]>
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" | |
4f2cad2c | 62 | #include "unixctl.h" |
064af421 | 63 | #include "vconn.h" |
5136ce49 | 64 | #include "vlog.h" |
064af421 | 65 | |
d98e6007 | 66 | VLOG_DEFINE_THIS_MODULE(ofproto); |
064af421 | 67 | |
cc01d0bb BP |
68 | COVERAGE_DEFINE(facet_changed_rule); |
69 | COVERAGE_DEFINE(facet_revalidate); | |
d76f09ea | 70 | COVERAGE_DEFINE(odp_overflow); |
d76f09ea BP |
71 | COVERAGE_DEFINE(ofproto_agg_request); |
72 | COVERAGE_DEFINE(ofproto_costly_flags); | |
73 | COVERAGE_DEFINE(ofproto_ctlr_action); | |
cc01d0bb | 74 | COVERAGE_DEFINE(ofproto_del_rule); |
d76f09ea BP |
75 | COVERAGE_DEFINE(ofproto_error); |
76 | COVERAGE_DEFINE(ofproto_expiration); | |
77 | COVERAGE_DEFINE(ofproto_expired); | |
78 | COVERAGE_DEFINE(ofproto_flows_req); | |
79 | COVERAGE_DEFINE(ofproto_flush); | |
80 | COVERAGE_DEFINE(ofproto_invalidated); | |
d76f09ea | 81 | COVERAGE_DEFINE(ofproto_no_packet_in); |
d76f09ea BP |
82 | COVERAGE_DEFINE(ofproto_ofconn_stuck); |
83 | COVERAGE_DEFINE(ofproto_ofp2odp); | |
84 | COVERAGE_DEFINE(ofproto_packet_in); | |
85 | COVERAGE_DEFINE(ofproto_packet_out); | |
86 | COVERAGE_DEFINE(ofproto_queue_req); | |
87 | COVERAGE_DEFINE(ofproto_recv_openflow); | |
88 | COVERAGE_DEFINE(ofproto_reinit_ports); | |
d76f09ea BP |
89 | COVERAGE_DEFINE(ofproto_unexpected_rule); |
90 | COVERAGE_DEFINE(ofproto_uninstallable); | |
91 | COVERAGE_DEFINE(ofproto_update_port); | |
92 | ||
72b06300 | 93 | #include "sflow_api.h" |
064af421 | 94 | |
f29152ca BP |
95 | struct rule; |
96 | ||
064af421 | 97 | struct ofport { |
ca0f572c | 98 | struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */ |
064af421 BP |
99 | struct netdev *netdev; |
100 | struct ofp_phy_port opp; /* In host byte order. */ | |
ca0f572c | 101 | uint16_t odp_port; |
064af421 BP |
102 | }; |
103 | ||
104 | static void ofport_free(struct ofport *); | |
105 | static void hton_ofp_phy_port(struct ofp_phy_port *); | |
106 | ||
f29152ca BP |
107 | struct action_xlate_ctx { |
108 | /* action_xlate_ctx_init() initializes these members. */ | |
109 | ||
110 | /* The ofproto. */ | |
111 | struct ofproto *ofproto; | |
112 | ||
113 | /* Flow to which the OpenFlow actions apply. xlate_actions() will modify | |
114 | * this flow when actions change header fields. */ | |
115 | struct flow flow; | |
116 | ||
117 | /* The packet corresponding to 'flow', or a null pointer if we are | |
118 | * revalidating without a packet to refer to. */ | |
119 | const struct ofpbuf *packet; | |
120 | ||
7aa697dd BP |
121 | /* If nonnull, called just before executing a resubmit action. |
122 | * | |
123 | * This is normally null so the client has to set it manually after | |
124 | * calling action_xlate_ctx_init(). */ | |
125 | void (*resubmit_hook)(struct action_xlate_ctx *, const struct rule *); | |
126 | ||
f29152ca BP |
127 | /* xlate_actions() initializes and uses these members. The client might want |
128 | * to look at them after it returns. */ | |
129 | ||
cdee00fd | 130 | struct ofpbuf *odp_actions; /* Datapath actions. */ |
f29152ca BP |
131 | tag_type tags; /* Tags associated with OFPP_NORMAL actions. */ |
132 | bool may_set_up_flow; /* True ordinarily; false if the actions must | |
133 | * be reassessed for every packet. */ | |
134 | uint16_t nf_output_iface; /* Output interface index for NetFlow. */ | |
135 | ||
136 | /* xlate_actions() initializes and uses these members, but the client has no | |
137 | * reason to look at them. */ | |
138 | ||
139 | int recurse; /* Recursion level, via xlate_table_action. */ | |
cdee00fd BP |
140 | int last_pop_priority; /* Offset in 'odp_actions' just past most |
141 | * recently added ODPAT_SET_PRIORITY. */ | |
f29152ca BP |
142 | }; |
143 | ||
144 | static void action_xlate_ctx_init(struct action_xlate_ctx *, | |
145 | struct ofproto *, const struct flow *, | |
146 | const struct ofpbuf *); | |
cdee00fd BP |
147 | static struct ofpbuf *xlate_actions(struct action_xlate_ctx *, |
148 | const union ofp_action *in, size_t n_in); | |
064af421 | 149 | |
bcf84111 | 150 | /* An OpenFlow flow. */ |
064af421 | 151 | struct rule { |
0c43ad9a | 152 | long long int used; /* Time last used; time created if not used. */ |
064af421 | 153 | long long int created; /* Creation time. */ |
064af421 | 154 | |
bcf84111 | 155 | /* These statistics: |
064af421 | 156 | * |
bcf84111 BP |
157 | * - Do include packets and bytes from facets that have been deleted or |
158 | * whose own statistics have been folded into the rule. | |
79eee1eb | 159 | * |
bcf84111 BP |
160 | * - Do include packets and bytes sent "by hand" that were accounted to |
161 | * the rule without any facet being involved (this is a rare corner | |
162 | * case in rule_execute()). | |
064af421 | 163 | * |
bcf84111 BP |
164 | * - Do not include packet or bytes that can be obtained from any facet's |
165 | * packet_count or byte_count member or that can be obtained from the | |
166 | * datapath by, e.g., dpif_flow_get() for any facet. | |
167 | */ | |
168 | uint64_t packet_count; /* Number of packets received. */ | |
169 | uint64_t byte_count; /* Number of bytes received. */ | |
170 | ||
171 | ovs_be64 flow_cookie; /* Controller-issued identifier. */ | |
172 | ||
173 | struct cls_rule cr; /* In owning ofproto's classifier. */ | |
174 | uint16_t idle_timeout; /* In seconds from time of last use. */ | |
175 | uint16_t hard_timeout; /* In seconds from time of creation. */ | |
176 | bool send_flow_removed; /* Send a flow removed message? */ | |
177 | int n_actions; /* Number of elements in actions[]. */ | |
178 | union ofp_action *actions; /* OpenFlow actions. */ | |
179 | struct list facets; /* List of "struct facet"s. */ | |
064af421 BP |
180 | }; |
181 | ||
bcf84111 BP |
182 | static struct rule *rule_from_cls_rule(const struct cls_rule *); |
183 | static bool rule_is_hidden(const struct rule *); | |
064af421 | 184 | |
bcf84111 | 185 | static struct rule *rule_create(const struct cls_rule *, |
0193b2af | 186 | const union ofp_action *, size_t n_actions, |
ca069229 | 187 | uint16_t idle_timeout, uint16_t hard_timeout, |
8054fc48 | 188 | ovs_be64 flow_cookie, bool send_flow_removed); |
064af421 | 189 | static void rule_destroy(struct ofproto *, struct rule *); |
bcf84111 BP |
190 | static void rule_free(struct rule *); |
191 | ||
192 | static struct rule *rule_lookup(struct ofproto *, const struct flow *); | |
afe75089 | 193 | static void rule_insert(struct ofproto *, struct rule *); |
064af421 | 194 | static void rule_remove(struct ofproto *, struct rule *); |
bcf84111 BP |
195 | |
196 | static void rule_send_removed(struct ofproto *, struct rule *, uint8_t reason); | |
197 | ||
198 | /* An exact-match instantiation of an OpenFlow flow. */ | |
199 | struct facet { | |
200 | long long int used; /* Time last used; time created if not used. */ | |
201 | ||
202 | /* These statistics: | |
203 | * | |
204 | * - Do include packets and bytes sent "by hand", e.g. with | |
205 | * dpif_execute(). | |
206 | * | |
207 | * - Do include packets and bytes that were obtained from the datapath | |
208 | * when a flow was deleted (e.g. dpif_flow_del()) or when its | |
209 | * statistics were reset (e.g. dpif_flow_put() with ODPPF_ZERO_STATS). | |
210 | * | |
211 | * - Do not include any packets or bytes that can currently be obtained | |
212 | * from the datapath by, e.g., dpif_flow_get(). | |
213 | */ | |
214 | uint64_t packet_count; /* Number of packets received. */ | |
215 | uint64_t byte_count; /* Number of bytes received. */ | |
216 | ||
217 | /* Number of bytes passed to account_cb. This may include bytes that can | |
218 | * currently obtained from the datapath (thus, it can be greater than | |
219 | * byte_count). */ | |
220 | uint64_t accounted_bytes; | |
221 | ||
222 | struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */ | |
223 | struct list list_node; /* In owning rule's 'facets' list. */ | |
224 | struct rule *rule; /* Owning rule. */ | |
225 | struct flow flow; /* Exact-match flow. */ | |
226 | bool installed; /* Installed in datapath? */ | |
227 | bool may_install; /* True ordinarily; false if actions must | |
228 | * be reassessed for every packet. */ | |
cf22f8cb | 229 | size_t actions_len; /* Number of bytes in actions[]. */ |
cdee00fd | 230 | struct nlattr *actions; /* Datapath actions. */ |
bcf84111 BP |
231 | tag_type tags; /* Tags (set only by hooks). */ |
232 | struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ | |
233 | }; | |
234 | ||
235 | static struct facet *facet_create(struct ofproto *, struct rule *, | |
236 | const struct flow *, | |
237 | const struct ofpbuf *packet); | |
238 | static void facet_remove(struct ofproto *, struct facet *); | |
239 | static void facet_free(struct facet *); | |
240 | ||
241 | static struct facet *facet_lookup_valid(struct ofproto *, const struct flow *); | |
242 | static bool facet_revalidate(struct ofproto *, struct facet *); | |
243 | ||
244 | static void facet_install(struct ofproto *, struct facet *, bool zero_stats); | |
245 | static void facet_uninstall(struct ofproto *, struct facet *); | |
d530fcd2 | 246 | static void facet_flush_stats(struct ofproto *, struct facet *); |
bcf84111 | 247 | |
7f7ae89d BP |
248 | static void facet_make_actions(struct ofproto *, struct facet *, |
249 | const struct ofpbuf *packet); | |
bcf84111 BP |
250 | static void facet_update_stats(struct ofproto *, struct facet *, |
251 | const struct odp_flow_stats *); | |
064af421 | 252 | |
76ce9432 BP |
253 | /* ofproto supports two kinds of OpenFlow connections: |
254 | * | |
5899143f BP |
255 | * - "Primary" connections to ordinary OpenFlow controllers. ofproto |
256 | * maintains persistent connections to these controllers and by default | |
257 | * sends them asynchronous messages such as packet-ins. | |
76ce9432 | 258 | * |
5899143f | 259 | * - "Service" connections, e.g. from ovs-ofctl. When these connections |
76ce9432 BP |
260 | * drop, it is the other side's responsibility to reconnect them if |
261 | * necessary. ofproto does not send them asynchronous messages by default. | |
7d674866 BP |
262 | * |
263 | * Currently, active (tcp, ssl, unix) connections are always "primary" | |
264 | * connections and passive (ptcp, pssl, punix) connections are always "service" | |
265 | * connections. There is no inherent reason for this, but it reflects the | |
266 | * common case. | |
76ce9432 BP |
267 | */ |
268 | enum ofconn_type { | |
5899143f BP |
269 | OFCONN_PRIMARY, /* An ordinary OpenFlow controller. */ |
270 | OFCONN_SERVICE /* A service connection, e.g. "ovs-ofctl". */ | |
76ce9432 | 271 | }; |
064af421 | 272 | |
7d674866 BP |
273 | /* A listener for incoming OpenFlow "service" connections. */ |
274 | struct ofservice { | |
275 | struct hmap_node node; /* In struct ofproto's "services" hmap. */ | |
276 | struct pvconn *pvconn; /* OpenFlow connection listener. */ | |
277 | ||
278 | /* These are not used by ofservice directly. They are settings for | |
279 | * accepted "struct ofconn"s from the pvconn. */ | |
280 | int probe_interval; /* Max idle time before probing, in seconds. */ | |
281 | int rate_limit; /* Max packet-in rate in packets per second. */ | |
282 | int burst_limit; /* Limit on accumulating packet credits. */ | |
283 | }; | |
284 | ||
285 | static struct ofservice *ofservice_lookup(struct ofproto *, | |
286 | const char *target); | |
287 | static int ofservice_create(struct ofproto *, | |
288 | const struct ofproto_controller *); | |
289 | static void ofservice_reconfigure(struct ofservice *, | |
290 | const struct ofproto_controller *); | |
291 | static void ofservice_destroy(struct ofproto *, struct ofservice *); | |
292 | ||
76ce9432 BP |
293 | /* An OpenFlow connection. */ |
294 | struct ofconn { | |
295 | struct ofproto *ofproto; /* The ofproto that owns this connection. */ | |
296 | struct list node; /* In struct ofproto's "all_conns" list. */ | |
297 | struct rconn *rconn; /* OpenFlow connection. */ | |
298 | enum ofconn_type type; /* Type. */ | |
492f7572 | 299 | enum nx_flow_format flow_format; /* Currently selected flow format. */ |
76ce9432 BP |
300 | |
301 | /* OFPT_PACKET_IN related data. */ | |
302 | struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */ | |
303 | struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */ | |
304 | struct pktbuf *pktbuf; /* OpenFlow packet buffers. */ | |
305 | int miss_send_len; /* Bytes to send of buffered packets. */ | |
306 | ||
307 | /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow | |
308 | * requests, and the maximum number before we stop reading OpenFlow | |
309 | * requests. */ | |
064af421 BP |
310 | #define OFCONN_REPLY_MAX 100 |
311 | struct rconn_packet_counter *reply_counter; | |
76ce9432 | 312 | |
5899143f | 313 | /* type == OFCONN_PRIMARY only. */ |
9deba63b | 314 | enum nx_role role; /* Role. */ |
76ce9432 BP |
315 | struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */ |
316 | struct discovery *discovery; /* Controller discovery object, if enabled. */ | |
317 | struct status_category *ss; /* Switch status category. */ | |
d2ede7bc | 318 | enum ofproto_band band; /* In-band or out-of-band? */ |
064af421 BP |
319 | }; |
320 | ||
76ce9432 BP |
321 | /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's |
322 | * "schedulers" array. Their values are 0 and 1, and their meanings and values | |
323 | * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In | |
324 | * case anything ever changes, check their values here. */ | |
325 | #define N_SCHEDULERS 2 | |
326 | BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0); | |
327 | BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR); | |
328 | BUILD_ASSERT_DECL(OFPR_ACTION == 1); | |
329 | BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR); | |
330 | ||
331 | static struct ofconn *ofconn_create(struct ofproto *, struct rconn *, | |
332 | enum ofconn_type); | |
c475ae67 | 333 | static void ofconn_destroy(struct ofconn *); |
3269c562 | 334 | static void ofconn_run(struct ofconn *); |
064af421 | 335 | static void ofconn_wait(struct ofconn *); |
c91248b3 | 336 | static bool ofconn_receives_async_msgs(const struct ofconn *); |
eb15cdbb | 337 | static char *ofconn_make_name(const struct ofproto *, const char *target); |
7d674866 | 338 | static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst); |
c91248b3 | 339 | |
064af421 BP |
340 | static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn, |
341 | struct rconn_packet_counter *counter); | |
342 | ||
856081f6 BP |
343 | static void send_packet_in(struct ofproto *, struct dpif_upcall *, |
344 | const struct flow *, bool clone); | |
345 | static void do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn); | |
76ce9432 | 346 | |
064af421 BP |
347 | struct ofproto { |
348 | /* Settings. */ | |
349 | uint64_t datapath_id; /* Datapath ID. */ | |
350 | uint64_t fallback_dpid; /* Datapath ID if no better choice found. */ | |
5a719c38 JP |
351 | char *mfr_desc; /* Manufacturer. */ |
352 | char *hw_desc; /* Hardware. */ | |
353 | char *sw_desc; /* Software version. */ | |
354 | char *serial_desc; /* Serial number. */ | |
8abc4ed7 | 355 | char *dp_desc; /* Datapath description. */ |
064af421 BP |
356 | |
357 | /* Datapath. */ | |
c228a364 | 358 | struct dpif *dpif; |
e9e28be3 | 359 | struct netdev_monitor *netdev_monitor; |
ca0f572c | 360 | struct hmap ports; /* Contains "struct ofport"s. */ |
064af421 BP |
361 | struct shash port_by_name; |
362 | uint32_t max_ports; | |
363 | ||
364 | /* Configuration. */ | |
365 | struct switch_status *switch_status; | |
064af421 | 366 | struct fail_open *fail_open; |
064af421 | 367 | struct netflow *netflow; |
72b06300 | 368 | struct ofproto_sflow *sflow; |
064af421 | 369 | |
d2ede7bc BP |
370 | /* In-band control. */ |
371 | struct in_band *in_band; | |
372 | long long int next_in_band_update; | |
917e50e1 BP |
373 | struct sockaddr_in *extra_in_band_remotes; |
374 | size_t n_extra_remotes; | |
b1da6250 | 375 | int in_band_queue; |
917e50e1 | 376 | |
064af421 BP |
377 | /* Flow table. */ |
378 | struct classifier cls; | |
064af421 | 379 | long long int next_expiration; |
bcf84111 BP |
380 | |
381 | /* Facets. */ | |
382 | struct hmap facets; | |
383 | bool need_revalidate; | |
064af421 BP |
384 | struct tag_set revalidate_set; |
385 | ||
386 | /* OpenFlow connections. */ | |
76ce9432 BP |
387 | struct hmap controllers; /* Controller "struct ofconn"s. */ |
388 | struct list all_conns; /* Contains "struct ofconn"s. */ | |
31681a5d | 389 | enum ofproto_fail_mode fail_mode; |
7d674866 BP |
390 | |
391 | /* OpenFlow listeners. */ | |
392 | struct hmap services; /* Contains "struct ofservice"s. */ | |
064af421 BP |
393 | struct pvconn **snoops; |
394 | size_t n_snoops; | |
395 | ||
396 | /* Hooks for ovs-vswitchd. */ | |
397 | const struct ofhooks *ofhooks; | |
398 | void *aux; | |
399 | ||
400 | /* Used by default ofhooks. */ | |
401 | struct mac_learning *ml; | |
402 | }; | |
403 | ||
7aa697dd BP |
404 | /* Map from dpif name to struct ofproto, for use by unixctl commands. */ |
405 | static struct shash all_ofprotos = SHASH_INITIALIZER(&all_ofprotos); | |
406 | ||
064af421 BP |
407 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); |
408 | ||
409 | static const struct ofhooks default_ofhooks; | |
410 | ||
fa60c019 | 411 | static uint64_t pick_datapath_id(const struct ofproto *); |
064af421 | 412 | static uint64_t pick_fallback_dpid(void); |
76ce9432 | 413 | |
0de7a4b4 | 414 | static int ofproto_expire(struct ofproto *); |
4a4cdb3b | 415 | |
856081f6 | 416 | static void handle_upcall(struct ofproto *, struct dpif_upcall *); |
064af421 | 417 | |
3269c562 | 418 | static void handle_openflow(struct ofconn *, struct ofpbuf *); |
064af421 | 419 | |
ca0f572c | 420 | static struct ofport *get_port(const struct ofproto *, uint16_t odp_port); |
064af421 BP |
421 | static void update_port(struct ofproto *, const char *devname); |
422 | static int init_ports(struct ofproto *); | |
423 | static void reinit_ports(struct ofproto *); | |
424 | ||
7aa697dd BP |
425 | static void ofproto_unixctl_init(void); |
426 | ||
064af421 | 427 | int |
1a6f1e2a JG |
428 | ofproto_create(const char *datapath, const char *datapath_type, |
429 | const struct ofhooks *ofhooks, void *aux, | |
064af421 BP |
430 | struct ofproto **ofprotop) |
431 | { | |
064af421 | 432 | struct ofproto *p; |
c228a364 | 433 | struct dpif *dpif; |
064af421 BP |
434 | int error; |
435 | ||
436 | *ofprotop = NULL; | |
437 | ||
7aa697dd BP |
438 | ofproto_unixctl_init(); |
439 | ||
064af421 | 440 | /* Connect to datapath and start listening for messages. */ |
1a6f1e2a | 441 | error = dpif_open(datapath, datapath_type, &dpif); |
064af421 BP |
442 | if (error) { |
443 | VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error)); | |
444 | return error; | |
445 | } | |
72b06300 | 446 | error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW); |
064af421 BP |
447 | if (error) { |
448 | VLOG_ERR("failed to listen on datapath %s: %s", | |
449 | datapath, strerror(error)); | |
c228a364 | 450 | dpif_close(dpif); |
064af421 BP |
451 | return error; |
452 | } | |
c228a364 | 453 | dpif_flow_flush(dpif); |
8f24562a | 454 | dpif_recv_purge(dpif); |
064af421 BP |
455 | |
456 | /* Initialize settings. */ | |
ec6fde61 | 457 | p = xzalloc(sizeof *p); |
064af421 | 458 | p->fallback_dpid = pick_fallback_dpid(); |
fa60c019 | 459 | p->datapath_id = p->fallback_dpid; |
5a719c38 JP |
460 | p->mfr_desc = xstrdup(DEFAULT_MFR_DESC); |
461 | p->hw_desc = xstrdup(DEFAULT_HW_DESC); | |
462 | p->sw_desc = xstrdup(DEFAULT_SW_DESC); | |
463 | p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC); | |
23ff2821 | 464 | p->dp_desc = xstrdup(DEFAULT_DP_DESC); |
064af421 BP |
465 | |
466 | /* Initialize datapath. */ | |
467 | p->dpif = dpif; | |
8b61709d | 468 | p->netdev_monitor = netdev_monitor_create(); |
ca0f572c | 469 | hmap_init(&p->ports); |
064af421 | 470 | shash_init(&p->port_by_name); |
996c1b3d | 471 | p->max_ports = dpif_get_max_ports(dpif); |
064af421 BP |
472 | |
473 | /* Initialize submodules. */ | |
474 | p->switch_status = switch_status_create(p); | |
064af421 | 475 | p->fail_open = NULL; |
064af421 | 476 | p->netflow = NULL; |
72b06300 | 477 | p->sflow = NULL; |
064af421 | 478 | |
b1da6250 BP |
479 | /* Initialize in-band control. */ |
480 | p->in_band = NULL; | |
481 | p->in_band_queue = -1; | |
482 | ||
064af421 BP |
483 | /* Initialize flow table. */ |
484 | classifier_init(&p->cls); | |
064af421 | 485 | p->next_expiration = time_msec() + 1000; |
bcf84111 BP |
486 | |
487 | /* Initialize facet table. */ | |
488 | hmap_init(&p->facets); | |
489 | p->need_revalidate = false; | |
064af421 BP |
490 | tag_set_init(&p->revalidate_set); |
491 | ||
492 | /* Initialize OpenFlow connections. */ | |
493 | list_init(&p->all_conns); | |
76ce9432 | 494 | hmap_init(&p->controllers); |
7d674866 | 495 | hmap_init(&p->services); |
064af421 BP |
496 | p->snoops = NULL; |
497 | p->n_snoops = 0; | |
498 | ||
499 | /* Initialize hooks. */ | |
500 | if (ofhooks) { | |
501 | p->ofhooks = ofhooks; | |
502 | p->aux = aux; | |
503 | p->ml = NULL; | |
504 | } else { | |
505 | p->ofhooks = &default_ofhooks; | |
506 | p->aux = p; | |
507 | p->ml = mac_learning_create(); | |
508 | } | |
509 | ||
fa60c019 BP |
510 | /* Pick final datapath ID. */ |
511 | p->datapath_id = pick_datapath_id(p); | |
b123cc3c | 512 | VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id); |
fa60c019 | 513 | |
7aa697dd BP |
514 | shash_add_once(&all_ofprotos, dpif_name(p->dpif), p); |
515 | ||
064af421 BP |
516 | *ofprotop = p; |
517 | return 0; | |
518 | } | |
519 | ||
520 | void | |
521 | ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id) | |
522 | { | |
523 | uint64_t old_dpid = p->datapath_id; | |
fa60c019 | 524 | p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p); |
064af421 | 525 | if (p->datapath_id != old_dpid) { |
b123cc3c | 526 | VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id); |
76ce9432 BP |
527 | |
528 | /* Force all active connections to reconnect, since there is no way to | |
529 | * notify a controller that the datapath ID has changed. */ | |
fa05809b | 530 | ofproto_reconnect_controllers(p); |
064af421 BP |
531 | } |
532 | } | |
533 | ||
76ce9432 BP |
534 | static bool |
535 | is_discovery_controller(const struct ofproto_controller *c) | |
536 | { | |
537 | return !strcmp(c->target, "discover"); | |
538 | } | |
539 | ||
540 | static bool | |
541 | is_in_band_controller(const struct ofproto_controller *c) | |
542 | { | |
543 | return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND; | |
544 | } | |
545 | ||
546 | /* Creates a new controller in 'ofproto'. Some of the settings are initially | |
547 | * drawn from 'c', but update_controller() needs to be called later to finish | |
548 | * the new ofconn's configuration. */ | |
549 | static void | |
550 | add_controller(struct ofproto *ofproto, const struct ofproto_controller *c) | |
551 | { | |
552 | struct discovery *discovery; | |
553 | struct ofconn *ofconn; | |
554 | ||
555 | if (is_discovery_controller(c)) { | |
556 | int error = discovery_create(c->accept_re, c->update_resolv_conf, | |
557 | ofproto->dpif, ofproto->switch_status, | |
558 | &discovery); | |
559 | if (error) { | |
560 | return; | |
561 | } | |
562 | } else { | |
563 | discovery = NULL; | |
564 | } | |
565 | ||
5899143f | 566 | ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY); |
76ce9432 BP |
567 | ofconn->pktbuf = pktbuf_create(); |
568 | ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN; | |
569 | if (discovery) { | |
570 | ofconn->discovery = discovery; | |
571 | } else { | |
eb15cdbb BP |
572 | char *name = ofconn_make_name(ofproto, c->target); |
573 | rconn_connect(ofconn->rconn, c->target, name); | |
574 | free(name); | |
76ce9432 BP |
575 | } |
576 | hmap_insert(&ofproto->controllers, &ofconn->hmap_node, | |
577 | hash_string(c->target, 0)); | |
578 | } | |
579 | ||
580 | /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's | |
581 | * target or turn discovery on or off (these are done by creating new ofconns | |
582 | * and deleting old ones), but it can update the rest of an ofconn's | |
583 | * settings. */ | |
584 | static void | |
585 | update_controller(struct ofconn *ofconn, const struct ofproto_controller *c) | |
064af421 | 586 | { |
76ce9432 | 587 | int probe_interval; |
79c9f2ee | 588 | |
d2ede7bc BP |
589 | ofconn->band = (is_in_band_controller(c) |
590 | ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND); | |
591 | ||
76ce9432 | 592 | rconn_set_max_backoff(ofconn->rconn, c->max_backoff); |
79c9f2ee | 593 | |
76ce9432 BP |
594 | probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0; |
595 | rconn_set_probe_interval(ofconn->rconn, probe_interval); | |
79c9f2ee | 596 | |
76ce9432 BP |
597 | if (ofconn->discovery) { |
598 | discovery_set_update_resolv_conf(ofconn->discovery, | |
599 | c->update_resolv_conf); | |
600 | discovery_set_accept_controller_re(ofconn->discovery, c->accept_re); | |
601 | } | |
79c9f2ee | 602 | |
7d674866 | 603 | ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit); |
76ce9432 | 604 | } |
79c9f2ee | 605 | |
76ce9432 BP |
606 | static const char * |
607 | ofconn_get_target(const struct ofconn *ofconn) | |
608 | { | |
eb15cdbb | 609 | return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn); |
76ce9432 BP |
610 | } |
611 | ||
612 | static struct ofconn * | |
613 | find_controller_by_target(struct ofproto *ofproto, const char *target) | |
614 | { | |
615 | struct ofconn *ofconn; | |
616 | ||
4e8e4213 | 617 | HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node, |
76ce9432 BP |
618 | hash_string(target, 0), &ofproto->controllers) { |
619 | if (!strcmp(ofconn_get_target(ofconn), target)) { | |
620 | return ofconn; | |
79c9f2ee | 621 | } |
064af421 | 622 | } |
76ce9432 BP |
623 | return NULL; |
624 | } | |
064af421 | 625 | |
d2ede7bc BP |
626 | static void |
627 | update_in_band_remotes(struct ofproto *ofproto) | |
628 | { | |
629 | const struct ofconn *ofconn; | |
630 | struct sockaddr_in *addrs; | |
917e50e1 | 631 | size_t max_addrs, n_addrs; |
d2ede7bc | 632 | bool discovery; |
917e50e1 | 633 | size_t i; |
d2ede7bc | 634 | |
917e50e1 BP |
635 | /* Allocate enough memory for as many remotes as we could possibly have. */ |
