count); |
2439 | assert(!facet->dp_packet_count); | |
2440 | ||
878ae780 | 2441 | facet_push_stats(ofproto, facet); |
bcf84111 | 2442 | facet_account(ofproto, facet, 0); |
064af421 | 2443 | |
bcf84111 | 2444 | if (ofproto->netflow && !facet_is_controller_flow(facet)) { |
064af421 | 2445 | struct ofexpired expired; |
bcf84111 BP |
2446 | expired.flow = facet->flow; |
2447 | expired.packet_count = facet->packet_count; | |
2448 | expired.byte_count = facet->byte_count; | |
2449 | expired.used = facet->used; | |
2450 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
2451 | } | |
2452 | ||
2453 | facet->rule->packet_count += facet->packet_count; | |
2454 | facet->rule->byte_count += facet->byte_count; | |
2455 | ||
2456 | /* Reset counters to prevent double counting if 'facet' ever gets | |
2457 | * reinstalled. */ | |
2458 | facet->packet_count = 0; | |
2459 | facet->byte_count = 0; | |
878ae780 EJ |
2460 | facet->rs_packet_count = 0; |
2461 | facet->rs_byte_count = 0; | |
bcf84111 BP |
2462 | facet->accounted_bytes = 0; |
2463 | ||
2464 | netflow_flow_clear(&facet->nf_flow); | |
2465 | } | |
2466 | ||
2467 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
2468 | * Returns it if found, otherwise a null pointer. | |
2469 | * | |
2470 | * The returned facet might need revalidation; use facet_lookup_valid() | |
2471 | * instead if that is important. */ | |
2472 | static struct facet * | |
2473 | facet_find(struct ofproto *ofproto, const struct flow *flow) | |
2474 | { | |
2475 | struct facet *facet; | |
2476 | ||
2477 | HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0), | |
2478 | &ofproto->facets) { | |
2479 | if (flow_equal(flow, &facet->flow)) { | |
2480 | return facet; | |
2481 | } | |
2482 | } | |
2483 | ||
2484 | return NULL; | |
2485 | } | |
2486 | ||
2487 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
2488 | * Returns it if found, otherwise a null pointer. | |
2489 | * | |
2490 | * The returned facet is guaranteed to be valid. */ | |
2491 | static struct facet * | |
2492 | facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow) | |
2493 | { | |
2494 | struct facet *facet = facet_find(ofproto, flow); | |
2495 | ||
2496 | /* The facet we found might not be valid, since we could be in need of | |
2497 | * revalidation. If it is not valid, don't return it. */ | |
2498 | if (facet | |
2499 | && ofproto->need_revalidate | |
2500 | && !facet_revalidate(ofproto, facet)) { | |
2501 | COVERAGE_INC(ofproto_invalidated); | |
2502 | return NULL; | |
064af421 | 2503 | } |
064af421 | 2504 | |
bcf84111 BP |
2505 | return facet; |
2506 | } | |
0193b2af | 2507 | |
bcf84111 BP |
2508 | /* Re-searches 'ofproto''s classifier for a rule matching 'facet': |
2509 | * | |
2510 | * - If the rule found is different from 'facet''s current rule, moves | |
2511 | * 'facet' to the new rule and recompiles its actions. | |
2512 | * | |
2513 | * - If the rule found is the same as 'facet''s current rule, leaves 'facet' | |
2514 | * where it is and recompiles its actions anyway. | |
2515 | * | |
2516 | * - If there is none, destroys 'facet'. | |
2517 | * | |
d530fcd2 | 2518 | * Returns true if 'facet' still exists, false if it has been destroyed. */ |
bcf84111 BP |
2519 | static bool |
2520 | facet_revalidate(struct ofproto *ofproto, struct facet *facet) | |
2521 | { | |
f29152ca | 2522 | struct action_xlate_ctx ctx; |
cdee00fd | 2523 | struct ofpbuf *odp_actions; |
d530fcd2 | 2524 | struct rule *new_rule; |
d530fcd2 | 2525 | bool actions_changed; |
bcf84111 BP |
2526 | |
2527 | COVERAGE_INC(facet_revalidate); | |
d530fcd2 BP |
2528 | |
2529 | /* Determine the new rule. */ | |
2530 | new_rule = rule_lookup(ofproto, &facet->flow); | |
2531 | if (!new_rule) { | |
2532 | /* No new rule, so delete the facet. */ | |
bcf84111 BP |
2533 | facet_remove(ofproto, facet); |
2534 | return false; | |
2535 | } | |
2536 | ||
d530fcd2 BP |
2537 | /* Calculate new ODP actions. |
2538 | * | |
cdee00fd BP |
2539 | * We do not modify any 'facet' state yet, because we might need to, e.g., |
2540 | * emit a NetFlow expiration and, if so, we need to have the old state | |
2541 | * around to properly compose it. */ | |
f29152ca | 2542 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL); |
cdee00fd BP |
2543 | odp_actions = xlate_actions(&ctx, new_rule->actions, new_rule->n_actions); |
2544 | actions_changed = (facet->actions_len != odp_actions->size | |
2545 | || memcmp(facet->actions, odp_actions->data, | |
2546 | facet->actions_len)); | |
d530fcd2 BP |
2547 | |
2548 | /* If the ODP actions changed or the installability changed, then we need | |
2549 | * to talk to the datapath. */ | |
e4add896 BP |
2550 | if (actions_changed || ctx.may_set_up_flow != facet->installed) { |
2551 | if (ctx.may_set_up_flow) { | |
c97fb132 | 2552 | struct dpif_flow_stats stats; |
d530fcd2 | 2553 | |
c62b0064 BP |
2554 | facet_put__(ofproto, facet, |
2555 | odp_actions->data, odp_actions->size, &stats); | |
feebdea2 | 2556 | facet_update_stats(ofproto, facet, &stats); |
d530fcd2 BP |
2557 | } else { |
2558 | facet_uninstall(ofproto, facet); | |
2559 | } | |
2560 | ||
2561 | /* The datapath flow is gone or has zeroed stats, so push stats out of | |
2562 | * 'facet' into 'rule'. */ | |
2563 | facet_flush_stats(ofproto, facet); | |
2564 | } | |
2565 | ||
2566 | /* Update 'facet' now that we've taken care of all the old state. */ | |
f29152ca BP |
2567 | facet->tags = ctx.tags; |
2568 | facet->nf_flow.output_iface = ctx.nf_output_iface; | |
2569 | facet->may_install = ctx.may_set_up_flow; | |
d530fcd2 BP |
2570 | if (actions_changed) { |
2571 | free(facet->actions); | |
cdee00fd BP |
2572 | facet->actions_len = odp_actions->size; |
2573 | facet->actions = xmemdup(odp_actions->data, odp_actions->size); | |
d530fcd2 BP |
2574 | } |
2575 | if (facet->rule != new_rule) { | |
bcf84111 BP |
2576 | COVERAGE_INC(facet_changed_rule); |
2577 | list_remove(&facet->list_node); | |
d530fcd2 BP |
2578 | list_push_back(&new_rule->facets, &facet->list_node); |
2579 | facet->rule = new_rule; | |
2580 | facet->used = new_rule->created; | |
5e1b3214 | 2581 | facet->rs_used = facet->used; |
0c0afbec | 2582 | } |
bcf84111 | 2583 | |
cdfcd496 BP |
2584 | ofpbuf_delete(odp_actions); |
2585 | ||
bcf84111 | 2586 | return true; |
064af421 BP |
2587 | } |
2588 | \f | |
2589 | static void | |
2590 | queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn, | |
2591 | struct rconn_packet_counter *counter) | |
2592 | { | |
2593 | update_openflow_length(msg); | |
2594 | if (rconn_send(ofconn->rconn, msg, counter)) { | |
2595 | ofpbuf_delete(msg); | |
2596 | } | |
2597 | } | |
2598 | ||
064af421 BP |
2599 | static void |
2600 | send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh, | |
2601 | int error) | |
2602 | { | |
dc4762ed | 2603 | struct ofpbuf *buf = ofputil_encode_error_msg(error, oh); |
26c112c2 BP |
2604 | if (buf) { |
2605 | COVERAGE_INC(ofproto_error); | |
2606 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
2607 | } | |
064af421 BP |
2608 | } |
2609 | ||
2610 | static void | |
2611 | hton_ofp_phy_port(struct ofp_phy_port *opp) | |
2612 | { | |
2613 | opp->port_no = htons(opp->port_no); | |
2614 | opp->config = htonl(opp->config); | |
2615 | opp->state = htonl(opp->state); | |
2616 | opp->curr = htonl(opp->curr); | |
2617 | opp->advertised = htonl(opp->advertised); | |
2618 | opp->supported = htonl(opp->supported); | |
2619 | opp->peer = htonl(opp->peer); | |
2620 | } | |
2621 | ||
2622 | static int | |
d1e2cf21 | 2623 | handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh) |
064af421 | 2624 | { |
d1e2cf21 | 2625 | queue_tx(make_echo_reply(oh), ofconn, ofconn->reply_counter); |
064af421 BP |
2626 | return 0; |
2627 | } | |
2628 | ||
2629 | static int | |
d1e2cf21 | 2630 | handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh) |
064af421 BP |
2631 | { |
2632 | struct ofp_switch_features *osf; | |
2633 | struct ofpbuf *buf; | |
064af421 BP |
2634 | struct ofport *port; |
2635 | ||
2636 | osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf); | |
3269c562 | 2637 | osf->datapath_id = htonll(ofconn->ofproto->datapath_id); |
064af421 BP |
2638 | osf->n_buffers = htonl(pktbuf_capacity()); |
2639 | osf->n_tables = 2; | |
2640 | osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS | | |
0254ae23 | 2641 | OFPC_PORT_STATS | OFPC_ARP_MATCH_IP); |
064af421 BP |
2642 | osf->actions = htonl((1u << OFPAT_OUTPUT) | |
2643 | (1u << OFPAT_SET_VLAN_VID) | | |
2644 | (1u << OFPAT_SET_VLAN_PCP) | | |
2645 | (1u << OFPAT_STRIP_VLAN) | | |
2646 | (1u << OFPAT_SET_DL_SRC) | | |
2647 | (1u << OFPAT_SET_DL_DST) | | |
2648 | (1u << OFPAT_SET_NW_SRC) | | |
2649 | (1u << OFPAT_SET_NW_DST) | | |
959a2ecd | 2650 | (1u << OFPAT_SET_NW_TOS) | |
064af421 | 2651 | (1u << OFPAT_SET_TP_SRC) | |
c1c9c9c4 BP |
2652 | (1u << OFPAT_SET_TP_DST) | |
2653 | (1u << OFPAT_ENQUEUE)); | |
064af421 | 2654 | |
3269c562 | 2655 | HMAP_FOR_EACH (port, hmap_node, &ofconn->ofproto->ports) { |
064af421 BP |
2656 | hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp)); |
2657 | } | |
2658 | ||
2659 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
2660 | return 0; | |
2661 | } | |
2662 | ||
2663 | static int | |
d1e2cf21 | 2664 | handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh) |
064af421 BP |
2665 | { |
2666 | struct ofpbuf *buf; | |
2667 | struct ofp_switch_config *osc; | |
2668 | uint16_t flags; | |
2669 | bool drop_frags; | |
2670 | ||
2671 | /* Figure out flags. */ | |
3269c562 | 2672 | dpif_get_drop_frags(ofconn->ofproto->dpif, &drop_frags); |
064af421 | 2673 | flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL; |
064af421 BP |
2674 | |
2675 | /* Send reply. */ | |
2676 | osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf); | |
2677 | osc->flags = htons(flags); | |
2678 | osc->miss_send_len = htons(ofconn->miss_send_len); | |
2679 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
2680 | ||
2681 | return 0; | |
2682 | } | |
2683 | ||
2684 | static int | |
d1e2cf21 | 2685 | handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc) |
064af421 | 2686 | { |
d1e2cf21 | 2687 | uint16_t flags = ntohs(osc->flags); |
064af421 | 2688 | |
5899143f | 2689 | if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) { |
064af421 BP |
2690 | switch (flags & OFPC_FRAG_MASK) { |
2691 | case OFPC_FRAG_NORMAL: | |
3269c562 | 2692 | dpif_set_drop_frags(ofconn->ofproto->dpif, false); |
064af421 BP |
2693 | break; |
2694 | case OFPC_FRAG_DROP: | |
3269c562 | 2695 | dpif_set_drop_frags(ofconn->ofproto->dpif, true); |
064af421 BP |
2696 | break; |
2697 | default: | |
2698 | VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")", | |
2699 | osc->flags); | |
2700 | break; | |
2701 | } | |
2702 | } | |
2703 | ||
064af421 BP |
2704 | ofconn->miss_send_len = ntohs(osc->miss_send_len); |
2705 | ||
2706 | return 0; | |
2707 | } | |
2708 | ||
064af421 BP |
2709 | static void do_xlate_actions(const union ofp_action *in, size_t n_in, |
2710 | struct action_xlate_ctx *ctx); | |
2711 | ||
2712 | static void | |
2713 | add_output_action(struct action_xlate_ctx *ctx, uint16_t port) | |
2714 | { | |
ca0f572c | 2715 | const struct ofport *ofport = get_port(ctx->ofproto, port); |
6cfaf517 BP |
2716 | |
2717 | if (ofport) { | |
2718 | if (ofport->opp.config & OFPPC_NO_FWD) { | |
2719 | /* Forwarding disabled on port. */ | |
2720 | return; | |
2721 | } | |
2722 | } else { | |
2723 | /* | |
2724 | * We don't have an ofport record for this port, but it doesn't hurt to | |
2725 | * allow forwarding to it anyhow. Maybe such a port will appear later | |
2726 | * and we're pre-populating the flow table. | |
2727 | */ | |
064af421 | 2728 | } |
6cfaf517 | 2729 | |
7aec165d | 2730 | nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, port); |
6a07af36 | 2731 | ctx->nf_output_iface = port; |
064af421 BP |
2732 | } |
2733 | ||
2734 | static struct rule * | |
bcf84111 | 2735 | rule_lookup(struct ofproto *ofproto, const struct flow *flow) |
064af421 | 2736 | { |
3c4486a5 | 2737 | return rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow)); |
064af421 BP |
2738 | } |
2739 | ||
2740 | static void | |
2741 | xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port) | |
2742 | { | |
5f8bfd69 | 2743 | if (ctx->recurse < MAX_RESUBMIT_RECURSION) { |
2c5d1389 | 2744 | uint16_t old_in_port; |
064af421 | 2745 | struct rule *rule; |
064af421 | 2746 | |
2c5d1389 BP |
2747 | /* Look up a flow with 'in_port' as the input port. Then restore the |
2748 | * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will | |
2749 | * have surprising behavior). */ | |
2750 | old_in_port = ctx->flow.in_port; | |
e18fe8a2 | 2751 | ctx->flow.in_port = in_port; |
bcf84111 | 2752 | rule = rule_lookup(ctx->ofproto, &ctx->flow); |
2c5d1389 BP |
2753 | ctx->flow.in_port = old_in_port; |
2754 | ||
7aa697dd BP |
2755 | if (ctx->resubmit_hook) { |
2756 | ctx->resubmit_hook(ctx, rule); | |
2757 | } | |
2758 | ||
064af421 | 2759 | if (rule) { |
064af421 BP |
2760 | ctx->recurse++; |
2761 | do_xlate_actions(rule->actions, rule->n_actions, ctx); | |
2762 | ctx->recurse--; | |
2763 | } | |
5f8bfd69 | 2764 | } else { |
db5ce514 | 2765 | static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1); |
5f8bfd69 BP |
2766 | |
2767 | VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times", | |
2768 | MAX_RESUBMIT_RECURSION); | |
064af421 BP |
2769 | } |
2770 | } | |
2771 | ||
f1588b1f BP |
2772 | static void |
2773 | flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, uint32_t mask, | |
cdee00fd | 2774 | uint16_t *nf_output_iface, struct ofpbuf *odp_actions) |
f1588b1f BP |
2775 | { |
2776 | struct ofport *ofport; | |
2777 | ||
2778 | HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) { | |
2779 | uint16_t odp_port = ofport->odp_port; | |
2780 | if (odp_port != odp_in_port && !(ofport->opp.config & mask)) { | |
7aec165d | 2781 | nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port); |
f1588b1f BP |
2782 | } |
2783 | } | |
2784 | *nf_output_iface = NF_OUT_FLOOD; | |
2785 | } | |
2786 | ||
064af421 | 2787 | static void |
aae51f53 BP |
2788 | xlate_output_action__(struct action_xlate_ctx *ctx, |
2789 | uint16_t port, uint16_t max_len) | |
064af421 BP |
2790 | { |
2791 | uint16_t odp_port; | |
