2 * Copyright (c) 2009, 2010, 2011 Nicira Networks.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
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:
9 * http://www.apache.org/licenses/LICENSE-2.0
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.
22 #include <sys/socket.h>
24 #include <netinet/in.h>
28 #include "byte-order.h"
30 #include "classifier.h"
34 #include "dynamic-string.h"
35 #include "fail-open.h"
39 #include "mac-learning.h"
40 #include "multipath.h"
46 #include "ofp-print.h"
48 #include "ofproto-sflow.h"
50 #include "openflow/nicira-ext.h"
51 #include "openflow/openflow.h"
52 #include "openvswitch/datapath-protocol.h"
56 #include "poll-loop.h"
61 #include "stream-ssl.h"
65 #include "unaligned.h"
70 VLOG_DEFINE_THIS_MODULE(ofproto);
72 COVERAGE_DEFINE(facet_changed_rule);
73 COVERAGE_DEFINE(facet_revalidate);
74 COVERAGE_DEFINE(odp_overflow);
75 COVERAGE_DEFINE(ofproto_agg_request);
76 COVERAGE_DEFINE(ofproto_costly_flags);
77 COVERAGE_DEFINE(ofproto_ctlr_action);
78 COVERAGE_DEFINE(ofproto_del_rule);
79 COVERAGE_DEFINE(ofproto_error);
80 COVERAGE_DEFINE(ofproto_expiration);
81 COVERAGE_DEFINE(ofproto_expired);
82 COVERAGE_DEFINE(ofproto_flows_req);
83 COVERAGE_DEFINE(ofproto_flush);
84 COVERAGE_DEFINE(ofproto_invalidated);
85 COVERAGE_DEFINE(ofproto_no_packet_in);
86 COVERAGE_DEFINE(ofproto_ofp2odp);
87 COVERAGE_DEFINE(ofproto_packet_in);
88 COVERAGE_DEFINE(ofproto_packet_out);
89 COVERAGE_DEFINE(ofproto_queue_req);
90 COVERAGE_DEFINE(ofproto_recv_openflow);
91 COVERAGE_DEFINE(ofproto_reinit_ports);
92 COVERAGE_DEFINE(ofproto_unexpected_rule);
93 COVERAGE_DEFINE(ofproto_uninstallable);
94 COVERAGE_DEFINE(ofproto_update_port);
96 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
97 * flow translation. */
98 #define MAX_RESUBMIT_RECURSION 16
103 struct ofproto *ofproto; /* Owning ofproto. */
104 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
105 struct netdev *netdev;
106 struct ofp_phy_port opp;
108 struct cfm *cfm; /* Connectivity Fault Management, if any. */
111 static void ofport_free(struct ofport *);
112 static void ofport_run(struct ofport *);
113 static void ofport_wait(struct ofport *);
115 struct action_xlate_ctx {
116 /* action_xlate_ctx_init() initializes these members. */
119 struct ofproto *ofproto;
121 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
122 * this flow when actions change header fields. */
125 /* The packet corresponding to 'flow', or a null pointer if we are
126 * revalidating without a packet to refer to. */
127 const struct ofpbuf *packet;
129 /* If nonnull, called just before executing a resubmit action.
131 * This is normally null so the client has to set it manually after
132 * calling action_xlate_ctx_init(). */
133 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule *);
135 /* If true, the speciality of 'flow' should be checked before executing
136 * its actions. If special_cb returns false on 'flow' rendered
137 * uninstallable and no actions will be executed. */
140 /* xlate_actions() initializes and uses these members. The client might want
141 * to look at them after it returns. */
143 struct ofpbuf *odp_actions; /* Datapath actions. */
144 tag_type tags; /* Tags associated with OFPP_NORMAL actions. */
145 bool may_set_up_flow; /* True ordinarily; false if the actions must
146 * be reassessed for every packet. */
147 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
149 /* xlate_actions() initializes and uses these members, but the client has no
150 * reason to look at them. */
152 int recurse; /* Recursion level, via xlate_table_action. */
153 int last_pop_priority; /* Offset in 'odp_actions' just past most
154 * recent ODP_ACTION_ATTR_SET_PRIORITY. */
157 static void action_xlate_ctx_init(struct action_xlate_ctx *,
158 struct ofproto *, const struct flow *,
159 const struct ofpbuf *);
160 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
161 const union ofp_action *in, size_t n_in);
163 /* An OpenFlow flow. */
165 long long int used; /* Time last used; time created if not used. */
166 long long int created; /* Creation time. */
170 * - Do include packets and bytes from facets that have been deleted or
171 * whose own statistics have been folded into the rule.
173 * - Do include packets and bytes sent "by hand" that were accounted to
174 * the rule without any facet being involved (this is a rare corner
175 * case in rule_execute()).
177 * - Do not include packet or bytes that can be obtained from any facet's
178 * packet_count or byte_count member or that can be obtained from the
179 * datapath by, e.g., dpif_flow_get() for any facet.
181 uint64_t packet_count; /* Number of packets received. */
182 uint64_t byte_count; /* Number of bytes received. */
184 ovs_be64 flow_cookie; /* Controller-issued identifier. */
186 struct cls_rule cr; /* In owning ofproto's classifier. */
187 uint16_t idle_timeout; /* In seconds from time of last use. */
188 uint16_t hard_timeout; /* In seconds from time of creation. */
189 bool send_flow_removed; /* Send a flow removed message? */
190 int n_actions; /* Number of elements in actions[]. */
191 union ofp_action *actions; /* OpenFlow actions. */
192 struct list facets; /* List of "struct facet"s. */
195 static struct rule *rule_from_cls_rule(const struct cls_rule *);
196 static bool rule_is_hidden(const struct rule *);
198 static struct rule *rule_create(const struct cls_rule *,
199 const union ofp_action *, size_t n_actions,
200 uint16_t idle_timeout, uint16_t hard_timeout,
201 ovs_be64 flow_cookie, bool send_flow_removed);
202 static void rule_destroy(struct ofproto *, struct rule *);
203 static void rule_free(struct rule *);
205 static struct rule *rule_lookup(struct ofproto *, const struct flow *);
206 static void rule_insert(struct ofproto *, struct rule *);
207 static void rule_remove(struct ofproto *, struct rule *);
209 static void rule_send_removed(struct ofproto *, struct rule *, uint8_t reason);
210 static void rule_get_stats(const struct rule *, uint64_t *packets,
213 /* An exact-match instantiation of an OpenFlow flow. */
215 long long int used; /* Time last used; time created if not used. */
219 * - Do include packets and bytes sent "by hand", e.g. with
222 * - Do include packets and bytes that were obtained from the datapath
223 * when a flow was deleted (e.g. dpif_flow_del()) or when its
224 * statistics were reset (e.g. dpif_flow_put() with
225 * DPIF_FP_ZERO_STATS).
227 * - Do not include any packets or bytes that can currently be obtained
228 * from the datapath by, e.g., dpif_flow_get().
230 uint64_t packet_count; /* Number of packets received. */
231 uint64_t byte_count; /* Number of bytes received. */
233 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
234 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
236 uint64_t rs_packet_count; /* Packets pushed to resubmit children. */
237 uint64_t rs_byte_count; /* Bytes pushed to resubmit children. */
238 long long int rs_used; /* Used time pushed to resubmit children. */
240 /* Number of bytes passed to account_cb. This may include bytes that can
241 * currently obtained from the datapath (thus, it can be greater than
243 uint64_t accounted_bytes;
245 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
246 struct list list_node; /* In owning rule's 'facets' list. */
247 struct rule *rule; /* Owning rule. */
248 struct flow flow; /* Exact-match flow. */
249 bool installed; /* Installed in datapath? */
250 bool may_install; /* True ordinarily; false if actions must
251 * be reassessed for every packet. */
252 size_t actions_len; /* Number of bytes in actions[]. */
253 struct nlattr *actions; /* Datapath actions. */
254 tag_type tags; /* Tags (set only by hooks). */
255 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
258 static struct facet *facet_create(struct ofproto *, struct rule *,
260 const struct ofpbuf *packet);
261 static void facet_remove(struct ofproto *, struct facet *);
262 static void facet_free(struct facet *);
264 static struct facet *facet_lookup_valid(struct ofproto *, const struct flow *);
265 static bool facet_revalidate(struct ofproto *, struct facet *);
267 static void facet_install(struct ofproto *, struct facet *, bool zero_stats);
268 static void facet_uninstall(struct ofproto *, struct facet *);
269 static void facet_flush_stats(struct ofproto *, struct facet *);
271 static void facet_make_actions(struct ofproto *, struct facet *,
272 const struct ofpbuf *packet);
273 static void facet_update_stats(struct ofproto *, struct facet *,
274 const struct dpif_flow_stats *);
275 static void facet_push_stats(struct ofproto *, struct facet *);
277 static void send_packet_in(struct ofproto *, struct dpif_upcall *,
278 const struct flow *, bool clone);
281 char *name; /* Datapath name. */
282 struct hmap_node hmap_node; /* In global 'all_ofprotos' hmap. */
285 uint64_t datapath_id; /* Datapath ID. */
286 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
287 char *mfr_desc; /* Manufacturer. */
288 char *hw_desc; /* Hardware. */
289 char *sw_desc; /* Software version. */
290 char *serial_desc; /* Serial number. */
291 char *dp_desc; /* Datapath description. */
295 struct netdev_monitor *netdev_monitor;
296 struct hmap ports; /* Contains "struct ofport"s. */
297 struct shash port_by_name;
301 struct netflow *netflow;
302 struct ofproto_sflow *sflow;
305 struct classifier cls;
306 struct timer next_expiration;
310 bool need_revalidate;
311 struct tag_set revalidate_set;
313 /* OpenFlow connections. */
314 struct connmgr *connmgr;
316 /* Hooks for ovs-vswitchd. */
317 const struct ofhooks *ofhooks;
320 /* Used by default ofhooks. */
321 struct mac_learning *ml;
324 /* Map from dpif name to struct ofproto, for use by unixctl commands. */
325 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
327 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
329 static const struct ofhooks default_ofhooks;
331 static uint64_t pick_datapath_id(const struct ofproto *);
332 static uint64_t pick_fallback_dpid(void);
334 static void ofproto_flush_flows__(struct ofproto *);
335 static int ofproto_expire(struct ofproto *);
336 static void flow_push_stats(struct ofproto *, const struct rule *,
337 struct flow *, uint64_t packets, uint64_t bytes,
340 static void handle_upcall(struct ofproto *, struct dpif_upcall *);
342 static void handle_openflow(struct ofconn *, struct ofpbuf *);
344 static struct ofport *get_port(const struct ofproto *, uint16_t odp_port);
345 static void update_port(struct ofproto *, const char *devname);
346 static int init_ports(struct ofproto *);
347 static void reinit_ports(struct ofproto *);
349 static void ofproto_unixctl_init(void);
352 ofproto_create(const char *datapath, const char *datapath_type,
353 const struct ofhooks *ofhooks, void *aux,
354 struct ofproto **ofprotop)
356 char local_name[IF_NAMESIZE];
363 ofproto_unixctl_init();
365 /* Connect to datapath and start listening for messages. */
366 error = dpif_create_and_open(datapath, datapath_type, &dpif);
368 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
371 error = dpif_recv_set_mask(dpif,
372 ((1u << DPIF_UC_MISS) |
373 (1u << DPIF_UC_ACTION) |
374 (1u << DPIF_UC_SAMPLE)));
376 VLOG_ERR("failed to listen on datapath %s: %s",
377 datapath, strerror(error));
381 dpif_flow_flush(dpif);
382 dpif_recv_purge(dpif);
384 error = dpif_port_get_name(dpif, ODPP_LOCAL,
385 local_name, sizeof local_name);
387 VLOG_ERR("%s: cannot get name of datapath local port (%s)",
388 datapath, strerror(error));
392 /* Initialize settings. */
393 p = xzalloc(sizeof *p);
394 p->name = xstrdup(dpif_name(dpif));