636 | max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers); | |
637 | addrs = xmalloc(max_addrs * sizeof *addrs); | |
d2ede7bc BP |
638 | n_addrs = 0; |
639 | ||
640 | /* Add all the remotes. */ | |
641 | discovery = false; | |
4e8e4213 | 642 | HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) { |
d2ede7bc BP |
643 | struct sockaddr_in *sin = &addrs[n_addrs]; |
644 | ||
487ec65f BP |
645 | if (ofconn->band == OFPROTO_OUT_OF_BAND) { |
646 | continue; | |
647 | } | |
648 | ||
d2ede7bc BP |
649 | sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn); |
650 | if (sin->sin_addr.s_addr) { | |
651 | sin->sin_port = rconn_get_remote_port(ofconn->rconn); | |
652 | n_addrs++; | |
653 | } | |
654 | if (ofconn->discovery) { | |
655 | discovery = true; | |
656 | } | |
657 | } | |
917e50e1 BP |
658 | for (i = 0; i < ofproto->n_extra_remotes; i++) { |
659 | addrs[n_addrs++] = ofproto->extra_in_band_remotes[i]; | |
660 | } | |
d2ede7bc BP |
661 | |
662 | /* Create or update or destroy in-band. | |
663 | * | |
664 | * Ordinarily we only enable in-band if there's at least one remote | |
665 | * address, but discovery needs the in-band rules for DHCP to be installed | |
666 | * even before we know any remote addresses. */ | |
667 | if (n_addrs || discovery) { | |
668 | if (!ofproto->in_band) { | |
669 | in_band_create(ofproto, ofproto->dpif, ofproto->switch_status, | |
670 | &ofproto->in_band); | |
671 | } | |
40cae670 BP |
672 | if (ofproto->in_band) { |
673 | in_band_set_remotes(ofproto->in_band, addrs, n_addrs); | |
674 | } | |
b1da6250 | 675 | in_band_set_queue(ofproto->in_band, ofproto->in_band_queue); |
d2ede7bc BP |
676 | ofproto->next_in_band_update = time_msec() + 1000; |
677 | } else { | |
678 | in_band_destroy(ofproto->in_band); | |
679 | ofproto->in_band = NULL; | |
680 | } | |
681 | ||
682 | /* Clean up. */ | |
683 | free(addrs); | |
684 | } | |
685 | ||
31681a5d JP |
686 | static void |
687 | update_fail_open(struct ofproto *p) | |
688 | { | |
689 | struct ofconn *ofconn; | |
690 | ||
691 | if (!hmap_is_empty(&p->controllers) | |
692 | && p->fail_mode == OFPROTO_FAIL_STANDALONE) { | |
693 | struct rconn **rconns; | |
694 | size_t n; | |
695 | ||
696 | if (!p->fail_open) { | |
697 | p->fail_open = fail_open_create(p, p->switch_status); | |
698 | } | |
699 | ||
700 | n = 0; | |
701 | rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns); | |
4e8e4213 | 702 | HMAP_FOR_EACH (ofconn, hmap_node, &p->controllers) { |
31681a5d JP |
703 | rconns[n++] = ofconn->rconn; |
704 | } | |
705 | ||
706 | fail_open_set_controllers(p->fail_open, rconns, n); | |
707 | /* p->fail_open takes ownership of 'rconns'. */ | |
708 | } else { | |
709 | fail_open_destroy(p->fail_open); | |
710 | p->fail_open = NULL; | |
711 | } | |
712 | } | |
713 | ||
76ce9432 BP |
714 | void |
715 | ofproto_set_controllers(struct ofproto *p, | |
716 | const struct ofproto_controller *controllers, | |
717 | size_t n_controllers) | |
718 | { | |
719 | struct shash new_controllers; | |
7d674866 BP |
720 | struct ofconn *ofconn, *next_ofconn; |
721 | struct ofservice *ofservice, *next_ofservice; | |
76ce9432 | 722 | bool ss_exists; |
76ce9432 | 723 | size_t i; |
79c9f2ee | 724 | |
7d674866 BP |
725 | /* Create newly configured controllers and services. |
726 | * Create a name to ofproto_controller mapping in 'new_controllers'. */ | |
76ce9432 BP |
727 | shash_init(&new_controllers); |
728 | for (i = 0; i < n_controllers; i++) { | |
729 | const struct ofproto_controller *c = &controllers[i]; | |
730 | ||
7d674866 BP |
731 | if (!vconn_verify_name(c->target) || !strcmp(c->target, "discover")) { |
732 | if (!find_controller_by_target(p, c->target)) { | |
733 | add_controller(p, c); | |
734 | } | |
735 | } else if (!pvconn_verify_name(c->target)) { | |
736 | if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) { | |
737 | continue; | |
738 | } | |
739 | } else { | |
740 | VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"", | |
741 | dpif_name(p->dpif), c->target); | |
742 | continue; | |
76ce9432 | 743 | } |
7d674866 BP |
744 | |
745 | shash_add_once(&new_controllers, c->target, &controllers[i]); | |
76ce9432 BP |
746 | } |
747 | ||
7d674866 BP |
748 | /* Delete controllers that are no longer configured. |
749 | * Update configuration of all now-existing controllers. */ | |
76ce9432 | 750 | ss_exists = false; |
4e8e4213 | 751 | HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &p->controllers) { |
76ce9432 BP |
752 | struct ofproto_controller *c; |
753 | ||
754 | c = shash_find_data(&new_controllers, ofconn_get_target(ofconn)); | |
755 | if (!c) { | |
756 | ofconn_destroy(ofconn); | |
79c9f2ee | 757 | } else { |
76ce9432 | 758 | update_controller(ofconn, c); |
76ce9432 BP |
759 | if (ofconn->ss) { |
760 | ss_exists = true; | |
761 | } | |
76ce9432 BP |
762 | } |
763 | } | |
7d674866 BP |
764 | |
765 | /* Delete services that are no longer configured. | |
766 | * Update configuration of all now-existing services. */ | |
4e8e4213 | 767 | HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) { |
7d674866 BP |
768 | struct ofproto_controller *c; |
769 | ||
770 | c = shash_find_data(&new_controllers, | |
771 | pvconn_get_name(ofservice->pvconn)); | |
772 | if (!c) { | |
773 | ofservice_destroy(p, ofservice); | |
774 | } else { | |
775 | ofservice_reconfigure(ofservice, c); | |
776 | } | |
777 | } | |
778 | ||
76ce9432 BP |
779 | shash_destroy(&new_controllers); |
780 | ||
d2ede7bc | 781 | update_in_band_remotes(p); |
31681a5d | 782 | update_fail_open(p); |
79c9f2ee | 783 | |
76ce9432 BP |
784 | if (!hmap_is_empty(&p->controllers) && !ss_exists) { |
785 | ofconn = CONTAINER_OF(hmap_first(&p->controllers), | |
786 | struct ofconn, hmap_node); | |
787 | ofconn->ss = switch_status_register(p->switch_status, "remote", | |
788 | rconn_status_cb, ofconn->rconn); | |
79c9f2ee | 789 | } |
064af421 BP |
790 | } |
791 | ||
31681a5d JP |
792 | void |
793 | ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode) | |
794 | { | |
795 | p->fail_mode = fail_mode; | |
796 | update_fail_open(p); | |
797 | } | |
798 | ||
fa05809b BP |
799 | /* Drops the connections between 'ofproto' and all of its controllers, forcing |
800 | * them to reconnect. */ | |
801 | void | |
802 | ofproto_reconnect_controllers(struct ofproto *ofproto) | |
803 | { | |
804 | struct ofconn *ofconn; | |
805 | ||
4e8e4213 | 806 | LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) { |
fa05809b BP |
807 | rconn_reconnect(ofconn->rconn); |
808 | } | |
809 | } | |
810 | ||
917e50e1 BP |
811 | static bool |
812 | any_extras_changed(const struct ofproto *ofproto, | |
813 | const struct sockaddr_in *extras, size_t n) | |
814 | { | |
815 | size_t i; | |
816 | ||
817 | if (n != ofproto->n_extra_remotes) { | |
818 | return true; | |
819 | } | |
820 | ||
821 | for (i = 0; i < n; i++) { | |
822 | const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i]; | |
823 | const struct sockaddr_in *new = &extras[i]; | |
824 | ||
825 | if (old->sin_addr.s_addr != new->sin_addr.s_addr || | |
826 | old->sin_port != new->sin_port) { | |
827 | return true; | |
828 | } | |
829 | } | |
830 | ||
831 | return false; | |
832 | } | |
833 | ||
834 | /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s | |
835 | * in-band control should guarantee access, in the same way that in-band | |
836 | * control guarantees access to OpenFlow controllers. */ | |
837 | void | |
838 | ofproto_set_extra_in_band_remotes(struct ofproto *ofproto, | |
839 | const struct sockaddr_in *extras, size_t n) | |
840 | { | |
841 | if (!any_extras_changed(ofproto, extras, n)) { | |
842 | return; | |
843 | } | |
844 | ||
845 | free(ofproto->extra_in_band_remotes); | |
846 | ofproto->n_extra_remotes = n; | |
847 | ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras); | |
848 | ||
849 | update_in_band_remotes(ofproto); | |
850 | } | |
851 | ||
b1da6250 BP |
852 | /* Sets the OpenFlow queue used by flows set up by in-band control on |
853 | * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control | |
854 | * flows will use the default queue. */ | |
855 | void | |
856 | ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id) | |
857 | { | |
858 | if (queue_id != ofproto->in_band_queue) { | |
859 | ofproto->in_band_queue = queue_id; | |
860 | update_in_band_remotes(ofproto); | |
861 | } | |
862 | } | |
863 | ||
064af421 BP |
864 | void |
865 | ofproto_set_desc(struct ofproto *p, | |
5a719c38 JP |
866 | const char *mfr_desc, const char *hw_desc, |
867 | const char *sw_desc, const char *serial_desc, | |
8abc4ed7 | 868 | const char *dp_desc) |
064af421 | 869 | { |
5a719c38 JP |
870 | struct ofp_desc_stats *ods; |
871 | ||
872 | if (mfr_desc) { | |
873 | if (strlen(mfr_desc) >= sizeof ods->mfr_desc) { | |
874 | VLOG_WARN("truncating mfr_desc, must be less than %zu characters", | |
875 | sizeof ods->mfr_desc); | |
876 | } | |
877 | free(p->mfr_desc); | |
878 | p->mfr_desc = xstrdup(mfr_desc); | |
064af421 | 879 | } |
5a719c38 JP |
880 | if (hw_desc) { |
881 | if (strlen(hw_desc) >= sizeof ods->hw_desc) { | |
882 | VLOG_WARN("truncating hw_desc, must be less than %zu characters", | |
883 | sizeof ods->hw_desc); | |
884 | } | |
885 | free(p->hw_desc); | |
886 | p->hw_desc = xstrdup(hw_desc); | |
064af421 | 887 | } |
5a719c38 JP |
888 | if (sw_desc) { |
889 | if (strlen(sw_desc) >= sizeof ods->sw_desc) { | |
890 | VLOG_WARN("truncating sw_desc, must be less than %zu characters", | |
891 | sizeof ods->sw_desc); | |
892 | } | |
893 | free(p->sw_desc); | |
894 | p->sw_desc = xstrdup(sw_desc); | |
895 | } | |
896 | if (serial_desc) { | |
897 | if (strlen(serial_desc) >= sizeof ods->serial_num) { | |
898 | VLOG_WARN("truncating serial_desc, must be less than %zu " | |
899 | "characters", | |
900 | sizeof ods->serial_num); | |
901 | } | |
902 | free(p->serial_desc); | |
903 | p->serial_desc = xstrdup(serial_desc); | |
064af421 | 904 | } |
8abc4ed7 | 905 | if (dp_desc) { |
5a719c38 JP |
906 | if (strlen(dp_desc) >= sizeof ods->dp_desc) { |
907 | VLOG_WARN("truncating dp_desc, must be less than %zu characters", | |
908 | sizeof ods->dp_desc); | |
909 | } | |
8abc4ed7 JP |
910 | free(p->dp_desc); |
911 | p->dp_desc = xstrdup(dp_desc); | |
912 | } | |
064af421 BP |
913 | } |
914 | ||
064af421 BP |
915 | static int |
916 | set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp, | |
917 | const struct svec *svec) | |
918 | { | |
919 | struct pvconn **pvconns = *pvconnsp; | |
920 | size_t n_pvconns = *n_pvconnsp; | |
921 | int retval = 0; | |
922 | size_t i; | |
923 | ||
924 | for (i = 0; i < n_pvconns; i++) { | |
925 | pvconn_close(pvconns[i]); | |
926 | } | |
927 | free(pvconns); | |
928 | ||
929 | pvconns = xmalloc(svec->n * sizeof *pvconns); | |
930 | n_pvconns = 0; | |
931 | for (i = 0; i < svec->n; i++) { | |
932 | const char *name = svec->names[i]; | |
933 | struct pvconn *pvconn; | |
934 | int error; | |
935 | ||
936 | error = pvconn_open(name, &pvconn); | |
937 | if (!error) { | |
938 | pvconns[n_pvconns++] = pvconn; | |
939 | } else { | |
940 | VLOG_ERR("failed to listen on %s: %s", name, strerror(error)); | |
941 | if (!retval) { | |
942 | retval = error; | |
943 | } | |
944 | } | |
945 | } | |
946 | ||
947 | *pvconnsp = pvconns; | |
948 | *n_pvconnsp = n_pvconns; | |
949 | ||
950 | return retval; | |
951 | } | |
952 | ||
064af421 BP |
953 | int |
954 | ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops) | |
955 | { | |
956 | return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops); | |
957 | } | |
958 | ||
959 | int | |
0193b2af JG |
960 | ofproto_set_netflow(struct ofproto *ofproto, |
961 | const struct netflow_options *nf_options) | |
064af421 | 962 | { |
76343538 | 963 | if (nf_options && nf_options->collectors.n) { |
064af421 BP |
964 | if (!ofproto->netflow) { |
965 | ofproto->netflow = netflow_create(); | |
966 | } | |
0193b2af | 967 | return netflow_set_options(ofproto->netflow, nf_options); |
064af421 BP |
968 | } else { |
969 | netflow_destroy(ofproto->netflow); | |
970 | ofproto->netflow = NULL; | |
971 | return 0; | |
972 | } | |
973 | } | |
974 | ||
72b06300 BP |
975 | void |
976 | ofproto_set_sflow(struct ofproto *ofproto, | |
977 | const struct ofproto_sflow_options *oso) | |
978 | { | |
979 | struct ofproto_sflow *os = ofproto->sflow; | |
980 | if (oso) { | |
981 | if (!os) { | |
982 | struct ofport *ofport; | |
72b06300 BP |
983 | |
984 | os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif); | |
4e8e4213 | 985 | HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) { |
ca0f572c | 986 | ofproto_sflow_add_port(os, ofport->odp_port, |
72b06300 BP |
987 | netdev_get_name(ofport->netdev)); |
988 | } | |
989 | } | |
990 | ofproto_sflow_set_options(os, oso); | |
991 | } else { | |
992 | ofproto_sflow_destroy(os); | |
993 | ofproto->sflow = NULL; | |
994 | } | |
995 | } | |
996 | ||
064af421 BP |
997 | uint64_t |
998 | ofproto_get_datapath_id(const struct ofproto *ofproto) | |
999 | { | |
1000 | return ofproto->datapath_id; | |
1001 | } | |
1002 | ||
76ce9432 | 1003 | bool |
7d674866 | 1004 | ofproto_has_primary_controller(const struct ofproto *ofproto) |
064af421 | 1005 | { |
76ce9432 | 1006 | return !hmap_is_empty(&ofproto->controllers); |
064af421 BP |
1007 | } |
1008 | ||
abdfe474 JP |
1009 | enum ofproto_fail_mode |
1010 | ofproto_get_fail_mode(const struct ofproto *p) | |
1011 | { | |
1012 | return p->fail_mode; | |
1013 | } | |
1014 | ||
064af421 BP |
1015 | void |
1016 | ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops) | |
1017 | { | |
1018 | size_t i; | |
1019 | ||
1020 | for (i = 0; i < ofproto->n_snoops; i++) { | |
1021 | svec_add(snoops, pvconn_get_name(ofproto->snoops[i])); | |
1022 | } | |
1023 | } | |
1024 | ||
1025 | void | |
1026 | ofproto_destroy(struct ofproto *p) | |
1027 | { | |
7d674866 | 1028 | struct ofservice *ofservice, *next_ofservice; |
064af421 | 1029 | struct ofconn *ofconn, *next_ofconn; |
ca0f572c | 1030 | struct ofport *ofport, *next_ofport; |
064af421 BP |
1031 | size_t i; |
1032 | ||
1033 | if (!p) { | |
1034 | return; | |
1035 | } | |
1036 | ||
7aa697dd BP |
1037 | shash_find_and_delete(&all_ofprotos, dpif_name(p->dpif)); |
1038 | ||
f7de2cdf | 1039 | /* Destroy fail-open and in-band early, since they touch the classifier. */ |
79c9f2ee BP |
1040 | fail_open_destroy(p->fail_open); |
1041 | p->fail_open = NULL; | |
1042 | ||
1043 | in_band_destroy(p->in_band); | |
1044 | p->in_band = NULL; | |
917e50e1 | 1045 | free(p->extra_in_band_remotes); |
2f6d3445 | 1046 | |
064af421 BP |
1047 | ofproto_flush_flows(p); |
1048 | classifier_destroy(&p->cls); | |
bcf84111 | 1049 | hmap_destroy(&p->facets); |
064af421 | 1050 | |
4e8e4213 | 1051 | LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) { |
c475ae67 | 1052 | ofconn_destroy(ofconn); |
064af421 | 1053 | } |
76ce9432 | 1054 | hmap_destroy(&p->controllers); |
064af421 | 1055 | |
c228a364 | 1056 | dpif_close(p->dpif); |
e9e28be3 | 1057 | netdev_monitor_destroy(p->netdev_monitor); |
4e8e4213 | 1058 | HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) { |
ca0f572c | 1059 | hmap_remove(&p->ports, &ofport->hmap_node); |
064af421 BP |
1060 | ofport_free(ofport); |
1061 | } | |
1062 | shash_destroy(&p->port_by_name); | |
1063 | ||
1064 | switch_status_destroy(p->switch_status); | |
064af421 | 1065 | netflow_destroy(p->netflow); |
72b06300 | 1066 | ofproto_sflow_destroy(p->sflow); |
064af421 | 1067 | |
4e8e4213 | 1068 | HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) { |
7d674866 | 1069 | ofservice_destroy(p, ofservice); |
064af421 | 1070 | } |
7d674866 | 1071 | hmap_destroy(&p->services); |
064af421 BP |
1072 | |
1073 | for (i = 0; i < p->n_snoops; i++) { | |
1074 | pvconn_close(p->snoops[i]); | |
1075 | } | |
1076 | free(p->snoops); | |
1077 | ||
1078 | mac_learning_destroy(p->ml); | |
1079 | ||
5a719c38 JP |
1080 | free(p->mfr_desc); |
1081 | free(p->hw_desc); | |
1082 | free(p->sw_desc); | |
1083 | free(p->serial_desc); | |
cb871ae0 JP |
1084 | free(p->dp_desc); |
1085 | ||
ca0f572c | 1086 | hmap_destroy(&p->ports); |
3b917492 | 1087 | |
064af421 BP |
1088 | free(p); |
1089 | } | |
1090 | ||
1091 | int | |
1092 | ofproto_run(struct ofproto *p) | |
1093 | { | |
1094 | int error = ofproto_run1(p); | |
1095 | if (!error) { | |
1096 | error = ofproto_run2(p, false); | |
1097 | } | |
1098 | return error; | |
1099 | } | |
1100 | ||
e9e28be3 BP |
1101 | static void |
1102 | process_port_change(struct ofproto *ofproto, int error, char *devname) | |
1103 | { | |
1104 | if (error == ENOBUFS) { | |
1105 | reinit_ports(ofproto); | |
1106 | } else if (!error) { | |
1107 | update_port(ofproto, devname); | |
1108 | free(devname); | |
1109 | } | |
1110 | } | |
1111 | ||
e2bfacb6 BP |
1112 | /* Returns a "preference level" for snooping 'ofconn'. A higher return value |
1113 | * means that 'ofconn' is more interesting for monitoring than a lower return | |
1114 | * value. */ | |
1115 | static int | |
1116 | snoop_preference(const struct ofconn *ofconn) | |
1117 | { | |
1118 | switch (ofconn->role) { | |
1119 | case NX_ROLE_MASTER: | |
1120 | return 3; | |
1121 | case NX_ROLE_OTHER: | |
1122 | return 2; | |
1123 | case NX_ROLE_SLAVE: | |
1124 | return 1; | |
1125 | default: | |
1126 | /* Shouldn't happen. */ | |
1127 | return 0; | |
1128 | } | |
1129 | } | |
1130 | ||
76ce9432 BP |
1131 | /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'. |
1132 | * Connects this vconn to a controller. */ | |
1133 | static void | |
1134 | add_snooper(struct ofproto *ofproto, struct vconn *vconn) | |
1135 | { | |
e2bfacb6 | 1136 | struct ofconn *ofconn, *best; |
76ce9432 | 1137 | |
e2bfacb6 BP |
1138 | /* Pick a controller for monitoring. */ |
1139 | best = NULL; | |
4e8e4213 | 1140 | LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) { |
5899143f | 1141 | if (ofconn->type == OFCONN_PRIMARY |
e2bfacb6 BP |
1142 | && (!best || snoop_preference(ofconn) > snoop_preference(best))) { |
1143 | best = ofconn; | |
76ce9432 | 1144 | } |
e2bfacb6 | 1145 | } |
76ce9432 | 1146 | |
e2bfacb6 BP |
1147 | if (best) { |
1148 | rconn_add_monitor(best->rconn, vconn); | |
1149 | } else { | |
1150 | VLOG_INFO_RL(&rl, "no controller connection to snoop"); | |
1151 | vconn_close(vconn); | |
76ce9432 | 1152 | } |
76ce9432 BP |
1153 | } |
1154 | ||
064af421 BP |
1155 | int |
1156 | ofproto_run1(struct ofproto *p) | |
1157 | { | |
1158 | struct ofconn *ofconn, *next_ofconn; | |
7d674866 | 1159 | struct ofservice *ofservice; |
064af421 BP |
1160 | char *devname; |
1161 | int error; | |
1162 | int i; | |
1163 | ||
149f577a JG |
1164 | if (shash_is_empty(&p->port_by_name)) { |
1165 | init_ports(p); | |
1166 | } | |
1167 | ||
064af421 | 1168 | for (i = 0; i < 50; i++) { |
856081f6 | 1169 | struct dpif_upcall packet; |
064af421 | 1170 | |
856081f6 | 1171 | error = dpif_recv(p->dpif, &packet); |
064af421 BP |
1172 | if (error) { |
1173 | if (error == ENODEV) { | |
1174 | /* Someone destroyed the datapath behind our back. The caller | |
1175 | * better destroy us and give up, because we're just going to | |
1176 | * spin from here on out. */ | |
39a559f2 BP |
1177 | static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5); |
1178 | VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally", | |
c228a364 | 1179 | dpif_name(p->dpif)); |
064af421 BP |
1180 | return ENODEV; |
1181 | } | |
1182 | break; | |
1183 | } | |
1184 | ||
856081f6 | 1185 | handle_upcall(p, &packet); |
064af421 BP |
1186 | } |
1187 | ||
e9e28be3 BP |
1188 | while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) { |
1189 | process_port_change(p, error, devname); | |
1190 | } | |
1191 | while ((error = netdev_monitor_poll(p->netdev_monitor, | |
1192 | &devname)) != EAGAIN) { | |
1193 | process_port_change(p, error, devname); | |
064af421 BP |
1194 | } |
1195 | ||
1196 | if (p->in_band) { | |
d2ede7bc BP |
1197 | if (time_msec() >= p->next_in_band_update) { |
1198 | update_in_band_remotes(p); | |
1199 | } | |
064af421 BP |
1200 | in_band_run(p->in_band); |
1201 | } | |
064af421 | 1202 | |
4e8e4213 | 1203 | LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) { |
3269c562 | 1204 | ofconn_run(ofconn); |
064af421 BP |
1205 | } |
1206 | ||
7778bd15 BP |
1207 | /* Fail-open maintenance. Do this after processing the ofconns since |
1208 | * fail-open checks the status of the controller rconn. */ | |
1209 | if (p->fail_open) { | |
1210 | fail_open_run(p->fail_open); | |
1211 | } | |
1212 | ||
4e8e4213 | 1213 | HMAP_FOR_EACH (ofservice, node, &p->services) { |
064af421 BP |
1214 | struct vconn *vconn; |
1215 | int retval; | |
1216 | ||
7d674866 | 1217 | retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn); |
064af421 | 1218 | if (!retval) { |
9794e806 | 1219 | struct rconn *rconn; |
eb15cdbb | 1220 | char *name; |
9794e806 | 1221 | |
7d674866 | 1222 | rconn = rconn_create(ofservice->probe_interval, 0); |
eb15cdbb BP |
1223 | name = ofconn_make_name(p, vconn_get_name(vconn)); |
1224 | rconn_connect_unreliably(rconn, vconn, name); | |
1225 | free(name); | |
1226 | ||
7d674866 BP |
1227 | ofconn = ofconn_create(p, rconn, OFCONN_SERVICE); |
1228 | ofconn_set_rate_limit(ofconn, ofservice->rate_limit, | |
1229 | ofservice->burst_limit); | |
064af421 BP |
1230 | } else if (retval != EAGAIN) { |
1231 | VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval)); | |
1232 | } | |
1233 | } | |
1234 | ||
1235 | for (i = 0; i < p->n_snoops; i++) { | |
1236 | struct vconn *vconn; | |
1237 | int retval; | |
1238 | ||
1239 | retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn); | |
1240 | if (!retval) { | |
76ce9432 | 1241 | add_snooper(p, vconn); |
064af421 BP |
1242 | } else if (retval != EAGAIN) { |
1243 | VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval)); | |
1244 | } | |
1245 | } | |
1246 | ||
1247 | if (time_msec() >= p->next_expiration) { | |
0de7a4b4 BP |
1248 | int delay = ofproto_expire(p); |
1249 | p->next_expiration = time_msec() + delay; | |
064af421 | 1250 | COVERAGE_INC(ofproto_expiration); |
064af421 BP |
1251 | } |
1252 | ||
1253 | if (p->netflow) { | |
1254 | netflow_run(p->netflow); | |
1255 | } | |
72b06300 BP |
1256 | if (p->sflow) { |
1257 | ofproto_sflow_run(p->sflow); | |
1258 | } | |
064af421 BP |
1259 | |
1260 | return 0; | |
1261 | } | |
1262 | ||
064af421 BP |
1263 | int |
1264 | ofproto_run2(struct ofproto *p, bool revalidate_all) | |
1265 | { | |
bcf84111 BP |
1266 | /* Figure out what we need to revalidate now, if anything. */ |
1267 | struct tag_set revalidate_set = p->revalidate_set; | |
1268 | if (p->need_revalidate) { | |
1269 | revalidate_all = true; | |
1270 | } | |
1271 | ||
1272 | /* Clear the revalidation flags. */ | |
1273 | tag_set_init(&p->revalidate_set); | |
1274 | p->need_revalidate = false; | |
1275 | ||
1276 | /* Now revalidate if there's anything to do. */ | |
1277 | if (revalidate_all || !tag_set_is_empty(&revalidate_set)) { | |
1278 | struct facet *facet, *next; | |
1279 | ||
1280 | HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &p->facets) { | |
1281 | if (revalidate_all | |
1282 | || tag_set_intersects(&revalidate_set, facet->tags)) { | |
1283 | facet_revalidate(p, facet); | |
1284 | } | |
1285 | } | |
064af421 BP |
1286 | } |
1287 | ||
1288 | return 0; | |
1289 | } | |
1290 | ||
1291 | void | |
1292 | ofproto_wait(struct ofproto *p) | |
1293 | { | |
7d674866 | 1294 | struct ofservice *ofservice; |
064af421 BP |
1295 | struct ofconn *ofconn; |
1296 | size_t i; | |
1297 | ||
c228a364 | 1298 | dpif_recv_wait(p->dpif); |
e9e28be3 BP |
1299 | dpif_port_poll_wait(p->dpif); |
1300 | netdev_monitor_poll_wait(p->netdev_monitor); | |
4e8e4213 | 1301 | LIST_FOR_EACH (ofconn, node, &p->all_conns) { |
064af421 BP |
1302 | ofconn_wait(ofconn); |
1303 | } | |
1304 | if (p->in_band) { | |
7cf8b266 | 1305 | poll_timer_wait_until(p->next_in_band_update); |
064af421 BP |
1306 | in_band_wait(p->in_band); |
1307 | } | |
064af421 BP |
1308 | if (p->fail_open) { |
1309 | fail_open_wait(p->fail_open); | |
1310 | } | |
72b06300 BP |
1311 | if (p->sflow) { |
1312 | ofproto_sflow_wait(p->sflow); | |
1313 | } | |
064af421 BP |
1314 | if (!tag_set_is_empty(&p->revalidate_set)) { |
1315 | poll_immediate_wake(); | |
1316 | } | |
1317 | if (p->need_revalidate) { | |
1318 | /* Shouldn't happen, but if it does just go around again. */ | |
1319 | VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); | |
1320 | poll_immediate_wake(); | |
1321 | } else if (p->next_expiration != LLONG_MAX) { | |
7cf8b266 | 1322 | poll_timer_wait_until(p->next_expiration); |
064af421 | 1323 | } |
4e8e4213 | 1324 | HMAP_FOR_EACH (ofservice, node, &p->services) { |
7d674866 | 1325 | pvconn_wait(ofservice->pvconn); |
064af421 BP |
1326 | } |
1327 | for (i = 0; i < p->n_snoops; i++) { | |
1328 | pvconn_wait(p->snoops[i]); | |
1329 | } | |
1330 | } | |
1331 | ||
1332 | void | |
1333 | ofproto_revalidate(struct ofproto *ofproto, tag_type tag) | |
1334 | { | |
1335 | tag_set_add(&ofproto->revalidate_set, tag); | |
1336 | } | |
1337 | ||
1338 | struct tag_set * | |
1339 | ofproto_get_revalidate_set(struct ofproto *ofproto) | |
1340 | { | |
1341 | return &ofproto->revalidate_set; | |
1342 | } | |
1343 | ||
1344 | bool | |
1345 | ofproto_is_alive(const struct ofproto *p) | |
1346 | { | |
76ce9432 | 1347 | return !hmap_is_empty(&p->controllers); |
064af421 BP |
1348 | } |
1349 | ||
bffc0589 AE |
1350 | void |
1351 | ofproto_get_ofproto_controller_info(const struct ofproto * ofproto, | |
1352 | struct shash *info) | |
1353 | { | |
1354 | const struct ofconn *ofconn; | |
1355 | ||
1356 | shash_init(info); | |
1357 | ||
1358 | HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) { | |
1359 | const struct rconn *rconn = ofconn->rconn; | |
1360 | const int last_error = rconn_get_last_error(rconn); | |
1361 | struct ofproto_controller_info *cinfo = xmalloc(sizeof *cinfo); | |
1362 | ||
1363 | shash_add(info, rconn_get_target(rconn), cinfo); | |
1364 | ||
1365 | cinfo->is_connected = rconn_is_connected(rconn); | |
1366 | cinfo->role = ofconn->role; | |
1367 | ||
1368 | cinfo->pairs.n = 0; | |
1369 | ||
1370 | if (last_error == EOF) { | |
1371 | cinfo->pairs.keys[cinfo->pairs.n] = "last_error"; | |
1372 | cinfo->pairs.values[cinfo->pairs.n++] = xstrdup("End of file"); | |
1373 | } else if (last_error > 0) { | |
1374 | cinfo->pairs.keys[cinfo->pairs.n] = "last_error"; | |