6a07af36 JG |
2792 | uint16_t prev_nf_output_iface = ctx->nf_output_iface; |
2793 | ||
2794 | ctx->nf_output_iface = NF_OUT_DROP; | |
064af421 | 2795 | |
aae51f53 | 2796 | switch (port) { |
064af421 | 2797 | case OFPP_IN_PORT: |
e18fe8a2 | 2798 | add_output_action(ctx, ctx->flow.in_port); |
064af421 BP |
2799 | break; |
2800 | case OFPP_TABLE: | |
e18fe8a2 | 2801 | xlate_table_action(ctx, ctx->flow.in_port); |
064af421 BP |
2802 | break; |
2803 | case OFPP_NORMAL: | |
e18fe8a2 | 2804 | if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet, |
cdee00fd | 2805 | ctx->odp_actions, &ctx->tags, |
6a07af36 | 2806 | &ctx->nf_output_iface, |
064af421 BP |
2807 | ctx->ofproto->aux)) { |
2808 | COVERAGE_INC(ofproto_uninstallable); | |
d6fbec6d | 2809 | ctx->may_set_up_flow = false; |
064af421 BP |
2810 | } |
2811 | break; | |
2812 | case OFPP_FLOOD: | |
f1588b1f | 2813 | flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD, |
cdee00fd | 2814 | &ctx->nf_output_iface, ctx->odp_actions); |
9628cd42 | 2815 | break; |
064af421 | 2816 | case OFPP_ALL: |
f1588b1f | 2817 | flood_packets(ctx->ofproto, ctx->flow.in_port, 0, |
cdee00fd | 2818 | &ctx->nf_output_iface, ctx->odp_actions); |
064af421 BP |
2819 | break; |
2820 | case OFPP_CONTROLLER: | |
7aec165d | 2821 | nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_CONTROLLER, max_len); |
064af421 BP |
2822 | break; |
2823 | case OFPP_LOCAL: | |
2824 | add_output_action(ctx, ODPP_LOCAL); | |
2825 | break; | |
2826 | default: | |
aae51f53 | 2827 | odp_port = ofp_port_to_odp_port(port); |
e18fe8a2 | 2828 | if (odp_port != ctx->flow.in_port) { |
064af421 BP |
2829 | add_output_action(ctx, odp_port); |
2830 | } | |
2831 | break; | |
2832 | } | |
6a07af36 JG |
2833 | |
2834 | if (prev_nf_output_iface == NF_OUT_FLOOD) { | |
2835 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
2836 | } else if (ctx->nf_output_iface == NF_OUT_DROP) { | |
2837 | ctx->nf_output_iface = prev_nf_output_iface; | |
2838 | } else if (prev_nf_output_iface != NF_OUT_DROP && | |
2839 | ctx->nf_output_iface != NF_OUT_FLOOD) { | |
2840 | ctx->nf_output_iface = NF_OUT_MULTI; | |
2841 | } | |
064af421 BP |
2842 | } |
2843 | ||
aae51f53 BP |
2844 | static void |
2845 | xlate_output_action(struct action_xlate_ctx *ctx, | |
2846 | const struct ofp_action_output *oao) | |
2847 | { | |
2848 | xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len)); | |
2849 | } | |
2850 | ||
c1c9c9c4 BP |
2851 | /* If the final ODP action in 'ctx' is "pop priority", drop it, as an |
2852 | * optimization, because we're going to add another action that sets the | |
2853 | * priority immediately after, or because there are no actions following the | |
2854 | * pop. */ | |
2855 | static void | |
2856 | remove_pop_action(struct action_xlate_ctx *ctx) | |
2857 | { | |
cdee00fd BP |
2858 | if (ctx->odp_actions->size == ctx->last_pop_priority) { |
2859 | ctx->odp_actions->size -= NLA_ALIGN(NLA_HDRLEN); | |
2860 | ctx->last_pop_priority = -1; | |
2861 | } | |
2862 | } | |
2863 | ||
2864 | static void | |
2865 | add_pop_action(struct action_xlate_ctx *ctx) | |
2866 | { | |
2867 | if (ctx->odp_actions->size != ctx->last_pop_priority) { | |
7aec165d | 2868 | nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_POP_PRIORITY); |
cdee00fd | 2869 | ctx->last_pop_priority = ctx->odp_actions->size; |
c1c9c9c4 BP |
2870 | } |
2871 | } | |
2872 | ||
2873 | static void | |
2874 | xlate_enqueue_action(struct action_xlate_ctx *ctx, | |
2875 | const struct ofp_action_enqueue *oae) | |
2876 | { | |
2877 | uint16_t ofp_port, odp_port; | |
aae51f53 BP |
2878 | uint32_t priority; |
2879 | int error; | |
2880 | ||
2881 | error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id), | |
2882 | &priority); | |
2883 | if (error) { | |
2884 | /* Fall back to ordinary output action. */ | |
2885 | xlate_output_action__(ctx, ntohs(oae->port), 0); | |
2886 | return; | |
2887 | } | |
c1c9c9c4 BP |
2888 | |
2889 | /* Figure out ODP output port. */ | |
2890 | ofp_port = ntohs(oae->port); | |
2891 | if (ofp_port != OFPP_IN_PORT) { | |
2892 | odp_port = ofp_port_to_odp_port(ofp_port); | |
2893 | } else { | |
2894 | odp_port = ctx->flow.in_port; | |
2895 | } | |
2896 | ||
2897 | /* Add ODP actions. */ | |
2898 | remove_pop_action(ctx); | |
7aec165d | 2899 | nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority); |
c1c9c9c4 | 2900 | add_output_action(ctx, odp_port); |
cdee00fd | 2901 | add_pop_action(ctx); |
c1c9c9c4 BP |
2902 | |
2903 | /* Update NetFlow output port. */ | |
2904 | if (ctx->nf_output_iface == NF_OUT_DROP) { | |
2905 | ctx->nf_output_iface = odp_port; | |
2906 | } else if (ctx->nf_output_iface != NF_OUT_FLOOD) { | |
2907 | ctx->nf_output_iface = NF_OUT_MULTI; | |
2908 | } | |
2909 | } | |
2910 | ||
eedc0097 JP |
2911 | static void |
2912 | xlate_set_queue_action(struct action_xlate_ctx *ctx, | |
2913 | const struct nx_action_set_queue *nasq) | |
2914 | { | |
2915 | uint32_t priority; | |
2916 | int error; | |
2917 | ||
2918 | error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id), | |
2919 | &priority); | |
2920 | if (error) { | |
2921 | /* Couldn't translate queue to a priority, so ignore. A warning | |
2922 | * has already been logged. */ | |
2923 | return; | |
2924 | } | |
2925 | ||
2926 | remove_pop_action(ctx); | |
7aec165d | 2927 | nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority); |
eedc0097 JP |
2928 | } |
2929 | ||
350a665f BP |
2930 | static void |
2931 | xlate_set_dl_tci(struct action_xlate_ctx *ctx) | |
2932 | { | |
66642cb4 BP |
2933 | ovs_be16 tci = ctx->flow.vlan_tci; |
2934 | if (!(tci & htons(VLAN_CFI))) { | |
7aec165d | 2935 | nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN); |
350a665f | 2936 | } else { |
7aec165d | 2937 | nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_TCI, |
cdee00fd | 2938 | tci & ~htons(VLAN_CFI)); |
350a665f BP |
2939 | } |
2940 | } | |
2941 | ||
7b064a79 BP |
2942 | struct xlate_reg_state { |
2943 | ovs_be16 vlan_tci; | |
2944 | ovs_be64 tun_id; | |
2945 | }; | |
2946 | ||
b6c9e612 | 2947 | static void |
7b064a79 BP |
2948 | save_reg_state(const struct action_xlate_ctx *ctx, |
2949 | struct xlate_reg_state *state) | |
b6c9e612 | 2950 | { |
7b064a79 BP |
2951 | state->vlan_tci = ctx->flow.vlan_tci; |
2952 | state->tun_id = ctx->flow.tun_id; | |
2953 | } | |
b6c9e612 | 2954 | |
7b064a79 BP |
2955 | static void |
2956 | update_reg_state(struct action_xlate_ctx *ctx, | |
2957 | const struct xlate_reg_state *state) | |
2958 | { | |
2959 | if (ctx->flow.vlan_tci != state->vlan_tci) { | |
b6c9e612 BP |
2960 | xlate_set_dl_tci(ctx); |
2961 | } | |
7b064a79 | 2962 | if (ctx->flow.tun_id != state->tun_id) { |
7aec165d BP |
2963 | nl_msg_put_be64(ctx->odp_actions, |
2964 | ODP_ACTION_ATTR_SET_TUNNEL, ctx->flow.tun_id); | |
926947e6 | 2965 | } |
b6c9e612 BP |
2966 | } |
2967 | ||
064af421 BP |
2968 | static void |
2969 | xlate_nicira_action(struct action_xlate_ctx *ctx, | |
2970 | const struct nx_action_header *nah) | |
2971 | { | |
2972 | const struct nx_action_resubmit *nar; | |
659586ef | 2973 | const struct nx_action_set_tunnel *nast; |
eedc0097 | 2974 | const struct nx_action_set_queue *nasq; |
53ddd40a | 2975 | const struct nx_action_multipath *nam; |
e41a9130 | 2976 | enum nx_action_subtype subtype = ntohs(nah->subtype); |
7b064a79 | 2977 | struct xlate_reg_state state; |
b9298d3f | 2978 | ovs_be64 tun_id; |
064af421 BP |
2979 | |
2980 | assert(nah->vendor == htonl(NX_VENDOR_ID)); | |
2981 | switch (subtype) { | |
2982 | case NXAST_RESUBMIT: | |
2983 | nar = (const struct nx_action_resubmit *) nah; | |
2984 | xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port))); | |
2985 | break; | |
2986 | ||
659586ef JG |
2987 | case NXAST_SET_TUNNEL: |
2988 | nast = (const struct nx_action_set_tunnel *) nah; | |
b9298d3f | 2989 | tun_id = htonll(ntohl(nast->tun_id)); |
7aec165d | 2990 | nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id); |
b9298d3f | 2991 | ctx->flow.tun_id = tun_id; |
659586ef JG |
2992 | break; |
2993 | ||
401eeb92 BP |
2994 | case NXAST_DROP_SPOOFED_ARP: |
2995 | if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) { | |
7aec165d BP |
2996 | nl_msg_put_flag(ctx->odp_actions, |
2997 | ODP_ACTION_ATTR_DROP_SPOOFED_ARP); | |
401eeb92 BP |
2998 | } |
2999 | break; | |
3000 | ||
eedc0097 JP |
3001 | case NXAST_SET_QUEUE: |
3002 | nasq = (const struct nx_action_set_queue *) nah; | |
3003 | xlate_set_queue_action(ctx, nasq); | |
3004 | break; | |
3005 | ||
3006 | case NXAST_POP_QUEUE: | |
cdee00fd | 3007 | add_pop_action(ctx); |
eedc0097 JP |
3008 | break; |
3009 | ||
b6c9e612 | 3010 | case NXAST_REG_MOVE: |
7b064a79 BP |
3011 | save_reg_state(ctx, &state); |
3012 | nxm_execute_reg_move((const struct nx_action_reg_move *) nah, | |
3013 | &ctx->flow); | |
3014 | update_reg_state(ctx, &state); | |
b6c9e612 BP |
3015 | break; |
3016 | ||
3017 | case NXAST_REG_LOAD: | |
7b064a79 | 3018 | save_reg_state(ctx, &state); |
b6c9e612 BP |
3019 | nxm_execute_reg_load((const struct nx_action_reg_load *) nah, |
3020 | &ctx->flow); | |
7b064a79 BP |
3021 | update_reg_state(ctx, &state); |
3022 | break; | |
96fc46e8 BP |
3023 | |
3024 | case NXAST_NOTE: | |
3025 | /* Nothing to do. */ | |
b6c9e612 BP |
3026 | break; |
3027 | ||
b9298d3f BP |
3028 | case NXAST_SET_TUNNEL64: |
3029 | tun_id = ((const struct nx_action_set_tunnel64 *) nah)->tun_id; | |
7aec165d | 3030 | nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id); |
b9298d3f BP |
3031 | ctx->flow.tun_id = tun_id; |
3032 | break; | |
3033 | ||
53ddd40a BP |
3034 | case NXAST_MULTIPATH: |
3035 | nam = (const struct nx_action_multipath *) nah; | |
3036 | multipath_execute(nam, &ctx->flow); | |
3037 | break; | |
3038 | ||
999f0d45 | 3039 | /* If you add a new action here that modifies flow data, don't forget to |
c1c9c9c4 | 3040 | * update the flow key in ctx->flow at the same time. */ |
999f0d45 | 3041 | |
e41a9130 | 3042 | case NXAST_SNAT__OBSOLETE: |
064af421 | 3043 | default: |
e41a9130 | 3044 | VLOG_DBG_RL(&rl, "unknown Nicira action type %d", (int) subtype); |
064af421 BP |
3045 | break; |
3046 | } | |
3047 | } | |
3048 | ||
3049 | static void | |
3050 | do_xlate_actions(const union ofp_action *in, size_t n_in, | |
3051 | struct action_xlate_ctx *ctx) | |
3052 | { | |
3053 | struct actions_iterator iter; | |
3054 | const union ofp_action *ia; | |
3055 | const struct ofport *port; | |
3056 | ||
ca0f572c | 3057 | port = get_port(ctx->ofproto, ctx->flow.in_port); |
064af421 | 3058 | if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) && |
ba186119 | 3059 | port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp) |
064af421 BP |
3060 | ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) { |
3061 | /* Drop this flow. */ | |
3062 | return; | |
3063 | } | |
3064 | ||
3065 | for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) { | |
e41a9130 | 3066 | enum ofp_action_type type = ntohs(ia->type); |
cdee00fd | 3067 | const struct ofp_action_dl_addr *oada; |
064af421 BP |
3068 | |
3069 | switch (type) { | |
3070 | case OFPAT_OUTPUT: | |
3071 | xlate_output_action(ctx, &ia->output); | |
3072 | break; | |
3073 | ||
3074 | case OFPAT_SET_VLAN_VID: | |
66642cb4 BP |
3075 | ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK); |
3076 | ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI); | |
350a665f | 3077 | xlate_set_dl_tci(ctx); |
064af421 BP |
3078 | break; |
3079 | ||
3080 | case OFPAT_SET_VLAN_PCP: | |
66642cb4 BP |
3081 | ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK); |
3082 | ctx->flow.vlan_tci |= htons( | |
3083 | (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI); | |
350a665f | 3084 | xlate_set_dl_tci(ctx); |
064af421 BP |
3085 | break; |
3086 | ||
3087 | case OFPAT_STRIP_VLAN: | |
66642cb4 | 3088 | ctx->flow.vlan_tci = htons(0); |
350a665f | 3089 | xlate_set_dl_tci(ctx); |
064af421 BP |
3090 | break; |
3091 | ||
3092 | case OFPAT_SET_DL_SRC: | |
cdee00fd | 3093 | oada = ((struct ofp_action_dl_addr *) ia); |
7aec165d | 3094 | nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_SRC, |
cdee00fd BP |
3095 | oada->dl_addr, ETH_ADDR_LEN); |
3096 | memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN); | |
064af421 BP |
3097 | break; |
3098 | ||
3099 | case OFPAT_SET_DL_DST: | |
cdee00fd | 3100 | oada = ((struct ofp_action_dl_addr *) ia); |
7aec165d | 3101 | nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_DST, |
cdee00fd BP |
3102 | oada->dl_addr, ETH_ADDR_LEN); |
3103 | memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN); | |
064af421 BP |
3104 | break; |
3105 | ||
3106 | case OFPAT_SET_NW_SRC: | |
7aec165d | 3107 | nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_SRC, |
cdee00fd BP |
3108 | ia->nw_addr.nw_addr); |
3109 | ctx->flow.nw_src = ia->nw_addr.nw_addr; | |
064af421 BP |
3110 | break; |
3111 | ||
2d70a31a | 3112 | case OFPAT_SET_NW_DST: |
7aec165d | 3113 | nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_DST, |
cdee00fd BP |
3114 | ia->nw_addr.nw_addr); |
3115 | ctx->flow.nw_dst = ia->nw_addr.nw_addr; | |
2d38e234 | 3116 | break; |
959a2ecd JP |
3117 | |
3118 | case OFPAT_SET_NW_TOS: | |
7aec165d | 3119 | nl_msg_put_u8(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_TOS, |
cdee00fd BP |
3120 | ia->nw_tos.nw_tos); |
3121 | ctx->flow.nw_tos = ia->nw_tos.nw_tos; | |
2d70a31a JP |
3122 | break; |
3123 | ||
064af421 | 3124 | case OFPAT_SET_TP_SRC: |
7aec165d | 3125 | nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_SRC, |
cdee00fd BP |
3126 | ia->tp_port.tp_port); |
3127 | ctx->flow.tp_src = ia->tp_port.tp_port; | |
064af421 BP |
3128 | break; |
3129 | ||
2d70a31a | 3130 | case OFPAT_SET_TP_DST: |
7aec165d | 3131 | nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_DST, |
cdee00fd BP |
3132 | ia->tp_port.tp_port); |
3133 | ctx->flow.tp_dst = ia->tp_port.tp_port; | |
2d70a31a JP |
3134 | break; |
3135 | ||
064af421 BP |
3136 | case OFPAT_VENDOR: |
3137 | xlate_nicira_action(ctx, (const struct nx_action_header *) ia); | |
3138 | break; | |
3139 | ||
c1c9c9c4 BP |
3140 | case OFPAT_ENQUEUE: |
3141 | xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia); | |
3142 | break; | |
3143 | ||
064af421 | 3144 | default: |
e41a9130 | 3145 | VLOG_DBG_RL(&rl, "unknown action type %d", (int) type); |