395 hmap_insert(&all_ofprotos, &p->hmap_node, hash_string(p->name, 0));
396 p->fallback_dpid = pick_fallback_dpid();
397 p->datapath_id = p->fallback_dpid;
398 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
399 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
400 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
401 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
402 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
404 /* Initialize datapath. */
406 p->netdev_monitor = netdev_monitor_create();
407 hmap_init(&p->ports);
408 shash_init(&p->port_by_name);
409 p->max_ports = dpif_get_max_ports(dpif);
411 /* Initialize submodules. */
415 /* Initialize flow table. */
416 classifier_init(&p->cls);
417 timer_set_duration(&p->next_expiration, 1000);
419 /* Initialize facet table. */
420 hmap_init(&p->facets);
421 p->need_revalidate = false;
422 tag_set_init(&p->revalidate_set);
424 /* Initialize hooks. */
426 p->ofhooks = ofhooks;
430 p->ofhooks = &default_ofhooks;
432 p->ml = mac_learning_create();
435 /* Pick final datapath ID. */
436 p->datapath_id = pick_datapath_id(p);
437 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
439 /* Initialize OpenFlow connections. */
440 p->connmgr = connmgr_create(p, datapath, local_name);
449 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
451 uint64_t old_dpid = p->datapath_id;
452 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
453 if (p->datapath_id != old_dpid) {
454 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
456 /* Force all active connections to reconnect, since there is no way to
457 * notify a controller that the datapath ID has changed. */
458 ofproto_reconnect_controllers(p);
463 ofproto_set_controllers(struct ofproto *p,
464 const struct ofproto_controller *controllers,
465 size_t n_controllers)
467 connmgr_set_controllers(p->connmgr, controllers, n_controllers);
471 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
473 connmgr_set_fail_mode(p->connmgr, fail_mode);
476 /* Drops the connections between 'ofproto' and all of its controllers, forcing
477 * them to reconnect. */
479 ofproto_reconnect_controllers(struct ofproto *ofproto)
481 connmgr_reconnect(ofproto->connmgr);
484 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
485 * in-band control should guarantee access, in the same way that in-band
486 * control guarantees access to OpenFlow controllers. */
488 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
489 const struct sockaddr_in *extras, size_t n)
491 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
494 /* Sets the OpenFlow queue used by flows set up by in-band control on
495 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
496 * flows will use the default queue. */
498 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
500 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
504 ofproto_set_desc(struct ofproto *p,
505 const char *mfr_desc, const char *hw_desc,
506 const char *sw_desc, const char *serial_desc,
509 struct ofp_desc_stats *ods;
512 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
513 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
514 sizeof ods->mfr_desc);
517 p->mfr_desc = xstrdup(mfr_desc);
520 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
521 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
522 sizeof ods->hw_desc);
525 p->hw_desc = xstrdup(hw_desc);
528 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
529 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
530 sizeof ods->sw_desc);
533 p->sw_desc = xstrdup(sw_desc);
536 if (strlen(serial_desc) >= sizeof ods->serial_num) {
537 VLOG_WARN("truncating serial_desc, must be less than %zu "
539 sizeof ods->serial_num);
541 free(p->serial_desc);
542 p->serial_desc = xstrdup(serial_desc);
545 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
546 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
547 sizeof ods->dp_desc);
550 p->dp_desc = xstrdup(dp_desc);
555 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
557 return connmgr_set_snoops(ofproto->connmgr, snoops);
561 ofproto_set_netflow(struct ofproto *ofproto,
562 const struct netflow_options *nf_options)
564 if (nf_options && !sset_is_empty(&nf_options->collectors)) {
565 if (!ofproto->netflow) {
566 ofproto->netflow = netflow_create();
568 return netflow_set_options(ofproto->netflow, nf_options);
570 netflow_destroy(ofproto->netflow);
571 ofproto->netflow = NULL;
577 ofproto_set_sflow(struct ofproto *ofproto,
578 const struct ofproto_sflow_options *oso)
580 struct ofproto_sflow *os = ofproto->sflow;
583 struct ofport *ofport;
585 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
586 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
587 ofproto_sflow_add_port(os, ofport->odp_port,
588 netdev_get_name(ofport->netdev));
591 ofproto_sflow_set_options(os, oso);
593 ofproto_sflow_destroy(os);
594 ofproto->sflow = NULL;
598 /* Connectivity Fault Management configuration. */
600 /* Clears the CFM configuration from 'port_no' on 'ofproto'. */
602 ofproto_port_clear_cfm(struct ofproto *ofproto, uint32_t port_no)
604 struct ofport *ofport = get_port(ofproto, port_no);
605 if (ofport && ofport->cfm){
606 cfm_destroy(ofport->cfm);
611 /* Configures connectivity fault management on 'port_no' in 'ofproto'. Takes
612 * basic configuration from the configuration members in 'cfm', and the set of
613 * remote maintenance points from the 'n_remote_mps' elements in 'remote_mps'.
614 * Ignores the statistics members of 'cfm'.
616 * This function has no effect if 'ofproto' does not have a port 'port_no'. */
618 ofproto_port_set_cfm(struct ofproto *ofproto, uint32_t port_no,
619 const struct cfm *cfm,
620 const uint16_t *remote_mps, size_t n_remote_mps)
622 struct ofport *ofport;
624 ofport = get_port(ofproto, port_no);
626 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu32,
627 ofproto->name, port_no);
632 ofport->cfm = cfm_create();
635 ofport->cfm->mpid = cfm->mpid;
636 ofport->cfm->interval = cfm->interval;
637 memcpy(ofport->cfm->maid, cfm->maid, CCM_MAID_LEN);
639 cfm_update_remote_mps(ofport->cfm, remote_mps, n_remote_mps);
641 if (!cfm_configure(ofport->cfm)) {
642 VLOG_WARN("%s: CFM configuration on port %"PRIu32" (%s) failed",
643 ofproto->name, port_no,
644 netdev_get_name(ofport->netdev));
645 cfm_destroy(ofport->cfm);
650 /* Returns the connectivity fault management object associated with 'port_no'
651 * within 'ofproto', or a null pointer if 'ofproto' does not have a port
652 * 'port_no' or if that port does not have CFM configured. The caller must not
653 * modify or destroy the returned object. */
655 ofproto_port_get_cfm(struct ofproto *ofproto, uint32_t port_no)
657 struct ofport *ofport = get_port(ofproto, port_no);
658 return ofport ? ofport->cfm : NULL;
662 ofproto_get_datapath_id(const struct ofproto *ofproto)
664 return ofproto->datapath_id;
667 enum ofproto_fail_mode
668 ofproto_get_fail_mode(const struct ofproto *p)
670 return connmgr_get_fail_mode(p->connmgr);
674 ofproto_has_snoops(const struct ofproto *ofproto)
676 return connmgr_has_snoops(ofproto->connmgr);
680 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
682 connmgr_get_snoops(ofproto->connmgr, snoops);
686 ofproto_destroy__(struct ofproto *p, bool delete)
688 struct ofport *ofport, *next_ofport;
694 hmap_remove(&all_ofprotos, &p->hmap_node);
696 ofproto_flush_flows__(p);
697 connmgr_destroy(p->connmgr);
698 classifier_destroy(&p->cls);
699 hmap_destroy(&p->facets);
702 int error = dpif_delete(p->dpif);
703 if (error && error != ENOENT) {
704 VLOG_ERR("bridge %s: failed to destroy (%s)",
705 p->name, strerror(error));
710 netdev_monitor_destroy(p->netdev_monitor);
711 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
712 hmap_remove(&p->ports, &ofport->hmap_node);
715 shash_destroy(&p->port_by_name);
717 netflow_destroy(p->netflow);
718 ofproto_sflow_destroy(p->sflow);
720 mac_learning_destroy(p->ml);
725 free(p->serial_desc);
728 hmap_destroy(&p->ports);
735 ofproto_destroy(struct ofproto *p)
737 ofproto_destroy__(p, false);
741 ofproto_destroy_and_delete(struct ofproto *p)
743 ofproto_destroy__(p, true);
747 ofproto_run(struct ofproto *p)
749 int error = ofproto_run1(p);
751 error = ofproto_run2(p, false);
757 process_port_change(struct ofproto *ofproto, int error, char *devname)
759 if (error == ENOBUFS) {
760 reinit_ports(ofproto);
762 update_port(ofproto, devname);
768 ofproto_run1(struct ofproto *p)
770 struct ofport *ofport;
775 for (i = 0; i < 50; i++) {
776 struct dpif_upcall packet;
778 error = dpif_recv(p->dpif, &packet);
780 if (error == ENODEV) {
781 /* Someone destroyed the datapath behind our back. The caller
782 * better destroy us and give up, because we're just going to
783 * spin from here on out. */
784 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
785 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
792 handle_upcall(p, &packet);
795 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
796 process_port_change(p, error, devname);
798 while ((error = netdev_monitor_poll(p->netdev_monitor,
799 &devname)) != EAGAIN) {
800 process_port_change(p, error, devname);
803 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
807 connmgr_run(p->connmgr, handle_openflow);
809 if (timer_expired(&p->next_expiration)) {
810 int delay = ofproto_expire(p);
811 timer_set_duration(&p->next_expiration, delay);
812 COVERAGE_INC(ofproto_expiration);
816 netflow_run(p->netflow);
819 ofproto_sflow_run(p->sflow);
826 ofproto_run2(struct ofproto *p, bool revalidate_all)
828 /* Figure out what we need to revalidate now, if anything. */
829 struct tag_set revalidate_set = p->revalidate_set;
830 if (p->need_revalidate) {
831 revalidate_all = true;
834 /* Clear the revalidation flags. */
835 tag_set_init(&p->revalidate_set);
836 p->need_revalidate = false;
838 /* Now revalidate if there's anything to do. */
839 if (revalidate_all || !tag_set_is_empty(&revalidate_set)) {
840 struct facet *facet, *next;
842 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &p->facets) {
844 || tag_set_intersects(&revalidate_set, facet->tags)) {
845 facet_revalidate(p, facet);
854 ofproto_wait(struct ofproto *p)
856 struct ofport *ofport;
858 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
861 dpif_recv_wait(p->dpif);
862 dpif_port_poll_wait(p->dpif);
863 netdev_monitor_poll_wait(p->netdev_monitor);
865 ofproto_sflow_wait(p->sflow);
867 if (!tag_set_is_empty(&p->revalidate_set)) {
868 poll_immediate_wake();
870 if (p->need_revalidate) {
871 /* Shouldn't happen, but if it does just go around again. */
872 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
873 poll_immediate_wake();
875 timer_wait(&p->next_expiration);
877 connmgr_wait(p->connmgr);
881 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
883 tag_set_add(&ofproto->revalidate_set, tag);
887 ofproto_get_revalidate_set(struct ofproto *ofproto)
889 return &ofproto->revalidate_set;
893 ofproto_is_alive(const struct ofproto *p)
895 return connmgr_has_controllers(p->connmgr);
899 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
902 connmgr_get_controller_info(ofproto->connmgr, info);
906 ofproto_free_ofproto_controller_info(struct shash *info)
908 struct shash_node *node;
910 SHASH_FOR_EACH (node, info) {
911 struct ofproto_controller_info *cinfo = node->data;
912 while (cinfo->pairs.n) {
913 free((char *) cinfo->pairs.values[--cinfo->pairs.n]);
920 /* Makes a deep copy of 'old' into 'port'. */
922 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
924 port->name = xstrdup(old->name);
925 port->type = xstrdup(old->type);
926 port->ofp_port = old->ofp_port;
929 /* Frees memory allocated to members of 'ofproto_port'.