1375 | cinfo->pairs.values[cinfo->pairs.n++] = | |
1376 | xstrdup(strerror(last_error)); | |
1377 | } | |
1378 | ||
1379 | cinfo->pairs.keys[cinfo->pairs.n] = "state"; | |
1380 | cinfo->pairs.values[cinfo->pairs.n++] = | |
1381 | xstrdup(rconn_get_state(rconn)); | |
1382 | ||
1383 | cinfo->pairs.keys[cinfo->pairs.n] = "time_in_state"; | |
1384 | cinfo->pairs.values[cinfo->pairs.n++] = | |
1385 | xasprintf("%u", rconn_get_state_elapsed(rconn)); | |
1386 | } | |
1387 | } | |
1388 | ||
1389 | void | |
1390 | ofproto_free_ofproto_controller_info(struct shash *info) | |
1391 | { | |
1392 | struct shash_node *node; | |
1393 | ||
1394 | SHASH_FOR_EACH (node, info) { | |
1395 | struct ofproto_controller_info *cinfo = node->data; | |
1396 | while (cinfo->pairs.n) { | |
1397 | free((char *) cinfo->pairs.values[--cinfo->pairs.n]); | |
1398 | } | |
1399 | free(cinfo); | |
1400 | } | |
1401 | shash_destroy(info); | |
1402 | } | |
1403 | ||
3a6ccc8c BP |
1404 | /* Deletes port number 'odp_port' from the datapath for 'ofproto'. |
1405 | * | |
1406 | * This is almost the same as calling dpif_port_del() directly on the | |
1407 | * datapath, but it also makes 'ofproto' close its open netdev for the port | |
1408 | * (if any). This makes it possible to create a new netdev of a different | |
1409 | * type under the same name, which otherwise the netdev library would refuse | |
1410 | * to do because of the conflict. (The netdev would eventually get closed on | |
1411 | * the next trip through ofproto_run(), but this interface is more direct.) | |
1412 | * | |
3a6ccc8c BP |
1413 | * Returns 0 if successful, otherwise a positive errno. */ |
1414 | int | |
1415 | ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port) | |
1416 | { | |
1417 | struct ofport *ofport = get_port(ofproto, odp_port); | |
0b61210e | 1418 | const char *name = ofport ? ofport->opp.name : "<unknown>"; |
3a6ccc8c BP |
1419 | int error; |
1420 | ||
1421 | error = dpif_port_del(ofproto->dpif, odp_port); | |
1422 | if (error) { | |
1423 | VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)", | |
1424 | dpif_name(ofproto->dpif), odp_port, name, strerror(error)); | |
1425 | } else if (ofport) { | |
1426 | /* 'name' is ofport->opp.name and update_port() is going to destroy | |
1427 | * 'ofport'. Just in case update_port() refers to 'name' after it | |
1428 | * destroys 'ofport', make a copy of it around the update_port() | |
1429 | * call. */ | |
1430 | char *devname = xstrdup(name); | |
1431 | update_port(ofproto, devname); | |
1432 | free(devname); | |
1433 | } | |
1434 | return error; | |
1435 | } | |
1436 | ||
a4e2e1f2 EJ |
1437 | /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns |
1438 | * true if 'odp_port' exists and should be included, false otherwise. */ | |
1439 | bool | |
1440 | ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port) | |
1441 | { | |
1442 | struct ofport *ofport = get_port(ofproto, odp_port); | |
1443 | return ofport && !(ofport->opp.config & OFPPC_NO_FLOOD); | |
1444 | } | |
1445 | ||
064af421 | 1446 | int |
ae412e7d | 1447 | ofproto_send_packet(struct ofproto *p, const struct flow *flow, |
064af421 BP |
1448 | const union ofp_action *actions, size_t n_actions, |
1449 | const struct ofpbuf *packet) | |
1450 | { | |
f29152ca | 1451 | struct action_xlate_ctx ctx; |
cdee00fd | 1452 | struct ofpbuf *odp_actions; |
064af421 | 1453 | |
f29152ca | 1454 | action_xlate_ctx_init(&ctx, p, flow, packet); |
cdee00fd | 1455 | odp_actions = xlate_actions(&ctx, actions, n_actions); |
064af421 BP |
1456 | |
1457 | /* XXX Should we translate the dpif_execute() errno value into an OpenFlow | |
1458 | * error code? */ | |
cdee00fd BP |
1459 | dpif_execute(p->dpif, odp_actions->data, odp_actions->size, packet); |
1460 | ||
1461 | ofpbuf_delete(odp_actions); | |
1462 | ||
064af421 BP |
1463 | return 0; |
1464 | } | |
1465 | ||
fa8b054f BP |
1466 | /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and |
1467 | * performs the 'n_actions' actions in 'actions'. The new flow will not | |
1468 | * timeout. | |
1469 | * | |
1470 | * If cls_rule->priority is in the range of priorities supported by OpenFlow | |
1471 | * (0...65535, inclusive) then the flow will be visible to OpenFlow | |
1472 | * controllers; otherwise, it will be hidden. | |
1473 | * | |
1474 | * The caller retains ownership of 'cls_rule' and 'actions'. */ | |
064af421 | 1475 | void |
cf3fad8a | 1476 | ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule, |
fa8b054f | 1477 | const union ofp_action *actions, size_t n_actions) |
064af421 BP |
1478 | { |
1479 | struct rule *rule; | |
bcf84111 | 1480 | rule = rule_create(cls_rule, actions, n_actions, 0, 0, 0, false); |
afe75089 | 1481 | rule_insert(p, rule); |
064af421 BP |
1482 | } |
1483 | ||
1484 | void | |
cf3fad8a | 1485 | ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target) |
064af421 BP |
1486 | { |
1487 | struct rule *rule; | |
1488 | ||
1489 | rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls, | |
cf3fad8a | 1490 | target)); |
064af421 BP |
1491 | if (rule) { |
1492 | rule_remove(ofproto, rule); | |
1493 | } | |
1494 | } | |
1495 | ||
064af421 BP |
1496 | void |
1497 | ofproto_flush_flows(struct ofproto *ofproto) | |
1498 | { | |
bcf84111 | 1499 | struct facet *facet, *next_facet; |
5ecc9d81 BP |
1500 | struct rule *rule, *next_rule; |
1501 | struct cls_cursor cursor; | |
bcf84111 | 1502 | |
064af421 | 1503 | COVERAGE_INC(ofproto_flush); |
bcf84111 BP |
1504 | |
1505 | HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) { | |
1506 | /* Mark the facet as not installed so that facet_remove() doesn't | |
1507 | * bother trying to uninstall it. There is no point in uninstalling it | |
1508 | * individually since we are about to blow away all the facets with | |
1509 | * dpif_flow_flush(). */ | |
1510 | facet->installed = false; | |
1511 | facet_remove(ofproto, facet); | |
1512 | } | |
5ecc9d81 BP |
1513 | |
1514 | cls_cursor_init(&cursor, &ofproto->cls, NULL); | |
1515 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) { | |
1516 | rule_remove(ofproto, rule); | |
1517 | } | |
1518 | ||
c228a364 | 1519 | dpif_flow_flush(ofproto->dpif); |
064af421 BP |
1520 | if (ofproto->in_band) { |
1521 | in_band_flushed(ofproto->in_band); | |
1522 | } | |
1523 | if (ofproto->fail_open) { | |
1524 | fail_open_flushed(ofproto->fail_open); | |
1525 | } | |
1526 | } | |
1527 | \f | |
1528 | static void | |
1529 | reinit_ports(struct ofproto *p) | |
1530 | { | |
b0ec0f27 | 1531 | struct dpif_port_dump dump; |
c77d9d13 BP |
1532 | struct shash_node *node; |
1533 | struct shash devnames; | |
064af421 | 1534 | struct ofport *ofport; |
4c738a8d | 1535 | struct dpif_port dpif_port; |
064af421 | 1536 | |
898bf89d JP |
1537 | COVERAGE_INC(ofproto_reinit_ports); |
1538 | ||
c77d9d13 | 1539 | shash_init(&devnames); |
4e8e4213 | 1540 | HMAP_FOR_EACH (ofport, hmap_node, &p->ports) { |
c77d9d13 | 1541 | shash_add_once (&devnames, ofport->opp.name, NULL); |
064af421 | 1542 | } |
4c738a8d BP |
1543 | DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) { |
1544 | shash_add_once (&devnames, dpif_port.name, NULL); | |
064af421 | 1545 | } |
064af421 | 1546 | |
c77d9d13 BP |
1547 | SHASH_FOR_EACH (node, &devnames) { |
1548 | update_port(p, node->name); | |
064af421 | 1549 | } |
c77d9d13 | 1550 | shash_destroy(&devnames); |
064af421 BP |
1551 | } |
1552 | ||
064af421 | 1553 | static struct ofport * |
4c738a8d | 1554 | make_ofport(const struct dpif_port *dpif_port) |
064af421 | 1555 | { |
149f577a | 1556 | struct netdev_options netdev_options; |
064af421 BP |
1557 | enum netdev_flags flags; |
1558 | struct ofport *ofport; | |
1559 | struct netdev *netdev; | |
064af421 BP |
1560 | int error; |
1561 | ||
149f577a | 1562 | memset(&netdev_options, 0, sizeof netdev_options); |
4c738a8d BP |
1563 | netdev_options.name = dpif_port->name; |
1564 | netdev_options.type = dpif_port->type; | |
149f577a | 1565 | netdev_options.ethertype = NETDEV_ETH_TYPE_NONE; |
149f577a JG |
1566 | |
1567 | error = netdev_open(&netdev_options, &netdev); | |
064af421 BP |
1568 | if (error) { |
1569 | VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s " | |
1570 | "cannot be opened (%s)", | |
4c738a8d BP |
1571 | dpif_port->name, dpif_port->port_no, |
1572 | dpif_port->name, strerror(error)); | |
064af421 BP |
1573 | return NULL; |
1574 | } | |
1575 | ||
1576 | ofport = xmalloc(sizeof *ofport); | |
1577 | ofport->netdev = netdev; | |
4c738a8d BP |
1578 | ofport->odp_port = dpif_port->port_no; |
1579 | ofport->opp.port_no = odp_port_to_ofp_port(dpif_port->port_no); | |
80992a35 | 1580 | netdev_get_etheraddr(netdev, ofport->opp.hw_addr); |
4c738a8d | 1581 | ovs_strlcpy(ofport->opp.name, dpif_port->name, sizeof ofport->opp.name); |
064af421 BP |
1582 | |
1583 | netdev_get_flags(netdev, &flags); | |
1584 | ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN; | |
1585 | ||
85da620e | 1586 | ofport->opp.state = netdev_get_carrier(netdev) ? 0 : OFPPS_LINK_DOWN; |
064af421 BP |
1587 | |
1588 | netdev_get_features(netdev, | |
1589 | &ofport->opp.curr, &ofport->opp.advertised, | |
1590 | &ofport->opp.supported, &ofport->opp.peer); | |
1591 | return ofport; | |
1592 | } | |
1593 | ||
1594 | static bool | |
4c738a8d | 1595 | ofport_conflicts(const struct ofproto *p, const struct dpif_port *dpif_port) |
064af421 | 1596 | { |
4c738a8d | 1597 | if (get_port(p, dpif_port->port_no)) { |
064af421 | 1598 | VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath", |
4c738a8d | 1599 | dpif_port->port_no); |
064af421 | 1600 | return true; |
4c738a8d | 1601 | } else if (shash_find(&p->port_by_name, dpif_port->name)) { |
064af421 | 1602 | VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath", |
4c738a8d | 1603 | dpif_port->name); |
064af421 BP |
1604 | return true; |
1605 | } else { | |
1606 | return false; | |
1607 | } | |
1608 | } | |
1609 | ||
1610 | static int | |
1611 | ofport_equal(const struct ofport *a_, const struct ofport *b_) | |
1612 | { | |
1613 | const struct ofp_phy_port *a = &a_->opp; | |
1614 | const struct ofp_phy_port *b = &b_->opp; | |
1615 | ||
1616 | BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */ | |
1617 | return (a->port_no == b->port_no | |
1618 | && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr) | |
0b61210e | 1619 | && !strcmp(a->name, b->name) |
064af421 BP |
1620 | && a->state == b->state |
1621 | && a->config == b->config | |
1622 | && a->curr == b->curr | |
1623 | && a->advertised == b->advertised | |
1624 | && a->supported == b->supported | |
1625 | && a->peer == b->peer); | |
1626 | } | |
1627 | ||
1628 | static void | |
1629 | send_port_status(struct ofproto *p, const struct ofport *ofport, | |
1630 | uint8_t reason) | |
1631 | { | |
1632 | /* XXX Should limit the number of queued port status change messages. */ | |
1633 | struct ofconn *ofconn; | |
4e8e4213 | 1634 | LIST_FOR_EACH (ofconn, node, &p->all_conns) { |
064af421 BP |
1635 | struct ofp_port_status *ops; |
1636 | struct ofpbuf *b; | |
1637 | ||
197a992f BP |
1638 | /* Primary controllers, even slaves, should always get port status |
1639 | updates. Otherwise obey ofconn_receives_async_msgs(). */ | |
1640 | if (ofconn->type != OFCONN_PRIMARY | |
1641 | && !ofconn_receives_async_msgs(ofconn)) { | |
9deba63b BP |
1642 | continue; |
1643 | } | |
1644 | ||
064af421 BP |
1645 | ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b); |
1646 | ops->reason = reason; | |
1647 | ops->desc = ofport->opp; | |
1648 | hton_ofp_phy_port(&ops->desc); | |
1649 | queue_tx(b, ofconn, NULL); | |
1650 | } | |
064af421 BP |
1651 | } |
1652 | ||
1653 | static void | |
1654 | ofport_install(struct ofproto *p, struct ofport *ofport) | |
1655 | { | |
0b61210e | 1656 | const char *netdev_name = ofport->opp.name; |
72b06300 | 1657 | |
e9e28be3 | 1658 | netdev_monitor_add(p->netdev_monitor, ofport->netdev); |
ca0f572c | 1659 | hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0)); |
72b06300 BP |
1660 | shash_add(&p->port_by_name, netdev_name, ofport); |
1661 | if (p->sflow) { | |
ca0f572c | 1662 | ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name); |
72b06300 | 1663 | } |
064af421 BP |
1664 | } |
1665 | ||
1666 | static void | |
1667 | ofport_remove(struct ofproto *p, struct ofport *ofport) | |
1668 | { | |
e9e28be3 | 1669 | netdev_monitor_remove(p->netdev_monitor, ofport->netdev); |
ca0f572c | 1670 | hmap_remove(&p->ports, &ofport->hmap_node); |
064af421 | 1671 | shash_delete(&p->port_by_name, |
0b61210e | 1672 | shash_find(&p->port_by_name, ofport->opp.name)); |
72b06300 | 1673 | if (p->sflow) { |
ca0f572c | 1674 | ofproto_sflow_del_port(p->sflow, ofport->odp_port); |
72b06300 | 1675 | } |
064af421 BP |
1676 | } |
1677 | ||
1678 | static void | |
1679 | ofport_free(struct ofport *ofport) | |
1680 | { | |
1681 | if (ofport) { | |
1682 | netdev_close(ofport->netdev); | |
1683 | free(ofport); | |
1684 | } | |
1685 | } | |
1686 | ||
ca0f572c BP |
1687 | static struct ofport * |
1688 | get_port(const struct ofproto *ofproto, uint16_t odp_port) | |
1689 | { | |
1690 | struct ofport *port; | |
1691 | ||
4e8e4213 | 1692 | HMAP_FOR_EACH_IN_BUCKET (port, hmap_node, |
ca0f572c BP |
1693 | hash_int(odp_port, 0), &ofproto->ports) { |
1694 | if (port->odp_port == odp_port) { | |
1695 | return port; | |
1696 | } | |
1697 | } | |
1698 | return NULL; | |
1699 | } | |
1700 | ||
064af421 BP |
1701 | static void |
1702 | update_port(struct ofproto *p, const char *devname) | |
1703 | { | |
4c738a8d | 1704 | struct dpif_port dpif_port; |
c874dc6d BP |
1705 | struct ofport *old_ofport; |
1706 | struct ofport *new_ofport; | |
064af421 BP |
1707 | int error; |
1708 | ||
1709 | COVERAGE_INC(ofproto_update_port); | |
c874dc6d BP |
1710 | |
1711 | /* Query the datapath for port information. */ | |
4c738a8d | 1712 | error = dpif_port_query_by_name(p->dpif, devname, &dpif_port); |
064af421 | 1713 | |
c874dc6d BP |
1714 | /* Find the old ofport. */ |
1715 | old_ofport = shash_find_data(&p->port_by_name, devname); | |
1716 | if (!error) { | |
1717 | if (!old_ofport) { | |
1718 | /* There's no port named 'devname' but there might be a port with | |
1719 | * the same port number. This could happen if a port is deleted | |
1720 | * and then a new one added in its place very quickly, or if a port | |
1721 | * is renamed. In the former case we want to send an OFPPR_DELETE | |
1722 | * and an OFPPR_ADD, and in the latter case we want to send a | |
1723 | * single OFPPR_MODIFY. We can distinguish the cases by comparing | |
1724 | * the old port's ifindex against the new port, or perhaps less | |
1725 | * reliably but more portably by comparing the old port's MAC | |
1726 | * against the new port's MAC. However, this code isn't that smart | |
1727 | * and always sends an OFPPR_MODIFY (XXX). */ | |
4c738a8d | 1728 | old_ofport = get_port(p, dpif_port.port_no); |
064af421 | 1729 | } |
c874dc6d | 1730 | } else if (error != ENOENT && error != ENODEV) { |
064af421 BP |
1731 | VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error " |
1732 | "%s", strerror(error)); | |
4c738a8d | 1733 | goto exit; |
064af421 | 1734 | } |
c874dc6d BP |
1735 | |
1736 | /* Create a new ofport. */ | |
4c738a8d | 1737 | new_ofport = !error ? make_ofport(&dpif_port) : NULL; |
c874dc6d BP |
1738 | |
1739 | /* Eliminate a few pathological cases. */ | |
1740 | if (!old_ofport && !new_ofport) { | |
4c738a8d | 1741 | goto exit; |
c874dc6d BP |
1742 | } else if (old_ofport && new_ofport) { |
1743 | /* Most of the 'config' bits are OpenFlow soft state, but | |
bc4a55cd BP |
1744 | * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the |
1745 | * OpenFlow bits from old_ofport. (make_ofport() only sets | |
1746 | * OFPPC_PORT_DOWN and leaves the other bits 0.) */ | |
c874dc6d BP |
1747 | new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN; |
1748 | ||
1749 | if (ofport_equal(old_ofport, new_ofport)) { | |
1750 | /* False alarm--no change. */ | |
1751 | ofport_free(new_ofport); | |
4c738a8d | 1752 | goto exit; |
c874dc6d BP |
1753 | } |
1754 | } | |
1755 | ||
1756 | /* Now deal with the normal cases. */ | |
1757 | if (old_ofport) { | |
1758 | ofport_remove(p, old_ofport); | |
1759 | } | |
1760 | if (new_ofport) { | |
1761 | ofport_install(p, new_ofport); | |
1762 | } | |
1763 | send_port_status(p, new_ofport ? new_ofport : old_ofport, | |
1764 | (!old_ofport ? OFPPR_ADD | |
1765 | : !new_ofport ? OFPPR_DELETE | |
1766 | : OFPPR_MODIFY)); | |
1767 | ofport_free(old_ofport); | |
4c738a8d BP |
1768 | |
1769 | exit: | |
1770 | dpif_port_destroy(&dpif_port); | |
064af421 BP |
1771 | } |
1772 | ||
1773 | static int | |
1774 | init_ports(struct ofproto *p) | |
1775 | { | |
b0ec0f27 | 1776 | struct dpif_port_dump dump; |
4c738a8d | 1777 | struct dpif_port dpif_port; |
064af421 | 1778 | |
4c738a8d BP |
1779 | DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) { |
1780 | if (!ofport_conflicts(p, &dpif_port)) { | |
1781 | struct ofport *ofport = make_ofport(&dpif_port); | |
064af421 BP |
1782 | if (ofport) { |
1783 | ofport_install(p, ofport); | |
1784 | } | |
1785 | } | |
1786 | } | |
b0ec0f27 | 1787 | |
064af421 BP |
1788 | return 0; |
1789 | } | |
1790 | \f | |
1791 | static struct ofconn * | |
76ce9432 | 1792 | ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type) |
064af421 | 1793 | { |
76ce9432 BP |
1794 | struct ofconn *ofconn = xzalloc(sizeof *ofconn); |
1795 | ofconn->ofproto = p; | |
064af421 BP |
1796 | list_push_back(&p->all_conns, &ofconn->node); |
1797 | ofconn->rconn = rconn; | |
76ce9432 | 1798 | ofconn->type = type; |
b70eac89 | 1799 | ofconn->flow_format = NXFF_OPENFLOW10; |
9deba63b | 1800 | ofconn->role = NX_ROLE_OTHER; |
76ce9432 | 1801 | ofconn->packet_in_counter = rconn_packet_counter_create (); |
064af421 | 1802 | ofconn->pktbuf = NULL; |
064af421 | 1803 | ofconn->miss_send_len = 0; |
064af421 BP |
1804 | ofconn->reply_counter = rconn_packet_counter_create (); |
1805 | return ofconn; | |
1806 | } | |
1807 | ||
1808 | static void | |
c475ae67 | 1809 | ofconn_destroy(struct ofconn *ofconn) |
064af421 | 1810 | { |
5899143f | 1811 | if (ofconn->type == OFCONN_PRIMARY) { |
76ce9432 BP |
1812 | hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node); |
1813 | } | |
1814 | discovery_destroy(ofconn->discovery); | |
1815 | ||
064af421 | 1816 | list_remove(&ofconn->node); |
76ce9432 | 1817 | switch_status_unregister(ofconn->ss); |
064af421 BP |
1818 | rconn_destroy(ofconn->rconn); |
1819 | rconn_packet_counter_destroy(ofconn->packet_in_counter); | |
1820 | rconn_packet_counter_destroy(ofconn->reply_counter); | |
1821 | pktbuf_destroy(ofconn->pktbuf); | |
1822 | free(ofconn); | |
1823 | } | |
1824 | ||
1825 | static void | |
3269c562 | 1826 | ofconn_run(struct ofconn *ofconn) |
064af421 | 1827 | { |
3269c562 | 1828 | struct ofproto *p = ofconn->ofproto; |
064af421 | 1829 | int iteration; |
76ce9432 BP |
1830 | size_t i; |
1831 | ||
1832 | if (ofconn->discovery) { | |
1833 | char *controller_name; | |
1834 | if (rconn_is_connectivity_questionable(ofconn->rconn)) { | |
1835 | discovery_question_connectivity(ofconn->discovery); | |
1836 | } | |
1837 | if (discovery_run(ofconn->discovery, &controller_name)) { | |
1838 | if (controller_name) { | |
eb15cdbb BP |
1839 | char *ofconn_name = ofconn_make_name(p, controller_name); |
1840 | rconn_connect(ofconn->rconn, controller_name, ofconn_name); | |
1841 | free(ofconn_name); | |
76ce9432 BP |
1842 | } else { |
1843 | rconn_disconnect(ofconn->rconn); | |
1844 | } | |
1845 | } | |
1846 | } | |
1847 | ||
1848 | for (i = 0; i < N_SCHEDULERS; i++) { | |
1849 | pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn); | |
1850 | } | |
064af421 BP |
1851 | |
1852 | rconn_run(ofconn->rconn); | |
1853 | ||
1854 | if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) { | |
1855 | /* Limit the number of iterations to prevent other tasks from | |
1856 | * starving. */ | |
1857 | for (iteration = 0; iteration < 50; iteration++) { | |
1858 | struct ofpbuf *of_msg = rconn_recv(ofconn->rconn); | |
1859 | if (!of_msg) { | |
1860 | break; | |
1861 | } | |
7778bd15 BP |
1862 | if (p->fail_open) { |
1863 | fail_open_maybe_recover(p->fail_open); | |
1864 | } | |
3269c562 | 1865 | handle_openflow(ofconn, of_msg); |
064af421 BP |
1866 | ofpbuf_delete(of_msg); |
1867 | } | |
1868 | } | |
1869 | ||
76ce9432 | 1870 | if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) { |
c475ae67 | 1871 | ofconn_destroy(ofconn); |
064af421 BP |
1872 | } |
1873 | } | |
1874 | ||
1875 | static void | |
1876 | ofconn_wait(struct ofconn *ofconn) | |
1877 | { | |
76ce9432 BP |
1878 | int i; |
1879 | ||
1880 | if (ofconn->discovery) { | |
1881 | discovery_wait(ofconn->discovery); | |
1882 | } | |
1883 | for (i = 0; i < N_SCHEDULERS; i++) { | |
1884 | pinsched_wait(ofconn->schedulers[i]); | |
1885 | } | |
064af421 BP |
1886 | rconn_run_wait(ofconn->rconn); |
1887 | if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) { | |
1888 | rconn_recv_wait(ofconn->rconn); | |
1889 | } else { | |
1890 | COVERAGE_INC(ofproto_ofconn_stuck); | |
1891 | } | |
1892 | } | |
c91248b3 BP |
1893 | |
1894 | /* Returns true if 'ofconn' should receive asynchronous messages. */ | |
1895 | static bool | |
1896 | ofconn_receives_async_msgs(const struct ofconn *ofconn) | |
1897 | { | |
5899143f BP |
1898 | if (ofconn->type == OFCONN_PRIMARY) { |
1899 | /* Primary controllers always get asynchronous messages unless they | |
c91248b3 BP |
1900 | * have configured themselves as "slaves". */ |
1901 | return ofconn->role != NX_ROLE_SLAVE; | |
1902 | } else { | |
5899143f BP |
1903 | /* Service connections don't get asynchronous messages unless they have |
1904 | * explicitly asked for them by setting a nonzero miss send length. */ | |
c91248b3 BP |
1905 | return ofconn->miss_send_len > 0; |
1906 | } | |
1907 | } | |
eb15cdbb BP |
1908 | |
1909 | /* Returns a human-readable name for an OpenFlow connection between 'ofproto' | |
1910 | * and 'target', suitable for use in log messages for identifying the | |
1911 | * connection. | |
1912 | * | |
1913 | * The name is dynamically allocated. The caller should free it (with free()) | |
1914 | * when it is no longer needed. */ | |
1915 | static char * | |
1916 | ofconn_make_name(const struct ofproto *ofproto, const char *target) | |
1917 | { | |
1918 | return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target); | |
1919 | } | |
7d674866 BP |
1920 | |
1921 | static void | |
1922 | ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst) | |
1923 | { | |
1924 | int i; | |
1925 | ||
1926 | for (i = 0; i < N_SCHEDULERS; i++) { | |
1927 | struct pinsched **s = &ofconn->schedulers[i]; | |
1928 | ||
1929 | if (rate > 0) { | |
1930 | if (!*s) { | |
1931 | *s = pinsched_create(rate, burst, | |
1932 | ofconn->ofproto->switch_status); | |
1933 | } else { | |
1934 | pinsched_set_limits(*s, rate, burst); | |
1935 | } | |
1936 | } else { | |
1937 | pinsched_destroy(*s); | |
1938 | *s = NULL; | |
1939 | } | |
1940 | } | |
1941 | } | |
1942 | \f | |
1943 | static void | |
1944 | ofservice_reconfigure(struct ofservice *ofservice, | |
1945 | const struct ofproto_controller *c) | |
1946 | { | |
1947 | ofservice->probe_interval = c->probe_interval; | |
1948 | ofservice->rate_limit = c->rate_limit; | |
1949 | ofservice->burst_limit = c->burst_limit; | |
1950 | } | |
1951 | ||
1952 | /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a | |
1953 | * positive errno value. */ | |
1954 | static int | |
1955 | ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c) | |
1956 | { | |
1957 | struct ofservice *ofservice; | |
1958 | struct pvconn *pvconn; | |
1959 | int error; | |
1960 | ||
1961 | error = pvconn_open(c->target, &pvconn); | |
1962 | if (error) { | |
1963 | return error; | |
1964 | } | |
1965 | ||
1966 | ofservice = xzalloc(sizeof *ofservice); | |
1967 | hmap_insert(&ofproto->services, &ofservice->node, | |
1968 | hash_string(c->target, 0)); | |
1969 | ofservice->pvconn = pvconn; | |
1970 | ||
1971 | ofservice_reconfigure(ofservice, c); | |
1972 | ||
1973 | return 0; | |
1974 | } | |
1975 | ||
1976 | static void | |
1977 | ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice) | |
1978 | { | |
1979 | hmap_remove(&ofproto->services, &ofservice->node); | |
1980 | pvconn_close(ofservice->pvconn); | |
1981 | free(ofservice); | |
1982 | } | |
1983 | ||
1984 | /* Finds and returns the ofservice within 'ofproto' that has the given | |
1985 | * 'target', or a null pointer if none exists. */ | |
1986 | static struct ofservice * | |
1987 | ofservice_lookup(struct ofproto *ofproto, const char *target) | |
1988 | { | |
1989 | struct ofservice *ofservice; | |
1990 | ||
4e8e4213 BP |
1991 | HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0), |
1992 | &ofproto->services) { | |
7d674866 BP |
1993 | if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) { |
1994 | return ofservice; | |
1995 | } | |
1996 | } | |
1997 | return NULL; | |
1998 | } | |
064af421 | 1999 | \f |
bcf84111 BP |
2000 | /* Returns true if 'rule' should be hidden from the controller. |
2001 | * | |
2002 | * Rules with priority higher than UINT16_MAX are set up by ofproto itself | |
2003 | * (e.g. by in-band control) and are intentionally hidden from the | |
2004 | * controller. */ | |
2005 | static bool | |
2006 | rule_is_hidden(const struct rule *rule) | |
2007 | { | |
2008 | return rule->cr.priority > UINT16_MAX; | |
2009 | } | |
2010 | ||
2011 | /* Creates and returns a new rule initialized as specified. | |
2012 | * | |
2013 | * The caller is responsible for inserting the rule into the classifier (with | |
2014 | * rule_insert()). */ | |
064af421 | 2015 | static struct rule * |
bcf84111 | 2016 | rule_create(const struct cls_rule *cls_rule, |
064af421 | 2017 | const union ofp_action *actions, size_t n_actions, |
ca069229 | 2018 | uint16_t idle_timeout, uint16_t hard_timeout, |
8054fc48 | 2019 | ovs_be64 flow_cookie, bool send_flow_removed) |
064af421 | 2020 | { |
ec6fde61 | 2021 | struct rule *rule = xzalloc(sizeof *rule); |
bcf84111 | 2022 | rule->cr = *cls_rule; |
064af421 BP |
2023 | rule->idle_timeout = idle_timeout; |
2024 | rule->hard_timeout = hard_timeout; | |
39997502 | 2025 | rule->flow_cookie = flow_cookie; |
064af421 | 2026 | rule->used = rule->created = time_msec(); |
ca069229 | 2027 | rule->send_flow_removed = send_flow_removed; |