064af421 BP |
3146 | break; |
3147 | } | |
3148 | } | |
3149 | } | |
3150 | ||
f29152ca BP |
3151 | static void |
3152 | action_xlate_ctx_init(struct action_xlate_ctx *ctx, | |
3153 | struct ofproto *ofproto, const struct flow *flow, | |
3154 | const struct ofpbuf *packet) | |
064af421 | 3155 | { |
f29152ca BP |
3156 | ctx->ofproto = ofproto; |
3157 | ctx->flow = *flow; | |
3158 | ctx->packet = packet; | |
7aa697dd | 3159 | ctx->resubmit_hook = NULL; |
ebe482fd | 3160 | ctx->check_special = true; |
f29152ca | 3161 | } |
1eb0942d | 3162 | |
cdee00fd | 3163 | static struct ofpbuf * |
f29152ca BP |
3164 | xlate_actions(struct action_xlate_ctx *ctx, |
3165 | const union ofp_action *in, size_t n_in) | |
3166 | { | |
064af421 | 3167 | COVERAGE_INC(ofproto_ofp2odp); |
cdee00fd BP |
3168 | |
3169 | ctx->odp_actions = ofpbuf_new(512); | |
f29152ca BP |
3170 | ctx->tags = 0; |
3171 | ctx->may_set_up_flow = true; | |
3172 | ctx->nf_output_iface = NF_OUT_DROP; | |
3173 | ctx->recurse = 0; | |
cdee00fd | 3174 | ctx->last_pop_priority = -1; |
ebe482fd EJ |
3175 | |
3176 | if (!ctx->check_special | |
7a7ae75a BP |
3177 | || !ctx->ofproto->ofhooks->special_cb |
3178 | || ctx->ofproto->ofhooks->special_cb(&ctx->flow, ctx->packet, | |
3179 | ctx->ofproto->aux)) { | |
ebe482fd EJ |
3180 | do_xlate_actions(in, n_in, ctx); |
3181 | } else { | |
3182 | ctx->may_set_up_flow = false; | |
3183 | } | |
3184 | ||
f29152ca | 3185 | remove_pop_action(ctx); |
0ad9b732 | 3186 | |
d6fbec6d | 3187 | /* Check with in-band control to see if we're allowed to set up this |
0ad9b732 | 3188 | * flow. */ |
cdee00fd BP |
3189 | if (!in_band_rule_check(ctx->ofproto->in_band, &ctx->flow, |
3190 | ctx->odp_actions->data, ctx->odp_actions->size)) { | |
f29152ca | 3191 | ctx->may_set_up_flow = false; |
0ad9b732 JP |
3192 | } |
3193 | ||
cdee00fd | 3194 | return ctx->odp_actions; |
064af421 BP |
3195 | } |
3196 | ||
9deba63b BP |
3197 | /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow |
3198 | * error message code (composed with ofp_mkerr()) for the caller to propagate | |
3199 | * upward. Otherwise, returns 0. | |
3200 | * | |
2228b50d | 3201 | * The log message mentions 'msg_type'. */ |
9deba63b | 3202 | static int |
2228b50d | 3203 | reject_slave_controller(struct ofconn *ofconn, const const char *msg_type) |
9deba63b | 3204 | { |
5899143f | 3205 | if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) { |
9deba63b | 3206 | static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5); |
9deba63b | 3207 | VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller", |
2228b50d | 3208 | msg_type); |
9deba63b BP |
3209 | |
3210 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM); | |
3211 | } else { | |
3212 | return 0; | |
3213 | } | |
3214 | } | |
3215 | ||
064af421 | 3216 | static int |
d1e2cf21 | 3217 | handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh) |
064af421 | 3218 | { |
3269c562 | 3219 | struct ofproto *p = ofconn->ofproto; |
064af421 BP |
3220 | struct ofp_packet_out *opo; |
3221 | struct ofpbuf payload, *buffer; | |
ac51afaf | 3222 | union ofp_action *ofp_actions; |
f29152ca | 3223 | struct action_xlate_ctx ctx; |
cdee00fd | 3224 | struct ofpbuf *odp_actions; |
ac51afaf | 3225 | struct ofpbuf request; |
ae412e7d | 3226 | struct flow flow; |
ac51afaf | 3227 | size_t n_ofp_actions; |
064af421 | 3228 | uint16_t in_port; |
064af421 BP |
3229 | int error; |
3230 | ||
ac51afaf BP |
3231 | COVERAGE_INC(ofproto_packet_out); |
3232 | ||
2228b50d | 3233 | error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT"); |
9deba63b BP |
3234 | if (error) { |
3235 | return error; | |
3236 | } | |
3237 | ||
ac51afaf | 3238 | /* Get ofp_packet_out. */ |
0bc9407d | 3239 | ofpbuf_use_const(&request, oh, ntohs(oh->length)); |
bbc32a88 | 3240 | opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions)); |
ac51afaf BP |
3241 | |
3242 | /* Get actions. */ | |
3243 | error = ofputil_pull_actions(&request, ntohs(opo->actions_len), | |
3244 | &ofp_actions, &n_ofp_actions); | |
064af421 BP |
3245 | if (error) { |
3246 | return error; | |
3247 | } | |
064af421 | 3248 | |
ac51afaf | 3249 | /* Get payload. */ |
064af421 BP |
3250 | if (opo->buffer_id != htonl(UINT32_MAX)) { |
3251 | error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id), | |
3252 | &buffer, &in_port); | |
7778bd15 | 3253 | if (error || !buffer) { |
064af421 BP |
3254 | return error; |
3255 | } | |
3256 | payload = *buffer; | |
3257 | } else { | |
ac51afaf | 3258 | payload = request; |
064af421 BP |
3259 | buffer = NULL; |
3260 | } | |
3261 | ||
ac51afaf BP |
3262 | /* Extract flow, check actions. */ |
3263 | flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), | |
3264 | &flow); | |
f1defbf9 | 3265 | error = validate_actions(ofp_actions, n_ofp_actions, &flow, p->max_ports); |
ac51afaf BP |
3266 | if (error) { |
3267 | goto exit; | |
3268 | } | |
3269 | ||
3270 | /* Send. */ | |
f29152ca | 3271 | action_xlate_ctx_init(&ctx, p, &flow, &payload); |
cdee00fd BP |
3272 | odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions); |
3273 | dpif_execute(p->dpif, odp_actions->data, odp_actions->size, &payload); | |
3274 | ofpbuf_delete(odp_actions); | |
064af421 | 3275 | |
ac51afaf BP |
3276 | exit: |
3277 | ofpbuf_delete(buffer); | |
3278 | return 0; | |
064af421 BP |
3279 | } |
3280 | ||
3281 | static void | |
3282 | update_port_config(struct ofproto *p, struct ofport *port, | |
3283 | uint32_t config, uint32_t mask) | |
3284 | { | |
3285 | mask &= config ^ port->opp.config; | |
3286 | if (mask & OFPPC_PORT_DOWN) { | |
3287 | if (config & OFPPC_PORT_DOWN) { | |
3288 | netdev_turn_flags_off(port->netdev, NETDEV_UP, true); | |
3289 | } else { | |
3290 | netdev_turn_flags_on(port->netdev, NETDEV_UP, true); | |
3291 | } | |
3292 | } | |
f1588b1f BP |
3293 | #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \ |
3294 | OFPPC_NO_FWD | OFPPC_NO_FLOOD) | |
064af421 BP |
3295 | if (mask & REVALIDATE_BITS) { |
3296 | COVERAGE_INC(ofproto_costly_flags); | |
3297 | port->opp.config ^= mask & REVALIDATE_BITS; | |
3298 | p->need_revalidate = true; | |
3299 | } | |
3300 | #undef REVALIDATE_BITS | |
064af421 BP |
3301 | if (mask & OFPPC_NO_PACKET_IN) { |
3302 | port->opp.config ^= OFPPC_NO_PACKET_IN; | |
3303 | } | |
3304 | } | |
3305 | ||
3306 | static int | |
d1e2cf21 | 3307 | handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh) |
064af421 | 3308 | { |
3269c562 | 3309 | struct ofproto *p = ofconn->ofproto; |
d1e2cf21 | 3310 | const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh; |
064af421 BP |
3311 | struct ofport *port; |
3312 | int error; | |
3313 | ||
2228b50d | 3314 | error = reject_slave_controller(ofconn, "OFPT_PORT_MOD"); |
9deba63b BP |
3315 | if (error) { |
3316 | return error; | |
3317 | } | |
064af421 | 3318 | |
ca0f572c | 3319 | port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no))); |
064af421 BP |
3320 | if (!port) { |
3321 | return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT); | |
3322 | } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) { | |
3323 | return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR); | |
3324 | } else { | |
3325 | update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask)); | |
3326 | if (opm->advertise) { | |
3327 | netdev_set_advertisements(port->netdev, ntohl(opm->advertise)); | |
3328 | } | |
3329 | } | |
3330 | return 0; | |
3331 | } | |
3332 | ||
3333 | static struct ofpbuf * | |
06a5e131 | 3334 | make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len) |
064af421 BP |
3335 | { |
3336 | struct ofp_stats_reply *osr; | |
3337 | struct ofpbuf *msg; | |
3338 | ||
3339 | msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX)); | |
3340 | osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg); | |
3341 | osr->type = type; | |
3342 | osr->flags = htons(0); | |
3343 | return msg; | |
3344 | } | |
3345 | ||
3346 | static struct ofpbuf * | |
d1e2cf21 | 3347 | start_ofp_stats_reply(const struct ofp_header *request, size_t body_len) |
064af421 | 3348 | { |
d1e2cf21 BP |
3349 | const struct ofp_stats_request *osr |
3350 | = (const struct ofp_stats_request *) request; | |
3351 | return make_ofp_stats_reply(osr->header.xid, osr->type, body_len); | |
064af421 BP |
3352 | } |
3353 | ||
3354 | static void * | |
06a5e131 BP |
3355 | append_ofp_stats_reply(size_t nbytes, struct ofconn *ofconn, |
3356 | struct ofpbuf **msgp) | |
064af421 BP |
3357 | { |
3358 | struct ofpbuf *msg = *msgp; | |
3359 | assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply)); | |
3360 | if (nbytes + msg->size > UINT16_MAX) { | |
3361 | struct ofp_stats_reply *reply = msg->data; | |
3362 | reply->flags = htons(OFPSF_REPLY_MORE); | |
06a5e131 | 3363 | *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes); |
064af421 BP |
3364 | queue_tx(msg, ofconn, ofconn->reply_counter); |
3365 | } | |
3366 | return ofpbuf_put_uninit(*msgp, nbytes); | |
3367 | } | |
3368 | ||
09246b99 BP |
3369 | static struct ofpbuf * |
3370 | make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len) | |
3371 | { | |
3372 | struct nicira_stats_msg *nsm; | |
3373 | struct ofpbuf *msg; | |
3374 | ||
3375 | msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX)); | |
3376 | nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg); | |
3377 | nsm->type = htons(OFPST_VENDOR); | |
3378 | nsm->flags = htons(0); | |
3379 | nsm->vendor = htonl(NX_VENDOR_ID); | |
d5f2379b | 3380 | nsm->subtype = subtype; |
09246b99 BP |
3381 | return msg; |
3382 | } | |
3383 | ||
3384 | static struct ofpbuf * | |
3385 | start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len) | |
3386 | { | |
3387 | return make_nxstats_reply(request->header.xid, request->subtype, body_len); | |
3388 | } | |
3389 | ||
3390 | static void | |
3391 | append_nxstats_reply(size_t nbytes, struct ofconn *ofconn, | |
3392 | struct ofpbuf **msgp) | |
3393 | { | |
3394 | struct ofpbuf *msg = *msgp; | |
3395 | assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg)); | |
3396 | if (nbytes + msg->size > UINT16_MAX) { | |
3397 | struct nicira_stats_msg *reply = msg->data; | |
3398 | reply->flags = htons(OFPSF_REPLY_MORE); | |
3399 | *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes); | |
3400 | queue_tx(msg, ofconn, ofconn->reply_counter); | |
3401 | } | |
3402 | ofpbuf_prealloc_tailroom(*msgp, nbytes); | |
3403 | } | |
3404 | ||
064af421 | 3405 | static int |
3269c562 | 3406 | handle_desc_stats_request(struct ofconn *ofconn, |
d1e2cf21 | 3407 | const struct ofp_header *request) |
064af421 | 3408 | { |
3269c562 | 3409 | struct ofproto *p = ofconn->ofproto; |
064af421 BP |
3410 | struct ofp_desc_stats *ods; |
3411 | struct ofpbuf *msg; | |
3412 | ||
06a5e131 BP |
3413 | msg = start_ofp_stats_reply(request, sizeof *ods); |
3414 | ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg); | |
5a719c38 JP |
3415 | memset(ods, 0, sizeof *ods); |
3416 | ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc); | |
3417 | ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc); | |
3418 | ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc); | |
3419 | ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num); | |
3420 | ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc); | |
064af421 BP |
3421 | queue_tx(msg, ofconn, ofconn->reply_counter); |
3422 | ||
3423 | return 0; | |
3424 | } | |
3425 | ||
064af421 | 3426 | static int |
3269c562 | 3427 | handle_table_stats_request(struct ofconn *ofconn, |
d1e2cf21 | 3428 | const struct ofp_header *request) |
064af421 | 3429 | { |
3269c562 | 3430 | struct ofproto *p = ofconn->ofproto; |
064af421 BP |
3431 | struct ofp_table_stats *ots; |
3432 | struct ofpbuf *msg; | |
064af421 | 3433 | |
06a5e131 | 3434 | msg = start_ofp_stats_reply(request, sizeof *ots * 2); |
064af421 | 3435 | |
064af421 | 3436 | /* Classifier table. */ |
06a5e131 | 3437 | ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg); |
064af421 | 3438 | memset(ots, 0, sizeof *ots); |
064af421 | 3439 | strcpy(ots->name, "classifier"); |
b70eac89 | 3440 | ots->wildcards = (ofconn->flow_format == NXFF_OPENFLOW10 |
f9bfea14 | 3441 | ? htonl(OFPFW_ALL) : htonl(OVSFW_ALL)); |
ad828225 | 3442 | ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */ |
bcf84111 | 3443 | ots->active_count = htonl(classifier_count(&p->cls)); |
c4617b3c BP |
3444 | put_32aligned_be64(&ots->lookup_count, htonll(0)); /* XXX */ |
3445 | put_32aligned_be64(&ots->matched_count, htonll(0)); /* XXX */ | |
064af421 BP |
3446 | |
3447 | queue_tx(msg, ofconn, ofconn->reply_counter); | |
3448 | return 0; | |
3449 | } | |
3450 | ||
abaad8cf | 3451 | static void |
ca0f572c | 3452 | append_port_stat(struct ofport *port, struct ofconn *ofconn, |
a4948b95 | 3453 | struct ofpbuf **msgp) |
abaad8cf JP |
3454 | { |
3455 | struct netdev_stats stats; | |
3456 | struct ofp_port_stats *ops; | |
3457 | ||
d295e8e9 JP |
3458 | /* Intentionally ignore return value, since errors will set |
3459 | * 'stats' to all-1s, which is correct for OpenFlow, and | |
abaad8cf JP |
3460 | * netdev_get_stats() will log errors. */ |
3461 | netdev_get_stats(port->netdev, &stats); | |
3462 | ||
06a5e131 | 3463 | ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp); |
ca0f572c | 3464 | ops->port_no = htons(port->opp.port_no); |
abaad8cf | 3465 | memset(ops->pad, 0, sizeof ops->pad); |
c4617b3c BP |
3466 | put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets)); |
3467 | put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets)); | |
3468 | put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes)); | |
3469 | put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes)); | |
3470 | put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped)); | |
3471 | put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped)); | |
3472 | put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors)); | |
3473 | put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors)); | |
3474 | put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors)); | |
3475 | put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors)); | |
3476 | put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors)); | |
3477 | put_32aligned_be64(&ops->collisions, htonll(stats.collisions)); | |