931 * Do not call this function on an ofproto_port obtained from
932 * ofproto_port_dump_next(): that function retains ownership of the data in the
935 ofproto_port_destroy(struct ofproto_port *ofproto_port)
937 free(ofproto_port->name);
938 free(ofproto_port->type);
941 /* Converts a dpif_port into an ofproto_port.
943 * This only makes a shallow copy, so make sure that the dpif_port doesn't get
944 * freed while the ofproto_port is still in use. You can choose to free the
945 * ofproto_port instead of the dpif_port. */
947 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
948 struct dpif_port *dpif_port)
950 ofproto_port->name = dpif_port->name;
951 ofproto_port->type = dpif_port->type;
952 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
955 /* Initializes 'dump' to begin dumping the ports in an ofproto.
957 * This function provides no status indication. An error status for the entire
958 * dump operation is provided when it is completed by calling
959 * ofproto_port_dump_done().
962 ofproto_port_dump_start(struct ofproto_port_dump *dump,
963 const struct ofproto *ofproto)
965 struct dpif_port_dump *dpif_dump;
967 dump->state = dpif_dump = xmalloc(sizeof *dpif_dump);
968 dpif_port_dump_start(dpif_dump, ofproto->dpif);
971 /* Attempts to retrieve another port from 'dump', which must have been created
972 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
973 * 'port' and returns true. On failure, returns false.
975 * Failure might indicate an actual error or merely that the last port has been
976 * dumped. An error status for the entire dump operation is provided when it
977 * is completed by calling ofproto_port_dump_done().
979 * The ofproto owns the data stored in 'port'. It will remain valid until at
980 * least the next time 'dump' is passed to ofproto_port_dump_next() or
981 * ofproto_port_dump_done(). */
983 ofproto_port_dump_next(struct ofproto_port_dump *dump,
984 struct ofproto_port *port)
986 struct dpif_port_dump *dpif_dump = dump->state;
987 struct dpif_port dpif_port;
990 ok = dpif_port_dump_next(dpif_dump, &dpif_port);
992 ofproto_port_from_dpif_port(port, &dpif_port);
997 /* Completes port table dump operation 'dump', which must have been created
998 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
999 * error-free, otherwise a positive errno value describing the problem. */
1001 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1003 struct dpif_port_dump *dpif_dump = dump->state;
1004 int error = dpif_port_dump_done(dpif_dump);
1009 /* Attempts to add 'netdev' as a port on 'ofproto'. If successful, returns 0
1010 * and sets '*ofp_portp' to the new port's port OpenFlow number (if 'ofp_portp'
1011 * is non-null). On failure, returns a positive errno value and sets
1012 * '*ofp_portp' to OFPP_NONE (if 'ofp_portp' is non-null). */
1014 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1015 uint16_t *ofp_portp)
1020 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
1022 update_port(ofproto, netdev_get_name(netdev));
1025 *ofp_portp = error ? OFPP_NONE : odp_port_to_ofp_port(odp_port);
1030 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1031 * initializes '*port' appropriately; on failure, returns a positive errno
1034 * The caller owns the data in 'port' and must free it with
1035 * ofproto_port_destroy() when it is no longer needed. */
1037 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1038 struct ofproto_port *port)
1040 struct dpif_port dpif_port;
1043 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
1045 ofproto_port_from_dpif_port(port, &dpif_port);
1050 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1052 * This is almost the same as calling dpif_port_del() directly on the
1053 * datapath, but it also makes 'ofproto' close its open netdev for the port
1054 * (if any). This makes it possible to create a new netdev of a different
1055 * type under the same name, which otherwise the netdev library would refuse
1056 * to do because of the conflict. (The netdev would eventually get closed on
1057 * the next trip through ofproto_run(), but this interface is more direct.)
1059 * Returns 0 if successful, otherwise a positive errno. */
1061 ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port)
1063 uint32_t odp_port = ofp_port_to_odp_port(ofp_port);
1064 struct ofport *ofport = get_port(ofproto, odp_port);
1065 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1068 error = dpif_port_del(ofproto->dpif, odp_port);
1070 VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)",
1071 ofproto->name, odp_port, name, strerror(error));
1072 } else if (ofport) {
1073 /* 'name' is the netdev's name and update_port() is going to close the
1074 * netdev. Just in case update_port() refers to 'name' after it
1075 * destroys 'ofport', make a copy of it around the update_port()
1077 char *devname = xstrdup(name);
1078 update_port(ofproto, devname);
1084 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
1085 * true if 'odp_port' exists and should be included, false otherwise. */
1087 ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port)
1089 struct ofport *ofport = get_port(ofproto, odp_port);
1090 return ofport && !(ofport->opp.config & htonl(OFPPC_NO_FLOOD));
1093 /* Sends 'packet' out of port 'port_no' within 'p'. If 'vlan_tci' is zero the
1094 * packet will not have any 802.1Q hader; if it is nonzero, then the packet
1095 * will be sent with the VLAN TCI specified by 'vlan_tci & ~VLAN_CFI'.
1097 * Returns 0 if successful, otherwise a positive errno value. */
1099 ofproto_send_packet(struct ofproto *ofproto,
1100 uint32_t port_no, uint16_t vlan_tci,
1101 const struct ofpbuf *packet)
1103 struct ofpbuf odp_actions;
1106 ofpbuf_init(&odp_actions, 32);
1107 if (vlan_tci != 0) {
1108 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
1109 ntohs(vlan_tci & ~VLAN_CFI));
1111 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, port_no);
1112 error = dpif_execute(ofproto->dpif, odp_actions.data, odp_actions.size,
1114 ofpbuf_uninit(&odp_actions);
1117 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
1118 ofproto->name, port_no, strerror(error));
1123 /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and
1124 * performs the 'n_actions' actions in 'actions'. The new flow will not
1127 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1128 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1129 * controllers; otherwise, it will be hidden.
1131 * The caller retains ownership of 'cls_rule' and 'actions'. */
1133 ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule,
1134 const union ofp_action *actions, size_t n_actions)
1137 rule = rule_create(cls_rule, actions, n_actions, 0, 0, 0, false);
1138 rule_insert(p, rule);
1142 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1146 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1149 rule_remove(ofproto, rule);
1154 ofproto_flush_flows__(struct ofproto *ofproto)
1156 struct facet *facet, *next_facet;
1157 struct rule *rule, *next_rule;
1158 struct cls_cursor cursor;
1160 COVERAGE_INC(ofproto_flush);
1162 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1163 /* Mark the facet as not installed so that facet_remove() doesn't
1164 * bother trying to uninstall it. There is no point in uninstalling it
1165 * individually since we are about to blow away all the facets with
1166 * dpif_flow_flush(). */
1167 facet->installed = false;
1168 facet->dp_packet_count = 0;
1169 facet->dp_byte_count = 0;
1170 facet_remove(ofproto, facet);
1173 cls_cursor_init(&cursor, &ofproto->cls, NULL);
1174 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1175 rule_remove(ofproto, rule);
1178 dpif_flow_flush(ofproto->dpif);
1182 ofproto_flush_flows(struct ofproto *ofproto)
1184 ofproto_flush_flows__(ofproto);
1185 connmgr_flushed(ofproto->connmgr);
1189 reinit_ports(struct ofproto *p)
1191 struct dpif_port_dump dump;
1192 struct sset devnames;
1193 struct ofport *ofport;
1194 struct dpif_port dpif_port;
1195 const char *devname;
1197 COVERAGE_INC(ofproto_reinit_ports);
1199 sset_init(&devnames);
1200 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1201 sset_add(&devnames, netdev_get_name(ofport->netdev));
1203 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1204 sset_add(&devnames, dpif_port.name);
1207 SSET_FOR_EACH (devname, &devnames) {
1208 update_port(p, devname);
1210 sset_destroy(&devnames);
1213 /* Opens and returns a netdev for 'dpif_port', or a null pointer if the netdev
1214 * cannot be opened. On success, also fills in 'opp'. */
1215 static struct netdev *
1216 ofport_open(const struct dpif_port *dpif_port, struct ofp_phy_port *opp)
1218 uint32_t curr, advertised, supported, peer;
1219 struct netdev_options netdev_options;
1220 enum netdev_flags flags;
1221 struct netdev *netdev;
1224 memset(&netdev_options, 0, sizeof netdev_options);
1225 netdev_options.name = dpif_port->name;
1226 netdev_options.type = dpif_port->type;
1227 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1229 error = netdev_open(&netdev_options, &netdev);
1231 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1232 "cannot be opened (%s)",
1233 dpif_port->name, dpif_port->port_no,
1234 dpif_port->name, strerror(error));
1238 netdev_get_flags(netdev, &flags);
1239 netdev_get_features(netdev, &curr, &advertised, &supported, &peer);
1241 opp->port_no = htons(odp_port_to_ofp_port(dpif_port->port_no));
1242 netdev_get_etheraddr(netdev, opp->hw_addr);
1243 ovs_strzcpy(opp->name, dpif_port->name, sizeof opp->name);
1244 opp->config = flags & NETDEV_UP ? 0 : htonl(OFPPC_PORT_DOWN);
1245 opp->state = netdev_get_carrier(netdev) ? 0 : htonl(OFPPS_LINK_DOWN);
1246 opp->curr = htonl(curr);
1247 opp->advertised = htonl(advertised);
1248 opp->supported = htonl(supported);
1249 opp->peer = htonl(peer);
1255 ofport_conflicts(const struct ofproto *p, const struct dpif_port *dpif_port)
1257 if (get_port(p, dpif_port->port_no)) {
1258 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1259 dpif_port->port_no);
1261 } else if (shash_find(&p->port_by_name, dpif_port->name)) {
1262 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1270 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1271 * port number, and 'config' bits other than OFPPC_PORT_DOWN are
1274 ofport_equal(const struct ofp_phy_port *a, const struct ofp_phy_port *b)
1276 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1277 return (!memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1278 && a->state == b->state
1279 && !((a->config ^ b->config) & htonl(OFPPC_PORT_DOWN))
1280 && a->curr == b->curr
1281 && a->advertised == b->advertised
1282 && a->supported == b->supported
1283 && a->peer == b->peer);
1286 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1287 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1288 * one with the same name or port number). */
1290 ofport_install(struct ofproto *p,
1291 struct netdev *netdev, const struct ofp_phy_port *opp)
1293 const char *netdev_name = netdev_get_name(netdev);
1294 struct ofport *ofport;
1296 connmgr_send_port_status(p->connmgr, opp, OFPPR_ADD);
1298 /* Create ofport. */
1299 ofport = xmalloc(sizeof *ofport);
1300 ofport->ofproto = p;
1301 ofport->netdev = netdev;
1303 ofport->odp_port = ofp_port_to_odp_port(ntohs(opp->port_no));
1306 /* Add port to 'p'. */
1307 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1308 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
1309 shash_add(&p->port_by_name, netdev_name, ofport);
1311 ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
1315 /* Removes 'ofport' from 'p' and destroys it. */
1317 ofport_remove(struct ofport *ofport)
1319 struct ofproto *p = ofport->ofproto;
1321 connmgr_send_port_status(p->connmgr, &ofport->opp, OFPPR_DELETE);
1323 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1324 hmap_remove(&p->ports, &ofport->hmap_node);
1325 shash_delete(&p->port_by_name,
1326 shash_find(&p->port_by_name,
1327 netdev_get_name(ofport->netdev)));
1329 ofproto_sflow_del_port(p->sflow, ofport->odp_port);
1332 ofport_free(ofport);
1335 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1338 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1340 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1342 ofport_remove(port);
1346 /* Updates 'port' within 'ofproto' with the new 'netdev' and 'opp'.