bcf84111 | 2028 | list_init(&rule->facets); |
3dffcf07 BP |
2029 | if (n_actions > 0) { |
2030 | rule->n_actions = n_actions; | |
2031 | rule->actions = xmemdup(actions, n_actions * sizeof *actions); | |
2032 | } | |
0193b2af | 2033 | |
064af421 BP |
2034 | return rule; |
2035 | } | |
2036 | ||
2037 | static struct rule * | |
2038 | rule_from_cls_rule(const struct cls_rule *cls_rule) | |
2039 | { | |
2040 | return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL; | |
2041 | } | |
2042 | ||
2043 | static void | |
2044 | rule_free(struct rule *rule) | |
2045 | { | |
2046 | free(rule->actions); | |
064af421 BP |
2047 | free(rule); |
2048 | } | |
2049 | ||
bcf84111 BP |
2050 | /* Destroys 'rule' and iterates through all of its facets and revalidates them, |
2051 | * destroying any that no longer has a rule (which is probably all of them). | |
064af421 | 2052 | * |
bcf84111 | 2053 | * The caller must have already removed 'rule' from the classifier. */ |
064af421 BP |
2054 | static void |
2055 | rule_destroy(struct ofproto *ofproto, struct rule *rule) | |
2056 | { | |
bcf84111 BP |
2057 | struct facet *facet, *next_facet; |
2058 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { | |
2059 | facet_revalidate(ofproto, facet); | |
064af421 BP |
2060 | } |
2061 | rule_free(rule); | |
2062 | } | |
2063 | ||
bcf84111 BP |
2064 | /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action |
2065 | * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't | |
2066 | * count). */ | |
064af421 | 2067 | static bool |
8054fc48 | 2068 | rule_has_out_port(const struct rule *rule, ovs_be16 out_port) |
064af421 BP |
2069 | { |
2070 | const union ofp_action *oa; | |
2071 | struct actions_iterator i; | |
2072 | ||
2073 | if (out_port == htons(OFPP_NONE)) { | |
2074 | return true; | |
2075 | } | |
2076 | for (oa = actions_first(&i, rule->actions, rule->n_actions); oa; | |
2077 | oa = actions_next(&i)) { | |
c1c9c9c4 | 2078 | if (action_outputs_to_port(oa, out_port)) { |
064af421 BP |
2079 | return true; |
2080 | } | |
2081 | } | |
2082 | return false; | |
2083 | } | |
2084 | ||
750638bb BP |
2085 | /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on |
2086 | * 'packet', which arrived on 'in_port'. | |
2087 | * | |
2088 | * Takes ownership of 'packet'. */ | |
9dbb9d5e | 2089 | static bool |
856081f6 | 2090 | execute_odp_actions(struct ofproto *ofproto, const struct flow *flow, |
cf22f8cb | 2091 | const struct nlattr *odp_actions, size_t actions_len, |
750638bb | 2092 | struct ofpbuf *packet) |
9dbb9d5e | 2093 | { |
b9298d3f | 2094 | if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t)) |
cdee00fd | 2095 | && odp_actions->nla_type == ODPAT_CONTROLLER) { |
9dbb9d5e BP |
2096 | /* As an optimization, avoid a round-trip from userspace to kernel to |
2097 | * userspace. This also avoids possibly filling up kernel packet | |
2098 | * buffers along the way. */ | |
856081f6 | 2099 | struct dpif_upcall upcall; |
9dbb9d5e | 2100 | |
856081f6 BP |
2101 | upcall.type = _ODPL_ACTION_NR; |
2102 | upcall.packet = packet; | |
2103 | upcall.key = NULL; | |
2104 | upcall.key_len = 0; | |
2105 | upcall.userdata = nl_attr_get_u64(odp_actions); | |
2106 | upcall.sample_pool = 0; | |
2107 | upcall.actions = NULL; | |
2108 | upcall.actions_len = 0; | |
9dbb9d5e | 2109 | |
856081f6 | 2110 | send_packet_in(ofproto, &upcall, flow, false); |
9dbb9d5e | 2111 | |
750638bb BP |
2112 | return true; |
2113 | } else { | |
2114 | int error; | |
9dbb9d5e | 2115 | |
cdee00fd | 2116 | error = dpif_execute(ofproto->dpif, odp_actions, actions_len, packet); |
750638bb BP |
2117 | ofpbuf_delete(packet); |
2118 | return !error; | |
2119 | } | |
9dbb9d5e BP |
2120 | } |
2121 | ||
bcf84111 BP |
2122 | /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s |
2123 | * statistics appropriately. 'packet' must have at least sizeof(struct | |
2124 | * ofp_packet_in) bytes of headroom. | |
064af421 | 2125 | * |
bcf84111 BP |
2126 | * For correct results, 'packet' must actually be in 'facet''s flow; that is, |
2127 | * applying flow_extract() to 'packet' would yield the same flow as | |
2128 | * 'facet->flow'. | |
064af421 | 2129 | * |
bcf84111 BP |
2130 | * 'facet' must have accurately composed ODP actions; that is, it must not be |
2131 | * in need of revalidation. | |
750638bb BP |
2132 | * |
2133 | * Takes ownership of 'packet'. */ | |
064af421 | 2134 | static void |
bcf84111 BP |
2135 | facet_execute(struct ofproto *ofproto, struct facet *facet, |
2136 | struct ofpbuf *packet) | |
064af421 | 2137 | { |
750638bb | 2138 | struct odp_flow_stats stats; |
bcf84111 BP |
2139 | |
2140 | assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in)); | |
2141 | ||
2142 | flow_extract_stats(&facet->flow, packet, &stats); | |
856081f6 | 2143 | if (execute_odp_actions(ofproto, &facet->flow, |
cdee00fd | 2144 | facet->actions, facet->actions_len, packet)) { |
bcf84111 BP |
2145 | facet_update_stats(ofproto, facet, &stats); |
2146 | facet->used = time_msec(); | |
2147 | netflow_flow_update_time(ofproto->netflow, | |
2148 | &facet->nf_flow, facet->used); | |
2149 | } | |
2150 | } | |
2151 | ||
2152 | /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s | |
2153 | * statistics (or the statistics for one of its facets) appropriately. | |
2154 | * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom. | |
2155 | * | |
2156 | * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited | |
2157 | * with statistics for 'packet' either way. | |
2158 | * | |
2159 | * Takes ownership of 'packet'. */ | |
2160 | static void | |
2161 | rule_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port, | |
2162 | struct ofpbuf *packet) | |
2163 | { | |
f29152ca | 2164 | struct action_xlate_ctx ctx; |
cdee00fd | 2165 | struct ofpbuf *odp_actions; |
bcf84111 | 2166 | struct facet *facet; |
bcf84111 BP |
2167 | struct flow flow; |
2168 | size_t size; | |
064af421 | 2169 | |
750638bb BP |
2170 | assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in)); |
2171 | ||
bcf84111 BP |
2172 | flow_extract(packet, 0, in_port, &flow); |
2173 | ||
2174 | /* First look for a related facet. If we find one, account it to that. */ | |
2175 | facet = facet_lookup_valid(ofproto, &flow); | |
2176 | if (facet && facet->rule == rule) { | |
2177 | facet_execute(ofproto, facet, packet); | |
2178 | return; | |
064af421 BP |
2179 | } |
2180 | ||
bcf84111 BP |
2181 | /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then |
2182 | * create a new facet for it and use that. */ | |
2183 | if (rule_lookup(ofproto, &flow) == rule) { | |
2184 | facet = facet_create(ofproto, rule, &flow, packet); | |
2185 | facet_execute(ofproto, facet, packet); | |
2186 | facet_install(ofproto, facet, true); | |
2187 | return; | |
2188 | } | |
2189 | ||
2190 | /* We can't account anything to a facet. If we were to try, then that | |
2191 | * facet would have a non-matching rule, busting our invariants. */ | |
f29152ca | 2192 | action_xlate_ctx_init(&ctx, ofproto, &flow, packet); |
cdee00fd | 2193 | odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions); |
bcf84111 | 2194 | size = packet->size; |
856081f6 | 2195 | if (execute_odp_actions(ofproto, &flow, odp_actions->data, |
cdee00fd | 2196 | odp_actions->size, packet)) { |
064af421 | 2197 | rule->used = time_msec(); |
bcf84111 BP |
2198 | rule->packet_count++; |
2199 | rule->byte_count += size; | |
064af421 | 2200 | } |
cdee00fd | 2201 | ofpbuf_delete(odp_actions); |
064af421 BP |
2202 | } |
2203 | ||
afe75089 | 2204 | /* Inserts 'rule' into 'p''s flow table. */ |
064af421 | 2205 | static void |
afe75089 | 2206 | rule_insert(struct ofproto *p, struct rule *rule) |
064af421 BP |
2207 | { |
2208 | struct rule *displaced_rule; | |
2209 | ||
064af421 | 2210 | displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr)); |
bcf84111 BP |
2211 | if (displaced_rule) { |
2212 | rule_destroy(p, displaced_rule); | |
064af421 | 2213 | } |
bcf84111 | 2214 | p->need_revalidate = true; |
064af421 BP |
2215 | } |
2216 | ||
bcf84111 BP |
2217 | /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a |
2218 | * 'flow' and an example 'packet' within that flow. | |
2219 | * | |
2220 | * The caller must already have determined that no facet with an identical | |
2221 | * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in | |
2222 | * 'ofproto''s classifier table. */ | |
2223 | static struct facet * | |
2224 | facet_create(struct ofproto *ofproto, struct rule *rule, | |
2225 | const struct flow *flow, const struct ofpbuf *packet) | |
064af421 | 2226 | { |
bcf84111 | 2227 | struct facet *facet; |
fbb2ea0b | 2228 | |
bcf84111 BP |
2229 | facet = xzalloc(sizeof *facet); |
2230 | facet->used = time_msec(); | |
2231 | hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0)); | |
2232 | list_push_back(&rule->facets, &facet->list_node); | |
2233 | facet->rule = rule; | |
2234 | facet->flow = *flow; | |
2235 | netflow_flow_init(&facet->nf_flow); | |
2236 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used); | |
2237 | ||
2238 | facet_make_actions(ofproto, facet, packet); | |
064af421 | 2239 | |
bcf84111 BP |
2240 | return facet; |
2241 | } | |
2242 | ||
2243 | static void | |
2244 | facet_free(struct facet *facet) | |
2245 | { | |
2246 | free(facet->actions); | |
2247 | free(facet); | |
064af421 BP |
2248 | } |
2249 | ||
431d4707 | 2250 | /* Remove 'rule' from 'ofproto' and free up the associated memory: |
431d4707 BP |
2251 | * |
2252 | * - Removes 'rule' from the classifier. | |
2253 | * | |
bcf84111 BP |
2254 | * - If 'rule' has facets, revalidates them (and possibly uninstalls and |
2255 | * destroys them), via rule_destroy(). | |
431d4707 | 2256 | */ |
064af421 BP |
2257 | static void |
2258 | rule_remove(struct ofproto *ofproto, struct rule *rule) | |
2259 | { | |
bcf84111 BP |
2260 | COVERAGE_INC(ofproto_del_rule); |
2261 | ofproto->need_revalidate = true; | |
064af421 BP |
2262 | classifier_remove(&ofproto->cls, &rule->cr); |
2263 | rule_destroy(ofproto, rule); | |
2264 | } | |
2265 | ||
bcf84111 BP |
2266 | /* Remove 'facet' from 'ofproto' and free up the associated memory: |
2267 | * | |
2268 | * - If 'facet' was installed in the datapath, uninstalls it and updates its | |
2269 | * rule's statistics, via facet_uninstall(). | |
2270 | * | |
2271 | * - Removes 'facet' from its rule and from ofproto->facets. | |
2272 | */ | |
2273 | static void | |
2274 | facet_remove(struct ofproto *ofproto, struct facet *facet) | |
2275 | { | |
2276 | facet_uninstall(ofproto, facet); | |
d530fcd2 | 2277 | facet_flush_stats(ofproto, facet); |
bcf84111 BP |
2278 | hmap_remove(&ofproto->facets, &facet->hmap_node); |
2279 | list_remove(&facet->list_node); | |
2280 | facet_free(facet); | |
2281 | } | |
2282 | ||
7f7ae89d BP |
2283 | /* Composes the ODP actions for 'facet' based on its rule's actions. */ |
2284 | static void | |
bcf84111 BP |
2285 | facet_make_actions(struct ofproto *p, struct facet *facet, |
2286 | const struct ofpbuf *packet) | |
064af421 | 2287 | { |
bcf84111 | 2288 | const struct rule *rule = facet->rule; |
cdee00fd | 2289 | struct ofpbuf *odp_actions; |
f29152ca | 2290 | struct action_xlate_ctx ctx; |
064af421 | 2291 | |
f29152ca | 2292 | action_xlate_ctx_init(&ctx, p, &facet->flow, packet); |
cdee00fd | 2293 | odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions); |
19739aee JP |
2294 | facet->tags = ctx.tags; |
2295 | facet->may_install = ctx.may_set_up_flow; | |
2296 | facet->nf_flow.output_iface = ctx.nf_output_iface; | |
064af421 | 2297 | |
cdee00fd BP |
2298 | if (facet->actions_len != odp_actions->size |
2299 | || memcmp(facet->actions, odp_actions->data, odp_actions->size)) { | |
7f7ae89d | 2300 | free(facet->actions); |
cdee00fd BP |
2301 | facet->actions_len = odp_actions->size; |
2302 | facet->actions = xmemdup(odp_actions->data, odp_actions->size); | |
064af421 | 2303 | } |
cdee00fd BP |
2304 | |
2305 | ofpbuf_delete(odp_actions); | |
064af421 BP |
2306 | } |
2307 | ||
2308 | static int | |
bcf84111 | 2309 | facet_put__(struct ofproto *ofproto, struct facet *facet, int flags, |
064af421 BP |
2310 | struct odp_flow_put *put) |
2311 | { | |
36956a7d BP |
2312 | uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S]; |
2313 | struct ofpbuf key; | |
2314 | ||
2315 | ofpbuf_use_stack(&key, keybuf, sizeof keybuf); | |
2316 | odp_flow_key_from_flow(&key, &facet->flow); | |
2317 | assert(key.base == keybuf); | |
2318 | ||
064af421 | 2319 | memset(&put->flow.stats, 0, sizeof put->flow.stats); |
36956a7d BP |
2320 | put->flow.key = key.data; |
2321 | put->flow.key_len = key.size; | |
bcf84111 | 2322 | put->flow.actions = facet->actions; |
cdee00fd | 2323 | put->flow.actions_len = facet->actions_len; |
ab48643b | 2324 | put->flow.flags = 0; |
064af421 | 2325 | put->flags = flags; |
c228a364 | 2326 | return dpif_flow_put(ofproto->dpif, put); |
064af421 BP |
2327 | } |
2328 | ||
bcf84111 BP |
2329 | /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If |
2330 | * 'zero_stats' is true, clears any existing statistics from the datapath for | |
2331 | * 'facet'. */ | |
064af421 | 2332 | static void |
bcf84111 | 2333 | facet_install(struct ofproto *p, struct facet *facet, bool zero_stats) |
064af421 | 2334 | { |
bcf84111 | 2335 | if (facet->may_install) { |
064af421 | 2336 | struct odp_flow_put put; |
bcf84111 | 2337 | int flags; |
064af421 | 2338 | |
bcf84111 BP |
2339 | flags = ODPPF_CREATE | ODPPF_MODIFY; |
2340 | if (zero_stats) { | |
2341 | flags |= ODPPF_ZERO_STATS; | |
2342 | } | |
2343 | if (!facet_put__(p, facet, flags, &put)) { | |
2344 | facet->installed = true; | |
2345 | } | |
064af421 BP |
2346 | } |
2347 | } | |
2348 | ||
bcf84111 BP |
2349 | /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up |
2350 | * to the accounting hook function in the ofhooks structure. */ | |
064af421 | 2351 | static void |
bcf84111 BP |
2352 | facet_account(struct ofproto *ofproto, |
2353 | struct facet *facet, uint64_t extra_bytes) | |
064af421 | 2354 | { |
bcf84111 | 2355 | uint64_t total_bytes = facet->byte_count + extra_bytes; |
064af421 BP |
2356 | |
2357 | if (ofproto->ofhooks->account_flow_cb | |
bcf84111 | 2358 | && total_bytes > facet->accounted_bytes) |
064af421 BP |
2359 | { |
2360 | ofproto->ofhooks->account_flow_cb( | |
cdee00fd | 2361 | &facet->flow, facet->tags, facet->actions, facet->actions_len, |
bcf84111 BP |
2362 | total_bytes - facet->accounted_bytes, ofproto->aux); |
2363 | facet->accounted_bytes = total_bytes; | |
064af421 BP |
2364 | } |
2365 | } | |
2366 | ||
d530fcd2 | 2367 | /* If 'rule' is installed in the datapath, uninstalls it. */ |
064af421 | 2368 | static void |
bcf84111 | 2369 | facet_uninstall(struct ofproto *p, struct facet *facet) |
064af421 | 2370 | { |
bcf84111 | 2371 | if (facet->installed) { |
36956a7d | 2372 | uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S]; |
064af421 | 2373 | struct odp_flow odp_flow; |
36956a7d | 2374 | struct ofpbuf key; |
064af421 | 2375 | |
36956a7d BP |
2376 | ofpbuf_use_stack(&key, keybuf, sizeof keybuf); |
2377 | odp_flow_key_from_flow(&key, &facet->flow); | |
2378 | assert(key.base == keybuf); | |
2379 | ||
2380 | odp_flow.key = key.data; | |
2381 | odp_flow.key_len = key.size; | |
064af421 | 2382 | odp_flow.actions = NULL; |
cdee00fd | 2383 | odp_flow.actions_len = 0; |
ab48643b | 2384 | odp_flow.flags = 0; |
c228a364 | 2385 | if (!dpif_flow_del(p->dpif, &odp_flow)) { |
bcf84111 | 2386 | facet_update_stats(p, facet, &odp_flow.stats); |
064af421 | 2387 | } |
bcf84111 | 2388 | facet->installed = false; |
064af421 BP |
2389 | } |
2390 | } | |
2391 | ||
bcf84111 BP |
2392 | /* Returns true if the only action for 'facet' is to send to the controller. |
2393 | * (We don't report NetFlow expiration messages for such facets because they | |
2394 | * are just part of the control logic for the network, not real traffic). */ | |
0193b2af | 2395 | static bool |
bcf84111 | 2396 | facet_is_controller_flow(struct facet *facet) |
0193b2af | 2397 | { |
bcf84111 BP |
2398 | return (facet |
2399 | && facet->rule->n_actions == 1 | |
2400 | && action_outputs_to_port(&facet->rule->actions[0], | |
c1c9c9c4 | 2401 | htons(OFPP_CONTROLLER))); |
0193b2af JG |
2402 | } |
2403 | ||
bcf84111 BP |
2404 | /* Folds all of 'facet''s statistics into its rule. Also updates the |
2405 | * accounting ofhook and emits a NetFlow expiration if appropriate. */ | |
064af421 | 2406 | static void |
d530fcd2 | 2407 | facet_flush_stats(struct ofproto *ofproto, struct facet *facet) |
064af421 | 2408 | { |
bcf84111 | 2409 | facet_account(ofproto, facet, 0); |
064af421 | 2410 | |
bcf84111 | 2411 | if (ofproto->netflow && !facet_is_controller_flow(facet)) { |
064af421 | 2412 | struct ofexpired expired; |
bcf84111 BP |
2413 | expired.flow = facet->flow; |
2414 | expired.packet_count = facet->packet_count; | |
2415 | expired.byte_count = facet->byte_count; | |
2416 | expired.used = facet->used; | |
2417 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
2418 | } | |
2419 | ||
2420 | facet->rule->packet_count += facet->packet_count; | |
2421 | facet->rule->byte_count += facet->byte_count; | |
2422 | ||
2423 | /* Reset counters to prevent double counting if 'facet' ever gets | |
2424 | * reinstalled. */ | |
2425 | facet->packet_count = 0; | |
2426 | facet->byte_count = 0; | |
2427 | facet->accounted_bytes = 0; | |
2428 | ||
2429 | netflow_flow_clear(&facet->nf_flow); | |
2430 | } | |
2431 | ||
2432 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
2433 | * Returns it if found, otherwise a null pointer. | |
2434 | * | |
2435 | * The returned facet might need revalidation; use facet_lookup_valid() | |
2436 | * instead if that is important. */ | |
2437 | static struct facet * | |
2438 | facet_find(struct ofproto *ofproto, const struct flow *flow) | |
2439 | { | |
2440 | struct facet *facet; | |
2441 | ||
2442 | HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0), | |
2443 | &ofproto->facets) { | |
2444 | if (flow_equal(flow, &facet->flow)) { | |
2445 | return facet; | |
2446 | } | |
2447 | } | |
2448 | ||
2449 | return NULL; | |
2450 | } | |
2451 | ||
2452 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
2453 | * Returns it if found, otherwise a null pointer. | |
2454 | * | |
2455 | * The returned facet is guaranteed to be valid. */ | |
2456 | static struct facet * | |
2457 | facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow) | |
2458 | { | |
2459 | struct facet *facet = facet_find(ofproto, flow); | |
2460 | ||
2461 | /* The facet we found might not be valid, since we could be in need of | |
2462 | * revalidation. If it is not valid, don't return it. */ | |
2463 | if (facet | |
2464 | && ofproto->need_revalidate | |
2465 | && !facet_revalidate(ofproto, facet)) { | |
2466 | COVERAGE_INC(ofproto_invalidated); | |
2467 | return NULL; | |
064af421 | 2468 | } |
064af421 | 2469 | |
bcf84111 BP |
2470 | return facet; |
2471 | } | |
0193b2af | 2472 | |
bcf84111 BP |
2473 | /* Re-searches 'ofproto''s classifier for a rule matching 'facet': |
2474 | * | |
2475 | * - If the rule found is different from 'facet''s current rule, moves | |
2476 | * 'facet' to the new rule and recompiles its actions. | |
2477 | * | |
2478 | * - If the rule found is the same as 'facet''s current rule, leaves 'facet' | |
2479 | * where it is and recompiles its actions anyway. | |
2480 | * | |
2481 | * - If there is none, destroys 'facet'. | |
2482 | * | |
d530fcd2 | 2483 | * Returns true if 'facet' still exists, false if it has been destroyed. */ |
bcf84111 BP |
2484 | static bool |
2485 | facet_revalidate(struct ofproto *ofproto, struct facet *facet) | |
2486 | { | |
f29152ca | 2487 | struct action_xlate_ctx ctx; |
cdee00fd | 2488 | struct ofpbuf *odp_actions; |
d530fcd2 | 2489 | struct rule *new_rule; |
d530fcd2 | 2490 | bool actions_changed; |
bcf84111 BP |
2491 | |
2492 | COVERAGE_INC(facet_revalidate); | |
d530fcd2 BP |
2493 | |
2494 | /* Determine the new rule. */ | |
2495 | new_rule = rule_lookup(ofproto, &facet->flow); | |
2496 | if (!new_rule) { | |
2497 | /* No new rule, so delete the facet. */ | |
bcf84111 BP |
2498 | facet_remove(ofproto, facet); |
2499 | return false; | |
2500 | } | |
2501 | ||
d530fcd2 BP |
2502 | /* Calculate new ODP actions. |
2503 | * | |
cdee00fd BP |
2504 | * We do not modify any 'facet' state yet, because we might need to, e.g., |
2505 | * emit a NetFlow expiration and, if so, we need to have the old state | |
2506 | * around to properly compose it. */ | |
f29152ca | 2507 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL); |
cdee00fd BP |
2508 | odp_actions = xlate_actions(&ctx, new_rule->actions, new_rule->n_actions); |
2509 | actions_changed = (facet->actions_len != odp_actions->size | |
2510 | || memcmp(facet->actions, odp_actions->data, | |
2511 | facet->actions_len)); | |
d530fcd2 BP |
2512 | |
2513 | /* If the ODP actions changed or the installability changed, then we need | |
2514 | * to talk to the datapath. */ | |
2515 | if (actions_changed || facet->may_install != facet->installed) { | |
2516 | if (facet->may_install) { | |
36956a7d | 2517 | uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S]; |
d530fcd2 | 2518 | struct odp_flow_put put; |
36956a7d BP |
2519 | struct ofpbuf key; |
2520 | ||
2521 | ofpbuf_use_stack(&key, keybuf, sizeof keybuf); | |
2522 | odp_flow_key_from_flow(&key, &facet->flow); | |
d530fcd2 BP |
2523 | |
2524 | memset(&put.flow.stats, 0, sizeof put.flow.stats); | |
36956a7d BP |
2525 | put.flow.key = key.data; |
2526 | put.flow.key_len = key.size; | |
cdee00fd BP |
2527 | put.flow.actions = odp_actions->data; |
2528 | put.flow.actions_len = odp_actions->size; | |
d530fcd2 BP |
2529 | put.flow.flags = 0; |
2530 | put.flags = ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS; | |
2531 | dpif_flow_put(ofproto->dpif, &put); | |
2532 | ||
2533 | facet_update_stats(ofproto, facet, &put.flow.stats); | |
2534 | } else { | |
2535 | facet_uninstall(ofproto, facet); | |
2536 | } | |
2537 | ||
2538 | /* The datapath flow is gone or has zeroed stats, so push stats out of | |
2539 | * 'facet' into 'rule'. */ | |
2540 | facet_flush_stats(ofproto, facet); | |
2541 | } | |
2542 | ||
2543 | /* Update 'facet' now that we've taken care of all the old state. */ | |
f29152ca BP |
2544 | facet->tags = ctx.tags; |
2545 | facet->nf_flow.output_iface = ctx.nf_output_iface; | |
2546 | facet->may_install = ctx.may_set_up_flow; | |
d530fcd2 BP |
2547 | if (actions_changed) { |
2548 | free(facet->actions); | |
cdee00fd BP |
2549 | facet->actions_len = odp_actions->size; |
2550 | facet->actions = xmemdup(odp_actions->data, odp_actions->size); | |
d530fcd2 BP |
2551 | } |
2552 | if (facet->rule != new_rule) { | |
bcf84111 BP |
2553 | COVERAGE_INC(facet_changed_rule); |
2554 | list_remove(&facet->list_node); | |
d530fcd2 BP |
2555 | list_push_back(&new_rule->facets, &facet->list_node); |
2556 | facet->rule = new_rule; | |
2557 | facet->used = new_rule->created; | |
0c0afbec | 2558 | } |
bcf84111 | 2559 | |
cdfcd496 BP |
2560 | ofpbuf_delete(odp_actions); |
2561 | ||
bcf84111 | 2562 | return true; |
064af421 BP |
2563 | } |
2564 | \f | |
2565 | static void | |
2566 | queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn, | |
2567 | struct rconn_packet_counter *counter) | |
2568 | { | |
2569 | update_openflow_length(msg); | |
2570 | if (rconn_send(ofconn->rconn, msg, counter)) { | |
2571 | ofpbuf_delete(msg); | |
2572 | } | |
2573 | } | |
2574 | ||
064af421 BP |
2575 | static void |
2576 | send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh, | |
2577 | int error) | |
2578 | { | |
dc4762ed | 2579 | struct ofpbuf *buf = ofputil_encode_error_msg(error, oh); |
26c112c2 BP |
2580 | if (buf) { |
2581 | COVERAGE_INC(ofproto_error); | |
2582 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
2583 | } | |
064af421 BP |
2584 | } |
2585 | ||
2586 | static void | |
2587 | hton_ofp_phy_port(struct ofp_phy_port *opp) | |
2588 | { | |
2589 | opp->port_no = htons(opp->port_no); | |
2590 | opp->config = htonl(opp->config); | |
2591 | opp->state = htonl(opp->state); | |
2592 | opp->curr = htonl(opp->curr); | |
2593 | opp->advertised = htonl(opp->advertised); | |
2594 | opp->supported = htonl(opp->supported); | |
2595 | opp->peer = htonl(opp->peer); | |
2596 | } | |
2597 | ||
2598 | static int | |
d1e2cf21 | 2599 | handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh) |
064af421 | 2600 | { |
d1e2cf21 | 2601 | queue_tx(make_echo_reply(oh), ofconn, ofconn->reply_counter); |
064af421 BP |
2602 | return 0; |
2603 | } | |
2604 | ||
2605 | static int | |
d1e2cf21 | 2606 | handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh) |
064af421 BP |
2607 | { |
2608 | struct ofp_switch_features *osf; | |
2609 | struct ofpbuf *buf; | |
064af421 BP |
2610 | struct ofport *port; |
2611 | ||
2612 | osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf); | |
3269c562 | 2613 | osf->datapath_id = htonll(ofconn->ofproto->datapath_id); |
064af421 BP |
2614 | osf->n_buffers = htonl(pktbuf_capacity()); |
2615 | osf->n_tables = 2; | |
2616 | osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS | | |
0254ae23 | 2617 | OFPC_PORT_STATS | OFPC_ARP_MATCH_IP); |
064af421 BP |
2618 | osf->actions = htonl((1u << OFPAT_OUTPUT) | |
2619 | (1u << OFPAT_SET_VLAN_VID) | | |
2620 | (1u << OFPAT_SET_VLAN_PCP) | | |
2621 | (1u << OFPAT_STRIP_VLAN) | | |
2622 | (1u << OFPAT_SET_DL_SRC) | | |
2623 | (1u << OFPAT_SET_DL_DST) | | |
2624 | (1u << OFPAT_SET_NW_SRC) | | |
2625 | (1u << OFPAT_SET_NW_DST) | | |
959a2ecd | 2626 | (1u << OFPAT_SET_NW_TOS) | |
064af421 | 2627 | (1u << OFPAT_SET_TP_SRC) | |
c1c9c9c4 BP |
2628 | (1u << OFPAT_SET_TP_DST) | |
2629 | (1u << OFPAT_ENQUEUE)); | |
064af421 | 2630 | |
3269c562 | 2631 | HMAP_FOR_EACH (port, hmap_node, &ofconn->ofproto->ports) { |
064af421 BP |
2632 | hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp)); |
2633 | } | |
2634 | ||
2635 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
2636 | return 0; | |
2637 | } | |
2638 | ||
2639 | static int | |
d1e2cf21 | 2640 | handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh) |
064af421 BP |
2641 | { |
2642 | struct ofpbuf *buf; | |
2643 | struct ofp_switch_config *osc; | |
2644 | uint16_t flags; | |
2645 | bool drop_frags; | |
2646 | ||
2647 | /* Figure out flags. */ | |
3269c562 | 2648 | dpif_get_drop_frags(ofconn->ofproto->dpif, &drop_frags); |