abaad8cf JP |
3478 | } |
3479 | ||
064af421 | 3480 | static int |
d1e2cf21 | 3481 | handle_port_stats_request(struct ofconn *ofconn, const struct ofp_header *oh) |
064af421 | 3482 | { |
3269c562 | 3483 | struct ofproto *p = ofconn->ofproto; |
d1e2cf21 | 3484 | const struct ofp_port_stats_request *psr = ofputil_stats_body(oh); |
064af421 BP |
3485 | struct ofp_port_stats *ops; |
3486 | struct ofpbuf *msg; | |
3487 | struct ofport *port; | |
064af421 | 3488 | |
d1e2cf21 | 3489 | msg = start_ofp_stats_reply(oh, sizeof *ops * 16); |
abaad8cf | 3490 | if (psr->port_no != htons(OFPP_NONE)) { |
ca0f572c | 3491 | port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no))); |
abaad8cf | 3492 | if (port) { |
ca0f572c | 3493 | append_port_stat(port, ofconn, &msg); |
abaad8cf JP |
3494 | } |
3495 | } else { | |
4e8e4213 | 3496 | HMAP_FOR_EACH (port, hmap_node, &p->ports) { |
ca0f572c | 3497 | append_port_stat(port, ofconn, &msg); |
abaad8cf | 3498 | } |
064af421 BP |
3499 | } |
3500 | ||
3501 | queue_tx(msg, ofconn, ofconn->reply_counter); | |
3502 | return 0; | |
3503 | } | |
3504 | ||
c6ebb8fb BP |
3505 | static void |
3506 | calc_flow_duration(long long int start, ovs_be32 *sec, ovs_be32 *nsec) | |
3507 | { | |
3508 | long long int msecs = time_msec() - start; | |
3509 | *sec = htonl(msecs / 1000); | |
3510 | *nsec = htonl((msecs % 1000) * (1000 * 1000)); | |
3511 | } | |
3512 | ||
064af421 | 3513 | static void |
5ecc9d81 BP |
3514 | put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule, |
3515 | ovs_be16 out_port, struct ofpbuf **replyp) | |
064af421 | 3516 | { |
064af421 BP |
3517 | struct ofp_flow_stats *ofs; |
3518 | uint64_t packet_count, byte_count; | |
c4617b3c | 3519 | ovs_be64 cookie; |
064af421 BP |
3520 | size_t act_len, len; |
3521 | ||
5ecc9d81 | 3522 | if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) { |
064af421 BP |
3523 | return; |
3524 | } | |
3525 | ||
3526 | act_len = sizeof *rule->actions * rule->n_actions; | |
3527 | len = offsetof(struct ofp_flow_stats, actions) + act_len; | |
3528 | ||
3394b5b6 | 3529 | rule_get_stats(rule, &packet_count, &byte_count); |
064af421 | 3530 | |
5ecc9d81 | 3531 | ofs = append_ofp_stats_reply(len, ofconn, replyp); |
064af421 | 3532 | ofs->length = htons(len); |
ad828225 | 3533 | ofs->table_id = 0; |
064af421 | 3534 | ofs->pad = 0; |
ff9d3826 | 3535 | ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofs->match, |
c4617b3c BP |
3536 | rule->flow_cookie, &cookie); |
3537 | put_32aligned_be64(&ofs->cookie, cookie); | |
c6ebb8fb | 3538 | calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec); |
064af421 BP |
3539 | ofs->priority = htons(rule->cr.priority); |
3540 | ofs->idle_timeout = htons(rule->idle_timeout); | |
3541 | ofs->hard_timeout = htons(rule->hard_timeout); | |
39997502 | 3542 | memset(ofs->pad2, 0, sizeof ofs->pad2); |
c4617b3c BP |
3543 | put_32aligned_be64(&ofs->packet_count, htonll(packet_count)); |
3544 | put_32aligned_be64(&ofs->byte_count, htonll(byte_count)); | |
3dffcf07 BP |
3545 | if (rule->n_actions > 0) { |
3546 | memcpy(ofs->actions, rule->actions, act_len); | |
3547 | } | |
064af421 BP |
3548 | } |
3549 | ||
3c4486a5 BP |
3550 | static bool |
3551 | is_valid_table(uint8_t table_id) | |
064af421 | 3552 | { |
a02e5331 BP |
3553 | if (table_id == 0 || table_id == 0xff) { |
3554 | return true; | |
3555 | } else { | |
3556 | /* It would probably be better to reply with an error but there doesn't | |
3557 | * seem to be any appropriate value, so that might just be | |
3558 | * confusing. */ | |
3559 | VLOG_WARN_RL(&rl, "controller asked for invalid table %"PRIu8, | |
3560 | table_id); | |
3561 | return false; | |
3562 | } | |
064af421 BP |
3563 | } |
3564 | ||
3565 | static int | |
d1e2cf21 | 3566 | handle_flow_stats_request(struct ofconn *ofconn, const struct ofp_header *oh) |
064af421 | 3567 | { |
d1e2cf21 | 3568 | const struct ofp_flow_stats_request *fsr = ofputil_stats_body(oh); |
5ecc9d81 | 3569 | struct ofpbuf *reply; |
064af421 | 3570 | |
064af421 | 3571 | COVERAGE_INC(ofproto_flows_req); |
d1e2cf21 | 3572 | reply = start_ofp_stats_reply(oh, 1024); |
3c4486a5 | 3573 | if (is_valid_table(fsr->table_id)) { |
5ecc9d81 | 3574 | struct cls_cursor cursor; |
3c4486a5 | 3575 | struct cls_rule target; |
5ecc9d81 | 3576 | struct rule *rule; |
3c4486a5 | 3577 | |
d8ae4d67 BP |
3578 | ofputil_cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0, |
3579 | &target); | |
5ecc9d81 BP |
3580 | cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target); |
3581 | CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { | |
3582 | put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply); | |
3583 | } | |
3c4486a5 | 3584 | } |
5ecc9d81 | 3585 | queue_tx(reply, ofconn, ofconn->reply_counter); |
3c4486a5 | 3586 | |
064af421 BP |
3587 | return 0; |
3588 | } | |
3589 | ||
09246b99 | 3590 | static void |
5ecc9d81 BP |
3591 | put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule, |
3592 | ovs_be16 out_port, struct ofpbuf **replyp) | |
09246b99 | 3593 | { |
09246b99 BP |
3594 | struct nx_flow_stats *nfs; |
3595 | uint64_t packet_count, byte_count; | |
3596 | size_t act_len, start_len; | |
5ecc9d81 | 3597 | struct ofpbuf *reply; |
09246b99 | 3598 | |
5ecc9d81 | 3599 | if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) { |
09246b99 BP |
3600 | return; |
3601 | } | |
3602 | ||
3394b5b6 | 3603 | rule_get_stats(rule, &packet_count, &byte_count); |
09246b99 BP |
3604 | |
3605 | act_len = sizeof *rule->actions * rule->n_actions; | |
3606 | ||
5ecc9d81 | 3607 | append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, ofconn, replyp); |
1dfee98d | 3608 | start_len = (*replyp)->size; |
5ecc9d81 BP |
3609 | reply = *replyp; |
3610 | ||
3611 | nfs = ofpbuf_put_uninit(reply, sizeof *nfs); | |
09246b99 BP |
3612 | nfs->table_id = 0; |
3613 | nfs->pad = 0; | |
3614 | calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec); | |
3615 | nfs->cookie = rule->flow_cookie; | |
3616 | nfs->priority = htons(rule->cr.priority); | |
3617 | nfs->idle_timeout = htons(rule->idle_timeout); | |
3618 | nfs->hard_timeout = htons(rule->hard_timeout); | |
5ecc9d81 | 3619 | nfs->match_len = htons(nx_put_match(reply, &rule->cr)); |
09246b99 BP |
3620 | memset(nfs->pad2, 0, sizeof nfs->pad2); |
3621 | nfs->packet_count = htonll(packet_count); | |
3622 | nfs->byte_count = htonll(byte_count); | |
3623 | if (rule->n_actions > 0) { | |
5ecc9d81 | 3624 | ofpbuf_put(reply, rule->actions, act_len); |
09246b99 | 3625 | } |
5ecc9d81 | 3626 | nfs->length = htons(reply->size - start_len); |
09246b99 BP |
3627 | } |
3628 | ||
3629 | static int | |
d1e2cf21 | 3630 | handle_nxst_flow(struct ofconn *ofconn, const struct ofp_header *oh) |
09246b99 BP |
3631 | { |
3632 | struct nx_flow_stats_request *nfsr; | |
09246b99 | 3633 | struct cls_rule target; |
5ecc9d81 | 3634 | struct ofpbuf *reply; |
d1e2cf21 | 3635 | struct ofpbuf b; |
09246b99 BP |
3636 | int error; |
3637 | ||
0bc9407d | 3638 | ofpbuf_use_const(&b, oh, ntohs(oh->length)); |
d1e2cf21 | 3639 | |
09246b99 | 3640 | /* Dissect the message. */ |
bbc32a88 | 3641 | nfsr = ofpbuf_pull(&b, sizeof *nfsr); |
d1e2cf21 | 3642 | error = nx_pull_match(&b, ntohs(nfsr->match_len), 0, &target); |
09246b99 BP |
3643 | if (error) { |
3644 | return error; | |
3645 | } | |
d1e2cf21 BP |
3646 | if (b.size) { |
3647 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
3648 | } | |
09246b99 BP |
3649 | |
3650 | COVERAGE_INC(ofproto_flows_req); | |
5ecc9d81 | 3651 | reply = start_nxstats_reply(&nfsr->nsm, 1024); |
3c4486a5 | 3652 | if (is_valid_table(nfsr->table_id)) { |
5ecc9d81 BP |
3653 | struct cls_cursor cursor; |
3654 | struct rule *rule; | |
3655 | ||
3656 | cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target); | |
3657 | CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { | |
3658 | put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply); | |
3659 | } | |
3c4486a5 | 3660 | } |
5ecc9d81 BP |
3661 | queue_tx(reply, ofconn, ofconn->reply_counter); |
3662 | ||
09246b99 BP |
3663 | return 0; |
3664 | } | |
3665 | ||
4f2cad2c | 3666 | static void |
3394b5b6 | 3667 | flow_stats_ds(struct rule *rule, struct ds *results) |
4f2cad2c | 3668 | { |
4f2cad2c JP |
3669 | uint64_t packet_count, byte_count; |
3670 | size_t act_len = sizeof *rule->actions * rule->n_actions; | |
3671 | ||
3394b5b6 | 3672 | rule_get_stats(rule, &packet_count, &byte_count); |
4f2cad2c JP |
3673 | |
3674 | ds_put_format(results, "duration=%llds, ", | |
3675 | (time_msec() - rule->created) / 1000); | |
3693a563 | 3676 | ds_put_format(results, "idle=%.3fs, ", (time_msec() - rule->used) / 1000.0); |
52ae00b3 | 3677 | ds_put_format(results, "priority=%u, ", rule->cr.priority); |
4f2cad2c JP |
3678 | ds_put_format(results, "n_packets=%"PRIu64", ", packet_count); |
3679 | ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count); | |
cb833cf6 | 3680 | cls_rule_format(&rule->cr, results); |
a5df0e72 | 3681 | ds_put_char(results, ','); |
3dffcf07 BP |
3682 | if (act_len > 0) { |
3683 | ofp_print_actions(results, &rule->actions->header, act_len); | |
3c8552c1 JP |
3684 | } else { |
3685 | ds_put_cstr(results, "drop"); | |
3dffcf07 | 3686 | } |
4f2cad2c JP |
3687 | ds_put_cstr(results, "\n"); |
3688 | } | |
3689 | ||
d295e8e9 | 3690 | /* Adds a pretty-printed description of all flows to 'results', including |
4f2cad2c JP |
3691 | * those marked hidden by secchan (e.g., by in-band control). */ |
3692 | void | |
3693 | ofproto_get_all_flows(struct ofproto *p, struct ds *results) | |
3694 | { | |
5ecc9d81 BP |
3695 | struct cls_cursor cursor; |
3696 | struct rule *rule; | |
064af421 | 3697 | |
5ecc9d81 BP |
3698 | cls_cursor_init(&cursor, &p->cls, NULL); |
3699 | CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { | |
3394b5b6 | 3700 | flow_stats_ds(rule, results); |
064af421 | 3701 | } |
064af421 BP |
3702 | } |
3703 | ||
27d34fce BP |
3704 | static void |
3705 | query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target, | |
734bbeb4 | 3706 | ovs_be16 out_port, uint8_t table_id, |
27d34fce BP |
3707 | struct ofp_aggregate_stats_reply *oasr) |
3708 | { | |
5ecc9d81 BP |
3709 | uint64_t total_packets = 0; |
3710 | uint64_t total_bytes = 0; | |
3711 | int n_flows = 0; | |
27d34fce BP |
3712 | |
3713 | COVERAGE_INC(ofproto_agg_request); | |
5ecc9d81 | 3714 | |
3c4486a5 | 3715 | if (is_valid_table(table_id)) { |
5ecc9d81 BP |
3716 | struct cls_cursor cursor; |
3717 | struct rule *rule; | |
3c4486a5 | 3718 | |
5ecc9d81 BP |
3719 | cls_cursor_init(&cursor, &ofproto->cls, target); |
3720 | CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { | |
3721 | if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) { | |
3722 | uint64_t packet_count; | |
3723 | uint64_t byte_count; | |
3724 | ||
3394b5b6 | 3725 | rule_get_stats(rule, &packet_count, &byte_count); |
5ecc9d81 BP |
3726 | |
3727 | total_packets += packet_count; | |
3728 | total_bytes += byte_count; | |
3729 | n_flows++; | |
3730 | } | |
3731 | } | |
3c4486a5 | 3732 | } |
27d34fce | 3733 | |
5ecc9d81 | 3734 | oasr->flow_count = htonl(n_flows); |
c4617b3c BP |
3735 | put_32aligned_be64(&oasr->packet_count, htonll(total_packets)); |
3736 | put_32aligned_be64(&oasr->byte_count, htonll(total_bytes)); | |
27d34fce BP |
3737 | memset(oasr->pad, 0, sizeof oasr->pad); |
3738 | } | |
3739 | ||
064af421 | 3740 | static int |
3269c562 | 3741 | handle_aggregate_stats_request(struct ofconn *ofconn, |
d1e2cf21 | 3742 | const struct ofp_header *oh) |
064af421 | 3743 | { |
d1e2cf21 | 3744 | const struct ofp_aggregate_stats_request *request = ofputil_stats_body(oh); |
064af421 | 3745 | struct ofp_aggregate_stats_reply *reply; |
064af421 BP |
3746 | struct cls_rule target; |
3747 | struct ofpbuf *msg; | |
3748 | ||
d8ae4d67 BP |
3749 | ofputil_cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0, |
3750 | &target); | |
064af421 | 3751 | |
d1e2cf21 | 3752 | msg = start_ofp_stats_reply(oh, sizeof *reply); |
06a5e131 | 3753 | reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg); |
27d34fce BP |
3754 | query_aggregate_stats(ofconn->ofproto, &target, request->out_port, |
3755 | request->table_id, reply); | |
064af421 BP |
3756 | queue_tx(msg, ofconn, ofconn->reply_counter); |
3757 | return 0; | |
3758 | } | |
3759 | ||
09246b99 | 3760 | static int |
d1e2cf21 | 3761 | handle_nxst_aggregate(struct ofconn *ofconn, const struct ofp_header *oh) |
09246b99 BP |
3762 | { |
3763 | struct nx_aggregate_stats_request *request; | |
3764 | struct ofp_aggregate_stats_reply *reply; | |
3765 | struct cls_rule target; | |
d1e2cf21 | 3766 | struct ofpbuf b; |
09246b99 BP |
3767 | struct ofpbuf *buf; |
3768 | int error; | |
3769 | ||
0bc9407d | 3770 | ofpbuf_use_const(&b, oh, ntohs(oh->length)); |
d1e2cf21 | 3771 | |
09246b99 | 3772 | /* Dissect the message. */ |
bbc32a88 | 3773 | request = ofpbuf_pull(&b, sizeof *request); |
d1e2cf21 | 3774 | error = nx_pull_match(&b, ntohs(request->match_len), 0, &target); |
09246b99 BP |
3775 | if (error) { |
3776 | return error; | |
3777 | } | |
d1e2cf21 BP |
3778 | if (b.size) { |
3779 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
3780 | } | |
09246b99 BP |
3781 | |
3782 | /* Reply. */ | |
3783 | COVERAGE_INC(ofproto_flows_req); | |
3784 | buf = start_nxstats_reply(&request->nsm, sizeof *reply); | |
3785 | reply = ofpbuf_put_uninit(buf, sizeof *reply); | |
3786 | query_aggregate_stats(ofconn->ofproto, &target, request->out_port, | |
3787 | request->table_id, reply); | |
3788 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
3789 | ||
3790 | return 0; | |
3791 | } | |
3792 | ||
c1c9c9c4 BP |
3793 | struct queue_stats_cbdata { |
3794 | struct ofconn *ofconn; | |
ca0f572c | 3795 | struct ofport *ofport; |
c1c9c9c4 | 3796 | struct ofpbuf *msg; |
c1c9c9c4 BP |
3797 | }; |
3798 | ||
3799 | static void | |
db9220c3 | 3800 | put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id, |
c1c9c9c4 BP |
3801 | const struct netdev_queue_stats *stats) |
3802 | { | |
3803 | struct ofp_queue_stats *reply; | |
3804 | ||
06a5e131 | 3805 | reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg); |
ca0f572c | 3806 | reply->port_no = htons(cbdata->ofport->opp.port_no); |
c1c9c9c4 BP |
3807 | memset(reply->pad, 0, sizeof reply->pad); |