1348 * Does not handle a name or port number change. The caller must implement
1349 * such a change as a delete followed by an add. */
1351 ofport_modified(struct ofport *port,
1352 struct netdev *netdev, struct ofp_phy_port *opp)
1354 struct ofproto *ofproto = port->ofproto;
1356 memcpy(port->opp.hw_addr, opp->hw_addr, ETH_ADDR_LEN);
1357 port->opp.config = ((port->opp.config & ~htonl(OFPPC_PORT_DOWN))
1358 | (opp->config & htonl(OFPPC_PORT_DOWN)));
1359 port->opp.state = opp->state;
1360 port->opp.curr = opp->curr;
1361 port->opp.advertised = opp->advertised;
1362 port->opp.supported = opp->supported;
1363 port->opp.peer = opp->peer;
1365 netdev_monitor_remove(ofproto->netdev_monitor, port->netdev);
1366 netdev_monitor_add(ofproto->netdev_monitor, netdev);
1368 netdev_close(port->netdev);
1369 port->netdev = netdev;
1371 connmgr_send_port_status(ofproto->connmgr, &port->opp, OFPPR_MODIFY);
1375 ofport_run(struct ofport *ofport)
1378 cfm_run(ofport->cfm);
1380 if (cfm_should_send_ccm(ofport->cfm)) {
1381 struct ofpbuf packet;
1384 ofpbuf_init(&packet, 0);
1385 ccm = eth_compose(&packet, eth_addr_ccm, ofport->opp.hw_addr,
1386 ETH_TYPE_CFM, sizeof *ccm);
1387 cfm_compose_ccm(ofport->cfm, ccm);
1388 ofproto_send_packet(ofport->ofproto, ofport->odp_port, 0, &packet);
1389 ofpbuf_uninit(&packet);
1395 ofport_wait(struct ofport *ofport)
1398 cfm_wait(ofport->cfm);
1403 ofport_free(struct ofport *ofport)
1406 cfm_destroy(ofport->cfm);
1407 netdev_close(ofport->netdev);
1412 static struct ofport *
1413 get_port(const struct ofproto *ofproto, uint16_t odp_port)
1415 struct ofport *port;
1417 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1418 hash_int(odp_port, 0), &ofproto->ports) {
1419 if (port->odp_port == odp_port) {
1427 update_port(struct ofproto *ofproto, const char *name)
1429 struct dpif_port dpif_port;
1430 struct ofp_phy_port opp;
1431 struct netdev *netdev;
1432 struct ofport *port;
1434 COVERAGE_INC(ofproto_update_port);
1436 /* Fetch 'name''s location and properties from the datapath. */
1437 netdev = (!dpif_port_query_by_name(ofproto->dpif, name, &dpif_port)
1438 ? ofport_open(&dpif_port, &opp)
1441 port = get_port(ofproto, dpif_port.port_no);
1442 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
1443 /* 'name' hasn't changed location. Any properties changed? */
1444 if (!ofport_equal(&port->opp, &opp)) {
1445 ofport_modified(port, netdev, &opp);
1447 netdev_close(netdev);
1450 /* If 'port' is nonnull then its name differs from 'name' and thus
1451 * we should delete it. If we think there's a port named 'name'
1452 * then its port number must be wrong now so delete it too. */
1454 ofport_remove(port);
1456 ofport_remove_with_name(ofproto, name);
1457 ofport_install(ofproto, netdev, &opp);
1460 /* Any port named 'name' is gone now. */
1461 ofport_remove_with_name(ofproto, name);
1463 dpif_port_destroy(&dpif_port);
1467 init_ports(struct ofproto *p)
1469 struct dpif_port_dump dump;
1470 struct dpif_port dpif_port;
1472 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1473 if (!ofport_conflicts(p, &dpif_port)) {
1474 struct ofp_phy_port opp;
1475 struct netdev *netdev;
1477 netdev = ofport_open(&dpif_port, &opp);
1479 ofport_install(p, netdev, &opp);
1487 /* Returns true if 'rule' should be hidden from the controller.
1489 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1490 * (e.g. by in-band control) and are intentionally hidden from the
1493 rule_is_hidden(const struct rule *rule)
1495 return rule->cr.priority > UINT16_MAX;
1498 /* Creates and returns a new rule initialized as specified.
1500 * The caller is responsible for inserting the rule into the classifier (with
1501 * rule_insert()). */
1502 static struct rule *
1503 rule_create(const struct cls_rule *cls_rule,
1504 const union ofp_action *actions, size_t n_actions,
1505 uint16_t idle_timeout, uint16_t hard_timeout,
1506 ovs_be64 flow_cookie, bool send_flow_removed)
1508 struct rule *rule = xzalloc(sizeof *rule);
1509 rule->cr = *cls_rule;
1510 rule->idle_timeout = idle_timeout;
1511 rule->hard_timeout = hard_timeout;
1512 rule->flow_cookie = flow_cookie;
1513 rule->used = rule->created = time_msec();
1514 rule->send_flow_removed = send_flow_removed;
1515 list_init(&rule->facets);
1516 if (n_actions > 0) {
1517 rule->n_actions = n_actions;
1518 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1524 static struct rule *
1525 rule_from_cls_rule(const struct cls_rule *cls_rule)
1527 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1531 rule_free(struct rule *rule)
1533 free(rule->actions);
1537 /* Destroys 'rule' and iterates through all of its facets and revalidates them,
1538 * destroying any that no longer has a rule (which is probably all of them).
1540 * The caller must have already removed 'rule' from the classifier. */
1542 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1544 struct facet *facet, *next_facet;
1545 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
1546 facet_revalidate(ofproto, facet);
1551 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
1552 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
1555 rule_has_out_port(const struct rule *rule, ovs_be16 out_port)
1557 const union ofp_action *oa;
1558 struct actions_iterator i;
1560 if (out_port == htons(OFPP_NONE)) {
1563 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1564 oa = actions_next(&i)) {
1565 if (action_outputs_to_port(oa, out_port)) {
1572 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
1573 * 'packet', which arrived on 'in_port'.
1575 * Takes ownership of 'packet'. */
1577 execute_odp_actions(struct ofproto *ofproto, const struct flow *flow,
1578 const struct nlattr *odp_actions, size_t actions_len,
1579 struct ofpbuf *packet)
1581 if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t))
1582 && odp_actions->nla_type == ODP_ACTION_ATTR_CONTROLLER) {
1583 /* As an optimization, avoid a round-trip from userspace to kernel to
1584 * userspace. This also avoids possibly filling up kernel packet
1585 * buffers along the way. */
1586 struct dpif_upcall upcall;
1588 upcall.type = DPIF_UC_ACTION;
1589 upcall.packet = packet;
1592 upcall.userdata = nl_attr_get_u64(odp_actions);
1593 upcall.sample_pool = 0;
1594 upcall.actions = NULL;
1595 upcall.actions_len = 0;
1597 send_packet_in(ofproto, &upcall, flow, false);
1603 error = dpif_execute(ofproto->dpif, odp_actions, actions_len, packet);
1604 ofpbuf_delete(packet);
1609 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
1610 * statistics appropriately. 'packet' must have at least sizeof(struct
1611 * ofp_packet_in) bytes of headroom.
1613 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
1614 * applying flow_extract() to 'packet' would yield the same flow as
1617 * 'facet' must have accurately composed ODP actions; that is, it must not be
1618 * in need of revalidation.
1620 * Takes ownership of 'packet'. */
1622 facet_execute(struct ofproto *ofproto, struct facet *facet,
1623 struct ofpbuf *packet)
1625 struct dpif_flow_stats stats;
1627 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1629 flow_extract_stats(&facet->flow, packet, &stats);
1630 stats.used = time_msec();
1631 if (execute_odp_actions(ofproto, &facet->flow,
1632 facet->actions, facet->actions_len, packet)) {
1633 facet_update_stats(ofproto, facet, &stats);
1637 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
1638 * statistics (or the statistics for one of its facets) appropriately.
1639 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom.
1641 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
1642 * with statistics for 'packet' either way.
1644 * Takes ownership of 'packet'. */
1646 rule_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port,
1647 struct ofpbuf *packet)
1649 struct action_xlate_ctx ctx;
1650 struct ofpbuf *odp_actions;
1651 struct facet *facet;
1655 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1657 flow_extract(packet, 0, in_port, &flow);
1659 /* First look for a related facet. If we find one, account it to that. */
1660 facet = facet_lookup_valid(ofproto, &flow);
1661 if (facet && facet->rule == rule) {
1662 facet_execute(ofproto, facet, packet);
1666 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
1667 * create a new facet for it and use that. */
1668 if (rule_lookup(ofproto, &flow) == rule) {
1669 facet = facet_create(ofproto, rule, &flow, packet);
1670 facet_execute(ofproto, facet, packet);
1671 facet_install(ofproto, facet, true);
1675 /* We can't account anything to a facet. If we were to try, then that
1676 * facet would have a non-matching rule, busting our invariants. */
1677 action_xlate_ctx_init(&ctx, ofproto, &flow, packet);
1678 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
1679 size = packet->size;
1680 if (execute_odp_actions(ofproto, &flow, odp_actions->data,
1681 odp_actions->size, packet)) {
1682 rule->used = time_msec();
1683 rule->packet_count++;
1684 rule->byte_count += size;
1685 flow_push_stats(ofproto, rule, &flow, 1, size, rule->used);
1687 ofpbuf_delete(odp_actions);
1690 /* Inserts 'rule' into 'p''s flow table. */
1692 rule_insert(struct ofproto *p, struct rule *rule)
1694 struct rule *displaced_rule;
1696 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1697 if (displaced_rule) {
1698 rule_destroy(p, displaced_rule);
1700 p->need_revalidate = true;
1703 /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a
1704 * 'flow' and an example 'packet' within that flow.
1706 * The caller must already have determined that no facet with an identical
1707 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
1708 * 'ofproto''s classifier table. */
1709 static struct facet *
1710 facet_create(struct ofproto *ofproto, struct rule *rule,
1711 const struct flow *flow, const struct ofpbuf *packet)
1713 struct facet *facet;
1715 facet = xzalloc(sizeof *facet);
1716 facet->used = time_msec();
1717 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
1718 list_push_back(&rule->facets, &facet->list_node);
1720 facet->flow = *flow;
1721 netflow_flow_init(&facet->nf_flow);
1722 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
1724 facet_make_actions(ofproto, facet, packet);
1730 facet_free(struct facet *facet)
1732 free(facet->actions);
1736 /* Remove 'rule' from 'ofproto' and free up the associated memory:
1738 * - Removes 'rule' from the classifier.