064af421 | 2649 | flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL; |
064af421 BP |
2650 | |
2651 | /* Send reply. */ | |
2652 | osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf); | |
2653 | osc->flags = htons(flags); | |
2654 | osc->miss_send_len = htons(ofconn->miss_send_len); | |
2655 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
2656 | ||
2657 | return 0; | |
2658 | } | |
2659 | ||
2660 | static int | |
d1e2cf21 | 2661 | handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc) |
064af421 | 2662 | { |
d1e2cf21 | 2663 | uint16_t flags = ntohs(osc->flags); |
064af421 | 2664 | |
5899143f | 2665 | if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) { |
064af421 BP |
2666 | switch (flags & OFPC_FRAG_MASK) { |
2667 | case OFPC_FRAG_NORMAL: | |
3269c562 | 2668 | dpif_set_drop_frags(ofconn->ofproto->dpif, false); |
064af421 BP |
2669 | break; |
2670 | case OFPC_FRAG_DROP: | |
3269c562 | 2671 | dpif_set_drop_frags(ofconn->ofproto->dpif, true); |
064af421 BP |
2672 | break; |
2673 | default: | |
2674 | VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")", | |
2675 | osc->flags); | |
2676 | break; | |
2677 | } | |
2678 | } | |
2679 | ||
064af421 BP |
2680 | ofconn->miss_send_len = ntohs(osc->miss_send_len); |
2681 | ||
2682 | return 0; | |
2683 | } | |
2684 | ||
5f8bfd69 BP |
2685 | /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a |
2686 | * flow translation. */ | |
33b5304b | 2687 | #define MAX_RESUBMIT_RECURSION 16 |
5f8bfd69 | 2688 | |
064af421 BP |
2689 | static void do_xlate_actions(const union ofp_action *in, size_t n_in, |
2690 | struct action_xlate_ctx *ctx); | |
2691 | ||
2692 | static void | |
2693 | add_output_action(struct action_xlate_ctx *ctx, uint16_t port) | |
2694 | { | |
ca0f572c | 2695 | const struct ofport *ofport = get_port(ctx->ofproto, port); |
6cfaf517 BP |
2696 | |
2697 | if (ofport) { | |
2698 | if (ofport->opp.config & OFPPC_NO_FWD) { | |
2699 | /* Forwarding disabled on port. */ | |
2700 | return; | |
2701 | } | |
2702 | } else { | |
2703 | /* | |
2704 | * We don't have an ofport record for this port, but it doesn't hurt to | |
2705 | * allow forwarding to it anyhow. Maybe such a port will appear later | |
2706 | * and we're pre-populating the flow table. | |
2707 | */ | |
064af421 | 2708 | } |
6cfaf517 | 2709 | |
cdee00fd | 2710 | nl_msg_put_u32(ctx->odp_actions, ODPAT_OUTPUT, port); |
6a07af36 | 2711 | ctx->nf_output_iface = port; |
064af421 BP |
2712 | } |
2713 | ||
2714 | static struct rule * | |
bcf84111 | 2715 | rule_lookup(struct ofproto *ofproto, const struct flow *flow) |
064af421 | 2716 | { |
3c4486a5 | 2717 | return rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow)); |
064af421 BP |
2718 | } |
2719 | ||
2720 | static void | |
2721 | xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port) | |
2722 | { | |
5f8bfd69 | 2723 | if (ctx->recurse < MAX_RESUBMIT_RECURSION) { |
2c5d1389 | 2724 | uint16_t old_in_port; |
064af421 | 2725 | struct rule *rule; |
064af421 | 2726 | |
2c5d1389 BP |
2727 | /* Look up a flow with 'in_port' as the input port. Then restore the |
2728 | * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will | |
2729 | * have surprising behavior). */ | |
2730 | old_in_port = ctx->flow.in_port; | |
e18fe8a2 | 2731 | ctx->flow.in_port = in_port; |
bcf84111 | 2732 | rule = rule_lookup(ctx->ofproto, &ctx->flow); |
2c5d1389 BP |
2733 | ctx->flow.in_port = old_in_port; |
2734 | ||
7aa697dd BP |
2735 | if (ctx->resubmit_hook) { |
2736 | ctx->resubmit_hook(ctx, rule); | |
2737 | } | |
2738 | ||
064af421 | 2739 | if (rule) { |
064af421 BP |
2740 | ctx->recurse++; |
2741 | do_xlate_actions(rule->actions, rule->n_actions, ctx); | |
2742 | ctx->recurse--; | |
2743 | } | |
5f8bfd69 | 2744 | } else { |
db5ce514 | 2745 | static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1); |
5f8bfd69 BP |
2746 | |
2747 | VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times", | |
2748 | MAX_RESUBMIT_RECURSION); | |
064af421 BP |
2749 | } |
2750 | } | |
2751 | ||
f1588b1f BP |
2752 | static void |
2753 | flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, uint32_t mask, | |
cdee00fd | 2754 | uint16_t *nf_output_iface, struct ofpbuf *odp_actions) |
f1588b1f BP |
2755 | { |
2756 | struct ofport *ofport; | |
2757 | ||
2758 | HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) { | |
2759 | uint16_t odp_port = ofport->odp_port; | |
2760 | if (odp_port != odp_in_port && !(ofport->opp.config & mask)) { | |
cdee00fd | 2761 | nl_msg_put_u32(odp_actions, ODPAT_OUTPUT, odp_port); |
f1588b1f BP |
2762 | } |
2763 | } | |
2764 | *nf_output_iface = NF_OUT_FLOOD; | |
2765 | } | |
2766 | ||
064af421 | 2767 | static void |
aae51f53 BP |
2768 | xlate_output_action__(struct action_xlate_ctx *ctx, |
2769 | uint16_t port, uint16_t max_len) | |
064af421 BP |
2770 | { |
2771 | uint16_t odp_port; | |
6a07af36 JG |
2772 | uint16_t prev_nf_output_iface = ctx->nf_output_iface; |
2773 | ||
2774 | ctx->nf_output_iface = NF_OUT_DROP; | |
064af421 | 2775 | |
aae51f53 | 2776 | switch (port) { |
064af421 | 2777 | case OFPP_IN_PORT: |
e18fe8a2 | 2778 | add_output_action(ctx, ctx->flow.in_port); |
064af421 BP |
2779 | break; |
2780 | case OFPP_TABLE: | |
e18fe8a2 | 2781 | xlate_table_action(ctx, ctx->flow.in_port); |
064af421 BP |
2782 | break; |
2783 | case OFPP_NORMAL: | |
e18fe8a2 | 2784 | if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet, |
cdee00fd | 2785 | ctx->odp_actions, &ctx->tags, |
6a07af36 | 2786 | &ctx->nf_output_iface, |
064af421 BP |
2787 | ctx->ofproto->aux)) { |
2788 | COVERAGE_INC(ofproto_uninstallable); | |
d6fbec6d | 2789 | ctx->may_set_up_flow = false; |
064af421 BP |
2790 | } |
2791 | break; | |
2792 | case OFPP_FLOOD: | |
f1588b1f | 2793 | flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD, |
cdee00fd | 2794 | &ctx->nf_output_iface, ctx->odp_actions); |
9628cd42 | 2795 | break; |
064af421 | 2796 | case OFPP_ALL: |
f1588b1f | 2797 | flood_packets(ctx->ofproto, ctx->flow.in_port, 0, |
cdee00fd | 2798 | &ctx->nf_output_iface, ctx->odp_actions); |
064af421 BP |
2799 | break; |
2800 | case OFPP_CONTROLLER: | |
b9298d3f | 2801 | nl_msg_put_u64(ctx->odp_actions, ODPAT_CONTROLLER, max_len); |
064af421 BP |
2802 | break; |
2803 | case OFPP_LOCAL: | |
2804 | add_output_action(ctx, ODPP_LOCAL); | |
2805 | break; | |
2806 | default: | |
aae51f53 | 2807 | odp_port = ofp_port_to_odp_port(port); |
e18fe8a2 | 2808 | if (odp_port != ctx->flow.in_port) { |
064af421 BP |
2809 | add_output_action(ctx, odp_port); |
2810 | } | |
2811 | break; | |
2812 | } | |
6a07af36 JG |
2813 | |
2814 | if (prev_nf_output_iface == NF_OUT_FLOOD) { | |
2815 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
2816 | } else if (ctx->nf_output_iface == NF_OUT_DROP) { | |
2817 | ctx->nf_output_iface = prev_nf_output_iface; | |
2818 | } else if (prev_nf_output_iface != NF_OUT_DROP && | |
2819 | ctx->nf_output_iface != NF_OUT_FLOOD) { | |
2820 | ctx->nf_output_iface = NF_OUT_MULTI; | |
2821 | } | |
064af421 BP |
2822 | } |
2823 | ||
aae51f53 BP |
2824 | static void |
2825 | xlate_output_action(struct action_xlate_ctx *ctx, | |
2826 | const struct ofp_action_output *oao) | |
2827 | { | |
2828 | xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len)); | |
2829 | } | |
2830 | ||
c1c9c9c4 BP |
2831 | /* If the final ODP action in 'ctx' is "pop priority", drop it, as an |
2832 | * optimization, because we're going to add another action that sets the | |
2833 | * priority immediately after, or because there are no actions following the | |
2834 | * pop. */ | |
2835 | static void | |
2836 | remove_pop_action(struct action_xlate_ctx *ctx) | |
2837 | { | |
cdee00fd BP |
2838 | if (ctx->odp_actions->size == ctx->last_pop_priority) { |
2839 | ctx->odp_actions->size -= NLA_ALIGN(NLA_HDRLEN); | |
2840 | ctx->last_pop_priority = -1; | |
2841 | } | |
2842 | } | |
2843 | ||
2844 | static void | |
2845 | add_pop_action(struct action_xlate_ctx *ctx) | |
2846 | { | |
2847 | if (ctx->odp_actions->size != ctx->last_pop_priority) { | |
2848 | nl_msg_put_flag(ctx->odp_actions, ODPAT_POP_PRIORITY); | |
2849 | ctx->last_pop_priority = ctx->odp_actions->size; | |
c1c9c9c4 BP |
2850 | } |
2851 | } | |
2852 | ||
2853 | static void | |
2854 | xlate_enqueue_action(struct action_xlate_ctx *ctx, | |
2855 | const struct ofp_action_enqueue *oae) | |
2856 | { | |
2857 | uint16_t ofp_port, odp_port; | |
aae51f53 BP |
2858 | uint32_t priority; |
2859 | int error; | |
2860 | ||
2861 | error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id), | |
2862 | &priority); | |
2863 | if (error) { | |
2864 | /* Fall back to ordinary output action. */ | |
2865 | xlate_output_action__(ctx, ntohs(oae->port), 0); | |
2866 | return; | |
2867 | } | |
c1c9c9c4 BP |
2868 | |
2869 | /* Figure out ODP output port. */ | |
2870 | ofp_port = ntohs(oae->port); | |
2871 | if (ofp_port != OFPP_IN_PORT) { | |
2872 | odp_port = ofp_port_to_odp_port(ofp_port); | |
2873 | } else { | |
2874 | odp_port = ctx->flow.in_port; | |
2875 | } | |
2876 | ||
2877 | /* Add ODP actions. */ | |
2878 | remove_pop_action(ctx); | |
cdee00fd | 2879 | nl_msg_put_u32(ctx->odp_actions, ODPAT_SET_PRIORITY, priority); |
c1c9c9c4 | 2880 | add_output_action(ctx, odp_port); |
cdee00fd | 2881 | add_pop_action(ctx); |
c1c9c9c4 BP |
2882 | |
2883 | /* Update NetFlow output port. */ | |
2884 | if (ctx->nf_output_iface == NF_OUT_DROP) { | |
2885 | ctx->nf_output_iface = odp_port; | |
2886 | } else if (ctx->nf_output_iface != NF_OUT_FLOOD) { | |
2887 | ctx->nf_output_iface = NF_OUT_MULTI; | |
2888 | } | |
2889 | } | |
2890 | ||
eedc0097 JP |
2891 | static void |
2892 | xlate_set_queue_action(struct action_xlate_ctx *ctx, | |
2893 | const struct nx_action_set_queue *nasq) | |
2894 | { | |
2895 | uint32_t priority; | |
2896 | int error; | |
2897 | ||
2898 | error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id), | |
2899 | &priority); | |
2900 | if (error) { | |
2901 | /* Couldn't translate queue to a priority, so ignore. A warning | |
2902 | * has already been logged. */ | |
2903 | return; | |
2904 | } | |
2905 | ||
2906 | remove_pop_action(ctx); | |
cdee00fd | 2907 | nl_msg_put_u32(ctx->odp_actions, ODPAT_SET_PRIORITY, priority); |
eedc0097 JP |
2908 | } |
2909 | ||
350a665f BP |
2910 | static void |
2911 | xlate_set_dl_tci(struct action_xlate_ctx *ctx) | |
2912 | { | |
66642cb4 BP |
2913 | ovs_be16 tci = ctx->flow.vlan_tci; |
2914 | if (!(tci & htons(VLAN_CFI))) { | |
cdee00fd | 2915 | nl_msg_put_flag(ctx->odp_actions, ODPAT_STRIP_VLAN); |
350a665f | 2916 | } else { |
cdee00fd BP |
2917 | nl_msg_put_be16(ctx->odp_actions, ODPAT_SET_DL_TCI, |
2918 | tci & ~htons(VLAN_CFI)); | |
350a665f BP |
2919 | } |
2920 | } | |
2921 | ||
7b064a79 BP |
2922 | struct xlate_reg_state { |
2923 | ovs_be16 vlan_tci; | |
2924 | ovs_be64 tun_id; | |
2925 | }; | |
2926 | ||
b6c9e612 | 2927 | static void |
7b064a79 BP |
2928 | save_reg_state(const struct action_xlate_ctx *ctx, |
2929 | struct xlate_reg_state *state) | |
b6c9e612 | 2930 | { |
7b064a79 BP |
2931 | state->vlan_tci = ctx->flow.vlan_tci; |
2932 | state->tun_id = ctx->flow.tun_id; | |
2933 | } | |
b6c9e612 | 2934 | |
7b064a79 BP |
2935 | static void |
2936 | update_reg_state(struct action_xlate_ctx *ctx, | |
2937 | const struct xlate_reg_state *state) | |
2938 | { | |
2939 | if (ctx->flow.vlan_tci != state->vlan_tci) { | |
b6c9e612 BP |
2940 | xlate_set_dl_tci(ctx); |
2941 | } | |
7b064a79 | 2942 | if (ctx->flow.tun_id != state->tun_id) { |
926947e6 BP |
2943 | nl_msg_put_be64(ctx->odp_actions, ODPAT_SET_TUNNEL, ctx->flow.tun_id); |
2944 | } | |
b6c9e612 BP |
2945 | } |
2946 | ||
064af421 BP |
2947 | static void |
2948 | xlate_nicira_action(struct action_xlate_ctx *ctx, | |
2949 | const struct nx_action_header *nah) | |
2950 | { | |
2951 | const struct nx_action_resubmit *nar; | |
659586ef | 2952 | const struct nx_action_set_tunnel *nast; |
eedc0097 | 2953 | const struct nx_action_set_queue *nasq; |
53ddd40a | 2954 | const struct nx_action_multipath *nam; |
e41a9130 | 2955 | enum nx_action_subtype subtype = ntohs(nah->subtype); |
7b064a79 | 2956 | struct xlate_reg_state state; |
b9298d3f | 2957 | ovs_be64 tun_id; |
064af421 BP |
2958 | |
2959 | assert(nah->vendor == htonl(NX_VENDOR_ID)); | |
2960 | switch (subtype) { | |
2961 | case NXAST_RESUBMIT: | |
2962 | nar = (const struct nx_action_resubmit *) nah; | |
2963 | xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port))); | |
2964 | break; | |
2965 | ||
659586ef JG |
2966 | case NXAST_SET_TUNNEL: |
2967 | nast = (const struct nx_action_set_tunnel *) nah; | |
b9298d3f BP |
2968 | tun_id = htonll(ntohl(nast->tun_id)); |
2969 | nl_msg_put_be64(ctx->odp_actions, ODPAT_SET_TUNNEL, tun_id); | |
2970 | ctx->flow.tun_id = tun_id; | |
659586ef JG |
2971 | break; |
2972 | ||
401eeb92 BP |
2973 | case NXAST_DROP_SPOOFED_ARP: |
2974 | if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) { | |
cdee00fd | 2975 | nl_msg_put_flag(ctx->odp_actions, ODPAT_DROP_SPOOFED_ARP); |
401eeb92 BP |
2976 | } |
2977 | break; | |
2978 | ||
eedc0097 JP |
2979 | case NXAST_SET_QUEUE: |
2980 | nasq = (const struct nx_action_set_queue *) nah; | |
2981 | xlate_set_queue_action(ctx, nasq); | |
2982 | break; | |
2983 | ||
2984 | case NXAST_POP_QUEUE: | |
cdee00fd | 2985 | add_pop_action(ctx); |
eedc0097 JP |
2986 | break; |
2987 | ||
b6c9e612 | 2988 | case NXAST_REG_MOVE: |
7b064a79 BP |
2989 | save_reg_state(ctx, &state); |
2990 | nxm_execute_reg_move((const struct nx_action_reg_move *) nah, | |
2991 | &ctx->flow); | |
2992 | update_reg_state(ctx, &state); | |
b6c9e612 BP |
2993 | break; |
2994 | ||
2995 | case NXAST_REG_LOAD: | |
7b064a79 | 2996 | save_reg_state(ctx, &state); |
b6c9e612 BP |
2997 | nxm_execute_reg_load((const struct nx_action_reg_load *) nah, |
2998 | &ctx->flow); | |
7b064a79 BP |
2999 | update_reg_state(ctx, &state); |
3000 | break; | |
96fc46e8 BP |
3001 | |
3002 | case NXAST_NOTE: | |
3003 | /* Nothing to do. */ | |
b6c9e612 BP |
3004 | break; |
3005 | ||
b9298d3f BP |
3006 | case NXAST_SET_TUNNEL64: |
3007 | tun_id = ((const struct nx_action_set_tunnel64 *) nah)->tun_id; | |
3008 | nl_msg_put_be64(ctx->odp_actions, ODPAT_SET_TUNNEL, tun_id); | |
3009 | ctx->flow.tun_id = tun_id; | |
3010 | break; | |
3011 | ||
53ddd40a BP |
3012 | case NXAST_MULTIPATH: |
3013 | nam = (const struct nx_action_multipath *) nah; | |
3014 | multipath_execute(nam, &ctx->flow); | |
3015 | break; | |
3016 | ||
999f0d45 | 3017 | /* If you add a new action here that modifies flow data, don't forget to |
c1c9c9c4 | 3018 | * update the flow key in ctx->flow at the same time. */ |
999f0d45 | 3019 | |
e41a9130 | 3020 | case NXAST_SNAT__OBSOLETE: |
064af421 | 3021 | default: |
e41a9130 | 3022 | VLOG_DBG_RL(&rl, "unknown Nicira action type %d", (int) subtype); |
064af421 BP |
3023 | break; |
3024 | } | |
3025 | } | |
3026 | ||
3027 | static void | |
3028 | do_xlate_actions(const union ofp_action *in, size_t n_in, | |
3029 | struct action_xlate_ctx *ctx) | |
3030 | { | |
3031 | struct actions_iterator iter; | |
3032 | const union ofp_action *ia; | |
3033 | const struct ofport *port; | |
3034 | ||
ca0f572c | 3035 | port = get_port(ctx->ofproto, ctx->flow.in_port); |
064af421 | 3036 | if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) && |
ba186119 | 3037 | port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp) |
064af421 BP |
3038 | ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) { |
3039 | /* Drop this flow. */ | |
3040 | return; | |
3041 | } | |
3042 | ||
3043 | for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) { | |
e41a9130 | 3044 | enum ofp_action_type type = ntohs(ia->type); |
cdee00fd | 3045 | const struct ofp_action_dl_addr *oada; |
064af421 BP |
3046 | |
3047 | switch (type) { | |
3048 | case OFPAT_OUTPUT: | |
3049 | xlate_output_action(ctx, &ia->output); | |
3050 | break; | |
3051 | ||
3052 | case OFPAT_SET_VLAN_VID: | |
66642cb4 BP |
3053 | ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK); |
3054 | ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI); | |
350a665f | 3055 | xlate_set_dl_tci(ctx); |
064af421 BP |
3056 | break; |
3057 | ||
3058 | case OFPAT_SET_VLAN_PCP: | |
66642cb4 BP |
3059 | ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK); |
3060 | ctx->flow.vlan_tci |= htons( | |
3061 | (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI); | |
350a665f | 3062 | xlate_set_dl_tci(ctx); |
064af421 BP |
3063 | break; |
3064 | ||
3065 | case OFPAT_STRIP_VLAN: | |
66642cb4 | 3066 | ctx->flow.vlan_tci = htons(0); |
350a665f | 3067 | xlate_set_dl_tci(ctx); |
064af421 BP |
3068 | break; |
3069 | ||
3070 | case OFPAT_SET_DL_SRC: | |
cdee00fd BP |
3071 | oada = ((struct ofp_action_dl_addr *) ia); |
3072 | nl_msg_put_unspec(ctx->odp_actions, ODPAT_SET_DL_SRC, | |
3073 | oada->dl_addr, ETH_ADDR_LEN); | |
3074 | memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN); | |
064af421 BP |
3075 | break; |
3076 | ||
3077 | case OFPAT_SET_DL_DST: | |
cdee00fd BP |
3078 | oada = ((struct ofp_action_dl_addr *) ia); |
3079 | nl_msg_put_unspec(ctx->odp_actions, ODPAT_SET_DL_DST, | |
3080 | oada->dl_addr, ETH_ADDR_LEN); | |
3081 | memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN); | |
064af421 BP |
3082 | break; |
3083 | ||
3084 | case OFPAT_SET_NW_SRC: | |
cdee00fd BP |
3085 | nl_msg_put_be32(ctx->odp_actions, ODPAT_SET_NW_SRC, |
3086 | ia->nw_addr.nw_addr); | |
3087 | ctx->flow.nw_src = ia->nw_addr.nw_addr; | |
064af421 BP |
3088 | break; |
3089 | ||
2d70a31a | 3090 | case OFPAT_SET_NW_DST: |
cdee00fd BP |
3091 | nl_msg_put_be32(ctx->odp_actions, ODPAT_SET_NW_DST, |
3092 | ia->nw_addr.nw_addr); | |
3093 | ctx->flow.nw_dst = ia->nw_addr.nw_addr; | |
2d38e234 | 3094 | break; |
959a2ecd JP |
3095 | |
3096 | case OFPAT_SET_NW_TOS: | |
cdee00fd BP |
3097 | nl_msg_put_u8(ctx->odp_actions, ODPAT_SET_NW_TOS, |
3098 | ia->nw_tos.nw_tos); | |
3099 | ctx->flow.nw_tos = ia->nw_tos.nw_tos; | |
2d70a31a JP |
3100 | break; |
3101 | ||
064af421 | 3102 | case OFPAT_SET_TP_SRC: |
cdee00fd BP |
3103 | nl_msg_put_be16(ctx->odp_actions, ODPAT_SET_TP_SRC, |
3104 | ia->tp_port.tp_port); | |
3105 | ctx->flow.tp_src = ia->tp_port.tp_port; | |
064af421 BP |
3106 | break; |
3107 | ||
2d70a31a | 3108 | case OFPAT_SET_TP_DST: |
cdee00fd BP |
3109 | nl_msg_put_be16(ctx->odp_actions, ODPAT_SET_TP_DST, |
3110 | ia->tp_port.tp_port); | |
3111 | ctx->flow.tp_dst = ia->tp_port.tp_port; | |
2d70a31a JP |
3112 | break; |
3113 | ||
064af421 BP |
3114 | case OFPAT_VENDOR: |
3115 | xlate_nicira_action(ctx, (const struct nx_action_header *) ia); | |
3116 | break; | |
3117 | ||
c1c9c9c4 BP |
3118 | case OFPAT_ENQUEUE: |
3119 | xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia); | |
3120 | break; | |
3121 | ||
064af421 | 3122 | default: |
e41a9130 | 3123 | VLOG_DBG_RL(&rl, "unknown action type %d", (int) type); |
064af421 BP |
3124 | break; |
3125 | } | |
3126 | } | |
3127 | } | |
3128 | ||
f29152ca BP |
3129 | static void |
3130 | action_xlate_ctx_init(struct action_xlate_ctx *ctx, | |
3131 | struct ofproto *ofproto, const struct flow *flow, | |
3132 | const struct ofpbuf *packet) | |
064af421 | 3133 | { |
f29152ca BP |
3134 | ctx->ofproto = ofproto; |
3135 | ctx->flow = *flow; | |
3136 | ctx->packet = packet; | |
7aa697dd | 3137 | ctx->resubmit_hook = NULL; |
f29152ca | 3138 | } |
1eb0942d | 3139 | |
cdee00fd | 3140 | static struct ofpbuf * |
f29152ca BP |
3141 | xlate_actions(struct action_xlate_ctx *ctx, |
3142 | const union ofp_action *in, size_t n_in) | |
3143 | { | |
064af421 | 3144 | COVERAGE_INC(ofproto_ofp2odp); |
cdee00fd BP |
3145 | |
3146 | ctx->odp_actions = ofpbuf_new(512); | |
f29152ca BP |
3147 | ctx->tags = 0; |
3148 | ctx->may_set_up_flow = true; | |
3149 | ctx->nf_output_iface = NF_OUT_DROP; | |
3150 | ctx->recurse = 0; | |
cdee00fd | 3151 | ctx->last_pop_priority = -1; |
f29152ca BP |
3152 | do_xlate_actions(in, n_in, ctx); |
3153 | remove_pop_action(ctx); | |
0ad9b732 | 3154 | |
d6fbec6d | 3155 | /* Check with in-band control to see if we're allowed to set up this |
0ad9b732 | 3156 | * flow. */ |
cdee00fd BP |
3157 | if (!in_band_rule_check(ctx->ofproto->in_band, &ctx->flow, |
3158 | ctx->odp_actions->data, ctx->odp_actions->size)) { | |
f29152ca | 3159 | ctx->may_set_up_flow = false; |
0ad9b732 JP |
3160 | } |
3161 | ||
cdee00fd | 3162 | return ctx->odp_actions; |
064af421 BP |
3163 | } |
3164 | ||
9deba63b BP |
3165 | /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow |
3166 | * error message code (composed with ofp_mkerr()) for the caller to propagate | |
3167 | * upward. Otherwise, returns 0. | |
3168 | * | |
2228b50d | 3169 | * The log message mentions 'msg_type'. */ |
9deba63b | 3170 | static int |
2228b50d | 3171 | reject_slave_controller(struct ofconn *ofconn, const const char *msg_type) |
9deba63b | 3172 | { |
5899143f | 3173 | if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) { |
9deba63b | 3174 | static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5); |
9deba63b | 3175 | VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller", |
2228b50d | 3176 | msg_type); |
9deba63b BP |
3177 | |
3178 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM); | |
3179 | } else { | |
3180 | return 0; | |
3181 | } | |
3182 | } | |
3183 | ||
064af421 | 3184 | static int |
d1e2cf21 | 3185 | handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh) |
064af421 | 3186 | { |
3269c562 | 3187 | struct ofproto *p = ofconn->ofproto; |
064af421 BP |
3188 | struct ofp_packet_out *opo; |
3189 | struct ofpbuf payload, *buffer; | |
ac51afaf | 3190 | union ofp_action *ofp_actions; |
f29152ca | 3191 | struct action_xlate_ctx ctx; |
cdee00fd | 3192 | struct ofpbuf *odp_actions; |
ac51afaf | 3193 | struct ofpbuf request; |
ae412e7d | 3194 | struct flow flow; |
ac51afaf | 3195 | size_t n_ofp_actions; |
064af421 | 3196 | uint16_t in_port; |
064af421 BP |
3197 | int error; |
3198 | ||
ac51afaf BP |
3199 | COVERAGE_INC(ofproto_packet_out); |
3200 | ||
2228b50d | 3201 | error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT"); |
9deba63b BP |
3202 | if (error) { |
3203 | return error; | |
3204 | } | |
3205 | ||
ac51afaf | 3206 | /* Get ofp_packet_out. */ |
0bc9407d | 3207 | ofpbuf_use_const(&request, oh, ntohs(oh->length)); |
bbc32a88 | 3208 | opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions)); |
ac51afaf BP |
3209 | |
3210 | /* Get actions. */ | |
3211 | error = ofputil_pull_actions(&request, ntohs(opo->actions_len), | |
3212 | &ofp_actions, &n_ofp_actions); | |
064af421 BP |
3213 | if (error) { |
3214 | return error; | |
3215 | } | |
064af421 | 3216 | |
ac51afaf | 3217 | /* Get payload. */ |
064af421 BP |
3218 | if (opo->buffer_id != htonl(UINT32_MAX)) { |
3219 | error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id), | |
3220 | &buffer, &in_port); | |
7778bd15 | 3221 | if (error || !buffer) { |
064af421 BP |
3222 | return error; |
3223 | } | |
3224 | payload = *buffer; | |
3225 | } else { | |
ac51afaf | 3226 | payload = request; |
064af421 BP |
3227 | buffer = NULL; |
3228 | } | |
3229 | ||
ac51afaf BP |
3230 | /* Extract flow, check actions. */ |
3231 | flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), | |
3232 | &flow); | |
f1defbf9 | 3233 | error = validate_actions(ofp_actions, n_ofp_actions, &flow, p->max_ports); |
ac51afaf BP |
3234 | if (error) { |
3235 | goto exit; | |
3236 | } | |
3237 | ||
3238 | /* Send. */ | |
f29152ca | 3239 | action_xlate_ctx_init(&ctx, p, &flow, &payload); |
cdee00fd BP |
3240 | odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions); |
3241 | dpif_execute(p->dpif, odp_actions->data, odp_actions->size, &payload); | |
3242 | ofpbuf_delete(odp_actions); | |
064af421 | 3243 | |
ac51afaf BP |
3244 | exit: |
3245 | ofpbuf_delete(buffer); | |
3246 | return 0; | |
064af421 BP |
3247 | } |
3248 | ||
3249 | static void | |
3250 | update_port_config(struct ofproto *p, struct ofport *port, | |
3251 | uint32_t config, uint32_t mask) | |
3252 | { | |
3253 | mask &= config ^ port->opp.config; | |
3254 | if (mask & OFPPC_PORT_DOWN) { | |
3255 | if (config & OFPPC_PORT_DOWN) { | |
3256 | netdev_turn_flags_off(port->netdev, NETDEV_UP, true); | |
3257 | } else { | |
3258 | netdev_turn_flags_on(port->netdev, NETDEV_UP, true); | |
3259 | } | |
3260 | } | |
f1588b1f BP |
3261 | #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \ |
3262 | OFPPC_NO_FWD | OFPPC_NO_FLOOD) | |
064af421 BP |
3263 | if (mask & REVALIDATE_BITS) { |
3264 | COVERAGE_INC(ofproto_costly_flags); | |
3265 | port->opp.config ^= mask & REVALIDATE_BITS; | |
3266 | p->need_revalidate = true; | |
3267 | } | |
3268 | #undef REVALIDATE_BITS | |
064af421 BP |
3269 | if (mask & OFPPC_NO_PACKET_IN) { |
3270 | port->opp.config ^= OFPPC_NO_PACKET_IN; | |
3271 | } | |
3272 | } | |
3273 | ||
3274 | static int | |
d1e2cf21 | 3275 | handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh) |
064af421 | 3276 | { |
3269c562 | 3277 | struct ofproto *p = ofconn->ofproto; |
d1e2cf21 | 3278 | const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh; |
064af421 BP |
3279 | struct ofport *port; |
3280 | int error; | |
3281 | ||
2228b50d | 3282 | error = reject_slave_controller(ofconn, "OFPT_PORT_MOD"); |
9deba63b BP |
3283 | if (error) { |
3284 | return error; | |
3285 | } | |
064af421 | 3286 | |
ca0f572c | 3287 | port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no))); |
064af421 BP |
3288 | if (!port) { |
3289 | return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT); | |
3290 | } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) { | |
3291 | return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR); | |
3292 | } else { | |
3293 | update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask)); | |
3294 | if (opm->advertise) { | |
3295 | netdev_set_advertisements(port->netdev, ntohl(opm->advertise)); | |
3296 | } | |
3297 | } | |
3298 | return 0; | |
3299 | } | |
3300 | ||
3301 | static struct ofpbuf * | |
06a5e131 | 3302 | make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len) |
064af421 BP |
3303 | { |
3304 | struct ofp_stats_reply *osr; | |
3305 | struct ofpbuf *msg; | |
3306 | ||
3307 | msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX)); | |
3308 | osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg); | |
3309 | osr->type = type; | |
3310 | osr->flags = htons(0); | |