3808 | reply->queue_id = htonl(queue_id); | |
c4617b3c BP |
3809 | put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes)); |
3810 | put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets)); | |
3811 | put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors)); | |
c1c9c9c4 BP |
3812 | } |
3813 | ||
3814 | static void | |
db9220c3 | 3815 | handle_queue_stats_dump_cb(uint32_t queue_id, |
c1c9c9c4 BP |
3816 | struct netdev_queue_stats *stats, |
3817 | void *cbdata_) | |
3818 | { | |
3819 | struct queue_stats_cbdata *cbdata = cbdata_; | |
3820 | ||
3821 | put_queue_stats(cbdata, queue_id, stats); | |
3822 | } | |
3823 | ||
3824 | static void | |
ca0f572c | 3825 | handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id, |
c1c9c9c4 BP |
3826 | struct queue_stats_cbdata *cbdata) |
3827 | { | |
ca0f572c | 3828 | cbdata->ofport = port; |
c1c9c9c4 BP |
3829 | if (queue_id == OFPQ_ALL) { |
3830 | netdev_dump_queue_stats(port->netdev, | |
3831 | handle_queue_stats_dump_cb, cbdata); | |
3832 | } else { | |
3833 | struct netdev_queue_stats stats; | |
3834 | ||
1ac788f6 BP |
3835 | if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) { |
3836 | put_queue_stats(cbdata, queue_id, &stats); | |
3837 | } | |
c1c9c9c4 BP |
3838 | } |
3839 | } | |
3840 | ||
3841 | static int | |
d1e2cf21 | 3842 | handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh) |
c1c9c9c4 | 3843 | { |
3269c562 | 3844 | struct ofproto *ofproto = ofconn->ofproto; |
d1e2cf21 | 3845 | const struct ofp_queue_stats_request *qsr; |
c1c9c9c4 BP |
3846 | struct queue_stats_cbdata cbdata; |
3847 | struct ofport *port; | |
3848 | unsigned int port_no; | |
3849 | uint32_t queue_id; | |
3850 | ||
d1e2cf21 BP |
3851 | qsr = ofputil_stats_body(oh); |
3852 | if (!qsr) { | |
c1c9c9c4 BP |
3853 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); |
3854 | } | |
c1c9c9c4 BP |
3855 | |
3856 | COVERAGE_INC(ofproto_queue_req); | |
3857 | ||
3858 | cbdata.ofconn = ofconn; | |
d1e2cf21 | 3859 | cbdata.msg = start_ofp_stats_reply(oh, 128); |
c1c9c9c4 BP |
3860 | |
3861 | port_no = ntohs(qsr->port_no); | |
3862 | queue_id = ntohl(qsr->queue_id); | |
3863 | if (port_no == OFPP_ALL) { | |
4e8e4213 | 3864 | HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) { |
ca0f572c | 3865 | handle_queue_stats_for_port(port, queue_id, &cbdata); |
c1c9c9c4 BP |
3866 | } |
3867 | } else if (port_no < ofproto->max_ports) { | |
ca0f572c | 3868 | port = get_port(ofproto, ofp_port_to_odp_port(port_no)); |
c1c9c9c4 | 3869 | if (port) { |
ca0f572c | 3870 | handle_queue_stats_for_port(port, queue_id, &cbdata); |
c1c9c9c4 BP |
3871 | } |
3872 | } else { | |
3873 | ofpbuf_delete(cbdata.msg); | |
3874 | return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT); | |
3875 | } | |
3876 | queue_tx(cbdata.msg, ofconn, ofconn->reply_counter); | |
3877 | ||
3878 | return 0; | |
3879 | } | |
3880 | ||
878ae780 EJ |
3881 | /* Updates 'facet''s used time. Caller is responsible for calling |
3882 | * facet_push_stats() to update the flows which 'facet' resubmits into. */ | |
064af421 | 3883 | static void |
bcf84111 | 3884 | facet_update_time(struct ofproto *ofproto, struct facet *facet, |
3394b5b6 | 3885 | long long int used) |
064af421 | 3886 | { |
bcf84111 BP |
3887 | if (used > facet->used) { |
3888 | facet->used = used; | |
3889 | if (used > facet->rule->used) { | |
3890 | facet->rule->used = used; | |
4836f9f2 | 3891 | } |
bcf84111 | 3892 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used); |
064af421 BP |
3893 | } |
3894 | } | |
3895 | ||
bcf84111 BP |
3896 | /* Folds the statistics from 'stats' into the counters in 'facet'. |
3897 | * | |
3898 | * Because of the meaning of a facet's counters, it only makes sense to do this | |
3899 | * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a | |
3900 | * packet that was sent by hand or if it represents statistics that have been | |
3901 | * cleared out of the datapath. */ | |
064af421 | 3902 | static void |
bcf84111 | 3903 | facet_update_stats(struct ofproto *ofproto, struct facet *facet, |
c97fb132 | 3904 | const struct dpif_flow_stats *stats) |
064af421 | 3905 | { |
0b13821f | 3906 | if (stats->n_packets || stats->used > facet->used) { |
3394b5b6 | 3907 | facet_update_time(ofproto, facet, stats->used); |
bcf84111 BP |
3908 | facet->packet_count += stats->n_packets; |
3909 | facet->byte_count += stats->n_bytes; | |
878ae780 | 3910 | facet_push_stats(ofproto, facet); |
bcf84111 | 3911 | netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags); |
064af421 BP |
3912 | } |
3913 | } | |
3914 | ||
878ae780 EJ |
3915 | static void |
3916 | facet_push_stats(struct ofproto *ofproto, struct facet *facet) | |
3917 | { | |
3918 | uint64_t rs_packets, rs_bytes; | |
3919 | ||
3920 | assert(facet->packet_count >= facet->rs_packet_count); | |
3921 | assert(facet->byte_count >= facet->rs_byte_count); | |
3922 | assert(facet->used >= facet->rs_used); | |
3923 | ||
3924 | rs_packets = facet->packet_count - facet->rs_packet_count; | |
3925 | rs_bytes = facet->byte_count - facet->rs_byte_count; | |
3926 | ||
3927 | if (rs_packets || rs_bytes || facet->used > facet->rs_used) { | |
3928 | facet->rs_packet_count = facet->packet_count; | |
3929 | facet->rs_byte_count = facet->byte_count; | |
3930 | facet->rs_used = facet->used; | |
3931 | ||
3932 | flow_push_stats(ofproto, facet->rule, &facet->flow, | |
3933 | rs_packets, rs_bytes, facet->used); | |
3934 | } | |
3935 | } | |
3936 | ||
3937 | struct ofproto_push { | |
3938 | struct action_xlate_ctx ctx; | |
3939 | uint64_t packets; | |
3940 | uint64_t bytes; | |
3941 | long long int used; | |
3942 | }; | |
3943 | ||
3944 | static void | |
3945 | push_resubmit(struct action_xlate_ctx *ctx, struct rule *rule) | |
3946 | { | |
3947 | struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx); | |
3948 | ||
3949 | if (rule) { | |
3950 | rule->packet_count += push->packets; | |
3951 | rule->byte_count += push->bytes; | |
3952 | rule->used = MAX(push->used, rule->used); | |
3953 | } | |
3954 | } | |
3955 | ||
3956 | /* Pushes flow statistics to the rules which 'flow' resubmits into given | |
3957 | * 'rule''s actions. */ | |
3958 | static void | |
3959 | flow_push_stats(struct ofproto *ofproto, const struct rule *rule, | |
3960 | struct flow *flow, uint64_t packets, uint64_t bytes, | |
3961 | long long int used) | |
3962 | { | |
3963 | struct ofproto_push push; | |
3964 | ||
3965 | push.packets = packets; | |
3966 | push.bytes = bytes; | |
3967 | push.used = used; | |
3968 | ||
3969 | action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL); | |
3970 | push.ctx.resubmit_hook = push_resubmit; | |
3971 | ofpbuf_delete(xlate_actions(&push.ctx, rule->actions, rule->n_actions)); | |
3972 | } | |
3973 | ||
79eee1eb BP |
3974 | /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT |
3975 | * in which no matching flow already exists in the flow table. | |
3976 | * | |
3977 | * Adds the flow specified by 'ofm', which is followed by 'n_actions' | |
3269c562 BP |
3978 | * ofp_actions, to ofconn->ofproto's flow table. Returns 0 on success or an |
3979 | * OpenFlow error code as encoded by ofp_mkerr() on failure. | |
79eee1eb BP |
3980 | * |
3981 | * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, | |
3982 | * if any. */ | |
064af421 | 3983 | static int |
3052b0c5 | 3984 | add_flow(struct ofconn *ofconn, struct flow_mod *fm) |
064af421 | 3985 | { |
3269c562 | 3986 | struct ofproto *p = ofconn->ofproto; |
064af421 BP |
3987 | struct ofpbuf *packet; |
3988 | struct rule *rule; | |
3989 | uint16_t in_port; | |
3990 | int error; | |
3991 | ||
3052b0c5 BP |
3992 | if (fm->flags & OFPFF_CHECK_OVERLAP |
3993 | && classifier_rule_overlaps(&p->cls, &fm->cr)) { | |
3994 | return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP); | |
49bdc010 JP |
3995 | } |
3996 | ||
064af421 | 3997 | error = 0; |
3052b0c5 BP |
3998 | if (fm->buffer_id != UINT32_MAX) { |
3999 | error = pktbuf_retrieve(ofconn->pktbuf, fm->buffer_id, | |
064af421 | 4000 | &packet, &in_port); |
212fe71c BP |
4001 | } else { |
4002 | packet = NULL; | |
165cd8a3 | 4003 | in_port = UINT16_MAX; |
064af421 BP |
4004 | } |
4005 | ||
bcf84111 BP |
4006 | rule = rule_create(&fm->cr, fm->actions, fm->n_actions, |
4007 | fm->idle_timeout, fm->hard_timeout, fm->cookie, | |
4008 | fm->flags & OFPFF_SEND_FLOW_REM); | |
afe75089 BP |
4009 | rule_insert(p, rule); |
4010 | if (packet) { | |
4011 | rule_execute(p, rule, in_port, packet); | |
4012 | } | |
064af421 BP |
4013 | return error; |
4014 | } | |
4015 | ||
79eee1eb | 4016 | static struct rule * |
3052b0c5 | 4017 | find_flow_strict(struct ofproto *p, const struct flow_mod *fm) |
064af421 | 4018 | { |
3052b0c5 | 4019 | return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr)); |
79eee1eb | 4020 | } |
064af421 | 4021 | |
79eee1eb | 4022 | static int |
3269c562 | 4023 | send_buffered_packet(struct ofconn *ofconn, |
3052b0c5 | 4024 | struct rule *rule, uint32_t buffer_id) |
79eee1eb BP |
4025 | { |
4026 | struct ofpbuf *packet; | |
4027 | uint16_t in_port; | |
79eee1eb | 4028 | int error; |
064af421 | 4029 | |
3052b0c5 | 4030 | if (buffer_id == UINT32_MAX) { |
79eee1eb | 4031 | return 0; |
064af421 | 4032 | } |
79eee1eb | 4033 | |
3052b0c5 | 4034 | error = pktbuf_retrieve(ofconn->pktbuf, buffer_id, &packet, &in_port); |
79eee1eb BP |
4035 | if (error) { |
4036 | return error; | |
4037 | } | |
4038 | ||
bcf84111 | 4039 | rule_execute(ofconn->ofproto, rule, in_port, packet); |
79eee1eb | 4040 | |
064af421 BP |
4041 | return 0; |
4042 | } | |
79eee1eb BP |
4043 | \f |
4044 | /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */ | |
064af421 BP |
4045 | |
4046 | struct modify_flows_cbdata { | |
4047 | struct ofproto *ofproto; | |
3052b0c5 | 4048 | const struct flow_mod *fm; |
79eee1eb | 4049 | struct rule *match; |
064af421 BP |
4050 | }; |
4051 | ||
3052b0c5 BP |
4052 | static int modify_flow(struct ofproto *, const struct flow_mod *, |
4053 | struct rule *); | |
79eee1eb BP |
4054 | |
4055 | /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as | |
4056 | * encoded by ofp_mkerr() on failure. | |
4057 | * | |
4058 | * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, | |
4059 | * if any. */ | |
4060 | static int | |
3052b0c5 | 4061 | modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm) |
79eee1eb | 4062 | { |
5ecc9d81 BP |
4063 | struct ofproto *p = ofconn->ofproto; |
4064 | struct rule *match = NULL; | |
4065 | struct cls_cursor cursor; | |
4066 | struct rule *rule; | |
79eee1eb | 4067 | |
5ecc9d81 BP |
4068 | cls_cursor_init(&cursor, &p->cls, &fm->cr); |
4069 | CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { | |
4070 | if (!rule_is_hidden(rule)) { | |
4071 | match = rule; | |
4072 | modify_flow(p, fm, rule); | |
4073 | } | |
4074 | } | |
79eee1eb | 4075 | |
5ecc9d81 | 4076 | if (match) { |
d6302b0f BP |
4077 | /* This credits the packet to whichever flow happened to match last. |
4078 | * That's weird. Maybe we should do a lookup for the flow that | |
4079 | * actually matches the packet? Who knows. */ | |
5ecc9d81 | 4080 | send_buffered_packet(ofconn, match, fm->buffer_id); |
79eee1eb BP |
4081 | return 0; |
4082 | } else { | |
3052b0c5 | 4083 | return add_flow(ofconn, fm); |
79eee1eb BP |
4084 | } |
4085 | } | |
4086 | ||
4087 | /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error | |
4088 | * code as encoded by ofp_mkerr() on failure. | |
4089 | * | |
4090 | * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, | |
4091 | * if any. */ | |
4092 | static int | |
3052b0c5 | 4093 | modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm) |
79eee1eb | 4094 | { |
3052b0c5 BP |
4095 | struct ofproto *p = ofconn->ofproto; |
4096 | struct rule *rule = find_flow_strict(p, fm); | |
79eee1eb | 4097 | if (rule && !rule_is_hidden(rule)) { |
3052b0c5 BP |
4098 | modify_flow(p, fm, rule); |
4099 | return send_buffered_packet(ofconn, rule, fm->buffer_id); | |
79eee1eb | 4100 | } else { |
3052b0c5 | 4101 | return add_flow(ofconn, fm); |
79eee1eb BP |
4102 | } |
4103 | } | |
4104 | ||
79eee1eb BP |
4105 | /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has |
4106 | * been identified as a flow in 'p''s flow table to be modified, by changing | |
4107 | * the rule's actions to match those in 'ofm' (which is followed by 'n_actions' | |
4108 | * ofp_action[] structures). */ | |
064af421 | 4109 | static int |
3052b0c5 | 4110 | modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule) |
064af421 | 4111 | { |
3052b0c5 | 4112 | size_t actions_len = fm->n_actions * sizeof *rule->actions; |
79eee1eb | 4113 | |
3052b0c5 | 4114 | rule->flow_cookie = fm->cookie; |
79eee1eb BP |
4115 | |
4116 | /* If the actions are the same, do nothing. */ | |
3052b0c5 BP |
4117 | if (fm->n_actions == rule->n_actions |
4118 | && (!fm->n_actions | |
4119 | || !memcmp(fm->actions, rule->actions, actions_len))) { | |
79eee1eb BP |
4120 | return 0; |
4121 | } | |
4122 | ||
4123 | /* Replace actions. */ | |
4124 | free(rule->actions); | |
3052b0c5 BP |
4125 | rule->actions = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL; |
4126 | rule->n_actions = fm->n_actions; | |
79eee1eb | 4127 | |
bcf84111 | 4128 | p->need_revalidate = true; |
79eee1eb BP |
4129 | |
4130 | return 0; | |
4131 | } | |
4132 | \f | |
4133 | /* OFPFC_DELETE implementation. */ | |
4134 | ||
8054fc48 | 4135 | static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port); |
79eee1eb BP |
4136 | |
4137 | /* Implements OFPFC_DELETE. */ | |
4138 | static void | |
3052b0c5 | 4139 | delete_flows_loose(struct ofproto *p, const struct flow_mod *fm) |
79eee1eb | 4140 | { |
5ecc9d81 BP |
4141 | struct rule *rule, *next_rule; |
4142 | struct cls_cursor cursor; | |
064af421 | 4143 | |
5ecc9d81 BP |
4144 | cls_cursor_init(&cursor, &p->cls, &fm->cr); |
4145 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) { | |
4146 | delete_flow(p, rule, htons(fm->out_port)); | |
4147 | } | |
064af421 BP |
4148 | } |
4149 | ||
79eee1eb BP |
4150 | /* Implements OFPFC_DELETE_STRICT. */ |
4151 | static void | |
3052b0c5 | 4152 | delete_flow_strict(struct ofproto *p, struct flow_mod *fm) |
79eee1eb | 4153 | { |
3052b0c5 | 4154 | struct rule *rule = find_flow_strict(p, fm); |
79eee1eb | 4155 | if (rule) { |