1740 * - If 'rule' has facets, revalidates them (and possibly uninstalls and
1741 * destroys them), via rule_destroy().
1744 rule_remove(struct ofproto *ofproto, struct rule *rule)
1746 COVERAGE_INC(ofproto_del_rule);
1747 ofproto->need_revalidate = true;
1748 classifier_remove(&ofproto->cls, &rule->cr);
1749 rule_destroy(ofproto, rule);
1752 /* Remove 'facet' from 'ofproto' and free up the associated memory:
1754 * - If 'facet' was installed in the datapath, uninstalls it and updates its
1755 * rule's statistics, via facet_uninstall().
1757 * - Removes 'facet' from its rule and from ofproto->facets.
1760 facet_remove(struct ofproto *ofproto, struct facet *facet)
1762 facet_uninstall(ofproto, facet);
1763 facet_flush_stats(ofproto, facet);
1764 hmap_remove(&ofproto->facets, &facet->hmap_node);
1765 list_remove(&facet->list_node);
1769 /* Composes the ODP actions for 'facet' based on its rule's actions. */
1771 facet_make_actions(struct ofproto *p, struct facet *facet,
1772 const struct ofpbuf *packet)
1774 const struct rule *rule = facet->rule;
1775 struct ofpbuf *odp_actions;
1776 struct action_xlate_ctx ctx;
1778 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
1779 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
1780 facet->tags = ctx.tags;
1781 facet->may_install = ctx.may_set_up_flow;
1782 facet->nf_flow.output_iface = ctx.nf_output_iface;
1784 if (facet->actions_len != odp_actions->size
1785 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
1786 free(facet->actions);
1787 facet->actions_len = odp_actions->size;
1788 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
1791 ofpbuf_delete(odp_actions);
1795 facet_put__(struct ofproto *ofproto, struct facet *facet,
1796 const struct nlattr *actions, size_t actions_len,
1797 struct dpif_flow_stats *stats)
1799 struct odputil_keybuf keybuf;
1800 enum dpif_flow_put_flags flags;
1803 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
1805 flags |= DPIF_FP_ZERO_STATS;
1806 facet->dp_packet_count = 0;
1807 facet->dp_byte_count = 0;
1810 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1811 odp_flow_key_from_flow(&key, &facet->flow);
1813 return dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
1814 actions, actions_len, stats);
1817 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
1818 * 'zero_stats' is true, clears any existing statistics from the datapath for
1821 facet_install(struct ofproto *p, struct facet *facet, bool zero_stats)
1823 struct dpif_flow_stats stats;
1825 if (facet->may_install
1826 && !facet_put__(p, facet, facet->actions, facet->actions_len,
1827 zero_stats ? &stats : NULL)) {
1828 facet->installed = true;
1832 /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up
1833 * to the accounting hook function in the ofhooks structure. */
1835 facet_account(struct ofproto *ofproto,
1836 struct facet *facet, uint64_t extra_bytes)
1838 uint64_t total_bytes = facet->byte_count + extra_bytes;
1840 if (ofproto->ofhooks->account_flow_cb
1841 && total_bytes > facet->accounted_bytes)
1843 ofproto->ofhooks->account_flow_cb(
1844 &facet->flow, facet->tags, facet->actions, facet->actions_len,
1845 total_bytes - facet->accounted_bytes, ofproto->aux);
1846 facet->accounted_bytes = total_bytes;
1850 /* If 'rule' is installed in the datapath, uninstalls it. */
1852 facet_uninstall(struct ofproto *p, struct facet *facet)
1854 if (facet->installed) {
1855 struct odputil_keybuf keybuf;
1856 struct dpif_flow_stats stats;
1859 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1860 odp_flow_key_from_flow(&key, &facet->flow);
1862 if (!dpif_flow_del(p->dpif, key.data, key.size, &stats)) {
1863 facet_update_stats(p, facet, &stats);
1865 facet->installed = false;
1866 facet->dp_packet_count = 0;
1867 facet->dp_byte_count = 0;
1869 assert(facet->dp_packet_count == 0);
1870 assert(facet->dp_byte_count == 0);
1874 /* Returns true if the only action for 'facet' is to send to the controller.
1875 * (We don't report NetFlow expiration messages for such facets because they
1876 * are just part of the control logic for the network, not real traffic). */
1878 facet_is_controller_flow(struct facet *facet)
1881 && facet->rule->n_actions == 1
1882 && action_outputs_to_port(&facet->rule->actions[0],
1883 htons(OFPP_CONTROLLER)));
1886 /* Folds all of 'facet''s statistics into its rule. Also updates the
1887 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
1888 * 'facet''s statistics in the datapath should have been zeroed and folded into
1889 * its packet and byte counts before this function is called. */
1891 facet_flush_stats(struct ofproto *ofproto, struct facet *facet)
1893 assert(!facet->dp_byte_count);
1894 assert(!facet->dp_packet_count);
1896 facet_push_stats(ofproto, facet);
1897 facet_account(ofproto, facet, 0);
1899 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
1900 struct ofexpired expired;
1901 expired.flow = facet->flow;
1902 expired.packet_count = facet->packet_count;
1903 expired.byte_count = facet->byte_count;
1904 expired.used = facet->used;
1905 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
1908 facet->rule->packet_count += facet->packet_count;
1909 facet->rule->byte_count += facet->byte_count;
1911 /* Reset counters to prevent double counting if 'facet' ever gets
1913 facet->packet_count = 0;
1914 facet->byte_count = 0;
1915 facet->rs_packet_count = 0;
1916 facet->rs_byte_count = 0;
1917 facet->accounted_bytes = 0;
1919 netflow_flow_clear(&facet->nf_flow);
1922 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
1923 * Returns it if found, otherwise a null pointer.
1925 * The returned facet might need revalidation; use facet_lookup_valid()
1926 * instead if that is important. */
1927 static struct facet *
1928 facet_find(struct ofproto *ofproto, const struct flow *flow)
1930 struct facet *facet;
1932 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
1934 if (flow_equal(flow, &facet->flow)) {
1942 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
1943 * Returns it if found, otherwise a null pointer.
1945 * The returned facet is guaranteed to be valid. */
1946 static struct facet *
1947 facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow)
1949 struct facet *facet = facet_find(ofproto, flow);
1951 /* The facet we found might not be valid, since we could be in need of
1952 * revalidation. If it is not valid, don't return it. */
1954 && ofproto->need_revalidate
1955 && !facet_revalidate(ofproto, facet)) {
1956 COVERAGE_INC(ofproto_invalidated);
1963 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
1965 * - If the rule found is different from 'facet''s current rule, moves
1966 * 'facet' to the new rule and recompiles its actions.
1968 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
1969 * where it is and recompiles its actions anyway.
1971 * - If there is none, destroys 'facet'.
1973 * Returns true if 'facet' still exists, false if it has been destroyed. */
1975 facet_revalidate(struct ofproto *ofproto, struct facet *facet)
1977 struct action_xlate_ctx ctx;
1978 struct ofpbuf *odp_actions;
1979 struct rule *new_rule;
1980 bool actions_changed;
1982 COVERAGE_INC(facet_revalidate);
1984 /* Determine the new rule. */
1985 new_rule = rule_lookup(ofproto, &facet->flow);
1987 /* No new rule, so delete the facet. */
1988 facet_remove(ofproto, facet);
1992 /* Calculate new ODP actions.
1994 * We do not modify any 'facet' state yet, because we might need to, e.g.,
1995 * emit a NetFlow expiration and, if so, we need to have the old state
1996 * around to properly compose it. */
1997 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
1998 odp_actions = xlate_actions(&ctx, new_rule->actions, new_rule->n_actions);
1999 actions_changed = (facet->actions_len != odp_actions->size
2000 || memcmp(facet->actions, odp_actions->data,
2001 facet->actions_len));
2003 /* If the ODP actions changed or the installability changed, then we need
2004 * to talk to the datapath. */
2005 if (actions_changed || ctx.may_set_up_flow != facet->installed) {
2006 if (ctx.may_set_up_flow) {
2007 struct dpif_flow_stats stats;
2009 facet_put__(ofproto, facet,
2010 odp_actions->data, odp_actions->size, &stats);
2011 facet_update_stats(ofproto, facet, &stats);
2013 facet_uninstall(ofproto, facet);
2016 /* The datapath flow is gone or has zeroed stats, so push stats out of
2017 * 'facet' into 'rule'. */
2018 facet_flush_stats(ofproto, facet);
2021 /* Update 'facet' now that we've taken care of all the old state. */
2022 facet->tags = ctx.tags;
2023 facet->nf_flow.output_iface = ctx.nf_output_iface;
2024 facet->may_install = ctx.may_set_up_flow;
2025 if (actions_changed) {
2026 free(facet->actions);
2027 facet->actions_len = odp_actions->size;
2028 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2030 if (facet->rule != new_rule) {
2031 COVERAGE_INC(facet_changed_rule);
2032 list_remove(&facet->list_node);
2033 list_push_back(&new_rule->facets, &facet->list_node);
2034 facet->rule = new_rule;
2035 facet->used = new_rule->created;
2036 facet->rs_used = facet->used;
2039 ofpbuf_delete(odp_actions);
2045 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2048 struct ofpbuf *buf = ofputil_encode_error_msg(error, oh);
2050 COVERAGE_INC(ofproto_error);
2051 ofconn_send_reply(ofconn, buf);
2056 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2058 ofconn_send_reply(ofconn, make_echo_reply(oh));
2063 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2065 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2066 struct ofp_switch_features *osf;
2068 struct ofport *port;
2070 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2071 osf->datapath_id = htonll(ofproto->datapath_id);
2072 osf->n_buffers = htonl(pktbuf_capacity());
2074 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2075 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2076 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2077 (1u << OFPAT_SET_VLAN_VID) |
2078 (1u << OFPAT_SET_VLAN_PCP) |
2079 (1u << OFPAT_STRIP_VLAN) |
2080 (1u << OFPAT_SET_DL_SRC) |
2081 (1u << OFPAT_SET_DL_DST) |
2082 (1u << OFPAT_SET_NW_SRC) |
2083 (1u << OFPAT_SET_NW_DST) |
2084 (1u << OFPAT_SET_NW_TOS) |
2085 (1u << OFPAT_SET_TP_SRC) |
2086 (1u << OFPAT_SET_TP_DST) |
2087 (1u << OFPAT_ENQUEUE));
2089 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2090 ofpbuf_put(buf, &port->opp, sizeof port->opp);
2093 ofconn_send_reply(ofconn, buf);
2098 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2100 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2102 struct ofp_switch_config *osc;
2106 /* Figure out flags. */
2107 dpif_get_drop_frags(ofproto->dpif, &drop_frags);
2108 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2111 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2112 osc->flags = htons(flags);
2113 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2114 ofconn_send_reply(ofconn, buf);
2120 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
2122 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2123 uint16_t flags = ntohs(osc->flags);
2125 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2126 && ofconn_get_role(ofconn) != NX_ROLE_SLAVE) {
2127 switch (flags & OFPC_FRAG_MASK) {
2128 case OFPC_FRAG_NORMAL:
2129 dpif_set_drop_frags(ofproto->dpif, false);
2131 case OFPC_FRAG_DROP:
2132 dpif_set_drop_frags(ofproto->dpif, true);
2135 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2141 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2146 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2147 struct action_xlate_ctx *ctx);
2150 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2152 const struct ofport *ofport = get_port(ctx->ofproto, port);
2155 if (ofport->opp.config & htonl(OFPPC_NO_FWD)) {
2156 /* Forwarding disabled on port. */
2161 * We don't have an ofport record for this port, but it doesn't hurt to
2162 * allow forwarding to it anyhow. Maybe such a port will appear later
2163 * and we're pre-populating the flow table.