3311 | return msg; | |
3312 | } | |
3313 | ||
3314 | static struct ofpbuf * | |
d1e2cf21 | 3315 | start_ofp_stats_reply(const struct ofp_header *request, size_t body_len) |
064af421 | 3316 | { |
d1e2cf21 BP |
3317 | const struct ofp_stats_request *osr |
3318 | = (const struct ofp_stats_request *) request; | |
3319 | return make_ofp_stats_reply(osr->header.xid, osr->type, body_len); | |
064af421 BP |
3320 | } |
3321 | ||
3322 | static void * | |
06a5e131 BP |
3323 | append_ofp_stats_reply(size_t nbytes, struct ofconn *ofconn, |
3324 | struct ofpbuf **msgp) | |
064af421 BP |
3325 | { |
3326 | struct ofpbuf *msg = *msgp; | |
3327 | assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply)); | |
3328 | if (nbytes + msg->size > UINT16_MAX) { | |
3329 | struct ofp_stats_reply *reply = msg->data; | |
3330 | reply->flags = htons(OFPSF_REPLY_MORE); | |
06a5e131 | 3331 | *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes); |
064af421 BP |
3332 | queue_tx(msg, ofconn, ofconn->reply_counter); |
3333 | } | |
3334 | return ofpbuf_put_uninit(*msgp, nbytes); | |
3335 | } | |
3336 | ||
09246b99 BP |
3337 | static struct ofpbuf * |
3338 | make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len) | |
3339 | { | |
3340 | struct nicira_stats_msg *nsm; | |
3341 | struct ofpbuf *msg; | |
3342 | ||
3343 | msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX)); | |
3344 | nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg); | |
3345 | nsm->type = htons(OFPST_VENDOR); | |
3346 | nsm->flags = htons(0); | |
3347 | nsm->vendor = htonl(NX_VENDOR_ID); | |
d5f2379b | 3348 | nsm->subtype = subtype; |
09246b99 BP |
3349 | return msg; |
3350 | } | |
3351 | ||
3352 | static struct ofpbuf * | |
3353 | start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len) | |
3354 | { | |
3355 | return make_nxstats_reply(request->header.xid, request->subtype, body_len); | |
3356 | } | |
3357 | ||
3358 | static void | |
3359 | append_nxstats_reply(size_t nbytes, struct ofconn *ofconn, | |
3360 | struct ofpbuf **msgp) | |
3361 | { | |
3362 | struct ofpbuf *msg = *msgp; | |
3363 | assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg)); | |
3364 | if (nbytes + msg->size > UINT16_MAX) { | |
3365 | struct nicira_stats_msg *reply = msg->data; | |
3366 | reply->flags = htons(OFPSF_REPLY_MORE); | |
3367 | *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes); | |
3368 | queue_tx(msg, ofconn, ofconn->reply_counter); | |
3369 | } | |
3370 | ofpbuf_prealloc_tailroom(*msgp, nbytes); | |
3371 | } | |
3372 | ||
064af421 | 3373 | static int |
3269c562 | 3374 | handle_desc_stats_request(struct ofconn *ofconn, |
d1e2cf21 | 3375 | const struct ofp_header *request) |
064af421 | 3376 | { |
3269c562 | 3377 | struct ofproto *p = ofconn->ofproto; |
064af421 BP |
3378 | struct ofp_desc_stats *ods; |
3379 | struct ofpbuf *msg; | |
3380 | ||
06a5e131 BP |
3381 | msg = start_ofp_stats_reply(request, sizeof *ods); |
3382 | ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg); | |
5a719c38 JP |
3383 | memset(ods, 0, sizeof *ods); |
3384 | ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc); | |
3385 | ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc); | |
3386 | ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc); | |
3387 | ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num); | |
3388 | ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc); | |
064af421 BP |
3389 | queue_tx(msg, ofconn, ofconn->reply_counter); |
3390 | ||
3391 | return 0; | |
3392 | } | |
3393 | ||
064af421 | 3394 | static int |
3269c562 | 3395 | handle_table_stats_request(struct ofconn *ofconn, |
d1e2cf21 | 3396 | const struct ofp_header *request) |
064af421 | 3397 | { |
3269c562 | 3398 | struct ofproto *p = ofconn->ofproto; |
064af421 BP |
3399 | struct ofp_table_stats *ots; |
3400 | struct ofpbuf *msg; | |
064af421 | 3401 | |
06a5e131 | 3402 | msg = start_ofp_stats_reply(request, sizeof *ots * 2); |
064af421 | 3403 | |
064af421 | 3404 | /* Classifier table. */ |
06a5e131 | 3405 | ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg); |
064af421 | 3406 | memset(ots, 0, sizeof *ots); |
064af421 | 3407 | strcpy(ots->name, "classifier"); |
b70eac89 | 3408 | ots->wildcards = (ofconn->flow_format == NXFF_OPENFLOW10 |
f9bfea14 | 3409 | ? htonl(OFPFW_ALL) : htonl(OVSFW_ALL)); |
ad828225 | 3410 | ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */ |
bcf84111 | 3411 | ots->active_count = htonl(classifier_count(&p->cls)); |
064af421 BP |
3412 | ots->lookup_count = htonll(0); /* XXX */ |
3413 | ots->matched_count = htonll(0); /* XXX */ | |
3414 | ||
3415 | queue_tx(msg, ofconn, ofconn->reply_counter); | |
3416 | return 0; | |
3417 | } | |
3418 | ||
abaad8cf | 3419 | static void |
ca0f572c | 3420 | append_port_stat(struct ofport *port, struct ofconn *ofconn, |
a4948b95 | 3421 | struct ofpbuf **msgp) |
abaad8cf JP |
3422 | { |
3423 | struct netdev_stats stats; | |
3424 | struct ofp_port_stats *ops; | |
3425 | ||
d295e8e9 JP |
3426 | /* Intentionally ignore return value, since errors will set |
3427 | * 'stats' to all-1s, which is correct for OpenFlow, and | |
abaad8cf JP |
3428 | * netdev_get_stats() will log errors. */ |
3429 | netdev_get_stats(port->netdev, &stats); | |
3430 | ||
06a5e131 | 3431 | ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp); |
ca0f572c | 3432 | ops->port_no = htons(port->opp.port_no); |
abaad8cf JP |
3433 | memset(ops->pad, 0, sizeof ops->pad); |
3434 | ops->rx_packets = htonll(stats.rx_packets); | |
3435 | ops->tx_packets = htonll(stats.tx_packets); | |
3436 | ops->rx_bytes = htonll(stats.rx_bytes); | |
3437 | ops->tx_bytes = htonll(stats.tx_bytes); | |
3438 | ops->rx_dropped = htonll(stats.rx_dropped); | |
3439 | ops->tx_dropped = htonll(stats.tx_dropped); | |
3440 | ops->rx_errors = htonll(stats.rx_errors); | |
3441 | ops->tx_errors = htonll(stats.tx_errors); | |
3442 | ops->rx_frame_err = htonll(stats.rx_frame_errors); | |
3443 | ops->rx_over_err = htonll(stats.rx_over_errors); | |
3444 | ops->rx_crc_err = htonll(stats.rx_crc_errors); | |
3445 | ops->collisions = htonll(stats.collisions); | |
3446 | } | |
3447 | ||
064af421 | 3448 | static int |
d1e2cf21 | 3449 | handle_port_stats_request(struct ofconn *ofconn, const struct ofp_header *oh) |
064af421 | 3450 | { |
3269c562 | 3451 | struct ofproto *p = ofconn->ofproto; |
d1e2cf21 | 3452 | const struct ofp_port_stats_request *psr = ofputil_stats_body(oh); |
064af421 BP |
3453 | struct ofp_port_stats *ops; |
3454 | struct ofpbuf *msg; | |
3455 | struct ofport *port; | |
064af421 | 3456 | |
d1e2cf21 | 3457 | msg = start_ofp_stats_reply(oh, sizeof *ops * 16); |
abaad8cf | 3458 | if (psr->port_no != htons(OFPP_NONE)) { |
ca0f572c | 3459 | port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no))); |
abaad8cf | 3460 | if (port) { |
ca0f572c | 3461 | append_port_stat(port, ofconn, &msg); |
abaad8cf JP |
3462 | } |
3463 | } else { | |
4e8e4213 | 3464 | HMAP_FOR_EACH (port, hmap_node, &p->ports) { |
ca0f572c | 3465 | append_port_stat(port, ofconn, &msg); |
abaad8cf | 3466 | } |
064af421 BP |
3467 | } |
3468 | ||
3469 | queue_tx(msg, ofconn, ofconn->reply_counter); | |
3470 | return 0; | |
3471 | } | |
3472 | ||
01149cfd | 3473 | /* Obtains statistic counters for 'rule' within 'p' and stores them into |
bcf84111 BP |
3474 | * '*packet_countp' and '*byte_countp'. The returned statistics include |
3475 | * statistics for all of 'rule''s facets. */ | |
064af421 BP |
3476 | static void |
3477 | query_stats(struct ofproto *p, struct rule *rule, | |
3478 | uint64_t *packet_countp, uint64_t *byte_countp) | |
3479 | { | |
36956a7d | 3480 | uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S]; |
064af421 | 3481 | uint64_t packet_count, byte_count; |
bcf84111 | 3482 | struct facet *facet; |
36956a7d | 3483 | struct ofpbuf key; |
064af421 | 3484 | |
01149cfd | 3485 | /* Start from historical data for 'rule' itself that are no longer tracked |
bcf84111 | 3486 | * by the datapath. This counts, for example, facets that have expired. */ |
b3137fe8 JG |
3487 | packet_count = rule->packet_count; |
3488 | byte_count = rule->byte_count; | |
3489 | ||
36956a7d BP |
3490 | /* Ask the datapath for statistics on all of the rule's facets. (We could |
3491 | * batch up statistics requests using dpif_flow_get_multiple(), but that is | |
3492 | * not yet implemented.) | |
01149cfd BP |
3493 | * |
3494 | * Also, add any statistics that are not tracked by the datapath for each | |
bcf84111 | 3495 | * facet. This includes, for example, statistics for packets that were |
01149cfd | 3496 | * executed "by hand" by ofproto via dpif_execute() but must be accounted |
bcf84111 | 3497 | * to a rule. */ |
36956a7d | 3498 | ofpbuf_use_stack(&key, keybuf, sizeof keybuf); |
bcf84111 | 3499 | LIST_FOR_EACH (facet, list_node, &rule->facets) { |
36956a7d | 3500 | struct odp_flow odp_flow; |
064af421 | 3501 | |
36956a7d BP |
3502 | ofpbuf_clear(&key); |
3503 | odp_flow_key_from_flow(&key, &facet->flow); | |
bcf84111 | 3504 | |
36956a7d BP |
3505 | odp_flow.key = key.data; |
3506 | odp_flow.key_len = key.size; | |
3507 | odp_flow.actions = NULL; | |
3508 | odp_flow.actions_len = 0; | |
3509 | odp_flow.flags = 0; | |
3510 | if (!dpif_flow_get(p->dpif, &odp_flow)) { | |
3511 | packet_count += odp_flow.stats.n_packets; | |
3512 | byte_count += odp_flow.stats.n_bytes; | |
064af421 | 3513 | } |
36956a7d BP |
3514 | |
3515 | packet_count += facet->packet_count; | |
3516 | byte_count += facet->byte_count; | |
064af421 | 3517 | } |
064af421 | 3518 | |
01149cfd | 3519 | /* Return the stats to the caller. */ |
064af421 BP |
3520 | *packet_countp = packet_count; |
3521 | *byte_countp = byte_count; | |
3522 | } | |
3523 | ||
c6ebb8fb BP |
3524 | static void |
3525 | calc_flow_duration(long long int start, ovs_be32 *sec, ovs_be32 *nsec) | |
3526 | { | |
3527 | long long int msecs = time_msec() - start; | |
3528 | *sec = htonl(msecs / 1000); | |
3529 | *nsec = htonl((msecs % 1000) * (1000 * 1000)); | |
3530 | } | |
3531 | ||
064af421 | 3532 | static void |
5ecc9d81 BP |
3533 | put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule, |
3534 | ovs_be16 out_port, struct ofpbuf **replyp) | |
064af421 | 3535 | { |
064af421 BP |
3536 | struct ofp_flow_stats *ofs; |
3537 | uint64_t packet_count, byte_count; | |
3538 | size_t act_len, len; | |
3539 | ||
5ecc9d81 | 3540 | if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) { |
064af421 BP |
3541 | return; |
3542 | } | |
3543 | ||
3544 | act_len = sizeof *rule->actions * rule->n_actions; | |
3545 | len = offsetof(struct ofp_flow_stats, actions) + act_len; | |
3546 | ||
5ecc9d81 | 3547 | query_stats(ofconn->ofproto, rule, &packet_count, &byte_count); |
064af421 | 3548 | |
5ecc9d81 | 3549 | ofs = append_ofp_stats_reply(len, ofconn, replyp); |
064af421 | 3550 | ofs->length = htons(len); |
ad828225 | 3551 | ofs->table_id = 0; |
064af421 | 3552 | ofs->pad = 0; |
ff9d3826 BP |
3553 | ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofs->match, |
3554 | rule->flow_cookie, &ofs->cookie); | |
c6ebb8fb | 3555 | calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec); |
064af421 BP |
3556 | ofs->priority = htons(rule->cr.priority); |
3557 | ofs->idle_timeout = htons(rule->idle_timeout); | |
3558 | ofs->hard_timeout = htons(rule->hard_timeout); | |
39997502 | 3559 | memset(ofs->pad2, 0, sizeof ofs->pad2); |
064af421 BP |
3560 | ofs->packet_count = htonll(packet_count); |
3561 | ofs->byte_count = htonll(byte_count); | |
3dffcf07 BP |
3562 | if (rule->n_actions > 0) { |
3563 | memcpy(ofs->actions, rule->actions, act_len); | |
3564 | } | |
064af421 BP |
3565 | } |
3566 | ||
3c4486a5 BP |
3567 | static bool |
3568 | is_valid_table(uint8_t table_id) | |
064af421 | 3569 | { |
3c4486a5 | 3570 | return table_id == 0 || table_id == 0xff; |
064af421 BP |
3571 | } |
3572 | ||
3573 | static int | |
d1e2cf21 | 3574 | handle_flow_stats_request(struct ofconn *ofconn, const struct ofp_header *oh) |
064af421 | 3575 | { |
d1e2cf21 | 3576 | const struct ofp_flow_stats_request *fsr = ofputil_stats_body(oh); |
5ecc9d81 | 3577 | struct ofpbuf *reply; |
064af421 | 3578 | |
064af421 | 3579 | COVERAGE_INC(ofproto_flows_req); |
d1e2cf21 | 3580 | reply = start_ofp_stats_reply(oh, 1024); |
3c4486a5 | 3581 | if (is_valid_table(fsr->table_id)) { |
5ecc9d81 | 3582 | struct cls_cursor cursor; |
3c4486a5 | 3583 | struct cls_rule target; |
5ecc9d81 | 3584 | struct rule *rule; |
3c4486a5 | 3585 | |
d8ae4d67 BP |
3586 | ofputil_cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0, |
3587 | &target); | |
5ecc9d81 BP |
3588 | cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target); |
3589 | CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { | |
3590 | put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply); | |
3591 | } | |
3c4486a5 | 3592 | } |
5ecc9d81 | 3593 | queue_tx(reply, ofconn, ofconn->reply_counter); |
3c4486a5 | 3594 | |
064af421 BP |
3595 | return 0; |
3596 | } | |
3597 | ||
09246b99 | 3598 | static void |
5ecc9d81 BP |
3599 | put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule, |
3600 | ovs_be16 out_port, struct ofpbuf **replyp) | |
09246b99 | 3601 | { |
09246b99 BP |
3602 | struct nx_flow_stats *nfs; |
3603 | uint64_t packet_count, byte_count; | |
3604 | size_t act_len, start_len; | |
5ecc9d81 | 3605 | struct ofpbuf *reply; |
09246b99 | 3606 | |
5ecc9d81 | 3607 | if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) { |
09246b99 BP |
3608 | return; |
3609 | } | |
3610 | ||
5ecc9d81 | 3611 | query_stats(ofconn->ofproto, rule, &packet_count, &byte_count); |
09246b99 BP |
3612 | |
3613 | act_len = sizeof *rule->actions * rule->n_actions; | |
3614 | ||
5ecc9d81 | 3615 | append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, ofconn, replyp); |
1dfee98d | 3616 | start_len = (*replyp)->size; |
5ecc9d81 BP |
3617 | reply = *replyp; |
3618 | ||
3619 | nfs = ofpbuf_put_uninit(reply, sizeof *nfs); | |
09246b99 BP |
3620 | nfs->table_id = 0; |
3621 | nfs->pad = 0; | |
3622 | calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec); | |
3623 | nfs->cookie = rule->flow_cookie; | |
3624 | nfs->priority = htons(rule->cr.priority); | |
3625 | nfs->idle_timeout = htons(rule->idle_timeout); | |
3626 | nfs->hard_timeout = htons(rule->hard_timeout); | |
5ecc9d81 | 3627 | nfs->match_len = htons(nx_put_match(reply, &rule->cr)); |
09246b99 BP |
3628 | memset(nfs->pad2, 0, sizeof nfs->pad2); |
3629 | nfs->packet_count = htonll(packet_count); | |
3630 | nfs->byte_count = htonll(byte_count); | |
3631 | if (rule->n_actions > 0) { | |
5ecc9d81 | 3632 | ofpbuf_put(reply, rule->actions, act_len); |
09246b99 | 3633 | } |
5ecc9d81 | 3634 | nfs->length = htons(reply->size - start_len); |
09246b99 BP |
3635 | } |
3636 | ||
3637 | static int | |
d1e2cf21 | 3638 | handle_nxst_flow(struct ofconn *ofconn, const struct ofp_header *oh) |
09246b99 BP |
3639 | { |
3640 | struct nx_flow_stats_request *nfsr; | |
09246b99 | 3641 | struct cls_rule target; |
5ecc9d81 | 3642 | struct ofpbuf *reply; |
d1e2cf21 | 3643 | struct ofpbuf b; |
09246b99 BP |
3644 | int error; |
3645 | ||
0bc9407d | 3646 | ofpbuf_use_const(&b, oh, ntohs(oh->length)); |
d1e2cf21 | 3647 | |
09246b99 | 3648 | /* Dissect the message. */ |
bbc32a88 | 3649 | nfsr = ofpbuf_pull(&b, sizeof *nfsr); |
d1e2cf21 | 3650 | error = nx_pull_match(&b, ntohs(nfsr->match_len), 0, &target); |
09246b99 BP |
3651 | if (error) { |
3652 | return error; | |
3653 | } | |
d1e2cf21 BP |
3654 | if (b.size) { |
3655 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
3656 | } | |
09246b99 BP |
3657 | |
3658 | COVERAGE_INC(ofproto_flows_req); | |
5ecc9d81 | 3659 | reply = start_nxstats_reply(&nfsr->nsm, 1024); |
3c4486a5 | 3660 | if (is_valid_table(nfsr->table_id)) { |
5ecc9d81 BP |
3661 | struct cls_cursor cursor; |
3662 | struct rule *rule; | |
3663 | ||
3664 | cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target); | |
3665 | CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { | |
3666 | put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply); | |
3667 | } | |
3c4486a5 | 3668 | } |
5ecc9d81 BP |
3669 | queue_tx(reply, ofconn, ofconn->reply_counter); |
3670 | ||
09246b99 BP |
3671 | return 0; |
3672 | } | |
3673 | ||
4f2cad2c | 3674 | static void |
5ecc9d81 | 3675 | flow_stats_ds(struct ofproto *ofproto, struct rule *rule, struct ds *results) |
4f2cad2c | 3676 | { |
4f2cad2c JP |
3677 | uint64_t packet_count, byte_count; |
3678 | size_t act_len = sizeof *rule->actions * rule->n_actions; | |
3679 | ||
5ecc9d81 | 3680 | query_stats(ofproto, rule, &packet_count, &byte_count); |
4f2cad2c JP |
3681 | |
3682 | ds_put_format(results, "duration=%llds, ", | |
3683 | (time_msec() - rule->created) / 1000); | |
52ae00b3 | 3684 | ds_put_format(results, "priority=%u, ", rule->cr.priority); |
4f2cad2c JP |
3685 | ds_put_format(results, "n_packets=%"PRIu64", ", packet_count); |
3686 | ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count); | |
cb833cf6 | 3687 | cls_rule_format(&rule->cr, results); |
3dffcf07 BP |
3688 | if (act_len > 0) { |
3689 | ofp_print_actions(results, &rule->actions->header, act_len); | |
3c8552c1 JP |
3690 | } else { |
3691 | ds_put_cstr(results, "drop"); | |
3dffcf07 | 3692 | } |
4f2cad2c JP |
3693 | ds_put_cstr(results, "\n"); |
3694 | } | |
3695 | ||
d295e8e9 | 3696 | /* Adds a pretty-printed description of all flows to 'results', including |
4f2cad2c JP |
3697 | * those marked hidden by secchan (e.g., by in-band control). */ |
3698 | void | |
3699 | ofproto_get_all_flows(struct ofproto *p, struct ds *results) | |
3700 | { | |
5ecc9d81 BP |
3701 | struct cls_cursor cursor; |
3702 | struct rule *rule; | |
064af421 | 3703 | |
5ecc9d81 BP |
3704 | cls_cursor_init(&cursor, &p->cls, NULL); |
3705 | CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { | |
3706 | flow_stats_ds(p, rule, results); | |
064af421 | 3707 | } |
064af421 BP |
3708 | } |
3709 | ||
27d34fce BP |
3710 | static void |
3711 | query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target, | |
734bbeb4 | 3712 | ovs_be16 out_port, uint8_t table_id, |
27d34fce BP |
3713 | struct ofp_aggregate_stats_reply *oasr) |
3714 | { | |
5ecc9d81 BP |
3715 | uint64_t total_packets = 0; |
3716 | uint64_t total_bytes = 0; | |
3717 | int n_flows = 0; | |
27d34fce BP |
3718 | |
3719 | COVERAGE_INC(ofproto_agg_request); | |
5ecc9d81 | 3720 | |
3c4486a5 | 3721 | if (is_valid_table(table_id)) { |
5ecc9d81 BP |
3722 | struct cls_cursor cursor; |
3723 | struct rule *rule; | |
3c4486a5 | 3724 | |
5ecc9d81 BP |
3725 | cls_cursor_init(&cursor, &ofproto->cls, target); |
3726 | CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { | |
3727 | if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) { | |
3728 | uint64_t packet_count; | |
3729 | uint64_t byte_count; | |
3730 | ||
3731 | query_stats(ofproto, rule, &packet_count, &byte_count); | |
3732 | ||
3733 | total_packets += packet_count; | |
3734 | total_bytes += byte_count; | |
3735 | n_flows++; | |
3736 | } | |
3737 | } | |
3c4486a5 | 3738 | } |
27d34fce | 3739 | |
5ecc9d81 BP |
3740 | oasr->flow_count = htonl(n_flows); |
3741 | oasr->packet_count = htonll(total_packets); | |
3742 | oasr->byte_count = htonll(total_bytes); | |
27d34fce BP |
3743 | memset(oasr->pad, 0, sizeof oasr->pad); |
3744 | } | |
3745 | ||
064af421 | 3746 | static int |
3269c562 | 3747 | handle_aggregate_stats_request(struct ofconn *ofconn, |
d1e2cf21 | 3748 | const struct ofp_header *oh) |
064af421 | 3749 | { |
d1e2cf21 | 3750 | const struct ofp_aggregate_stats_request *request = ofputil_stats_body(oh); |
064af421 | 3751 | struct ofp_aggregate_stats_reply *reply; |
064af421 BP |
3752 | struct cls_rule target; |
3753 | struct ofpbuf *msg; | |
3754 | ||
d8ae4d67 BP |
3755 | ofputil_cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0, |
3756 | &target); | |
064af421 | 3757 | |
d1e2cf21 | 3758 | msg = start_ofp_stats_reply(oh, sizeof *reply); |
06a5e131 | 3759 | reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg); |
27d34fce BP |
3760 | query_aggregate_stats(ofconn->ofproto, &target, request->out_port, |
3761 | request->table_id, reply); | |
064af421 BP |
3762 | queue_tx(msg, ofconn, ofconn->reply_counter); |
3763 | return 0; | |
3764 | } | |
3765 | ||
09246b99 | 3766 | static int |
d1e2cf21 | 3767 | handle_nxst_aggregate(struct ofconn *ofconn, const struct ofp_header *oh) |
09246b99 BP |
3768 | { |
3769 | struct nx_aggregate_stats_request *request; | |
3770 | struct ofp_aggregate_stats_reply *reply; | |
3771 | struct cls_rule target; | |
d1e2cf21 | 3772 | struct ofpbuf b; |
09246b99 BP |
3773 | struct ofpbuf *buf; |
3774 | int error; | |
3775 | ||
0bc9407d | 3776 | ofpbuf_use_const(&b, oh, ntohs(oh->length)); |
d1e2cf21 | 3777 | |
09246b99 | 3778 | /* Dissect the message. */ |
bbc32a88 | 3779 | request = ofpbuf_pull(&b, sizeof *request); |
d1e2cf21 | 3780 | error = nx_pull_match(&b, ntohs(request->match_len), 0, &target); |
09246b99 BP |
3781 | if (error) { |
3782 | return error; | |
3783 | } | |
d1e2cf21 BP |
3784 | if (b.size) { |
3785 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
3786 | } | |
09246b99 BP |
3787 | |
3788 | /* Reply. */ | |
3789 | COVERAGE_INC(ofproto_flows_req); | |
3790 | buf = start_nxstats_reply(&request->nsm, sizeof *reply); | |
3791 | reply = ofpbuf_put_uninit(buf, sizeof *reply); | |
3792 | query_aggregate_stats(ofconn->ofproto, &target, request->out_port, | |
3793 | request->table_id, reply); | |
3794 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
3795 | ||
3796 | return 0; | |
3797 | } | |
3798 | ||
c1c9c9c4 BP |
3799 | struct queue_stats_cbdata { |
3800 | struct ofconn *ofconn; | |
ca0f572c | 3801 | struct ofport *ofport; |
c1c9c9c4 | 3802 | struct ofpbuf *msg; |
c1c9c9c4 BP |
3803 | }; |
3804 | ||
3805 | static void | |
db9220c3 | 3806 | put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id, |
c1c9c9c4 BP |
3807 | const struct netdev_queue_stats *stats) |
3808 | { | |
3809 | struct ofp_queue_stats *reply; | |
3810 | ||
06a5e131 | 3811 | reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg); |
ca0f572c | 3812 | reply->port_no = htons(cbdata->ofport->opp.port_no); |
c1c9c9c4 BP |
3813 | memset(reply->pad, 0, sizeof reply->pad); |
3814 | reply->queue_id = htonl(queue_id); | |
3815 | reply->tx_bytes = htonll(stats->tx_bytes); | |
3816 | reply->tx_packets = htonll(stats->tx_packets); | |
3817 | reply->tx_errors = htonll(stats->tx_errors); | |
3818 | } | |
3819 | ||
3820 | static void | |
db9220c3 | 3821 | handle_queue_stats_dump_cb(uint32_t queue_id, |
c1c9c9c4 BP |
3822 | struct netdev_queue_stats *stats, |
3823 | void *cbdata_) | |
3824 | { | |
3825 | struct queue_stats_cbdata *cbdata = cbdata_; | |
3826 | ||
3827 | put_queue_stats(cbdata, queue_id, stats); | |
3828 | } | |
3829 | ||
3830 | static void | |
ca0f572c | 3831 | handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id, |
c1c9c9c4 BP |
3832 | struct queue_stats_cbdata *cbdata) |
3833 | { | |
ca0f572c | 3834 | cbdata->ofport = port; |
c1c9c9c4 BP |
3835 | if (queue_id == OFPQ_ALL) { |
3836 | netdev_dump_queue_stats(port->netdev, | |
3837 | handle_queue_stats_dump_cb, cbdata); | |
3838 | } else { | |
3839 | struct netdev_queue_stats stats; | |
3840 | ||
1ac788f6 BP |
3841 | if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) { |
3842 | put_queue_stats(cbdata, queue_id, &stats); | |
3843 | } | |
c1c9c9c4 BP |
3844 | } |
3845 | } | |
3846 | ||
3847 | static int | |
d1e2cf21 | 3848 | handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh) |
c1c9c9c4 | 3849 | { |
3269c562 | 3850 | struct ofproto *ofproto = ofconn->ofproto; |
d1e2cf21 | 3851 | const struct ofp_queue_stats_request *qsr; |
c1c9c9c4 BP |
3852 | struct queue_stats_cbdata cbdata; |
3853 | struct ofport *port; | |
3854 | unsigned int port_no; | |
3855 | uint32_t queue_id; | |
3856 | ||
d1e2cf21 BP |
3857 | qsr = ofputil_stats_body(oh); |
3858 | if (!qsr) { | |
c1c9c9c4 BP |
3859 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); |
3860 | } | |
c1c9c9c4 BP |
3861 | |
3862 | COVERAGE_INC(ofproto_queue_req); | |
3863 | ||
3864 | cbdata.ofconn = ofconn; | |
d1e2cf21 | 3865 | cbdata.msg = start_ofp_stats_reply(oh, 128); |
c1c9c9c4 BP |
3866 | |
3867 | port_no = ntohs(qsr->port_no); | |
3868 | queue_id = ntohl(qsr->queue_id); | |
3869 | if (port_no == OFPP_ALL) { | |
4e8e4213 | 3870 | HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) { |
ca0f572c | 3871 | handle_queue_stats_for_port(port, queue_id, &cbdata); |
c1c9c9c4 BP |
3872 | } |
3873 | } else if (port_no < ofproto->max_ports) { | |
ca0f572c | 3874 | port = get_port(ofproto, ofp_port_to_odp_port(port_no)); |
c1c9c9c4 | 3875 | if (port) { |
ca0f572c | 3876 | handle_queue_stats_for_port(port, queue_id, &cbdata); |
c1c9c9c4 BP |
3877 | } |
3878 | } else { | |
3879 | ofpbuf_delete(cbdata.msg); | |
3880 | return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT); | |
3881 | } | |
3882 | queue_tx(cbdata.msg, ofconn, ofconn->reply_counter); | |
3883 | ||
3884 | return 0; | |
3885 | } | |
3886 | ||
064af421 BP |
3887 | static long long int |
3888 | msec_from_nsec(uint64_t sec, uint32_t nsec) | |
3889 | { | |
3890 | return !sec ? 0 : sec * 1000 + nsec / 1000000; | |
3891 | } | |
3892 | ||
3893 | static void | |
bcf84111 BP |
3894 | facet_update_time(struct ofproto *ofproto, struct facet *facet, |
3895 | const struct odp_flow_stats *stats) | |
064af421 BP |
3896 | { |
3897 | long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec); | |
bcf84111 BP |
3898 | if (used > facet->used) { |
3899 | facet->used = used; | |
3900 | if (used > facet->rule->used) { | |
3901 | facet->rule->used = used; | |
4836f9f2 | 3902 | } |
bcf84111 | 3903 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used); |
064af421 BP |
3904 | } |
3905 | } | |
3906 | ||
bcf84111 BP |
3907 | /* Folds the statistics from 'stats' into the counters in 'facet'. |
3908 | * | |
3909 | * Because of the meaning of a facet's counters, it only makes sense to do this | |
3910 | * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a | |
3911 | * packet that was sent by hand or if it represents statistics that have been | |
3912 | * cleared out of the datapath. */ | |
064af421 | 3913 | static void |
bcf84111 BP |
3914 | facet_update_stats(struct ofproto *ofproto, struct facet *facet, |
3915 | const struct odp_flow_stats *stats) | |
064af421 | 3916 | { |