3052b0c5 | 4156 | delete_flow(p, rule, htons(fm->out_port)); |
79eee1eb BP |
4157 | } |
4158 | } | |
4159 | ||
79eee1eb BP |
4160 | /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has |
4161 | * been identified as a flow to delete from 'p''s flow table, by deleting the | |
4162 | * flow and sending out a OFPT_FLOW_REMOVED message to any interested | |
4163 | * controller. | |
4164 | * | |
4165 | * Will not delete 'rule' if it is hidden. Will delete 'rule' only if | |
4166 | * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the | |
4167 | * specified 'out_port'. */ | |
4168 | static void | |
8054fc48 | 4169 | delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port) |
79eee1eb BP |
4170 | { |
4171 | if (rule_is_hidden(rule)) { | |
4172 | return; | |
4173 | } | |
4174 | ||
4175 | if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) { | |
4176 | return; | |
4177 | } | |
4178 | ||
bcf84111 | 4179 | rule_send_removed(p, rule, OFPRR_DELETE); |
79eee1eb BP |
4180 | rule_remove(p, rule); |
4181 | } | |
4182 | \f | |
064af421 | 4183 | static int |
2e4f5fcf | 4184 | handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh) |
064af421 | 4185 | { |
3052b0c5 | 4186 | struct ofproto *p = ofconn->ofproto; |
2e4f5fcf | 4187 | struct flow_mod fm; |
064af421 BP |
4188 | int error; |
4189 | ||
3052b0c5 | 4190 | error = reject_slave_controller(ofconn, "flow_mod"); |
9deba63b BP |
4191 | if (error) { |
4192 | return error; | |
4193 | } | |
3052b0c5 | 4194 | |
2e4f5fcf | 4195 | error = ofputil_decode_flow_mod(&fm, oh, ofconn->flow_format); |
064af421 BP |
4196 | if (error) { |
4197 | return error; | |
4198 | } | |
4199 | ||
2e4f5fcf BP |
4200 | /* We do not support the emergency flow cache. It will hopefully get |
4201 | * dropped from OpenFlow in the near future. */ | |
4202 | if (fm.flags & OFPFF_EMERG) { | |
49bdc010 JP |
4203 | /* There isn't a good fit for an error code, so just state that the |
4204 | * flow table is full. */ | |
4205 | return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL); | |
4206 | } | |
4207 | ||
2e4f5fcf BP |
4208 | error = validate_actions(fm.actions, fm.n_actions, |
4209 | &fm.cr.flow, p->max_ports); | |
4210 | if (error) { | |
4211 | return error; | |
4212 | } | |
4213 | ||
4214 | switch (fm.command) { | |
3052b0c5 | 4215 | case OFPFC_ADD: |
2e4f5fcf | 4216 | return add_flow(ofconn, &fm); |
3052b0c5 BP |
4217 | |
4218 | case OFPFC_MODIFY: | |
2e4f5fcf | 4219 | return modify_flows_loose(ofconn, &fm); |
3052b0c5 BP |
4220 | |
4221 | case OFPFC_MODIFY_STRICT: | |
2e4f5fcf | 4222 | return modify_flow_strict(ofconn, &fm); |
3052b0c5 BP |
4223 | |
4224 | case OFPFC_DELETE: | |
2e4f5fcf | 4225 | delete_flows_loose(p, &fm); |
3052b0c5 BP |
4226 | return 0; |
4227 | ||
4228 | case OFPFC_DELETE_STRICT: | |
2e4f5fcf | 4229 | delete_flow_strict(p, &fm); |
3052b0c5 BP |
4230 | return 0; |
4231 | ||
4232 | default: | |
4233 | return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND); | |
4234 | } | |
4235 | } | |
4236 | ||
659586ef | 4237 | static int |
d1e2cf21 | 4238 | handle_tun_id_from_cookie(struct ofconn *ofconn, const struct ofp_header *oh) |
659586ef | 4239 | { |
d1e2cf21 BP |
4240 | const struct nxt_tun_id_cookie *msg |
4241 | = (const struct nxt_tun_id_cookie *) oh; | |
659586ef | 4242 | |
b70eac89 | 4243 | ofconn->flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10; |
659586ef JG |
4244 | return 0; |
4245 | } | |
4246 | ||
9deba63b | 4247 | static int |
d1e2cf21 | 4248 | handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh) |
9deba63b | 4249 | { |
d1e2cf21 | 4250 | struct nx_role_request *nrr = (struct nx_role_request *) oh; |
9deba63b BP |
4251 | struct nx_role_request *reply; |
4252 | struct ofpbuf *buf; | |
4253 | uint32_t role; | |
4254 | ||
5899143f | 4255 | if (ofconn->type != OFCONN_PRIMARY) { |
9deba63b BP |
4256 | VLOG_WARN_RL(&rl, "ignoring role request on non-controller " |
4257 | "connection"); | |
4258 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM); | |
4259 | } | |
4260 | ||
4261 | role = ntohl(nrr->role); | |
4262 | if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER | |
4263 | && role != NX_ROLE_SLAVE) { | |
4264 | VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role); | |
4265 | ||
4266 | /* There's no good error code for this. */ | |
4267 | return ofp_mkerr(OFPET_BAD_REQUEST, -1); | |
4268 | } | |
4269 | ||
4270 | if (role == NX_ROLE_MASTER) { | |
4271 | struct ofconn *other; | |
4272 | ||
3269c562 | 4273 | HMAP_FOR_EACH (other, hmap_node, &ofconn->ofproto->controllers) { |
9deba63b BP |
4274 | if (other->role == NX_ROLE_MASTER) { |
4275 | other->role = NX_ROLE_SLAVE; | |
4276 | } | |
4277 | } | |
4278 | } | |
4279 | ofconn->role = role; | |
4280 | ||
d1e2cf21 | 4281 | reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf); |
9deba63b BP |
4282 | reply->role = htonl(role); |
4283 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
4284 | ||
4285 | return 0; | |
4286 | } | |
4287 | ||
09246b99 | 4288 | static int |
d1e2cf21 | 4289 | handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh) |
09246b99 | 4290 | { |
d1e2cf21 BP |
4291 | const struct nxt_set_flow_format *msg |
4292 | = (const struct nxt_set_flow_format *) oh; | |
09246b99 | 4293 | uint32_t format; |
09246b99 BP |
4294 | |
4295 | format = ntohl(msg->format); | |
4296 | if (format == NXFF_OPENFLOW10 | |
4297 | || format == NXFF_TUN_ID_FROM_COOKIE | |
4298 | || format == NXFF_NXM) { | |
4299 | ofconn->flow_format = format; | |
4300 | return 0; | |
4301 | } else { | |
4302 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM); | |
4303 | } | |
4304 | } | |
4305 | ||
064af421 | 4306 | static int |
d1e2cf21 | 4307 | handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh) |
246e61ea JP |
4308 | { |
4309 | struct ofp_header *ob; | |
4310 | struct ofpbuf *buf; | |
4311 | ||
4312 | /* Currently, everything executes synchronously, so we can just | |
4313 | * immediately send the barrier reply. */ | |
4314 | ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf); | |
4315 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
4316 | return 0; | |
4317 | } | |
4318 | ||
d1e2cf21 BP |
4319 | static int |
4320 | handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg) | |
064af421 | 4321 | { |
d1e2cf21 BP |
4322 | const struct ofp_header *oh = msg->data; |
4323 | const struct ofputil_msg_type *type; | |
064af421 BP |
4324 | int error; |
4325 | ||
d1e2cf21 BP |
4326 | error = ofputil_decode_msg_type(oh, &type); |
4327 | if (error) { | |
4328 | return error; | |
4329 | } | |
064af421 | 4330 | |
d1e2cf21 BP |
4331 | switch (ofputil_msg_type_code(type)) { |
4332 | /* OpenFlow requests. */ | |
4333 | case OFPUTIL_OFPT_ECHO_REQUEST: | |
4334 | return handle_echo_request(ofconn, oh); | |
064af421 | 4335 | |
d1e2cf21 BP |
4336 | case OFPUTIL_OFPT_FEATURES_REQUEST: |
4337 | return handle_features_request(ofconn, oh); | |
064af421 | 4338 | |
d1e2cf21 BP |
4339 | case OFPUTIL_OFPT_GET_CONFIG_REQUEST: |
4340 | return handle_get_config_request(ofconn, oh); | |
064af421 | 4341 | |
d1e2cf21 BP |
4342 | case OFPUTIL_OFPT_SET_CONFIG: |
4343 | return handle_set_config(ofconn, msg->data); | |
064af421 | 4344 | |
d1e2cf21 BP |
4345 | case OFPUTIL_OFPT_PACKET_OUT: |
4346 | return handle_packet_out(ofconn, oh); | |
064af421 | 4347 | |
d1e2cf21 BP |
4348 | case OFPUTIL_OFPT_PORT_MOD: |
4349 | return handle_port_mod(ofconn, oh); | |
064af421 | 4350 | |
d1e2cf21 | 4351 | case OFPUTIL_OFPT_FLOW_MOD: |
2e4f5fcf | 4352 | return handle_flow_mod(ofconn, oh); |
064af421 | 4353 | |
d1e2cf21 BP |
4354 | case OFPUTIL_OFPT_BARRIER_REQUEST: |
4355 | return handle_barrier_request(ofconn, oh); | |
064af421 | 4356 | |
d1e2cf21 BP |
4357 | /* OpenFlow replies. */ |
4358 | case OFPUTIL_OFPT_ECHO_REPLY: | |
4359 | return 0; | |
246e61ea | 4360 | |
d1e2cf21 BP |
4361 | /* Nicira extension requests. */ |
4362 | case OFPUTIL_NXT_STATUS_REQUEST: | |
4363 | return switch_status_handle_request( | |
4364 | ofconn->ofproto->switch_status, ofconn->rconn, oh); | |
4365 | ||
4366 | case OFPUTIL_NXT_TUN_ID_FROM_COOKIE: | |
4367 | return handle_tun_id_from_cookie(ofconn, oh); | |
4368 | ||
4369 | case OFPUTIL_NXT_ROLE_REQUEST: | |
4370 | return handle_role_request(ofconn, oh); | |
4371 | ||
4372 | case OFPUTIL_NXT_SET_FLOW_FORMAT: | |
4373 | return handle_nxt_set_flow_format(ofconn, oh); | |
4374 | ||
4375 | case OFPUTIL_NXT_FLOW_MOD: | |
2e4f5fcf | 4376 | return handle_flow_mod(ofconn, oh); |
d1e2cf21 BP |
4377 | |
4378 | /* OpenFlow statistics requests. */ | |
4379 | case OFPUTIL_OFPST_DESC_REQUEST: | |
4380 | return handle_desc_stats_request(ofconn, oh); | |
4381 | ||
4382 | case OFPUTIL_OFPST_FLOW_REQUEST: | |
4383 | return handle_flow_stats_request(ofconn, oh); | |
4384 | ||
4385 | case OFPUTIL_OFPST_AGGREGATE_REQUEST: | |
4386 | return handle_aggregate_stats_request(ofconn, oh); | |
4387 | ||
4388 | case OFPUTIL_OFPST_TABLE_REQUEST: | |
4389 | return handle_table_stats_request(ofconn, oh); | |
4390 | ||
4391 | case OFPUTIL_OFPST_PORT_REQUEST: | |
4392 | return handle_port_stats_request(ofconn, oh); | |
4393 | ||
4394 | case OFPUTIL_OFPST_QUEUE_REQUEST: | |
4395 | return handle_queue_stats_request(ofconn, oh); | |
4396 | ||
4397 | /* Nicira extension statistics requests. */ | |
4398 | case OFPUTIL_NXST_FLOW_REQUEST: | |
4399 | return handle_nxst_flow(ofconn, oh); | |
4400 | ||
4401 | case OFPUTIL_NXST_AGGREGATE_REQUEST: | |
4402 | return handle_nxst_aggregate(ofconn, oh); | |
4403 | ||
4404 | case OFPUTIL_INVALID: | |
4405 | case OFPUTIL_OFPT_HELLO: | |
4406 | case OFPUTIL_OFPT_ERROR: | |
4407 | case OFPUTIL_OFPT_FEATURES_REPLY: | |
4408 | case OFPUTIL_OFPT_GET_CONFIG_REPLY: | |
4409 | case OFPUTIL_OFPT_PACKET_IN: | |
4410 | case OFPUTIL_OFPT_FLOW_REMOVED: | |
4411 | case OFPUTIL_OFPT_PORT_STATUS: | |
4412 | case OFPUTIL_OFPT_BARRIER_REPLY: | |
4413 | case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST: | |
4414 | case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY: | |
4415 | case OFPUTIL_OFPST_DESC_REPLY: | |
4416 | case OFPUTIL_OFPST_FLOW_REPLY: | |
4417 | case OFPUTIL_OFPST_QUEUE_REPLY: | |
4418 | case OFPUTIL_OFPST_PORT_REPLY: | |
4419 | case OFPUTIL_OFPST_TABLE_REPLY: | |
4420 | case OFPUTIL_OFPST_AGGREGATE_REPLY: | |
4421 | case OFPUTIL_NXT_STATUS_REPLY: | |
4422 | case OFPUTIL_NXT_ROLE_REPLY: | |
4423 | case OFPUTIL_NXT_FLOW_REMOVED: | |
4424 | case OFPUTIL_NXST_FLOW_REPLY: | |
4425 | case OFPUTIL_NXST_AGGREGATE_REPLY: | |
064af421 BP |
4426 | default: |
4427 | if (VLOG_IS_WARN_ENABLED()) { | |
4428 | char *s = ofp_to_string(oh, ntohs(oh->length), 2); | |
4429 | VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s); | |
4430 | free(s); | |
4431 | } | |
d1e2cf21 BP |
4432 | if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) { |
4433 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT); | |
4434 | } else { | |
4435 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE); | |
4436 | } | |
064af421 | 4437 | } |
d1e2cf21 | 4438 | } |
064af421 | 4439 | |
d1e2cf21 BP |
4440 | static void |
4441 | handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg) | |
4442 | { | |
4443 | int error = handle_openflow__(ofconn, ofp_msg); | |
064af421 BP |
4444 | if (error) { |
4445 | send_error_oh(ofconn, ofp_msg->data, error); | |
4446 | } | |
d1e2cf21 | 4447 | COVERAGE_INC(ofproto_recv_openflow); |
064af421 BP |
4448 | } |
4449 | \f | |
4450 | static void | |
856081f6 | 4451 | handle_miss_upcall(struct ofproto *p, struct dpif_upcall *upcall) |
064af421 | 4452 | { |
bcf84111 | 4453 | struct facet *facet; |
ae412e7d | 4454 | struct flow flow; |
064af421 | 4455 | |
856081f6 BP |
4456 | /* Obtain in_port and tun_id, at least. */ |
4457 | odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow); | |
064af421 | 4458 | |
856081f6 BP |
4459 | /* Set header pointers in 'flow'. */ |
4460 | flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow); | |
4617e2c1 | 4461 | |
ebe482fd EJ |
4462 | if (p->ofhooks->special_cb |
4463 | && !p->ofhooks->special_cb(&flow, upcall->packet, p->aux)) { | |
4464 | ofpbuf_delete(upcall->packet); | |
4465 | return; | |
4466 | } | |
4467 | ||
0ad9b732 JP |
4468 | /* Check with in-band control to see if this packet should be sent |
4469 | * to the local port regardless of the flow table. */ | |
856081f6 | 4470 | if (in_band_msg_in_hook(p->in_band, &flow, upcall->packet)) { |
cdee00fd | 4471 | struct ofpbuf odp_actions; |
0ad9b732 | 4472 | |
cdee00fd | 4473 | ofpbuf_init(&odp_actions, 32); |
7aec165d | 4474 | nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, ODPP_LOCAL); |
856081f6 BP |
4475 | dpif_execute(p->dpif, odp_actions.data, odp_actions.size, |
4476 | upcall->packet); | |
cdee00fd | 4477 | ofpbuf_uninit(&odp_actions); |
0ad9b732 JP |
4478 | } |
4479 | ||
bcf84111 BP |
4480 | facet = facet_lookup_valid(p, &flow); |
4481 | if (!facet) { | |
4482 | struct rule *rule = rule_lookup(p, &flow); | |
4483 | if (!rule) { | |
4484 | /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */ | |
856081f6 | 4485 | struct ofport *port = get_port(p, flow.in_port); |
bcf84111 BP |
4486 | if (port) { |
4487 | if (port->opp.config & OFPPC_NO_PACKET_IN) { | |
4488 | COVERAGE_INC(ofproto_no_packet_in); | |
4489 | /* XXX install 'drop' flow entry */ | |
856081f6 | 4490 | ofpbuf_delete(upcall->packet); |
bcf84111 BP |
4491 | return; |
4492 | } | |
4493 | } else { | |
4494 | VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, | |
856081f6 | 4495 | flow.in_port); |
064af421 | 4496 | } |
064af421 | 4497 | |
bcf84111 | 4498 | COVERAGE_INC(ofproto_packet_in); |
856081f6 | 4499 | send_packet_in(p, upcall, &flow, false); |
bcf84111 | 4500 | return; |
064af421 | 4501 | } |
bcf84111 | 4502 | |
856081f6 | 4503 | facet = facet_create(p, rule, &flow, upcall->packet); |
bcf84111 BP |
4504 | } else if (!facet->may_install) { |
4505 | /* The facet is not installable, that is, we need to process every | |
4506 | * packet, so process the current packet's actions into 'facet'. */ | |
856081f6 | 4507 | facet_make_actions(p, facet, upcall->packet); |
064af421 BP |
4508 | } |
4509 | ||
bcf84111 | 4510 | if (facet->rule->cr.priority == FAIL_OPEN_PRIORITY) { |
7778bd15 BP |
4511 | /* |
4512 | * Extra-special case for fail-open mode. | |