2167 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, port);
2168 ctx->nf_output_iface = port;
2171 static struct rule *
2172 rule_lookup(struct ofproto *ofproto, const struct flow *flow)
2174 return rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2178 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2180 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2181 uint16_t old_in_port;
2184 /* Look up a flow with 'in_port' as the input port. Then restore the
2185 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2186 * have surprising behavior). */
2187 old_in_port = ctx->flow.in_port;
2188 ctx->flow.in_port = in_port;
2189 rule = rule_lookup(ctx->ofproto, &ctx->flow);
2190 ctx->flow.in_port = old_in_port;
2192 if (ctx->resubmit_hook) {
2193 ctx->resubmit_hook(ctx, rule);
2198 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2202 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2204 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2205 MAX_RESUBMIT_RECURSION);
2210 flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, ovs_be32 mask,
2211 uint16_t *nf_output_iface, struct ofpbuf *odp_actions)
2213 struct ofport *ofport;
2215 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
2216 uint16_t odp_port = ofport->odp_port;
2217 if (odp_port != odp_in_port && !(ofport->opp.config & mask)) {
2218 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port);
2221 *nf_output_iface = NF_OUT_FLOOD;
2225 xlate_output_action__(struct action_xlate_ctx *ctx,
2226 uint16_t port, uint16_t max_len)
2229 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2231 ctx->nf_output_iface = NF_OUT_DROP;
2235 add_output_action(ctx, ctx->flow.in_port);
2238 xlate_table_action(ctx, ctx->flow.in_port);
2241 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2242 ctx->odp_actions, &ctx->tags,
2243 &ctx->nf_output_iface,
2244 ctx->ofproto->aux)) {
2245 COVERAGE_INC(ofproto_uninstallable);
2246 ctx->may_set_up_flow = false;
2250 flood_packets(ctx->ofproto, ctx->flow.in_port, htonl(OFPPC_NO_FLOOD),
2251 &ctx->nf_output_iface, ctx->odp_actions);
2254 flood_packets(ctx->ofproto, ctx->flow.in_port, htonl(0),
2255 &ctx->nf_output_iface, ctx->odp_actions);
2257 case OFPP_CONTROLLER:
2258 nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_CONTROLLER, max_len);
2261 add_output_action(ctx, ODPP_LOCAL);
2264 odp_port = ofp_port_to_odp_port(port);
2265 if (odp_port != ctx->flow.in_port) {
2266 add_output_action(ctx, odp_port);
2271 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2272 ctx->nf_output_iface = NF_OUT_FLOOD;
2273 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2274 ctx->nf_output_iface = prev_nf_output_iface;
2275 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2276 ctx->nf_output_iface != NF_OUT_FLOOD) {
2277 ctx->nf_output_iface = NF_OUT_MULTI;
2282 xlate_output_action(struct action_xlate_ctx *ctx,
2283 const struct ofp_action_output *oao)
2285 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2288 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2289 * optimization, because we're going to add another action that sets the
2290 * priority immediately after, or because there are no actions following the
2293 remove_pop_action(struct action_xlate_ctx *ctx)
2295 if (ctx->odp_actions->size == ctx->last_pop_priority) {
2296 ctx->odp_actions->size -= NLA_ALIGN(NLA_HDRLEN);
2297 ctx->last_pop_priority = -1;
2302 add_pop_action(struct action_xlate_ctx *ctx)
2304 if (ctx->odp_actions->size != ctx->last_pop_priority) {
2305 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_POP_PRIORITY);
2306 ctx->last_pop_priority = ctx->odp_actions->size;
2311 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2312 const struct ofp_action_enqueue *oae)
2314 uint16_t ofp_port, odp_port;
2318 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2321 /* Fall back to ordinary output action. */
2322 xlate_output_action__(ctx, ntohs(oae->port), 0);
2326 /* Figure out ODP output port. */
2327 ofp_port = ntohs(oae->port);
2328 if (ofp_port != OFPP_IN_PORT) {
2329 odp_port = ofp_port_to_odp_port(ofp_port);
2331 odp_port = ctx->flow.in_port;
2334 /* Add ODP actions. */
2335 remove_pop_action(ctx);
2336 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2337 add_output_action(ctx, odp_port);
2338 add_pop_action(ctx);
2340 /* Update NetFlow output port. */
2341 if (ctx->nf_output_iface == NF_OUT_DROP) {
2342 ctx->nf_output_iface = odp_port;
2343 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2344 ctx->nf_output_iface = NF_OUT_MULTI;
2349 xlate_set_queue_action(struct action_xlate_ctx *ctx,
2350 const struct nx_action_set_queue *nasq)
2355 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
2358 /* Couldn't translate queue to a priority, so ignore. A warning
2359 * has already been logged. */
2363 remove_pop_action(ctx);
2364 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2368 xlate_set_dl_tci(struct action_xlate_ctx *ctx)
2370 ovs_be16 tci = ctx->flow.vlan_tci;
2371 if (!(tci & htons(VLAN_CFI))) {
2372 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN);
2374 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
2375 tci & ~htons(VLAN_CFI));
2379 struct xlate_reg_state {
2385 save_reg_state(const struct action_xlate_ctx *ctx,
2386 struct xlate_reg_state *state)
2388 state->vlan_tci = ctx->flow.vlan_tci;
2389 state->tun_id = ctx->flow.tun_id;
2393 update_reg_state(struct action_xlate_ctx *ctx,
2394 const struct xlate_reg_state *state)
2396 if (ctx->flow.vlan_tci != state->vlan_tci) {
2397 xlate_set_dl_tci(ctx);
2399 if (ctx->flow.tun_id != state->tun_id) {
2400 nl_msg_put_be64(ctx->odp_actions,
2401 ODP_ACTION_ATTR_SET_TUNNEL, ctx->flow.tun_id);
2406 xlate_nicira_action(struct action_xlate_ctx *ctx,
2407 const struct nx_action_header *nah)
2409 const struct nx_action_resubmit *nar;
2410 const struct nx_action_set_tunnel *nast;
2411 const struct nx_action_set_queue *nasq;
2412 const struct nx_action_multipath *nam;
2413 const struct nx_action_autopath *naa;
2414 enum nx_action_subtype subtype = ntohs(nah->subtype);
2415 const struct ofhooks *ofhooks = ctx->ofproto->ofhooks;
2416 struct xlate_reg_state state;
2417 uint16_t autopath_port;
2420 assert(nah->vendor == htonl(NX_VENDOR_ID));
2422 case NXAST_RESUBMIT:
2423 nar = (const struct nx_action_resubmit *) nah;
2424 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2427 case NXAST_SET_TUNNEL:
2428 nast = (const struct nx_action_set_tunnel *) nah;
2429 tun_id = htonll(ntohl(nast->tun_id));
2430 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
2431 ctx->flow.tun_id = tun_id;
2434 case NXAST_DROP_SPOOFED_ARP:
2435 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
2436 nl_msg_put_flag(ctx->odp_actions,
2437 ODP_ACTION_ATTR_DROP_SPOOFED_ARP);
2441 case NXAST_SET_QUEUE:
2442 nasq = (const struct nx_action_set_queue *) nah;
2443 xlate_set_queue_action(ctx, nasq);
2446 case NXAST_POP_QUEUE:
2447 add_pop_action(ctx);
2450 case NXAST_REG_MOVE:
2451 save_reg_state(ctx, &state);
2452 nxm_execute_reg_move((const struct nx_action_reg_move *) nah,
2454 update_reg_state(ctx, &state);
2457 case NXAST_REG_LOAD:
2458 save_reg_state(ctx, &state);
2459 nxm_execute_reg_load((const struct nx_action_reg_load *) nah,
2461 update_reg_state(ctx, &state);
2465 /* Nothing to do. */
2468 case NXAST_SET_TUNNEL64:
2469 tun_id = ((const struct nx_action_set_tunnel64 *) nah)->tun_id;
2470 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
2471 ctx->flow.tun_id = tun_id;
2474 case NXAST_MULTIPATH:
2475 nam = (const struct nx_action_multipath *) nah;
2476 multipath_execute(nam, &ctx->flow);
2479 case NXAST_AUTOPATH:
2480 naa = (const struct nx_action_autopath *) nah;
2481 autopath_port = (ofhooks->autopath_cb
2482 ? ofhooks->autopath_cb(&ctx->flow, ntohl(naa->id),
2483 &ctx->tags, ctx->ofproto->aux)
2485 autopath_execute(naa, &ctx->flow, autopath_port);
2488 /* If you add a new action here that modifies flow data, don't forget to
2489 * update the flow key in ctx->flow at the same time. */
2491 case NXAST_SNAT__OBSOLETE:
2493 VLOG_DBG_RL(&rl, "unknown Nicira action type %d", (int) subtype);
2499 do_xlate_actions(const union ofp_action *in, size_t n_in,
2500 struct action_xlate_ctx *ctx)
2502 struct actions_iterator iter;
2503 const union ofp_action *ia;
2504 const struct ofport *port;
2506 port = get_port(ctx->ofproto, ctx->flow.in_port);
2507 if (port && port->opp.config & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2508 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
2509 ? htonl(OFPPC_NO_RECV_STP)
2510 : htonl(OFPPC_NO_RECV))) {
2511 /* Drop this flow. */
2515 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2516 enum ofp_action_type type = ntohs(ia->type);
2517 const struct ofp_action_dl_addr *oada;
2521 xlate_output_action(ctx, &ia->output);
2524 case OFPAT_SET_VLAN_VID:
2525 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
2526 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
2527 xlate_set_dl_tci(ctx);
2530 case OFPAT_SET_VLAN_PCP:
2531 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
2532 ctx->flow.vlan_tci |= htons(
2533 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
2534 xlate_set_dl_tci(ctx);
2537 case OFPAT_STRIP_VLAN:
2538 ctx->flow.vlan_tci = htons(0);
2539 xlate_set_dl_tci(ctx);
2542 case OFPAT_SET_DL_SRC:
2543 oada = ((struct ofp_action_dl_addr *) ia);
2544 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_SRC,
2545 oada->dl_addr, ETH_ADDR_LEN);
2546 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
2549 case OFPAT_SET_DL_DST:
2550 oada = ((struct ofp_action_dl_addr *) ia);
2551 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_DST,
2552 oada->dl_addr, ETH_ADDR_LEN);
2553 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
2556 case OFPAT_SET_NW_SRC:
2557 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_SRC,
2558 ia->nw_addr.nw_addr);
2559 ctx->flow.nw_src = ia->nw_addr.nw_addr;
2562 case OFPAT_SET_NW_DST:
2563 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_DST,
2564 ia->nw_addr.nw_addr);
2565 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
2568 case OFPAT_SET_NW_TOS:
2569 nl_msg_put_u8(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_TOS,
2571 ctx->flow.nw_tos = ia->nw_tos.nw_tos;
2574 case OFPAT_SET_TP_SRC:
2575 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_SRC,
2576 ia->tp_port.tp_port);
2577 ctx->flow.tp_src = ia->tp_port.tp_port;
2580 case OFPAT_SET_TP_DST:
2581 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_DST,
2582 ia->tp_port.tp_port);
2583 ctx->flow.tp_dst = ia->tp_port.tp_port;
2587 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2591 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
2595 VLOG_DBG_RL(&rl, "unknown action type %d", (int) type);
2602 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
2603 struct ofproto *ofproto, const struct flow *flow,
2604 const struct ofpbuf *packet)
2606 ctx->ofproto = ofproto;
2608 ctx->packet = packet;
2609 ctx->resubmit_hook = NULL;
2610 ctx->check_special = true;
2614 ofproto_process_cfm(struct ofproto *ofproto, const struct flow *flow,
2615 const struct ofpbuf *packet)
2617 struct ofport *ofport;
2619 ofport = get_port(ofproto, flow->in_port);
2620 if (ofport && ofport->cfm) {
2621 cfm_process_heartbeat(ofport->cfm, packet);
2625 static struct ofpbuf *
2626 xlate_actions(struct action_xlate_ctx *ctx,
2627 const union ofp_action *in, size_t n_in)
2629 COVERAGE_INC(ofproto_ofp2odp);
2631 ctx->odp_actions = ofpbuf_new(512);
2633 ctx->may_set_up_flow = true;
2634 ctx->nf_output_iface = NF_OUT_DROP;
2636 ctx->last_pop_priority = -1;
2638 if (ctx->check_special && cfm_should_process_flow(&ctx->flow)) {
2640 ofproto_process_cfm(ctx->ofproto, &ctx->flow, ctx->packet);
2642 ctx->may_set_up_flow = false;
2643 } else if (ctx->check_special
2644 && ctx->ofproto->ofhooks->special_cb
2645 && !ctx->ofproto->ofhooks->special_cb(&ctx->flow, ctx->packet,
2646 ctx->ofproto->aux)) {
2647 ctx->may_set_up_flow = false;
2649 do_xlate_actions(in, n_in, ctx);
2652 remove_pop_action(ctx);
2654 /* Check with in-band control to see if we're allowed to set up this
2656 if (!connmgr_may_set_up_flow(ctx->ofproto->connmgr, &ctx->flow,
2657 ctx->odp_actions->data,
2658 ctx->odp_actions->size)) {
2659 ctx->may_set_up_flow = false;
2662 return ctx->odp_actions;
2665 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2666 * error message code (composed with ofp_mkerr()) for the caller to propagate
2667 * upward. Otherwise, returns 0.