064af421 | 3917 | if (stats->n_packets) { |
bcf84111 BP |
3918 | facet_update_time(ofproto, facet, stats); |
3919 | facet->packet_count += stats->n_packets; | |
3920 | facet->byte_count += stats->n_bytes; | |
3921 | netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags); | |
064af421 BP |
3922 | } |
3923 | } | |
3924 | ||
79eee1eb BP |
3925 | /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT |
3926 | * in which no matching flow already exists in the flow table. | |
3927 | * | |
3928 | * Adds the flow specified by 'ofm', which is followed by 'n_actions' | |
3269c562 BP |
3929 | * ofp_actions, to ofconn->ofproto's flow table. Returns 0 on success or an |
3930 | * OpenFlow error code as encoded by ofp_mkerr() on failure. | |
79eee1eb BP |
3931 | * |
3932 | * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, | |
3933 | * if any. */ | |
064af421 | 3934 | static int |
3052b0c5 | 3935 | add_flow(struct ofconn *ofconn, struct flow_mod *fm) |
064af421 | 3936 | { |
3269c562 | 3937 | struct ofproto *p = ofconn->ofproto; |
064af421 BP |
3938 | struct ofpbuf *packet; |
3939 | struct rule *rule; | |
3940 | uint16_t in_port; | |
3941 | int error; | |
3942 | ||
3052b0c5 BP |
3943 | if (fm->flags & OFPFF_CHECK_OVERLAP |
3944 | && classifier_rule_overlaps(&p->cls, &fm->cr)) { | |
3945 | return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP); | |
49bdc010 JP |
3946 | } |
3947 | ||
064af421 | 3948 | error = 0; |
3052b0c5 BP |
3949 | if (fm->buffer_id != UINT32_MAX) { |
3950 | error = pktbuf_retrieve(ofconn->pktbuf, fm->buffer_id, | |
064af421 | 3951 | &packet, &in_port); |
212fe71c BP |
3952 | } else { |
3953 | packet = NULL; | |
165cd8a3 | 3954 | in_port = UINT16_MAX; |
064af421 BP |
3955 | } |
3956 | ||
bcf84111 BP |
3957 | rule = rule_create(&fm->cr, fm->actions, fm->n_actions, |
3958 | fm->idle_timeout, fm->hard_timeout, fm->cookie, | |
3959 | fm->flags & OFPFF_SEND_FLOW_REM); | |
afe75089 BP |
3960 | rule_insert(p, rule); |
3961 | if (packet) { | |
3962 | rule_execute(p, rule, in_port, packet); | |
3963 | } | |
064af421 BP |
3964 | return error; |
3965 | } | |
3966 | ||
79eee1eb | 3967 | static struct rule * |
3052b0c5 | 3968 | find_flow_strict(struct ofproto *p, const struct flow_mod *fm) |
064af421 | 3969 | { |
3052b0c5 | 3970 | return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr)); |
79eee1eb | 3971 | } |
064af421 | 3972 | |
79eee1eb | 3973 | static int |
3269c562 | 3974 | send_buffered_packet(struct ofconn *ofconn, |
3052b0c5 | 3975 | struct rule *rule, uint32_t buffer_id) |
79eee1eb BP |
3976 | { |
3977 | struct ofpbuf *packet; | |
3978 | uint16_t in_port; | |
79eee1eb | 3979 | int error; |
064af421 | 3980 | |
3052b0c5 | 3981 | if (buffer_id == UINT32_MAX) { |
79eee1eb | 3982 | return 0; |
064af421 | 3983 | } |
79eee1eb | 3984 | |
3052b0c5 | 3985 | error = pktbuf_retrieve(ofconn->pktbuf, buffer_id, &packet, &in_port); |
79eee1eb BP |
3986 | if (error) { |
3987 | return error; | |
3988 | } | |
3989 | ||
bcf84111 | 3990 | rule_execute(ofconn->ofproto, rule, in_port, packet); |
79eee1eb | 3991 | |
064af421 BP |
3992 | return 0; |
3993 | } | |
79eee1eb BP |
3994 | \f |
3995 | /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */ | |
064af421 BP |
3996 | |
3997 | struct modify_flows_cbdata { | |
3998 | struct ofproto *ofproto; | |
3052b0c5 | 3999 | const struct flow_mod *fm; |
79eee1eb | 4000 | struct rule *match; |
064af421 BP |
4001 | }; |
4002 | ||
3052b0c5 BP |
4003 | static int modify_flow(struct ofproto *, const struct flow_mod *, |
4004 | struct rule *); | |
79eee1eb BP |
4005 | |
4006 | /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as | |
4007 | * encoded by ofp_mkerr() on failure. | |
4008 | * | |
4009 | * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, | |
4010 | * if any. */ | |
4011 | static int | |
3052b0c5 | 4012 | modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm) |
79eee1eb | 4013 | { |
5ecc9d81 BP |
4014 | struct ofproto *p = ofconn->ofproto; |
4015 | struct rule *match = NULL; | |
4016 | struct cls_cursor cursor; | |
4017 | struct rule *rule; | |
79eee1eb | 4018 | |
5ecc9d81 BP |
4019 | cls_cursor_init(&cursor, &p->cls, &fm->cr); |
4020 | CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { | |
4021 | if (!rule_is_hidden(rule)) { | |
4022 | match = rule; | |
4023 | modify_flow(p, fm, rule); | |
4024 | } | |
4025 | } | |
79eee1eb | 4026 | |
5ecc9d81 | 4027 | if (match) { |
d6302b0f BP |
4028 | /* This credits the packet to whichever flow happened to match last. |
4029 | * That's weird. Maybe we should do a lookup for the flow that | |
4030 | * actually matches the packet? Who knows. */ | |
5ecc9d81 | 4031 | send_buffered_packet(ofconn, match, fm->buffer_id); |
79eee1eb BP |
4032 | return 0; |
4033 | } else { | |
3052b0c5 | 4034 | return add_flow(ofconn, fm); |
79eee1eb BP |
4035 | } |
4036 | } | |
4037 | ||
4038 | /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error | |
4039 | * code as encoded by ofp_mkerr() on failure. | |
4040 | * | |
4041 | * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, | |
4042 | * if any. */ | |
4043 | static int | |
3052b0c5 | 4044 | modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm) |
79eee1eb | 4045 | { |
3052b0c5 BP |
4046 | struct ofproto *p = ofconn->ofproto; |
4047 | struct rule *rule = find_flow_strict(p, fm); | |
79eee1eb | 4048 | if (rule && !rule_is_hidden(rule)) { |
3052b0c5 BP |
4049 | modify_flow(p, fm, rule); |
4050 | return send_buffered_packet(ofconn, rule, fm->buffer_id); | |
79eee1eb | 4051 | } else { |
3052b0c5 | 4052 | return add_flow(ofconn, fm); |
79eee1eb BP |
4053 | } |
4054 | } | |
4055 | ||
79eee1eb BP |
4056 | /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has |
4057 | * been identified as a flow in 'p''s flow table to be modified, by changing | |
4058 | * the rule's actions to match those in 'ofm' (which is followed by 'n_actions' | |
4059 | * ofp_action[] structures). */ | |
064af421 | 4060 | static int |
3052b0c5 | 4061 | modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule) |
064af421 | 4062 | { |
3052b0c5 | 4063 | size_t actions_len = fm->n_actions * sizeof *rule->actions; |
79eee1eb | 4064 | |
3052b0c5 | 4065 | rule->flow_cookie = fm->cookie; |
79eee1eb BP |
4066 | |
4067 | /* If the actions are the same, do nothing. */ | |
3052b0c5 BP |
4068 | if (fm->n_actions == rule->n_actions |
4069 | && (!fm->n_actions | |
4070 | || !memcmp(fm->actions, rule->actions, actions_len))) { | |
79eee1eb BP |
4071 | return 0; |
4072 | } | |
4073 | ||
4074 | /* Replace actions. */ | |
4075 | free(rule->actions); | |
3052b0c5 BP |
4076 | rule->actions = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL; |
4077 | rule->n_actions = fm->n_actions; | |
79eee1eb | 4078 | |
bcf84111 | 4079 | p->need_revalidate = true; |
79eee1eb BP |
4080 | |
4081 | return 0; | |
4082 | } | |
4083 | \f | |
4084 | /* OFPFC_DELETE implementation. */ | |
4085 | ||
8054fc48 | 4086 | static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port); |
79eee1eb BP |
4087 | |
4088 | /* Implements OFPFC_DELETE. */ | |
4089 | static void | |
3052b0c5 | 4090 | delete_flows_loose(struct ofproto *p, const struct flow_mod *fm) |
79eee1eb | 4091 | { |
5ecc9d81 BP |
4092 | struct rule *rule, *next_rule; |
4093 | struct cls_cursor cursor; | |
064af421 | 4094 | |
5ecc9d81 BP |
4095 | cls_cursor_init(&cursor, &p->cls, &fm->cr); |
4096 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) { | |
4097 | delete_flow(p, rule, htons(fm->out_port)); | |
4098 | } | |
064af421 BP |
4099 | } |
4100 | ||
79eee1eb BP |
4101 | /* Implements OFPFC_DELETE_STRICT. */ |
4102 | static void | |
3052b0c5 | 4103 | delete_flow_strict(struct ofproto *p, struct flow_mod *fm) |
79eee1eb | 4104 | { |
3052b0c5 | 4105 | struct rule *rule = find_flow_strict(p, fm); |
79eee1eb | 4106 | if (rule) { |
3052b0c5 | 4107 | delete_flow(p, rule, htons(fm->out_port)); |
79eee1eb BP |
4108 | } |
4109 | } | |
4110 | ||
79eee1eb BP |
4111 | /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has |
4112 | * been identified as a flow to delete from 'p''s flow table, by deleting the | |
4113 | * flow and sending out a OFPT_FLOW_REMOVED message to any interested | |
4114 | * controller. | |
4115 | * | |
4116 | * Will not delete 'rule' if it is hidden. Will delete 'rule' only if | |
4117 | * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the | |
4118 | * specified 'out_port'. */ | |
4119 | static void | |
8054fc48 | 4120 | delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port) |
79eee1eb BP |
4121 | { |
4122 | if (rule_is_hidden(rule)) { | |
4123 | return; | |
4124 | } | |
4125 | ||
4126 | if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) { | |
4127 | return; | |
4128 | } | |
4129 | ||
bcf84111 | 4130 | rule_send_removed(p, rule, OFPRR_DELETE); |
79eee1eb BP |
4131 | rule_remove(p, rule); |
4132 | } | |
4133 | \f | |
064af421 | 4134 | static int |
2e4f5fcf | 4135 | handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh) |
064af421 | 4136 | { |
3052b0c5 | 4137 | struct ofproto *p = ofconn->ofproto; |
2e4f5fcf | 4138 | struct flow_mod fm; |
064af421 BP |
4139 | int error; |
4140 | ||
3052b0c5 | 4141 | error = reject_slave_controller(ofconn, "flow_mod"); |
9deba63b BP |
4142 | if (error) { |
4143 | return error; | |
4144 | } | |
3052b0c5 | 4145 | |
2e4f5fcf | 4146 | error = ofputil_decode_flow_mod(&fm, oh, ofconn->flow_format); |
064af421 BP |
4147 | if (error) { |
4148 | return error; | |
4149 | } | |
4150 | ||
2e4f5fcf BP |
4151 | /* We do not support the emergency flow cache. It will hopefully get |
4152 | * dropped from OpenFlow in the near future. */ | |
4153 | if (fm.flags & OFPFF_EMERG) { | |
49bdc010 JP |
4154 | /* There isn't a good fit for an error code, so just state that the |
4155 | * flow table is full. */ | |
4156 | return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL); | |
4157 | } | |
4158 | ||
2e4f5fcf BP |
4159 | error = validate_actions(fm.actions, fm.n_actions, |
4160 | &fm.cr.flow, p->max_ports); | |
4161 | if (error) { | |
4162 | return error; | |
4163 | } | |
4164 | ||
4165 | switch (fm.command) { | |
3052b0c5 | 4166 | case OFPFC_ADD: |
2e4f5fcf | 4167 | return add_flow(ofconn, &fm); |
3052b0c5 BP |
4168 | |
4169 | case OFPFC_MODIFY: | |
2e4f5fcf | 4170 | return modify_flows_loose(ofconn, &fm); |
3052b0c5 BP |
4171 | |
4172 | case OFPFC_MODIFY_STRICT: | |
2e4f5fcf | 4173 | return modify_flow_strict(ofconn, &fm); |
3052b0c5 BP |
4174 | |
4175 | case OFPFC_DELETE: | |
2e4f5fcf | 4176 | delete_flows_loose(p, &fm); |
3052b0c5 BP |
4177 | return 0; |
4178 | ||
4179 | case OFPFC_DELETE_STRICT: | |
2e4f5fcf | 4180 | delete_flow_strict(p, &fm); |
3052b0c5 BP |
4181 | return 0; |
4182 | ||
4183 | default: | |
4184 | return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND); | |
4185 | } | |
4186 | } | |
4187 | ||
659586ef | 4188 | static int |
d1e2cf21 | 4189 | handle_tun_id_from_cookie(struct ofconn *ofconn, const struct ofp_header *oh) |
659586ef | 4190 | { |
d1e2cf21 BP |
4191 | const struct nxt_tun_id_cookie *msg |
4192 | = (const struct nxt_tun_id_cookie *) oh; | |
659586ef | 4193 | |
b70eac89 | 4194 | ofconn->flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10; |
659586ef JG |
4195 | return 0; |
4196 | } | |
4197 | ||
9deba63b | 4198 | static int |
d1e2cf21 | 4199 | handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh) |
9deba63b | 4200 | { |
d1e2cf21 | 4201 | struct nx_role_request *nrr = (struct nx_role_request *) oh; |
9deba63b BP |
4202 | struct nx_role_request *reply; |
4203 | struct ofpbuf *buf; | |
4204 | uint32_t role; | |
4205 | ||
5899143f | 4206 | if (ofconn->type != OFCONN_PRIMARY) { |
9deba63b BP |
4207 | VLOG_WARN_RL(&rl, "ignoring role request on non-controller " |
4208 | "connection"); | |
4209 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM); | |
4210 | } | |
4211 | ||
4212 | role = ntohl(nrr->role); | |
4213 | if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER | |
4214 | && role != NX_ROLE_SLAVE) { | |
4215 | VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role); | |
4216 | ||
4217 | /* There's no good error code for this. */ | |
4218 | return ofp_mkerr(OFPET_BAD_REQUEST, -1); | |
4219 | } | |
4220 | ||
4221 | if (role == NX_ROLE_MASTER) { | |
4222 | struct ofconn *other; | |
4223 | ||
3269c562 | 4224 | HMAP_FOR_EACH (other, hmap_node, &ofconn->ofproto->controllers) { |
9deba63b BP |
4225 | if (other->role == NX_ROLE_MASTER) { |
4226 | other->role = NX_ROLE_SLAVE; | |
4227 | } | |
4228 | } | |
4229 | } | |
4230 | ofconn->role = role; | |
4231 | ||
d1e2cf21 | 4232 | reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf); |
9deba63b BP |
4233 | reply->role = htonl(role); |
4234 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
4235 | ||
4236 | return 0; | |
4237 | } | |
4238 | ||
09246b99 | 4239 | static int |
d1e2cf21 | 4240 | handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh) |
09246b99 | 4241 | { |
d1e2cf21 BP |
4242 | const struct nxt_set_flow_format *msg |
4243 | = (const struct nxt_set_flow_format *) oh; | |
09246b99 | 4244 | uint32_t format; |
09246b99 BP |
4245 | |
4246 | format = ntohl(msg->format); | |
4247 | if (format == NXFF_OPENFLOW10 | |
4248 | || format == NXFF_TUN_ID_FROM_COOKIE | |
4249 | || format == NXFF_NXM) { | |
4250 | ofconn->flow_format = format; | |
4251 | return 0; | |
4252 | } else { | |
4253 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM); | |
4254 | } | |
4255 | } | |
4256 | ||
064af421 | 4257 | static int |
d1e2cf21 | 4258 | handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh) |
246e61ea JP |
4259 | { |
4260 | struct ofp_header *ob; | |
4261 | struct ofpbuf *buf; | |
4262 | ||
4263 | /* Currently, everything executes synchronously, so we can just | |
4264 | * immediately send the barrier reply. */ | |
4265 | ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf); | |
4266 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
4267 | return 0; | |
4268 | } | |
4269 | ||
d1e2cf21 BP |
4270 | static int |
4271 | handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg) | |
064af421 | 4272 | { |
d1e2cf21 BP |
4273 | const struct ofp_header *oh = msg->data; |
4274 | const struct ofputil_msg_type *type; | |
064af421 BP |
4275 | int error; |
4276 | ||
d1e2cf21 BP |
4277 | error = ofputil_decode_msg_type(oh, &type); |
4278 | if (error) { | |
4279 | return error; | |
4280 | } | |
064af421 | 4281 | |
d1e2cf21 BP |
4282 | switch (ofputil_msg_type_code(type)) { |
4283 | /* OpenFlow requests. */ | |
4284 | case OFPUTIL_OFPT_ECHO_REQUEST: | |
4285 | return handle_echo_request(ofconn, oh); | |
064af421 | 4286 | |
d1e2cf21 BP |
4287 | case OFPUTIL_OFPT_FEATURES_REQUEST: |
4288 | return handle_features_request(ofconn, oh); | |
064af421 | 4289 | |
d1e2cf21 BP |
4290 | case OFPUTIL_OFPT_GET_CONFIG_REQUEST: |
4291 | return handle_get_config_request(ofconn, oh); | |
064af421 | 4292 | |
d1e2cf21 BP |
4293 | case OFPUTIL_OFPT_SET_CONFIG: |
4294 | return handle_set_config(ofconn, msg->data); | |
064af421 | 4295 | |
d1e2cf21 BP |
4296 | case OFPUTIL_OFPT_PACKET_OUT: |
4297 | return handle_packet_out(ofconn, oh); | |
064af421 | 4298 | |
d1e2cf21 BP |
4299 | case OFPUTIL_OFPT_PORT_MOD: |
4300 | return handle_port_mod(ofconn, oh); | |
064af421 | 4301 | |
d1e2cf21 | 4302 | case OFPUTIL_OFPT_FLOW_MOD: |
2e4f5fcf | 4303 | return handle_flow_mod(ofconn, oh); |
064af421 | 4304 | |
d1e2cf21 BP |
4305 | case OFPUTIL_OFPT_BARRIER_REQUEST: |
4306 | return handle_barrier_request(ofconn, oh); | |
064af421 | 4307 | |
d1e2cf21 BP |
4308 | /* OpenFlow replies. */ |
4309 | case OFPUTIL_OFPT_ECHO_REPLY: | |
4310 | return 0; | |
246e61ea | 4311 | |
d1e2cf21 BP |
4312 | /* Nicira extension requests. */ |
4313 | case OFPUTIL_NXT_STATUS_REQUEST: | |
4314 | return switch_status_handle_request( | |
4315 | ofconn->ofproto->switch_status, ofconn->rconn, oh); | |
4316 | ||
4317 | case OFPUTIL_NXT_TUN_ID_FROM_COOKIE: | |
4318 | return handle_tun_id_from_cookie(ofconn, oh); | |
4319 | ||
4320 | case OFPUTIL_NXT_ROLE_REQUEST: | |
4321 | return handle_role_request(ofconn, oh); | |
4322 | ||
4323 | case OFPUTIL_NXT_SET_FLOW_FORMAT: | |
4324 | return handle_nxt_set_flow_format(ofconn, oh); | |
4325 | ||
4326 | case OFPUTIL_NXT_FLOW_MOD: | |
2e4f5fcf | 4327 | return handle_flow_mod(ofconn, oh); |
d1e2cf21 BP |
4328 | |
4329 | /* OpenFlow statistics requests. */ | |
4330 | case OFPUTIL_OFPST_DESC_REQUEST: | |
4331 | return handle_desc_stats_request(ofconn, oh); | |
4332 | ||
4333 | case OFPUTIL_OFPST_FLOW_REQUEST: | |
4334 | return handle_flow_stats_request(ofconn, oh); | |
4335 | ||
4336 | case OFPUTIL_OFPST_AGGREGATE_REQUEST: | |
4337 | return handle_aggregate_stats_request(ofconn, oh); | |
4338 | ||
4339 | case OFPUTIL_OFPST_TABLE_REQUEST: | |
4340 | return handle_table_stats_request(ofconn, oh); | |
4341 | ||
4342 | case OFPUTIL_OFPST_PORT_REQUEST: | |
4343 | return handle_port_stats_request(ofconn, oh); | |
4344 | ||
4345 | case OFPUTIL_OFPST_QUEUE_REQUEST: | |
4346 | return handle_queue_stats_request(ofconn, oh); | |
4347 | ||
4348 | /* Nicira extension statistics requests. */ | |
4349 | case OFPUTIL_NXST_FLOW_REQUEST: | |
4350 | return handle_nxst_flow(ofconn, oh); | |
4351 | ||
4352 | case OFPUTIL_NXST_AGGREGATE_REQUEST: | |
4353 | return handle_nxst_aggregate(ofconn, oh); | |
4354 | ||
4355 | case OFPUTIL_INVALID: | |
4356 | case OFPUTIL_OFPT_HELLO: | |
4357 | case OFPUTIL_OFPT_ERROR: | |
4358 | case OFPUTIL_OFPT_FEATURES_REPLY: | |
4359 | case OFPUTIL_OFPT_GET_CONFIG_REPLY: | |
4360 | case OFPUTIL_OFPT_PACKET_IN: | |
4361 | case OFPUTIL_OFPT_FLOW_REMOVED: | |
4362 | case OFPUTIL_OFPT_PORT_STATUS: | |
4363 | case OFPUTIL_OFPT_BARRIER_REPLY: | |
4364 | case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST: | |
4365 | case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY: | |
4366 | case OFPUTIL_OFPST_DESC_REPLY: | |
4367 | case OFPUTIL_OFPST_FLOW_REPLY: | |
4368 | case OFPUTIL_OFPST_QUEUE_REPLY: | |
4369 | case OFPUTIL_OFPST_PORT_REPLY: | |
4370 | case OFPUTIL_OFPST_TABLE_REPLY: | |
4371 | case OFPUTIL_OFPST_AGGREGATE_REPLY: | |
4372 | case OFPUTIL_NXT_STATUS_REPLY: | |
4373 | case OFPUTIL_NXT_ROLE_REPLY: | |
4374 | case OFPUTIL_NXT_FLOW_REMOVED: | |
4375 | case OFPUTIL_NXST_FLOW_REPLY: | |
4376 | case OFPUTIL_NXST_AGGREGATE_REPLY: | |
064af421 BP |
4377 | default: |
4378 | if (VLOG_IS_WARN_ENABLED()) { | |
4379 | char *s = ofp_to_string(oh, ntohs(oh->length), 2); | |
4380 | VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s); | |
4381 | free(s); | |
4382 | } | |
d1e2cf21 BP |
4383 | if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) { |
4384 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT); | |
4385 | } else { | |
4386 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE); | |
4387 | } | |
064af421 | 4388 | } |
d1e2cf21 | 4389 | } |
064af421 | 4390 | |
d1e2cf21 BP |
4391 | static void |
4392 | handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg) | |
4393 | { | |
4394 | int error = handle_openflow__(ofconn, ofp_msg); | |
064af421 BP |
4395 | if (error) { |
4396 | send_error_oh(ofconn, ofp_msg->data, error); | |
4397 | } | |
d1e2cf21 | 4398 | COVERAGE_INC(ofproto_recv_openflow); |
064af421 BP |
4399 | } |
4400 | \f | |
4401 | static void | |
856081f6 | 4402 | handle_miss_upcall(struct ofproto *p, struct dpif_upcall *upcall) |
064af421 | 4403 | { |
bcf84111 | 4404 | struct facet *facet; |
ae412e7d | 4405 | struct flow flow; |
064af421 | 4406 | |
856081f6 BP |
4407 | /* Obtain in_port and tun_id, at least. */ |
4408 | odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow); | |
064af421 | 4409 | |
856081f6 BP |
4410 | /* Set header pointers in 'flow'. */ |
4411 | flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow); | |
4617e2c1 | 4412 | |
0ad9b732 JP |
4413 | /* Check with in-band control to see if this packet should be sent |
4414 | * to the local port regardless of the flow table. */ | |
856081f6 | 4415 | if (in_band_msg_in_hook(p->in_band, &flow, upcall->packet)) { |
cdee00fd | 4416 | struct ofpbuf odp_actions; |
0ad9b732 | 4417 | |
cdee00fd BP |
4418 | ofpbuf_init(&odp_actions, 32); |
4419 | nl_msg_put_u32(&odp_actions, ODPAT_OUTPUT, ODPP_LOCAL); | |
856081f6 BP |
4420 | dpif_execute(p->dpif, odp_actions.data, odp_actions.size, |
4421 | upcall->packet); | |
cdee00fd | 4422 | ofpbuf_uninit(&odp_actions); |
0ad9b732 JP |
4423 | } |
4424 | ||
bcf84111 BP |
4425 | facet = facet_lookup_valid(p, &flow); |
4426 | if (!facet) { | |
4427 | struct rule *rule = rule_lookup(p, &flow); | |
4428 | if (!rule) { | |
4429 | /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */ | |
856081f6 | 4430 | struct ofport *port = get_port(p, flow.in_port); |
bcf84111 BP |
4431 | if (port) { |
4432 | if (port->opp.config & OFPPC_NO_PACKET_IN) { | |
4433 | COVERAGE_INC(ofproto_no_packet_in); | |
4434 | /* XXX install 'drop' flow entry */ | |
856081f6 | 4435 | ofpbuf_delete(upcall->packet); |
bcf84111 BP |
4436 | return; |
4437 | } | |
4438 | } else { | |
4439 | VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, | |
856081f6 | 4440 | flow.in_port); |
064af421 | 4441 | } |
064af421 | 4442 | |
bcf84111 | 4443 | COVERAGE_INC(ofproto_packet_in); |
856081f6 | 4444 | send_packet_in(p, upcall, &flow, false); |
bcf84111 | 4445 | return; |
064af421 | 4446 | } |
bcf84111 | 4447 | |
856081f6 | 4448 | facet = facet_create(p, rule, &flow, upcall->packet); |
bcf84111 BP |
4449 | } else if (!facet->may_install) { |
4450 | /* The facet is not installable, that is, we need to process every | |
4451 | * packet, so process the current packet's actions into 'facet'. */ | |
856081f6 | 4452 | facet_make_actions(p, facet, upcall->packet); |
064af421 BP |
4453 | } |
4454 | ||
bcf84111 | 4455 | if (facet->rule->cr.priority == FAIL_OPEN_PRIORITY) { |
7778bd15 BP |
4456 | /* |
4457 | * Extra-special case for fail-open mode. | |
4458 | * | |
4459 | * We are in fail-open mode and the packet matched the fail-open rule, | |
4460 | * but we are connected to a controller too. We should send the packet | |
4461 | * up to the controller in the hope that it will try to set up a flow | |
4462 | * and thereby allow us to exit fail-open. | |
4463 | * | |
4464 | * See the top-level comment in fail-open.c for more information. | |
4465 | */ | |
856081f6 | 4466 | send_packet_in(p, upcall, &flow, true); |
7778bd15 | 4467 | } |
750638bb | 4468 | |
856081f6 | 4469 | facet_execute(p, facet, upcall->packet); |
bcf84111 | 4470 | facet_install(p, facet, false); |
064af421 | 4471 | } |
72b06300 BP |
4472 | |
4473 | static void | |
856081f6 | 4474 | handle_upcall(struct ofproto *p, struct dpif_upcall *upcall) |
72b06300 | 4475 | { |
856081f6 | 4476 | struct flow flow; |
72b06300 | 4477 | |
856081f6 | 4478 | switch (upcall->type) { |
72b06300 BP |
4479 | case _ODPL_ACTION_NR: |
4480 | COVERAGE_INC(ofproto_ctlr_action); | |
856081f6 BP |
4481 | odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow); |
4482 | send_packet_in(p, upcall, &flow, false); | |
72b06300 BP |
4483 | break; |
4484 | ||
4485 | case _ODPL_SFLOW_NR: | |
4486 | if (p->sflow) { | |
856081f6 BP |
4487 | odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow); |
4488 | ofproto_sflow_received(p->sflow, upcall, &flow); | |
72b06300 | 4489 | } |
856081f6 | 4490 | ofpbuf_delete(upcall->packet); |
72b06300 BP |
4491 | break; |
4492 | ||
4493 | case _ODPL_MISS_NR: | |
856081f6 | 4494 | handle_miss_upcall(p, upcall); |
72b06300 BP |
4495 | break; |
4496 | ||
4497 | default: | |
856081f6 | 4498 | VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type); |
72b06300 BP |
4499 | break; |
4500 | } | |
4501 | } | |
064af421 | 4502 | \f |
4a4cdb3b BP |
4503 | /* Flow expiration. */ |
4504 | ||
0de7a4b4 | 4505 | static int ofproto_dp_max_idle(const struct ofproto *); |
4a4cdb3b | 4506 | static void ofproto_update_used(struct ofproto *); |
5ecc9d81 | 4507 | static void rule_expire(struct ofproto *, struct rule *); |
bcf84111 | 4508 | static void ofproto_expire_facets(struct ofproto *, int dp_max_idle); |
4a4cdb3b BP |
4509 | |
4510 | /* This function is called periodically by ofproto_run(). Its job is to | |
4511 | * collect updates for the flows that have been installed into the datapath, | |
4512 | * most importantly when they last were used, and then use that information to | |
0de7a4b4 BP |
4513 | * expire flows that have not been used recently. |
4514 | * | |
4515 | * Returns the number of milliseconds after which it should be called again. */ | |
4516 | static int | |
4a4cdb3b BP |
4517 | ofproto_expire(struct ofproto *ofproto) |
4518 | { | |
5ecc9d81 BP |
4519 | struct rule *rule, *next_rule; |
4520 | struct cls_cursor cursor; | |
4521 | int dp_max_idle; | |
4a4cdb3b BP |
4522 | |
4523 | /* Update 'used' for each flow in the datapath. */ | |
4524 | ofproto_update_used(ofproto); | |
4525 | ||
bcf84111 | 4526 | /* Expire facets that have been idle too long. */ |
5ecc9d81 BP |
4527 | dp_max_idle = ofproto_dp_max_idle(ofproto); |
4528 | ofproto_expire_facets(ofproto, dp_max_idle); | |
bcf84111 BP |
4529 | |
4530 | /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */ | |
5ecc9d81 BP |
4531 | cls_cursor_init(&cursor, &ofproto->cls, NULL); |
4532 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) { | |
4533 | rule_expire(ofproto, rule); | |
4534 | } | |
4a4cdb3b BP |
4535 | |
4536 | /* Let the hook know that we're at a stable point: all outstanding data | |
4537 | * in existing flows has been accounted to the account_cb. Thus, the | |
4538 | * hook can now reasonably do operations that depend on having accurate | |
4539 | * flow volume accounting (currently, that's just bond rebalancing). */ | |
4540 | if (ofproto->ofhooks->account_checkpoint_cb) { | |
4541 | ofproto->ofhooks->account_checkpoint_cb(ofproto->aux); | |
4542 | } | |
0de7a4b4 | 4543 | |
5ecc9d81 | 4544 | return MIN(dp_max_idle, 1000); |
4a4cdb3b BP |
4545 | } |
4546 | ||
bcf84111 | 4547 | /* Update 'used' member of installed facets. */ |
4a4cdb3b BP |
4548 | static void |
4549 | ofproto_update_used(struct ofproto *p) | |
4550 | { | |
704a1e09 | 4551 | struct dpif_flow_dump dump; |
4a4cdb3b | 4552 | |
704a1e09 | 4553 | dpif_flow_dump_start(&dump, p->dpif); |
36956a7d BP |
4554 | for (;;) { |
4555 | uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S]; | |
bcf84111 | 4556 | struct facet *facet; |
36956a7d | 4557 | struct odp_flow f; |