4513 | * | |
4514 | * We are in fail-open mode and the packet matched the fail-open rule, | |
4515 | * but we are connected to a controller too. We should send the packet | |
4516 | * up to the controller in the hope that it will try to set up a flow | |
4517 | * and thereby allow us to exit fail-open. | |
4518 | * | |
4519 | * See the top-level comment in fail-open.c for more information. | |
4520 | */ | |
856081f6 | 4521 | send_packet_in(p, upcall, &flow, true); |
7778bd15 | 4522 | } |
750638bb | 4523 | |
856081f6 | 4524 | facet_execute(p, facet, upcall->packet); |
bcf84111 | 4525 | facet_install(p, facet, false); |
064af421 | 4526 | } |
72b06300 BP |
4527 | |
4528 | static void | |
856081f6 | 4529 | handle_upcall(struct ofproto *p, struct dpif_upcall *upcall) |
72b06300 | 4530 | { |
856081f6 | 4531 | struct flow flow; |
72b06300 | 4532 | |
856081f6 | 4533 | switch (upcall->type) { |
82272ede | 4534 | case DPIF_UC_ACTION: |
72b06300 | 4535 | COVERAGE_INC(ofproto_ctlr_action); |
856081f6 BP |
4536 | odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow); |
4537 | send_packet_in(p, upcall, &flow, false); | |
72b06300 BP |
4538 | break; |
4539 | ||
82272ede | 4540 | case DPIF_UC_SAMPLE: |
72b06300 | 4541 | if (p->sflow) { |
856081f6 BP |
4542 | odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow); |
4543 | ofproto_sflow_received(p->sflow, upcall, &flow); | |
72b06300 | 4544 | } |
856081f6 | 4545 | ofpbuf_delete(upcall->packet); |
72b06300 BP |
4546 | break; |
4547 | ||
82272ede | 4548 | case DPIF_UC_MISS: |
856081f6 | 4549 | handle_miss_upcall(p, upcall); |
72b06300 BP |
4550 | break; |
4551 | ||
982b8810 | 4552 | case DPIF_N_UC_TYPES: |
72b06300 | 4553 | default: |
856081f6 | 4554 | VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type); |
72b06300 BP |
4555 | break; |
4556 | } | |
4557 | } | |
064af421 | 4558 | \f |
4a4cdb3b BP |
4559 | /* Flow expiration. */ |
4560 | ||
0de7a4b4 | 4561 | static int ofproto_dp_max_idle(const struct ofproto *); |
3394b5b6 | 4562 | static void ofproto_update_stats(struct ofproto *); |
5ecc9d81 | 4563 | static void rule_expire(struct ofproto *, struct rule *); |
bcf84111 | 4564 | static void ofproto_expire_facets(struct ofproto *, int dp_max_idle); |
4a4cdb3b BP |
4565 | |
4566 | /* This function is called periodically by ofproto_run(). Its job is to | |
4567 | * collect updates for the flows that have been installed into the datapath, | |
4568 | * most importantly when they last were used, and then use that information to | |
0de7a4b4 BP |
4569 | * expire flows that have not been used recently. |
4570 | * | |
4571 | * Returns the number of milliseconds after which it should be called again. */ | |
4572 | static int | |
4a4cdb3b BP |
4573 | ofproto_expire(struct ofproto *ofproto) |
4574 | { | |
5ecc9d81 BP |
4575 | struct rule *rule, *next_rule; |
4576 | struct cls_cursor cursor; | |
4577 | int dp_max_idle; | |
4a4cdb3b | 4578 | |
3394b5b6 EJ |
4579 | /* Update stats for each flow in the datapath. */ |
4580 | ofproto_update_stats(ofproto); | |
4a4cdb3b | 4581 | |
bcf84111 | 4582 | /* Expire facets that have been idle too long. */ |
5ecc9d81 BP |
4583 | dp_max_idle = ofproto_dp_max_idle(ofproto); |
4584 | ofproto_expire_facets(ofproto, dp_max_idle); | |
bcf84111 BP |
4585 | |
4586 | /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */ | |
5ecc9d81 BP |
4587 | cls_cursor_init(&cursor, &ofproto->cls, NULL); |
4588 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) { | |
4589 | rule_expire(ofproto, rule); | |
4590 | } | |
4a4cdb3b BP |
4591 | |
4592 | /* Let the hook know that we're at a stable point: all outstanding data | |
4593 | * in existing flows has been accounted to the account_cb. Thus, the | |
4594 | * hook can now reasonably do operations that depend on having accurate | |
4595 | * flow volume accounting (currently, that's just bond rebalancing). */ | |
4596 | if (ofproto->ofhooks->account_checkpoint_cb) { | |
4597 | ofproto->ofhooks->account_checkpoint_cb(ofproto->aux); | |
4598 | } | |
0de7a4b4 | 4599 | |
5ecc9d81 | 4600 | return MIN(dp_max_idle, 1000); |
4a4cdb3b BP |
4601 | } |
4602 | ||
3394b5b6 | 4603 | /* Update 'packet_count', 'byte_count', and 'used' members of installed facets. |
878ae780 EJ |
4604 | * |
4605 | * This function also pushes statistics updates to rules which each facet | |
4606 | * resubmits into. Generally these statistics will be accurate. However, if a | |
4607 | * facet changes the rule it resubmits into at some time in between | |
4608 | * ofproto_update_stats() runs, it is possible that statistics accrued to the | |
4609 | * old rule will be incorrectly attributed to the new rule. This could be | |
4610 | * avoided by calling ofproto_update_stats() whenever rules are created or | |
4611 | * deleted. However, the performance impact of making so many calls to the | |
4612 | * datapath do not justify the benefit of having perfectly accurate statistics. | |
3394b5b6 | 4613 | */ |
4a4cdb3b | 4614 | static void |
3394b5b6 | 4615 | ofproto_update_stats(struct ofproto *p) |
4a4cdb3b | 4616 | { |
c97fb132 | 4617 | const struct dpif_flow_stats *stats; |
704a1e09 | 4618 | struct dpif_flow_dump dump; |
feebdea2 BP |
4619 | const struct nlattr *key; |
4620 | size_t key_len; | |
4a4cdb3b | 4621 | |
704a1e09 | 4622 | dpif_flow_dump_start(&dump, p->dpif); |
feebdea2 | 4623 | while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) { |
bcf84111 | 4624 | struct facet *facet; |
ae412e7d | 4625 | struct flow flow; |
14608a15 | 4626 | |
feebdea2 | 4627 | if (odp_flow_key_to_flow(key, key_len, &flow)) { |
36956a7d BP |
4628 | struct ds s; |
4629 | ||
4630 | ds_init(&s); | |
feebdea2 | 4631 | odp_flow_key_format(key, key_len, &s); |
36956a7d BP |
4632 | VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s", |
4633 | ds_cstr(&s)); | |
4634 | ds_destroy(&s); | |
4635 | ||
4636 | continue; | |
4637 | } | |
bcf84111 | 4638 | facet = facet_find(p, &flow); |
4a4cdb3b | 4639 | |
bcf84111 | 4640 | if (facet && facet->installed) { |
3394b5b6 EJ |
4641 | |
4642 | if (stats->n_packets >= facet->dp_packet_count) { | |
4643 | facet->packet_count += stats->n_packets - facet->dp_packet_count; | |
4644 | } else { | |
4645 | VLOG_WARN_RL(&rl, "unexpected packet count from the datapath"); | |
4646 | } | |
4647 | ||
4648 | if (stats->n_bytes >= facet->dp_byte_count) { | |
4649 | facet->byte_count += stats->n_bytes - facet->dp_byte_count; | |
4650 | } else { | |
4651 | VLOG_WARN_RL(&rl, "unexpected byte count from datapath"); | |
4652 | } | |
4653 | ||
4654 | facet->dp_packet_count = stats->n_packets; | |
4655 | facet->dp_byte_count = stats->n_bytes; | |
4656 | ||
4657 | facet_update_time(p, facet, stats->used); | |
feebdea2 | 4658 | facet_account(p, facet, stats->n_bytes); |
878ae780 | 4659 | facet_push_stats(p, facet); |
4a4cdb3b BP |
4660 | } else { |
4661 | /* There's a flow in the datapath that we know nothing about. | |
4662 | * Delete it. */ | |
4663 | COVERAGE_INC(ofproto_unexpected_rule); | |
feebdea2 | 4664 | dpif_flow_del(p->dpif, key, key_len, NULL); |
4a4cdb3b | 4665 | } |
4a4cdb3b | 4666 | } |
704a1e09 | 4667 | dpif_flow_dump_done(&dump); |
4a4cdb3b BP |
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 | 4701 | * This requires a second pass through the facets, in addition to the pass |
3394b5b6 | 4702 | * made by ofproto_update_stats(), because the former function never looks |
bcf84111 | 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 | 4765 | struct ofexpired expired; |
feebdea2 | 4766 | |
bcf84111 | 4767 | if (facet->installed) { |
c97fb132 | 4768 | struct dpif_flow_stats stats; |
c33087b8 BP |
4769 | |
4770 | facet_put__(ofproto, facet, facet->actions, facet->actions_len, | |
4771 | &stats); | |
4772 | facet_update_stats(ofproto, facet, &stats); | |
4a4cdb3b BP |
4773 | } |
4774 | ||
c33087b8 BP |
4775 | expired.flow = facet->flow; |
4776 | expired.packet_count = facet->packet_count; | |
4777 | expired.byte_count = facet->byte_count; | |
4778 | expired.used = facet->used; | |
bcf84111 BP |
4779 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); |
4780 | } | |
4781 | } | |
4a4cdb3b | 4782 | |
bcf84111 BP |
4783 | static void |
4784 | ofproto_expire_facets(struct ofproto *ofproto, int dp_max_idle) | |
4785 | { | |
4786 | long long int cutoff = time_msec() - dp_max_idle; | |
4787 | struct facet *facet, *next_facet; | |
4788 | ||
4789 | HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) { | |
4790 | facet_active_timeout(ofproto, facet); | |
4791 | if (facet->used < cutoff) { | |
4792 | facet_remove(ofproto, facet); | |
4793 | } | |
4a4cdb3b BP |
4794 | } |
4795 | } | |
4796 | ||
5ecc9d81 BP |
4797 | /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules, |
4798 | * then delete it entirely. */ | |
4a4cdb3b | 4799 | static void |
5ecc9d81 | 4800 | rule_expire(struct ofproto *ofproto, struct rule *rule) |
4a4cdb3b | 4801 | { |
bcf84111 BP |
4802 | struct facet *facet, *next_facet; |
4803 | long long int now; | |
4804 | uint8_t reason; | |
4a4cdb3b | 4805 | |
bcf84111 | 4806 | /* Has 'rule' expired? */ |
4a4cdb3b | 4807 | now = time_msec(); |
bcf84111 BP |
4808 | if (rule->hard_timeout |
4809 | && now > rule->created + rule->hard_timeout * 1000) { | |
4810 | reason = OFPRR_HARD_TIMEOUT; | |
4811 | } else if (rule->idle_timeout && list_is_empty(&rule->facets) | |
4812 | && now >rule->used + rule->idle_timeout * 1000) { | |
4813 | reason = OFPRR_IDLE_TIMEOUT; | |
4a4cdb3b | 4814 | } else { |
bcf84111 | 4815 | return; |
4a4cdb3b | 4816 | } |
064af421 | 4817 | |
bcf84111 | 4818 | COVERAGE_INC(ofproto_expired); |
064af421 | 4819 | |
bcf84111 BP |
4820 | /* Update stats. (This is a no-op if the rule expired due to an idle |
4821 | * timeout, because that only happens when the rule has no facets left.) */ | |
4822 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { | |
5ecc9d81 | 4823 | facet_remove(ofproto, facet); |
064af421 BP |
4824 | } |
4825 | ||
bcf84111 BP |
4826 | /* Get rid of the rule. */ |
4827 | if (!rule_is_hidden(rule)) { | |
5ecc9d81 | 4828 | rule_send_removed(ofproto, rule, reason); |
bcf84111 | 4829 | } |
5ecc9d81 | 4830 | rule_remove(ofproto, rule); |
064af421 | 4831 | } |
bcf84111 | 4832 | \f |
064af421 | 4833 | static struct ofpbuf * |
09246b99 BP |
4834 | compose_ofp_flow_removed(struct ofconn *ofconn, const struct rule *rule, |
4835 | uint8_t reason) | |
064af421 | 4836 | { |
ca069229 | 4837 | struct ofp_flow_removed *ofr; |
064af421 BP |
4838 | struct ofpbuf *buf; |
4839 | ||
0224fbde | 4840 | ofr = make_openflow_xid(sizeof *ofr, OFPT_FLOW_REMOVED, htonl(0), &buf); |
ff9d3826 BP |
4841 | ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofr->match, |
4842 | rule->flow_cookie, &ofr->cookie); | |
ca069229 JP |
4843 | ofr->priority = htons(rule->cr.priority); |
4844 | ofr->reason = reason; | |
c6ebb8fb | 4845 | calc_flow_duration(rule->created, &ofr->duration_sec, &ofr->duration_nsec); |
ca069229 JP |
4846 | ofr->idle_timeout = htons(rule->idle_timeout); |
4847 | ofr->packet_count = htonll(rule->packet_count); | |
4848 | ofr->byte_count = htonll(rule->byte_count); | |
064af421 BP |
4849 | |
4850 | return buf; | |
4851 | } | |
4852 | ||
09246b99 BP |
4853 | static struct ofpbuf * |
4854 | compose_nx_flow_removed(const struct rule *rule, uint8_t reason) | |
4855 | { | |
4856 | struct nx_flow_removed *nfr; | |
4857 | struct ofpbuf *buf; | |
4858 | int match_len; | |
4859 | ||
0224fbde | 4860 | make_nxmsg_xid(sizeof *nfr, NXT_FLOW_REMOVED, htonl(0), &buf); |
09246b99 | 4861 | match_len = nx_put_match(buf, &rule->cr); |
0224fbde | 4862 | |
4d0ed519 | 4863 | nfr = buf->data; |
09246b99 BP |
4864 | nfr->cookie = rule->flow_cookie; |
4865 | nfr->priority = htons(rule->cr.priority); | |
4866 | nfr->reason = reason; | |
4867 | calc_flow_duration(rule->created, &nfr->duration_sec, &nfr->duration_nsec); | |
4868 | nfr->idle_timeout = htons(rule->idle_timeout); | |
4869 | nfr->match_len = htons(match_len); | |
4870 | nfr->packet_count = htonll(rule->packet_count); | |
4871 | nfr->byte_count = htonll(rule->byte_count); | |
4872 | ||
4873 | return buf; | |
4874 | } | |
4875 | ||
ca069229 | 4876 | static void |
bcf84111 | 4877 | rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason) |
064af421 BP |
4878 | { |
4879 | struct ofconn *ofconn; | |
064af421 | 4880 | |
3b587616 BP |
4881 | if (!rule->send_flow_removed) { |
4882 | return; | |
4883 | } | |
4884 | ||
4e8e4213 | 4885 | LIST_FOR_EACH (ofconn, node, &p->all_conns) { |
7a0efeb5 BP |
4886 | struct ofpbuf *msg; |
4887 | ||
4888 | if (!rconn_is_connected(ofconn->rconn) | |
4889 | || !ofconn_receives_async_msgs(ofconn)) { | |
4890 | continue; | |
064af421 | 4891 | } |
7a0efeb5 | 4892 | |
09246b99 BP |
4893 | msg = (ofconn->flow_format == NXFF_NXM |
4894 | ? compose_nx_flow_removed(rule, reason) | |
4895 | : compose_ofp_flow_removed(ofconn, rule, reason)); | |
6d6c7259 BP |
4896 | |
4897 | /* Account flow expirations under ofconn->reply_counter, the counter | |
4898 | * for replies to OpenFlow requests. That works because preventing | |
4899 | * OpenFlow requests from being processed also prevents new flows from | |
4900 | * being added (and expiring). (It also prevents processing OpenFlow | |
4901 | * requests that would not add new flows, so it is imperfect.) */ | |
7a0efeb5 | 4902 | queue_tx(msg, ofconn, ofconn->reply_counter); |
064af421 BP |
4903 | } |
4904 | } | |
4905 | ||
3394b5b6 EJ |
4906 | /* Obtains statistics for 'rule' and stores them in '*packets' and '*bytes'. |
4907 | * The returned statistics include statistics for all of 'rule''s facets. */ | |
4908 | static void | |
4909 | rule_get_stats(const struct rule *rule, uint64_t *packets, uint64_t *bytes) | |
4910 | { | |
4911 | uint64_t p, b; | |
4912 | struct facet *facet; | |
4913 | ||
4914 | /* Start from historical data for 'rule' itself that are no longer tracked | |
4915 | * in facets. This counts, for example, facets that have expired. */ | |
4916 | p = rule->packet_count; | |
4917 | b = rule->byte_count; | |
4918 | ||
4919 | /* Add any statistics that are tracked by facets. This includes | |
4920 | * statistical data recently updated by ofproto_update_stats() as well as | |
4921 | * stats for packets that were executed "by hand" via dpif_execute(). */ | |