2669 * The log message mentions 'msg_type'. */
2671 reject_slave_controller(struct ofconn *ofconn, const const char *msg_type)
2673 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2674 && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
2675 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
2676 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
2679 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2686 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2688 struct ofproto *p = ofconn_get_ofproto(ofconn);
2689 struct ofp_packet_out *opo;
2690 struct ofpbuf payload, *buffer;
2691 union ofp_action *ofp_actions;
2692 struct action_xlate_ctx ctx;
2693 struct ofpbuf *odp_actions;
2694 struct ofpbuf request;
2696 size_t n_ofp_actions;
2700 COVERAGE_INC(ofproto_packet_out);
2702 error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT");
2707 /* Get ofp_packet_out. */
2708 ofpbuf_use_const(&request, oh, ntohs(oh->length));
2709 opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions));
2712 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
2713 &ofp_actions, &n_ofp_actions);
2719 if (opo->buffer_id != htonl(UINT32_MAX)) {
2720 error = ofconn_pktbuf_retrieve(ofconn, ntohl(opo->buffer_id),
2722 if (error || !buffer) {
2731 /* Extract flow, check actions. */
2732 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)),
2734 error = validate_actions(ofp_actions, n_ofp_actions, &flow, p->max_ports);
2740 action_xlate_ctx_init(&ctx, p, &flow, &payload);
2741 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
2742 dpif_execute(p->dpif, odp_actions->data, odp_actions->size, &payload);
2743 ofpbuf_delete(odp_actions);
2746 ofpbuf_delete(buffer);
2751 update_port_config(struct ofproto *p, struct ofport *port,
2752 ovs_be32 config, ovs_be32 mask)
2754 mask &= config ^ port->opp.config;
2755 if (mask & htonl(OFPPC_PORT_DOWN)) {
2756 if (config & htonl(OFPPC_PORT_DOWN)) {
2757 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2759 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2762 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
2763 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
2764 if (mask & htonl(REVALIDATE_BITS)) {
2765 COVERAGE_INC(ofproto_costly_flags);
2766 port->opp.config ^= mask & htonl(REVALIDATE_BITS);
2767 p->need_revalidate = true;
2769 #undef REVALIDATE_BITS
2770 if (mask & htonl(OFPPC_NO_PACKET_IN)) {
2771 port->opp.config ^= htonl(OFPPC_NO_PACKET_IN);
2776 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2778 struct ofproto *p = ofconn_get_ofproto(ofconn);
2779 const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh;
2780 struct ofport *port;
2783 error = reject_slave_controller(ofconn, "OFPT_PORT_MOD");
2788 port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no)));
2790 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2791 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2792 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2794 update_port_config(p, port, opm->config, opm->mask);
2795 if (opm->advertise) {
2796 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2802 static struct ofpbuf *
2803 make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len)
2805 struct ofp_stats_reply *osr;
2808 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2809 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2811 osr->flags = htons(0);
2815 static struct ofpbuf *
2816 start_ofp_stats_reply(const struct ofp_header *request, size_t body_len)
2818 const struct ofp_stats_request *osr
2819 = (const struct ofp_stats_request *) request;
2820 return make_ofp_stats_reply(osr->header.xid, osr->type, body_len);
2824 append_ofp_stats_reply(size_t nbytes, struct ofconn *ofconn,
2825 struct ofpbuf **msgp)
2827 struct ofpbuf *msg = *msgp;
2828 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2829 if (nbytes + msg->size > UINT16_MAX) {
2830 struct ofp_stats_reply *reply = msg->data;
2831 reply->flags = htons(OFPSF_REPLY_MORE);
2832 *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes);
2833 ofconn_send_reply(ofconn, msg);
2835 return ofpbuf_put_uninit(*msgp, nbytes);
2838 static struct ofpbuf *
2839 make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len)
2841 struct nicira_stats_msg *nsm;
2844 msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX));
2845 nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg);
2846 nsm->type = htons(OFPST_VENDOR);
2847 nsm->flags = htons(0);
2848 nsm->vendor = htonl(NX_VENDOR_ID);
2849 nsm->subtype = subtype;
2853 static struct ofpbuf *
2854 start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len)
2856 return make_nxstats_reply(request->header.xid, request->subtype, body_len);
2860 append_nxstats_reply(size_t nbytes, struct ofconn *ofconn,
2861 struct ofpbuf **msgp)
2863 struct ofpbuf *msg = *msgp;
2864 assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg));
2865 if (nbytes + msg->size > UINT16_MAX) {
2866 struct nicira_stats_msg *reply = msg->data;
2867 reply->flags = htons(OFPSF_REPLY_MORE);
2868 *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes);
2869 ofconn_send_reply(ofconn, msg);
2871 ofpbuf_prealloc_tailroom(*msgp, nbytes);
2875 handle_desc_stats_request(struct ofconn *ofconn,
2876 const struct ofp_header *request)
2878 struct ofproto *p = ofconn_get_ofproto(ofconn);
2879 struct ofp_desc_stats *ods;
2882 msg = start_ofp_stats_reply(request, sizeof *ods);
2883 ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg);
2884 memset(ods, 0, sizeof *ods);
2885 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2886 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2887 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
2888 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
2889 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2890 ofconn_send_reply(ofconn, msg);
2896 handle_table_stats_request(struct ofconn *ofconn,
2897 const struct ofp_header *request)
2899 struct ofproto *p = ofconn_get_ofproto(ofconn);
2900 struct ofp_table_stats *ots;
2903 msg = start_ofp_stats_reply(request, sizeof *ots * 2);
2905 /* Classifier table. */
2906 ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg);
2907 memset(ots, 0, sizeof *ots);
2908 strcpy(ots->name, "classifier");
2909 ots->wildcards = (ofconn_get_flow_format(ofconn) == NXFF_OPENFLOW10
2910 ? htonl(OFPFW_ALL) : htonl(OVSFW_ALL));
2911 ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */
2912 ots->active_count = htonl(classifier_count(&p->cls));
2913 put_32aligned_be64(&ots->lookup_count, htonll(0)); /* XXX */
2914 put_32aligned_be64(&ots->matched_count, htonll(0)); /* XXX */
2916 ofconn_send_reply(ofconn, msg);
2921 append_port_stat(struct ofport *port, struct ofconn *ofconn,
2922 struct ofpbuf **msgp)
2924 struct netdev_stats stats;
2925 struct ofp_port_stats *ops;
2927 /* Intentionally ignore return value, since errors will set
2928 * 'stats' to all-1s, which is correct for OpenFlow, and
2929 * netdev_get_stats() will log errors. */
2930 netdev_get_stats(port->netdev, &stats);
2932 ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp);
2933 ops->port_no = port->opp.port_no;
2934 memset(ops->pad, 0, sizeof ops->pad);
2935 put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
2936 put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
2937 put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
2938 put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
2939 put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
2940 put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
2941 put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
2942 put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
2943 put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
2944 put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
2945 put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
2946 put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
2950 handle_port_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
2952 struct ofproto *p = ofconn_get_ofproto(ofconn);
2953 const struct ofp_port_stats_request *psr = ofputil_stats_body(oh);
2954 struct ofp_port_stats *ops;
2956 struct ofport *port;
2958 msg = start_ofp_stats_reply(oh, sizeof *ops * 16);
2959 if (psr->port_no != htons(OFPP_NONE)) {
2960 port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no)));
2962 append_port_stat(port, ofconn, &msg);
2965 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2966 append_port_stat(port, ofconn, &msg);
2970 ofconn_send_reply(ofconn, msg);
2975 calc_flow_duration__(long long int start, uint32_t *sec, uint32_t *nsec)
2977 long long int msecs = time_msec() - start;
2978 *sec = msecs / 1000;
2979 *nsec = (msecs % 1000) * (1000 * 1000);
2983 calc_flow_duration(long long int start, ovs_be32 *sec_be, ovs_be32 *nsec_be)
2987 calc_flow_duration__(start, &sec, &nsec);
2988 *sec_be = htonl(sec);
2989 *nsec_be = htonl(nsec);
2993 put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule,
2994 ovs_be16 out_port, struct ofpbuf **replyp)
2996 struct ofp_flow_stats *ofs;
2997 uint64_t packet_count, byte_count;
2999 size_t act_len, len;
3001 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3005 act_len = sizeof *rule->actions * rule->n_actions;
3006 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3008 rule_get_stats(rule, &packet_count, &byte_count);
3010 ofs = append_ofp_stats_reply(len, ofconn, replyp);
3011 ofs->length = htons(len);
3014 ofputil_cls_rule_to_match(&rule->cr, ofconn_get_flow_format(ofconn),
3015 &ofs->match, rule->flow_cookie, &cookie);
3016 put_32aligned_be64(&ofs->cookie, cookie);
3017 calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec);
3018 ofs->priority = htons(rule->cr.priority);
3019 ofs->idle_timeout = htons(rule->idle_timeout);
3020 ofs->hard_timeout = htons(rule->hard_timeout);
3021 memset(ofs->pad2, 0, sizeof ofs->pad2);
3022 put_32aligned_be64(&ofs->packet_count, htonll(packet_count));
3023 put_32aligned_be64(&ofs->byte_count, htonll(byte_count));
3024 if (rule->n_actions > 0) {
3025 memcpy(ofs->actions, rule->actions, act_len);
3030 is_valid_table(uint8_t table_id)
3032 if (table_id == 0 || table_id == 0xff) {
3035 /* It would probably be better to reply with an error but there doesn't
3036 * seem to be any appropriate value, so that might just be
3038 VLOG_WARN_RL(&rl, "controller asked for invalid table %"PRIu8,
3045 handle_flow_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3047 const struct ofp_flow_stats_request *fsr = ofputil_stats_body(oh);
3048 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3049 struct ofpbuf *reply;
3051 COVERAGE_INC(ofproto_flows_req);
3052 reply = start_ofp_stats_reply(oh, 1024);
3053 if (is_valid_table(fsr->table_id)) {
3054 struct cls_cursor cursor;
3055 struct cls_rule target;
3058 ofputil_cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0,