ae412e7d | 4558 | struct flow flow; |
14608a15 | 4559 | |
36956a7d BP |
4560 | memset(&f, 0, sizeof f); |
4561 | f.key = (struct nlattr *) keybuf; | |
4562 | f.key_len = sizeof keybuf; | |
4563 | if (!dpif_flow_dump_next(&dump, &f)) { | |
4564 | break; | |
4565 | } | |
4566 | ||
4567 | if (f.key_len > sizeof keybuf) { | |
4568 | VLOG_WARN_RL(&rl, "ODP flow key overflowed buffer"); | |
4569 | continue; | |
4570 | } | |
4571 | if (odp_flow_key_to_flow(f.key, f.key_len, &flow)) { | |
4572 | struct ds s; | |
4573 | ||
4574 | ds_init(&s); | |
4575 | odp_flow_key_format(f.key, f.key_len, &s); | |
4576 | VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s", | |
4577 | ds_cstr(&s)); | |
4578 | ds_destroy(&s); | |
4579 | ||
4580 | continue; | |
4581 | } | |
bcf84111 | 4582 | facet = facet_find(p, &flow); |
4a4cdb3b | 4583 | |
bcf84111 | 4584 | if (facet && facet->installed) { |
704a1e09 BP |
4585 | facet_update_time(p, facet, &f.stats); |
4586 | facet_account(p, facet, f.stats.n_bytes); | |
4a4cdb3b BP |
4587 | } else { |
4588 | /* There's a flow in the datapath that we know nothing about. | |
4589 | * Delete it. */ | |
4590 | COVERAGE_INC(ofproto_unexpected_rule); | |
704a1e09 | 4591 | dpif_flow_del(p->dpif, &f); |
4a4cdb3b | 4592 | } |
4a4cdb3b | 4593 | } |
704a1e09 | 4594 | dpif_flow_dump_done(&dump); |
4a4cdb3b BP |
4595 | } |
4596 | ||
0de7a4b4 | 4597 | /* Calculates and returns the number of milliseconds of idle time after which |
bcf84111 | 4598 | * facets should expire from the datapath and we should fold their statistics |
0de7a4b4 BP |
4599 | * into their parent rules in userspace. */ |
4600 | static int | |
4601 | ofproto_dp_max_idle(const struct ofproto *ofproto) | |
4602 | { | |
4603 | /* | |
4604 | * Idle time histogram. | |
4605 | * | |
bcf84111 | 4606 | * Most of the time a switch has a relatively small number of facets. When |
0de7a4b4 BP |
4607 | * this is the case we might as well keep statistics for all of them in |
4608 | * userspace and to cache them in the kernel datapath for performance as | |
4609 | * well. | |
4610 | * | |
bcf84111 | 4611 | * As the number of facets increases, the memory required to maintain |
0de7a4b4 | 4612 | * statistics about them in userspace and in the kernel becomes |
bcf84111 | 4613 | * significant. However, with a large number of facets it is likely that |
0de7a4b4 BP |
4614 | * only a few of them are "heavy hitters" that consume a large amount of |
4615 | * bandwidth. At this point, only heavy hitters are worth caching in the | |
bcf84111 | 4616 | * kernel and maintaining in userspaces; other facets we can discard. |
0de7a4b4 BP |
4617 | * |
4618 | * The technique used to compute the idle time is to build a histogram with | |
bcf84111 BP |
4619 | * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet |
4620 | * that is installed in the kernel gets dropped in the appropriate bucket. | |
0de7a4b4 | 4621 | * After the histogram has been built, we compute the cutoff so that only |
bcf84111 BP |
4622 | * the most-recently-used 1% of facets (but at least 1000 flows) are kept |
4623 | * cached. At least the most-recently-used bucket of facets is kept, so | |
4624 | * actually an arbitrary number of facets can be kept in any given | |
0de7a4b4 BP |
4625 | * expiration run (though the next run will delete most of those unless |
4626 | * they receive additional data). | |
4627 | * | |
bcf84111 BP |
4628 | * This requires a second pass through the facets, in addition to the pass |
4629 | * made by ofproto_update_used(), because the former function never looks | |
4630 | * at uninstallable facets. | |
0de7a4b4 BP |
4631 | */ |
4632 | enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) }; | |
4633 | enum { N_BUCKETS = 5000 / BUCKET_WIDTH }; | |
4634 | int buckets[N_BUCKETS] = { 0 }; | |
bcf84111 | 4635 | struct facet *facet; |
0de7a4b4 | 4636 | int total, bucket; |
0de7a4b4 BP |
4637 | long long int now; |
4638 | int i; | |
4639 | ||
bcf84111 | 4640 | total = hmap_count(&ofproto->facets); |
0de7a4b4 BP |
4641 | if (total <= 1000) { |
4642 | return N_BUCKETS * BUCKET_WIDTH; | |
4643 | } | |
4644 | ||
4645 | /* Build histogram. */ | |
4646 | now = time_msec(); | |
bcf84111 BP |
4647 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { |
4648 | long long int idle = now - facet->used; | |
0de7a4b4 BP |
4649 | int bucket = (idle <= 0 ? 0 |
4650 | : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1 | |
4651 | : (unsigned int) idle / BUCKET_WIDTH); | |
4652 | buckets[bucket]++; | |
4653 | } | |
4654 | ||
4655 | /* Find the first bucket whose flows should be expired. */ | |
4656 | for (bucket = 0; bucket < N_BUCKETS; bucket++) { | |
4657 | if (buckets[bucket]) { | |
4658 | int subtotal = 0; | |
4659 | do { | |
4660 | subtotal += buckets[bucket++]; | |
4661 | } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100)); | |
4662 | break; | |
4663 | } | |
4664 | } | |
4665 | ||
4666 | if (VLOG_IS_DBG_ENABLED()) { | |
4667 | struct ds s; | |
4668 | ||
4669 | ds_init(&s); | |
4670 | ds_put_cstr(&s, "keep"); | |
4671 | for (i = 0; i < N_BUCKETS; i++) { | |
4672 | if (i == bucket) { | |
4673 | ds_put_cstr(&s, ", drop"); | |
4674 | } | |
4675 | if (buckets[i]) { | |
4676 | ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]); | |
4677 | } | |
4678 | } | |
4679 | VLOG_INFO("%s: %s (msec:count)", | |
4680 | dpif_name(ofproto->dpif), ds_cstr(&s)); | |
4681 | ds_destroy(&s); | |
4682 | } | |
4683 | ||
4684 | return bucket * BUCKET_WIDTH; | |
4685 | } | |
4686 | ||
4a4cdb3b | 4687 | static void |
bcf84111 | 4688 | facet_active_timeout(struct ofproto *ofproto, struct facet *facet) |
4a4cdb3b | 4689 | { |
bcf84111 BP |
4690 | if (ofproto->netflow && !facet_is_controller_flow(facet) && |
4691 | netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) { | |
4a4cdb3b BP |
4692 | struct ofexpired expired; |
4693 | struct odp_flow odp_flow; | |
4694 | ||
4695 | /* Get updated flow stats. | |
4696 | * | |
4697 | * XXX We could avoid this call entirely if (1) ofproto_update_used() | |
4698 | * updated TCP flags and (2) the dpif_flow_list_all() in | |
4699 | * ofproto_update_used() zeroed TCP flags. */ | |
4700 | memset(&odp_flow, 0, sizeof odp_flow); | |
bcf84111 | 4701 | if (facet->installed) { |
36956a7d BP |
4702 | uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S]; |
4703 | struct ofpbuf key; | |
4704 | ||
4705 | ofpbuf_use_stack(&key, keybuf, sizeof keybuf); | |
4706 | odp_flow_key_from_flow(&key, &facet->flow); | |
4707 | ||
4708 | odp_flow.key = key.data; | |
4709 | odp_flow.key_len = key.size; | |
4a4cdb3b BP |
4710 | odp_flow.flags = ODPFF_ZERO_TCP_FLAGS; |
4711 | dpif_flow_get(ofproto->dpif, &odp_flow); | |
4712 | ||
4713 | if (odp_flow.stats.n_packets) { | |
bcf84111 BP |
4714 | facet_update_time(ofproto, facet, &odp_flow.stats); |
4715 | netflow_flow_update_flags(&facet->nf_flow, | |
4a4cdb3b BP |
4716 | odp_flow.stats.tcp_flags); |
4717 | } | |
4718 | } | |
4719 | ||
bcf84111 BP |
4720 | expired.flow = facet->flow; |
4721 | expired.packet_count = facet->packet_count + | |
4a4cdb3b | 4722 | odp_flow.stats.n_packets; |
bcf84111 BP |
4723 | expired.byte_count = facet->byte_count + odp_flow.stats.n_bytes; |
4724 | expired.used = facet->used; | |
4725 | ||
4726 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
4727 | } | |
4728 | } | |
4a4cdb3b | 4729 | |
bcf84111 BP |
4730 | static void |
4731 | ofproto_expire_facets(struct ofproto *ofproto, int dp_max_idle) | |
4732 | { | |
4733 | long long int cutoff = time_msec() - dp_max_idle; | |
4734 | struct facet *facet, *next_facet; | |
4735 | ||
4736 | HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) { | |
4737 | facet_active_timeout(ofproto, facet); | |
4738 | if (facet->used < cutoff) { | |
4739 | facet_remove(ofproto, facet); | |
4740 | } | |
4a4cdb3b BP |
4741 | } |
4742 | } | |
4743 | ||
5ecc9d81 BP |
4744 | /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules, |
4745 | * then delete it entirely. */ | |
4a4cdb3b | 4746 | static void |
5ecc9d81 | 4747 | rule_expire(struct ofproto *ofproto, struct rule *rule) |
4a4cdb3b | 4748 | { |
bcf84111 BP |
4749 | struct facet *facet, *next_facet; |
4750 | long long int now; | |
4751 | uint8_t reason; | |
4a4cdb3b | 4752 | |
bcf84111 | 4753 | /* Has 'rule' expired? */ |
4a4cdb3b | 4754 | now = time_msec(); |
bcf84111 BP |
4755 | if (rule->hard_timeout |
4756 | && now > rule->created + rule->hard_timeout * 1000) { | |
4757 | reason = OFPRR_HARD_TIMEOUT; | |
4758 | } else if (rule->idle_timeout && list_is_empty(&rule->facets) | |
4759 | && now >rule->used + rule->idle_timeout * 1000) { | |
4760 | reason = OFPRR_IDLE_TIMEOUT; | |
4a4cdb3b | 4761 | } else { |
bcf84111 | 4762 | return; |
4a4cdb3b | 4763 | } |
064af421 | 4764 | |
bcf84111 | 4765 | COVERAGE_INC(ofproto_expired); |
064af421 | 4766 | |
bcf84111 BP |
4767 | /* Update stats. (This is a no-op if the rule expired due to an idle |
4768 | * timeout, because that only happens when the rule has no facets left.) */ | |
4769 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { | |
5ecc9d81 | 4770 | facet_remove(ofproto, facet); |
064af421 BP |
4771 | } |
4772 | ||
bcf84111 BP |
4773 | /* Get rid of the rule. */ |
4774 | if (!rule_is_hidden(rule)) { | |
5ecc9d81 | 4775 | rule_send_removed(ofproto, rule, reason); |
bcf84111 | 4776 | } |
5ecc9d81 | 4777 | rule_remove(ofproto, rule); |
064af421 | 4778 | } |
bcf84111 | 4779 | \f |
064af421 | 4780 | static struct ofpbuf * |
09246b99 BP |
4781 | compose_ofp_flow_removed(struct ofconn *ofconn, const struct rule *rule, |
4782 | uint8_t reason) | |
064af421 | 4783 | { |
ca069229 | 4784 | struct ofp_flow_removed *ofr; |
064af421 BP |
4785 | struct ofpbuf *buf; |
4786 | ||
0224fbde | 4787 | ofr = make_openflow_xid(sizeof *ofr, OFPT_FLOW_REMOVED, htonl(0), &buf); |
ff9d3826 BP |
4788 | ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofr->match, |
4789 | rule->flow_cookie, &ofr->cookie); | |
ca069229 JP |
4790 | ofr->priority = htons(rule->cr.priority); |
4791 | ofr->reason = reason; | |
c6ebb8fb | 4792 | calc_flow_duration(rule->created, &ofr->duration_sec, &ofr->duration_nsec); |
ca069229 JP |
4793 | ofr->idle_timeout = htons(rule->idle_timeout); |
4794 | ofr->packet_count = htonll(rule->packet_count); | |
4795 | ofr->byte_count = htonll(rule->byte_count); | |
064af421 BP |
4796 | |
4797 | return buf; | |
4798 | } | |
4799 | ||
09246b99 BP |
4800 | static struct ofpbuf * |
4801 | compose_nx_flow_removed(const struct rule *rule, uint8_t reason) | |
4802 | { | |
4803 | struct nx_flow_removed *nfr; | |
4804 | struct ofpbuf *buf; | |
4805 | int match_len; | |
4806 | ||
0224fbde | 4807 | make_nxmsg_xid(sizeof *nfr, NXT_FLOW_REMOVED, htonl(0), &buf); |
09246b99 | 4808 | match_len = nx_put_match(buf, &rule->cr); |
0224fbde | 4809 | |
4d0ed519 | 4810 | nfr = buf->data; |
09246b99 BP |
4811 | nfr->cookie = rule->flow_cookie; |
4812 | nfr->priority = htons(rule->cr.priority); | |
4813 | nfr->reason = reason; | |
4814 | calc_flow_duration(rule->created, &nfr->duration_sec, &nfr->duration_nsec); | |
4815 | nfr->idle_timeout = htons(rule->idle_timeout); | |
4816 | nfr->match_len = htons(match_len); | |
4817 | nfr->packet_count = htonll(rule->packet_count); | |
4818 | nfr->byte_count = htonll(rule->byte_count); | |
4819 | ||
4820 | return buf; | |
4821 | } | |
4822 | ||
ca069229 | 4823 | static void |
bcf84111 | 4824 | rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason) |
064af421 BP |
4825 | { |
4826 | struct ofconn *ofconn; | |
064af421 | 4827 | |
3b587616 BP |
4828 | if (!rule->send_flow_removed) { |
4829 | return; | |
4830 | } | |
4831 | ||
4e8e4213 | 4832 | LIST_FOR_EACH (ofconn, node, &p->all_conns) { |
7a0efeb5 BP |
4833 | struct ofpbuf *msg; |
4834 | ||
4835 | if (!rconn_is_connected(ofconn->rconn) | |
4836 | || !ofconn_receives_async_msgs(ofconn)) { | |
4837 | continue; | |
064af421 | 4838 | } |
7a0efeb5 | 4839 | |
09246b99 BP |
4840 | msg = (ofconn->flow_format == NXFF_NXM |
4841 | ? compose_nx_flow_removed(rule, reason) | |
4842 | : compose_ofp_flow_removed(ofconn, rule, reason)); | |
6d6c7259 BP |
4843 | |
4844 | /* Account flow expirations under ofconn->reply_counter, the counter | |
4845 | * for replies to OpenFlow requests. That works because preventing | |
4846 | * OpenFlow requests from being processed also prevents new flows from | |
4847 | * being added (and expiring). (It also prevents processing OpenFlow | |
4848 | * requests that would not add new flows, so it is imperfect.) */ | |
7a0efeb5 | 4849 | queue_tx(msg, ofconn, ofconn->reply_counter); |
064af421 BP |
4850 | } |
4851 | } | |
4852 | ||
856081f6 | 4853 | /* pinsched callback for sending 'ofp_packet_in' on 'ofconn'. */ |
064af421 | 4854 | static void |
856081f6 | 4855 | do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn_) |
064af421 | 4856 | { |
76ce9432 | 4857 | struct ofconn *ofconn = ofconn_; |
43253595 | 4858 | |
856081f6 | 4859 | rconn_send_with_limit(ofconn->rconn, ofp_packet_in, |
43253595 BP |
4860 | ofconn->packet_in_counter, 100); |
4861 | } | |
4862 | ||
856081f6 BP |
4863 | /* Takes 'upcall', whose packet has the flow specified by 'flow', composes an |
4864 | * OpenFlow packet-in message from it, and passes it to 'ofconn''s packet | |
4865 | * scheduler for sending. | |
43253595 | 4866 | * |
856081f6 BP |
4867 | * If 'clone' is true, the caller retains ownership of 'upcall->packet'. |
4868 | * Otherwise, ownership is transferred to this function. */ | |
43253595 | 4869 | static void |
856081f6 BP |
4870 | schedule_packet_in(struct ofconn *ofconn, struct dpif_upcall *upcall, |
4871 | const struct flow *flow, bool clone) | |
43253595 | 4872 | { |
856081f6 | 4873 | enum { OPI_SIZE = offsetof(struct ofp_packet_in, data) }; |
76ce9432 | 4874 | struct ofproto *ofproto = ofconn->ofproto; |
856081f6 BP |
4875 | struct ofp_packet_in *opi; |
4876 | int total_len, send_len; | |
4877 | struct ofpbuf *packet; | |
76ce9432 | 4878 | uint32_t buffer_id; |
064af421 | 4879 | |
856081f6 BP |
4880 | /* Get OpenFlow buffer_id. */ |
4881 | if (upcall->type == _ODPL_ACTION_NR) { | |
76ce9432 | 4882 | buffer_id = UINT32_MAX; |
43253595 BP |
4883 | } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) { |
4884 | buffer_id = pktbuf_get_null(); | |
89b9612d BP |
4885 | } else if (!ofconn->pktbuf) { |
4886 | buffer_id = UINT32_MAX; | |
76ce9432 | 4887 | } else { |
856081f6 | 4888 | buffer_id = pktbuf_save(ofconn->pktbuf, upcall->packet, flow->in_port); |
76ce9432 | 4889 | } |
372179d4 | 4890 | |
43253595 | 4891 | /* Figure out how much of the packet to send. */ |
856081f6 | 4892 | total_len = send_len = upcall->packet->size; |
43253595 BP |
4893 | if (buffer_id != UINT32_MAX) { |
4894 | send_len = MIN(send_len, ofconn->miss_send_len); | |
4895 | } | |
856081f6 BP |
4896 | if (upcall->type == _ODPL_ACTION_NR) { |
4897 | send_len = MIN(send_len, upcall->userdata); | |
4898 | } | |
064af421 | 4899 | |
856081f6 | 4900 | /* Copy or steal buffer for OFPT_PACKET_IN. */ |
43253595 | 4901 | if (clone) { |
856081f6 BP |
4902 | packet = ofpbuf_clone_data_with_headroom(upcall->packet->data, |
4903 | send_len, OPI_SIZE); | |
43253595 | 4904 | } else { |
856081f6 BP |
4905 | packet = upcall->packet; |
4906 | packet->size = send_len; | |
43253595 BP |
4907 | } |
4908 | ||
856081f6 BP |
4909 | /* Add OFPT_PACKET_IN. */ |
4910 | opi = ofpbuf_push_zeros(packet, OPI_SIZE); | |
4911 | opi->header.version = OFP_VERSION; | |
4912 | opi->header.type = OFPT_PACKET_IN; | |
4913 | opi->total_len = htons(total_len); | |
4914 | opi->in_port = htons(odp_port_to_ofp_port(flow->in_port)); | |
4915 | opi->reason = upcall->type == _ODPL_MISS_NR ? OFPR_NO_MATCH : OFPR_ACTION; | |
43253595 BP |
4916 | opi->buffer_id = htonl(buffer_id); |
4917 | update_openflow_length(packet); | |
4918 | ||
4919 | /* Hand over to packet scheduler. It might immediately call into | |
4920 | * do_send_packet_in() or it might buffer it for a while (until a later | |
4921 | * call to pinsched_run()). */ | |
856081f6 | 4922 | pinsched_send(ofconn->schedulers[opi->reason], flow->in_port, |
43253595 | 4923 | packet, do_send_packet_in, ofconn); |
064af421 BP |
4924 | } |
4925 | ||
856081f6 BP |
4926 | /* Given 'upcall', of type _ODPL_ACTION_NR or _ODPL_MISS_NR, sends an |
4927 | * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to | |
4928 | * their individual configurations. | |
43253595 BP |
4929 | * |
4930 | * Takes ownership of 'packet'. */ | |
4931 | static void | |
856081f6 BP |
4932 | send_packet_in(struct ofproto *ofproto, struct dpif_upcall *upcall, |
4933 | const struct flow *flow, bool clone) | |
43253595 | 4934 | { |
76ce9432 | 4935 | struct ofconn *ofconn, *prev; |
76ce9432 BP |
4936 | |
4937 | prev = NULL; | |
4e8e4213 | 4938 | LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) { |
c91248b3 | 4939 | if (ofconn_receives_async_msgs(ofconn)) { |
9deba63b | 4940 | if (prev) { |
856081f6 | 4941 | schedule_packet_in(prev, upcall, flow, true); |
9deba63b BP |
4942 | } |
4943 | prev = ofconn; | |
064af421 | 4944 | } |
76ce9432 BP |
4945 | } |
4946 | if (prev) { | |
856081f6 BP |
4947 | schedule_packet_in(prev, upcall, flow, clone); |
4948 | } else if (!clone) { | |
4949 | ofpbuf_delete(upcall->packet); | |
064af421 | 4950 | } |
064af421 BP |
4951 | } |
4952 | ||
4953 | static uint64_t | |
fa60c019 | 4954 | pick_datapath_id(const struct ofproto *ofproto) |
064af421 | 4955 | { |
fa60c019 | 4956 | const struct ofport *port; |
064af421 | 4957 | |
ca0f572c | 4958 | port = get_port(ofproto, ODPP_LOCAL); |
fa60c019 BP |
4959 | if (port) { |
4960 | uint8_t ea[ETH_ADDR_LEN]; | |
4961 | int error; | |
4962 | ||
4963 | error = netdev_get_etheraddr(port->netdev, ea); | |
064af421 BP |
4964 | if (!error) { |
4965 | return eth_addr_to_uint64(ea); | |
4966 | } | |
4967 | VLOG_WARN("could not get MAC address for %s (%s)", | |
fa60c019 | 4968 | netdev_get_name(port->netdev), strerror(error)); |
064af421 | 4969 | } |
fa60c019 | 4970 | return ofproto->fallback_dpid; |
064af421 BP |
4971 | } |
4972 | ||
4973 | static uint64_t | |
4974 | pick_fallback_dpid(void) | |
4975 | { | |
4976 | uint8_t ea[ETH_ADDR_LEN]; | |
70150daf | 4977 | eth_addr_nicira_random(ea); |
064af421 BP |
4978 | return eth_addr_to_uint64(ea); |
4979 | } | |
4980 | \f | |
7aa697dd BP |
4981 | static void |
4982 | ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED, | |
4983 | void *aux OVS_UNUSED) | |
4984 | { | |
4985 | const struct shash_node *node; | |
4986 | struct ds results; | |
4987 | ||
4988 | ds_init(&results); | |
4989 | SHASH_FOR_EACH (node, &all_ofprotos) { | |
4990 | ds_put_format(&results, "%s\n", node->name); | |
4991 | } | |
4992 | unixctl_command_reply(conn, 200, ds_cstr(&results)); | |
4993 | ds_destroy(&results); | |
4994 | } | |
4995 | ||
4996 | struct ofproto_trace { | |
4997 | struct action_xlate_ctx ctx; | |
4998 | struct flow flow; | |
4999 | struct ds *result; | |
5000 | }; | |
5001 | ||
5002 | static void | |
5003 | trace_format_rule(struct ds *result, int level, const struct rule *rule) | |
5004 | { | |
5005 | ds_put_char_multiple(result, '\t', level); | |
5006 | if (!rule) { | |
5007 | ds_put_cstr(result, "No match\n"); | |
5008 | return; | |
5009 | } | |
5010 | ||
5011 | ds_put_format(result, "Rule: cookie=%#"PRIx64" ", | |
5012 | ntohll(rule->flow_cookie)); | |
5013 | cls_rule_format(&rule->cr, result); | |
5014 | ds_put_char(result, '\n'); | |
5015 | ||
5016 | ds_put_char_multiple(result, '\t', level); | |
5017 | ds_put_cstr(result, "OpenFlow "); | |
5018 | ofp_print_actions(result, (const struct ofp_action_header *) rule->actions, | |
5019 | rule->n_actions * sizeof *rule->actions); | |
5020 | ds_put_char(result, '\n'); | |
5021 | } | |
5022 | ||
5023 | static void | |
5024 | trace_format_flow(struct ds *result, int level, const char *title, | |
5025 | struct ofproto_trace *trace) | |
5026 | { | |
5027 | ds_put_char_multiple(result, '\t', level); | |
5028 | ds_put_format(result, "%s: ", title); | |
5029 | if (flow_equal(&trace->ctx.flow, &trace->flow)) { | |
5030 | ds_put_cstr(result, "unchanged"); | |
5031 | } else { | |
5032 | flow_format(result, &trace->ctx.flow); | |
5033 | trace->flow = trace->ctx.flow; | |
5034 | } | |
5035 | ds_put_char(result, '\n'); | |
5036 | } | |
5037 | ||
5038 | static void | |
5039 | trace_resubmit(struct action_xlate_ctx *ctx, const struct rule *rule) | |
5040 | { | |
5041 | struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx); | |
5042 | struct ds *result = trace->result; | |
5043 | ||
5044 | ds_put_char(result, '\n'); | |
5045 | trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace); | |
5046 | trace_format_rule(result, ctx->recurse + 1, rule); | |
5047 | } | |
5048 | ||
5049 | static void | |
5050 | ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_, | |
5051 | void *aux OVS_UNUSED) | |
5052 | { | |
5053 | char *dpname, *in_port_s, *tun_id_s, *packet_s; | |
5054 | char *args = xstrdup(args_); | |
5055 | char *save_ptr = NULL; | |
5056 | struct ofproto *ofproto; | |
5057 | struct ofpbuf packet; | |
5058 | struct rule *rule; | |
5059 | struct ds result; | |
5060 | struct flow flow; | |
5061 | uint16_t in_port; | |
11e6a15b | 5062 | ovs_be64 tun_id; |
7aa697dd BP |
5063 | char *s; |
5064 | ||
5065 | ofpbuf_init(&packet, strlen(args) / 2); | |
5066 | ds_init(&result); | |
5067 | ||
5068 | dpname = strtok_r(args, " ", &save_ptr); | |
5069 | tun_id_s = strtok_r(NULL, " ", &save_ptr); | |
5070 | in_port_s = strtok_r(NULL, " ", &save_ptr); | |
5071 | packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */ | |
5072 | if (!dpname || !in_port_s || !packet_s) { | |
5073 | unixctl_command_reply(conn, 501, "Bad command syntax"); | |
5074 | goto exit; | |
5075 | } | |
5076 | ||
5077 | ofproto = shash_find_data(&all_ofprotos, dpname); | |
5078 | if (!ofproto) { | |
5079 | unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list " | |
5080 | "for help)"); | |
5081 | goto exit; | |
5082 | } | |
5083 | ||
11e6a15b | 5084 | tun_id = htonll(strtoull(tun_id_s, NULL, 10)); |
7aa697dd BP |
5085 | in_port = ofp_port_to_odp_port(atoi(in_port_s)); |
5086 | ||
5087 | packet_s = ofpbuf_put_hex(&packet, packet_s, NULL); | |
5088 | packet_s += strspn(packet_s, " "); | |
5089 | if (*packet_s != '\0') { | |
5090 | unixctl_command_reply(conn, 501, "Trailing garbage in command"); | |
5091 | goto exit; | |
5092 | } | |
5093 | if (packet.size < ETH_HEADER_LEN) { | |
5094 | unixctl_command_reply(conn, 501, "Packet data too short for Ethernet"); | |
5095 | goto exit; | |
5096 | } | |
5097 | ||
5098 | ds_put_cstr(&result, "Packet: "); | |
5099 | s = ofp_packet_to_string(packet.data, packet.size, packet.size); | |
5100 | ds_put_cstr(&result, s); | |
5101 | free(s); | |
5102 | ||
5103 | flow_extract(&packet, tun_id, in_port, &flow); | |
5104 | ds_put_cstr(&result, "Flow: "); | |
5105 | flow_format(&result, &flow); | |
5106 | ds_put_char(&result, '\n'); | |
5107 | ||
5108 | rule = rule_lookup(ofproto, &flow); | |
5109 | trace_format_rule(&result, 0, rule); | |
5110 | if (rule) { | |
5111 | struct ofproto_trace trace; | |
cdee00fd | 5112 | struct ofpbuf *odp_actions; |
7aa697dd BP |
5113 | |
5114 | trace.result = &result; | |
5115 | trace.flow = flow; | |
5116 | action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet); | |
5117 | trace.ctx.resubmit_hook = trace_resubmit; | |
cdee00fd BP |
5118 | odp_actions = xlate_actions(&trace.ctx, |
5119 | rule->actions, rule->n_actions); | |
7aa697dd BP |
5120 | |
5121 | ds_put_char(&result, '\n'); | |
5122 | trace_format_flow(&result, 0, "Final flow", &trace); | |
5123 | ds_put_cstr(&result, "Datapath actions: "); | |
cdee00fd BP |
5124 | format_odp_actions(&result, odp_actions->data, odp_actions->size); |
5125 | ofpbuf_delete(odp_actions); | |
7aa697dd BP |
5126 | } |
5127 | ||
5128 | unixctl_command_reply(conn, 200, ds_cstr(&result)); | |
5129 | ||
5130 | exit: | |
5131 | ds_destroy(&result); | |
5132 | ofpbuf_uninit(&packet); | |
5133 | free(args); | |
5134 | } | |
5135 | ||
5136 | static void | |
5137 | ofproto_unixctl_init(void) | |
5138 | { | |
5139 | static bool registered; | |
5140 | if (registered) { | |
5141 | return; | |
5142 | } | |
5143 | registered = true; | |
5144 | ||
5145 | unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL); | |
5146 | unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL); | |
5147 | } | |
5148 | \f | |
064af421 | 5149 | static bool |
ae412e7d | 5150 | default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet, |
cdee00fd | 5151 | struct ofpbuf *odp_actions, tag_type *tags, |
6a07af36 | 5152 | uint16_t *nf_output_iface, void *ofproto_) |
064af421 BP |
5153 | { |
5154 | struct ofproto *ofproto = ofproto_; | |
5155 | int out_port; | |
5156 | ||
5157 | /* Drop frames for reserved multicast addresses. */ | |
5158 | if (eth_addr_is_reserved(flow->dl_dst)) { | |
5159 | return true; | |
5160 | } | |
5161 | ||
5162 | /* Learn source MAC (but don't try to learn from revalidation). */ | |
5163 | if (packet != NULL) { | |
5164 | tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src, | |
7febb910 JG |
5165 | 0, flow->in_port, |
5166 | GRAT_ARP_LOCK_NONE); | |
064af421 BP |
5167 | if (rev_tag) { |
5168 | /* The log messages here could actually be useful in debugging, | |
5169 | * so keep the rate limit relatively high. */ | |
5170 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); | |
5171 | VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16, | |
5172 | ETH_ADDR_ARGS(flow->dl_src), flow->in_port); | |
5173 | ofproto_revalidate(ofproto, rev_tag); | |
5174 | } | |
5175 | } | |
5176 | ||
5177 | /* Determine output port. */ | |
7febb910 JG |
5178 | out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags, |
5179 | NULL); | |
064af421 | 5180 | if (out_port < 0) { |
f1588b1f | 5181 | flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD, |
cdee00fd | 5182 | nf_output_iface, odp_actions); |
064af421 | 5183 | } else if (out_port != flow->in_port) { |
cdee00fd | 5184 | nl_msg_put_u32(odp_actions, ODPAT_OUTPUT, out_port); |
6a07af36 | 5185 | *nf_output_iface = out_port; |
064af421 BP |
5186 | } else { |
5187 | /* Drop. */ | |
5188 | } | |
5189 | ||
5190 | return true; | |
5191 | } | |
5192 | ||
5193 | static const struct ofhooks default_ofhooks = { | |
064af421 BP |
5194 | default_normal_ofhook_cb, |
5195 | NULL, | |
5196 | NULL | |
5197 | }; |