4922 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
4923 | p += facet->packet_count; | |
4924 | b += facet->byte_count; | |
4925 | } | |
4926 | ||
4927 | *packets = p; | |
4928 | *bytes = b; | |
4929 | } | |
4930 | ||
856081f6 | 4931 | /* pinsched callback for sending 'ofp_packet_in' on 'ofconn'. */ |
064af421 | 4932 | static void |
856081f6 | 4933 | do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn_) |
064af421 | 4934 | { |
76ce9432 | 4935 | struct ofconn *ofconn = ofconn_; |
43253595 | 4936 | |
856081f6 | 4937 | rconn_send_with_limit(ofconn->rconn, ofp_packet_in, |
43253595 BP |
4938 | ofconn->packet_in_counter, 100); |
4939 | } | |
4940 | ||
856081f6 BP |
4941 | /* Takes 'upcall', whose packet has the flow specified by 'flow', composes an |
4942 | * OpenFlow packet-in message from it, and passes it to 'ofconn''s packet | |
4943 | * scheduler for sending. | |
43253595 | 4944 | * |
856081f6 BP |
4945 | * If 'clone' is true, the caller retains ownership of 'upcall->packet'. |
4946 | * Otherwise, ownership is transferred to this function. */ | |
43253595 | 4947 | static void |
856081f6 BP |
4948 | schedule_packet_in(struct ofconn *ofconn, struct dpif_upcall *upcall, |
4949 | const struct flow *flow, bool clone) | |
43253595 | 4950 | { |
856081f6 | 4951 | enum { OPI_SIZE = offsetof(struct ofp_packet_in, data) }; |
76ce9432 | 4952 | struct ofproto *ofproto = ofconn->ofproto; |
856081f6 BP |
4953 | struct ofp_packet_in *opi; |
4954 | int total_len, send_len; | |
4955 | struct ofpbuf *packet; | |
76ce9432 | 4956 | uint32_t buffer_id; |
82272ede | 4957 | int idx; |
064af421 | 4958 | |
856081f6 | 4959 | /* Get OpenFlow buffer_id. */ |
82272ede | 4960 | if (upcall->type == DPIF_UC_ACTION) { |
76ce9432 | 4961 | buffer_id = UINT32_MAX; |
43253595 BP |
4962 | } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) { |
4963 | buffer_id = pktbuf_get_null(); | |
89b9612d BP |
4964 | } else if (!ofconn->pktbuf) { |
4965 | buffer_id = UINT32_MAX; | |
76ce9432 | 4966 | } else { |
856081f6 | 4967 | buffer_id = pktbuf_save(ofconn->pktbuf, upcall->packet, flow->in_port); |
76ce9432 | 4968 | } |
372179d4 | 4969 | |
43253595 | 4970 | /* Figure out how much of the packet to send. */ |
856081f6 | 4971 | total_len = send_len = upcall->packet->size; |
43253595 BP |
4972 | if (buffer_id != UINT32_MAX) { |
4973 | send_len = MIN(send_len, ofconn->miss_send_len); | |
4974 | } | |
82272ede | 4975 | if (upcall->type == DPIF_UC_ACTION) { |
856081f6 BP |
4976 | send_len = MIN(send_len, upcall->userdata); |
4977 | } | |
064af421 | 4978 | |
856081f6 | 4979 | /* Copy or steal buffer for OFPT_PACKET_IN. */ |
43253595 | 4980 | if (clone) { |
856081f6 BP |
4981 | packet = ofpbuf_clone_data_with_headroom(upcall->packet->data, |
4982 | send_len, OPI_SIZE); | |
43253595 | 4983 | } else { |
856081f6 BP |
4984 | packet = upcall->packet; |
4985 | packet->size = send_len; | |
43253595 BP |
4986 | } |
4987 | ||
856081f6 BP |
4988 | /* Add OFPT_PACKET_IN. */ |
4989 | opi = ofpbuf_push_zeros(packet, OPI_SIZE); | |
4990 | opi->header.version = OFP_VERSION; | |
4991 | opi->header.type = OFPT_PACKET_IN; | |
4992 | opi->total_len = htons(total_len); | |
4993 | opi->in_port = htons(odp_port_to_ofp_port(flow->in_port)); | |
82272ede | 4994 | opi->reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION; |
43253595 BP |
4995 | opi->buffer_id = htonl(buffer_id); |
4996 | update_openflow_length(packet); | |
4997 | ||
4998 | /* Hand over to packet scheduler. It might immediately call into | |
4999 | * do_send_packet_in() or it might buffer it for a while (until a later | |
5000 | * call to pinsched_run()). */ | |
82272ede BP |
5001 | idx = upcall->type == DPIF_UC_MISS ? 0 : 1; |
5002 | pinsched_send(ofconn->schedulers[idx], flow->in_port, | |
43253595 | 5003 | packet, do_send_packet_in, ofconn); |
064af421 BP |
5004 | } |
5005 | ||
82272ede | 5006 | /* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an |
856081f6 BP |
5007 | * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to |
5008 | * their individual configurations. | |
43253595 | 5009 | * |
2c00fecf BP |
5010 | * If 'clone' is true, the caller retains ownership of 'upcall->packet'. |
5011 | * Otherwise, ownership is transferred to this function. */ | |
43253595 | 5012 | static void |
856081f6 BP |
5013 | send_packet_in(struct ofproto *ofproto, struct dpif_upcall *upcall, |
5014 | const struct flow *flow, bool clone) | |
43253595 | 5015 | { |
76ce9432 | 5016 | struct ofconn *ofconn, *prev; |
76ce9432 BP |
5017 | |
5018 | prev = NULL; | |
4e8e4213 | 5019 | LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) { |
c91248b3 | 5020 | if (ofconn_receives_async_msgs(ofconn)) { |
9deba63b | 5021 | if (prev) { |
856081f6 | 5022 | schedule_packet_in(prev, upcall, flow, true); |
9deba63b BP |
5023 | } |
5024 | prev = ofconn; | |
064af421 | 5025 | } |
76ce9432 BP |
5026 | } |
5027 | if (prev) { | |
856081f6 BP |
5028 | schedule_packet_in(prev, upcall, flow, clone); |
5029 | } else if (!clone) { | |
5030 | ofpbuf_delete(upcall->packet); | |
064af421 | 5031 | } |
064af421 BP |
5032 | } |
5033 | ||
5034 | static uint64_t | |
fa60c019 | 5035 | pick_datapath_id(const struct ofproto *ofproto) |
064af421 | 5036 | { |
fa60c019 | 5037 | const struct ofport *port; |
064af421 | 5038 | |
ca0f572c | 5039 | port = get_port(ofproto, ODPP_LOCAL); |
fa60c019 BP |
5040 | if (port) { |
5041 | uint8_t ea[ETH_ADDR_LEN]; | |
5042 | int error; | |
5043 | ||
5044 | error = netdev_get_etheraddr(port->netdev, ea); | |
064af421 BP |
5045 | if (!error) { |
5046 | return eth_addr_to_uint64(ea); | |
5047 | } | |
5048 | VLOG_WARN("could not get MAC address for %s (%s)", | |
fa60c019 | 5049 | netdev_get_name(port->netdev), strerror(error)); |
064af421 | 5050 | } |
fa60c019 | 5051 | return ofproto->fallback_dpid; |
064af421 BP |
5052 | } |
5053 | ||
5054 | static uint64_t | |
5055 | pick_fallback_dpid(void) | |
5056 | { | |
5057 | uint8_t ea[ETH_ADDR_LEN]; | |
70150daf | 5058 | eth_addr_nicira_random(ea); |
064af421 BP |
5059 | return eth_addr_to_uint64(ea); |
5060 | } | |
5061 | \f | |
7aa697dd BP |
5062 | static void |
5063 | ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED, | |
5064 | void *aux OVS_UNUSED) | |
5065 | { | |
5066 | const struct shash_node *node; | |
5067 | struct ds results; | |
5068 | ||
5069 | ds_init(&results); | |
5070 | SHASH_FOR_EACH (node, &all_ofprotos) { | |
5071 | ds_put_format(&results, "%s\n", node->name); | |
5072 | } | |
5073 | unixctl_command_reply(conn, 200, ds_cstr(&results)); | |
5074 | ds_destroy(&results); | |
5075 | } | |
5076 | ||
5077 | struct ofproto_trace { | |
5078 | struct action_xlate_ctx ctx; | |
5079 | struct flow flow; | |
5080 | struct ds *result; | |
5081 | }; | |
5082 | ||
5083 | static void | |
5084 | trace_format_rule(struct ds *result, int level, const struct rule *rule) | |
5085 | { | |
5086 | ds_put_char_multiple(result, '\t', level); | |
5087 | if (!rule) { | |
5088 | ds_put_cstr(result, "No match\n"); | |
5089 | return; | |
5090 | } | |
5091 | ||
5092 | ds_put_format(result, "Rule: cookie=%#"PRIx64" ", | |
5093 | ntohll(rule->flow_cookie)); | |
5094 | cls_rule_format(&rule->cr, result); | |
5095 | ds_put_char(result, '\n'); | |
5096 | ||
5097 | ds_put_char_multiple(result, '\t', level); | |
5098 | ds_put_cstr(result, "OpenFlow "); | |
5099 | ofp_print_actions(result, (const struct ofp_action_header *) rule->actions, | |
5100 | rule->n_actions * sizeof *rule->actions); | |
5101 | ds_put_char(result, '\n'); | |
5102 | } | |
5103 | ||
5104 | static void | |
5105 | trace_format_flow(struct ds *result, int level, const char *title, | |
5106 | struct ofproto_trace *trace) | |
5107 | { | |
5108 | ds_put_char_multiple(result, '\t', level); | |
5109 | ds_put_format(result, "%s: ", title); | |
5110 | if (flow_equal(&trace->ctx.flow, &trace->flow)) { | |
5111 | ds_put_cstr(result, "unchanged"); | |
5112 | } else { | |
5113 | flow_format(result, &trace->ctx.flow); | |
5114 | trace->flow = trace->ctx.flow; | |
5115 | } | |
5116 | ds_put_char(result, '\n'); | |
5117 | } | |
5118 | ||
5119 | static void | |
878ae780 | 5120 | trace_resubmit(struct action_xlate_ctx *ctx, struct rule *rule) |
7aa697dd BP |
5121 | { |
5122 | struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx); | |
5123 | struct ds *result = trace->result; | |
5124 | ||
5125 | ds_put_char(result, '\n'); | |
5126 | trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace); | |
5127 | trace_format_rule(result, ctx->recurse + 1, rule); | |
5128 | } | |
5129 | ||
5130 | static void | |
5131 | ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_, | |
5132 | void *aux OVS_UNUSED) | |
5133 | { | |
5134 | char *dpname, *in_port_s, *tun_id_s, *packet_s; | |
5135 | char *args = xstrdup(args_); | |
5136 | char *save_ptr = NULL; | |
5137 | struct ofproto *ofproto; | |
5138 | struct ofpbuf packet; | |
5139 | struct rule *rule; | |
5140 | struct ds result; | |
5141 | struct flow flow; | |
5142 | uint16_t in_port; | |
11e6a15b | 5143 | ovs_be64 tun_id; |
7aa697dd BP |
5144 | char *s; |
5145 | ||
5146 | ofpbuf_init(&packet, strlen(args) / 2); | |
5147 | ds_init(&result); | |
5148 | ||
5149 | dpname = strtok_r(args, " ", &save_ptr); | |
5150 | tun_id_s = strtok_r(NULL, " ", &save_ptr); | |
5151 | in_port_s = strtok_r(NULL, " ", &save_ptr); | |
5152 | packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */ | |
5153 | if (!dpname || !in_port_s || !packet_s) { | |
5154 | unixctl_command_reply(conn, 501, "Bad command syntax"); | |
5155 | goto exit; | |
5156 | } | |
5157 | ||
5158 | ofproto = shash_find_data(&all_ofprotos, dpname); | |
5159 | if (!ofproto) { | |
5160 | unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list " | |
5161 | "for help)"); | |
5162 | goto exit; | |
5163 | } | |
5164 | ||
11e6a15b | 5165 | tun_id = htonll(strtoull(tun_id_s, NULL, 10)); |
7aa697dd BP |
5166 | in_port = ofp_port_to_odp_port(atoi(in_port_s)); |
5167 | ||
5168 | packet_s = ofpbuf_put_hex(&packet, packet_s, NULL); | |
5169 | packet_s += strspn(packet_s, " "); | |
5170 | if (*packet_s != '\0') { | |
5171 | unixctl_command_reply(conn, 501, "Trailing garbage in command"); | |
5172 | goto exit; | |
5173 | } | |
5174 | if (packet.size < ETH_HEADER_LEN) { | |
5175 | unixctl_command_reply(conn, 501, "Packet data too short for Ethernet"); | |
5176 | goto exit; | |
5177 | } | |
5178 | ||
5179 | ds_put_cstr(&result, "Packet: "); | |
5180 | s = ofp_packet_to_string(packet.data, packet.size, packet.size); | |
5181 | ds_put_cstr(&result, s); | |
5182 | free(s); | |
5183 | ||
5184 | flow_extract(&packet, tun_id, in_port, &flow); | |
5185 | ds_put_cstr(&result, "Flow: "); | |
5186 | flow_format(&result, &flow); | |
5187 | ds_put_char(&result, '\n'); | |
5188 | ||
5189 | rule = rule_lookup(ofproto, &flow); | |
5190 | trace_format_rule(&result, 0, rule); | |
5191 | if (rule) { | |
5192 | struct ofproto_trace trace; | |
cdee00fd | 5193 | struct ofpbuf *odp_actions; |
7aa697dd BP |
5194 | |
5195 | trace.result = &result; | |
5196 | trace.flow = flow; | |
5197 | action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet); | |
5198 | trace.ctx.resubmit_hook = trace_resubmit; | |
cdee00fd BP |
5199 | odp_actions = xlate_actions(&trace.ctx, |
5200 | rule->actions, rule->n_actions); | |
7aa697dd BP |
5201 | |
5202 | ds_put_char(&result, '\n'); | |
5203 | trace_format_flow(&result, 0, "Final flow", &trace); | |
5204 | ds_put_cstr(&result, "Datapath actions: "); | |
cdee00fd BP |
5205 | format_odp_actions(&result, odp_actions->data, odp_actions->size); |
5206 | ofpbuf_delete(odp_actions); | |
7aa697dd BP |
5207 | } |
5208 | ||
5209 | unixctl_command_reply(conn, 200, ds_cstr(&result)); | |
5210 | ||
5211 | exit: | |
5212 | ds_destroy(&result); | |
5213 | ofpbuf_uninit(&packet); | |
5214 | free(args); | |
5215 | } | |
5216 | ||
5217 | static void | |
5218 | ofproto_unixctl_init(void) | |
5219 | { | |
5220 | static bool registered; | |
5221 | if (registered) { | |
5222 | return; | |
5223 | } | |
5224 | registered = true; | |
5225 | ||
5226 | unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL); | |
5227 | unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL); | |
5228 | } | |
5229 | \f | |
064af421 | 5230 | static bool |
ae412e7d | 5231 | default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet, |
cdee00fd | 5232 | struct ofpbuf *odp_actions, tag_type *tags, |
6a07af36 | 5233 | uint16_t *nf_output_iface, void *ofproto_) |
064af421 BP |
5234 | { |
5235 | struct ofproto *ofproto = ofproto_; | |
5236 | int out_port; | |
5237 | ||
5238 | /* Drop frames for reserved multicast addresses. */ | |
5239 | if (eth_addr_is_reserved(flow->dl_dst)) { | |
5240 | return true; | |
5241 | } | |
5242 | ||
5243 | /* Learn source MAC (but don't try to learn from revalidation). */ | |
5244 | if (packet != NULL) { | |
5245 | tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src, | |
7febb910 JG |
5246 | 0, flow->in_port, |
5247 | GRAT_ARP_LOCK_NONE); | |
064af421 BP |
5248 | if (rev_tag) { |
5249 | /* The log messages here could actually be useful in debugging, | |
5250 | * so keep the rate limit relatively high. */ | |
5251 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); | |
5252 | VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16, | |
5253 | ETH_ADDR_ARGS(flow->dl_src), flow->in_port); | |
5254 | ofproto_revalidate(ofproto, rev_tag); | |
5255 | } | |
5256 | } | |
5257 | ||
5258 | /* Determine output port. */ | |
7febb910 JG |
5259 | out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags, |
5260 | NULL); | |
064af421 | 5261 | if (out_port < 0) { |
f1588b1f | 5262 | flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD, |
cdee00fd | 5263 | nf_output_iface, odp_actions); |
064af421 | 5264 | } else if (out_port != flow->in_port) { |
7aec165d | 5265 | nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, out_port); |
6a07af36 | 5266 | *nf_output_iface = out_port; |
064af421 BP |
5267 | } else { |
5268 | /* Drop. */ | |
5269 | } | |
5270 | ||
5271 | return true; | |
5272 | } | |
5273 | ||
5274 | static const struct ofhooks default_ofhooks = { | |
064af421 BP |
5275 | default_normal_ofhook_cb, |
5276 | NULL, | |
ebe482fd | 5277 | NULL, |
064af421 BP |
5278 | NULL |
5279 | }; |