3060 cls_cursor_init(&cursor, &ofproto->cls, &target);
3061 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3062 put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply);
3065 ofconn_send_reply(ofconn, reply);
3071 put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule,
3072 ovs_be16 out_port, struct ofpbuf **replyp)
3074 struct nx_flow_stats *nfs;
3075 uint64_t packet_count, byte_count;
3076 size_t act_len, start_len;
3077 struct ofpbuf *reply;
3079 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3083 rule_get_stats(rule, &packet_count, &byte_count);
3085 act_len = sizeof *rule->actions * rule->n_actions;
3087 append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, ofconn, replyp);
3088 start_len = (*replyp)->size;
3091 nfs = ofpbuf_put_uninit(reply, sizeof *nfs);
3094 calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec);
3095 nfs->cookie = rule->flow_cookie;
3096 nfs->priority = htons(rule->cr.priority);
3097 nfs->idle_timeout = htons(rule->idle_timeout);
3098 nfs->hard_timeout = htons(rule->hard_timeout);
3099 nfs->match_len = htons(nx_put_match(reply, &rule->cr));
3100 memset(nfs->pad2, 0, sizeof nfs->pad2);
3101 nfs->packet_count = htonll(packet_count);
3102 nfs->byte_count = htonll(byte_count);
3103 if (rule->n_actions > 0) {
3104 ofpbuf_put(reply, rule->actions, act_len);
3106 nfs->length = htons(reply->size - start_len);
3110 handle_nxst_flow(struct ofconn *ofconn, const struct ofp_header *oh)
3112 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3113 struct nx_flow_stats_request *nfsr;
3114 struct cls_rule target;
3115 struct ofpbuf *reply;
3119 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3121 /* Dissect the message. */
3122 nfsr = ofpbuf_pull(&b, sizeof *nfsr);
3123 error = nx_pull_match(&b, ntohs(nfsr->match_len), 0, &target);
3128 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3131 COVERAGE_INC(ofproto_flows_req);
3132 reply = start_nxstats_reply(&nfsr->nsm, 1024);
3133 if (is_valid_table(nfsr->table_id)) {
3134 struct cls_cursor cursor;
3137 cls_cursor_init(&cursor, &ofproto->cls, &target);
3138 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3139 put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply);
3142 ofconn_send_reply(ofconn, reply);
3148 flow_stats_ds(struct rule *rule, struct ds *results)
3150 uint64_t packet_count, byte_count;
3151 size_t act_len = sizeof *rule->actions * rule->n_actions;
3153 rule_get_stats(rule, &packet_count, &byte_count);
3155 ds_put_format(results, "duration=%llds, ",
3156 (time_msec() - rule->created) / 1000);
3157 ds_put_format(results, "idle=%.3fs, ", (time_msec() - rule->used) / 1000.0);
3158 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3159 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3160 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3161 cls_rule_format(&rule->cr, results);
3162 ds_put_char(results, ',');
3164 ofp_print_actions(results, &rule->actions->header, act_len);
3166 ds_put_cstr(results, "drop");
3168 ds_put_cstr(results, "\n");
3171 /* Adds a pretty-printed description of all flows to 'results', including
3172 * hidden flows (e.g., set up by in-band control). */
3174 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3176 struct cls_cursor cursor;
3179 cls_cursor_init(&cursor, &p->cls, NULL);
3180 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3181 flow_stats_ds(rule, results);
3185 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
3186 * '*engine_type' and '*engine_id', respectively. */
3188 ofproto_get_netflow_ids(const struct ofproto *ofproto,
3189 uint8_t *engine_type, uint8_t *engine_id)
3191 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
3195 query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target,
3196 ovs_be16 out_port, uint8_t table_id,
3197 struct ofp_aggregate_stats_reply *oasr)
3199 uint64_t total_packets = 0;
3200 uint64_t total_bytes = 0;
3203 COVERAGE_INC(ofproto_agg_request);
3205 if (is_valid_table(table_id)) {
3206 struct cls_cursor cursor;
3209 cls_cursor_init(&cursor, &ofproto->cls, target);
3210 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3211 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
3212 uint64_t packet_count;
3213 uint64_t byte_count;
3215 rule_get_stats(rule, &packet_count, &byte_count);
3217 total_packets += packet_count;
3218 total_bytes += byte_count;
3224 oasr->flow_count = htonl(n_flows);
3225 put_32aligned_be64(&oasr->packet_count, htonll(total_packets));
3226 put_32aligned_be64(&oasr->byte_count, htonll(total_bytes));
3227 memset(oasr->pad, 0, sizeof oasr->pad);
3231 handle_aggregate_stats_request(struct ofconn *ofconn,
3232 const struct ofp_header *oh)
3234 const struct ofp_aggregate_stats_request *request = ofputil_stats_body(oh);
3235 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3236 struct ofp_aggregate_stats_reply *reply;
3237 struct cls_rule target;
3240 ofputil_cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0,
3243 msg = start_ofp_stats_reply(oh, sizeof *reply);
3244 reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg);
3245 query_aggregate_stats(ofproto, &target, request->out_port,
3246 request->table_id, reply);
3247 ofconn_send_reply(ofconn, msg);
3252 handle_nxst_aggregate(struct ofconn *ofconn, const struct ofp_header *oh)
3254 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3255 struct nx_aggregate_stats_request *request;
3256 struct ofp_aggregate_stats_reply *reply;
3257 struct cls_rule target;
3262 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3264 /* Dissect the message. */
3265 request = ofpbuf_pull(&b, sizeof *request);
3266 error = nx_pull_match(&b, ntohs(request->match_len), 0, &target);
3271 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3275 COVERAGE_INC(ofproto_flows_req);
3276 buf = start_nxstats_reply(&request->nsm, sizeof *reply);
3277 reply = ofpbuf_put_uninit(buf, sizeof *reply);
3278 query_aggregate_stats(ofproto, &target, request->out_port,
3279 request->table_id, reply);
3280 ofconn_send_reply(ofconn, buf);
3285 struct queue_stats_cbdata {
3286 struct ofconn *ofconn;
3287 struct ofport *ofport;
3292 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3293 const struct netdev_queue_stats *stats)
3295 struct ofp_queue_stats *reply;
3297 reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3298 reply->port_no = cbdata->ofport->opp.port_no;
3299 memset(reply->pad, 0, sizeof reply->pad);
3300 reply->queue_id = htonl(queue_id);
3301 put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
3302 put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
3303 put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
3307 handle_queue_stats_dump_cb(uint32_t queue_id,
3308 struct netdev_queue_stats *stats,
3311 struct queue_stats_cbdata *cbdata = cbdata_;
3313 put_queue_stats(cbdata, queue_id, stats);
3317 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3318 struct queue_stats_cbdata *cbdata)
3320 cbdata->ofport = port;
3321 if (queue_id == OFPQ_ALL) {
3322 netdev_dump_queue_stats(port->netdev,
3323 handle_queue_stats_dump_cb, cbdata);
3325 struct netdev_queue_stats stats;
3327 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3328 put_queue_stats(cbdata, queue_id, &stats);
3334 handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3336 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3337 const struct ofp_queue_stats_request *qsr;
3338 struct queue_stats_cbdata cbdata;
3339 struct ofport *port;
3340 unsigned int port_no;
3343 qsr = ofputil_stats_body(oh);
3345 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3348 COVERAGE_INC(ofproto_queue_req);
3350 cbdata.ofconn = ofconn;
3351 cbdata.msg = start_ofp_stats_reply(oh, 128);
3353 port_no = ntohs(qsr->port_no);
3354 queue_id = ntohl(qsr->queue_id);
3355 if (port_no == OFPP_ALL) {
3356 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3357 handle_queue_stats_for_port(port, queue_id, &cbdata);
3359 } else if (port_no < ofproto->max_ports) {
3360 port = get_port(ofproto, ofp_port_to_odp_port(port_no));
3362 handle_queue_stats_for_port(port, queue_id, &cbdata);
3365 ofpbuf_delete(cbdata.msg);
3366 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3368 ofconn_send_reply(ofconn, cbdata.msg);
3373 /* Updates 'facet''s used time. Caller is responsible for calling
3374 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3376 facet_update_time(struct ofproto *ofproto, struct facet *facet,
3379 if (used > facet->used) {
3381 if (used > facet->rule->used) {
3382 facet->rule->used = used;
3384 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3388 /* Folds the statistics from 'stats' into the counters in 'facet'.
3390 * Because of the meaning of a facet's counters, it only makes sense to do this
3391 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3392 * packet that was sent by hand or if it represents statistics that have been
3393 * cleared out of the datapath. */
3395 facet_update_stats(struct ofproto *ofproto, struct facet *facet,
3396 const struct dpif_flow_stats *stats)
3398 if (stats->n_packets || stats->used > facet->used) {
3399 facet_update_time(ofproto, facet, stats->used);
3400 facet->packet_count += stats->n_packets;
3401 facet->byte_count += stats->n_bytes;
3402 facet_push_stats(ofproto, facet);
3403 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3408 facet_push_stats(struct ofproto *ofproto, struct facet *facet)
3410 uint64_t rs_packets, rs_bytes;
3412 assert(facet->packet_count >= facet->rs_packet_count);
3413 assert(facet->byte_count >= facet->rs_byte_count);
3414 assert(facet->used >= facet->rs_used);
3416 rs_packets = facet->packet_count - facet->rs_packet_count;
3417 rs_bytes = facet->byte_count - facet->rs_byte_count;
3419 if (rs_packets || rs_bytes || facet->used > facet->rs_used) {
3420 facet->rs_packet_count = facet->packet_count;
3421 facet->rs_byte_count = facet->byte_count;
3422 facet->rs_used = facet->used;
3424 flow_push_stats(ofproto, facet->rule, &facet->flow,
3425 rs_packets, rs_bytes, facet->used);
3429 struct ofproto_push {
3430 struct action_xlate_ctx ctx;
3437 push_resubmit(struct action_xlate_ctx *ctx, struct rule *rule)
3439 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3442 rule->packet_count += push->packets;
3443 rule->byte_count += push->bytes;
3444 rule->used = MAX(push->used, rule->used);
3448 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3449 * 'rule''s actions. */
3451 flow_push_stats(struct ofproto *ofproto, const struct rule *rule,
3452 struct flow *flow, uint64_t packets, uint64_t bytes,
3455 struct ofproto_push push;
3457 push.packets = packets;
3461 action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL);
3462 push.ctx.resubmit_hook = push_resubmit;
3463 ofpbuf_delete(xlate_actions(&push.ctx, rule->actions, rule->n_actions));
3466 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3467 * in which no matching flow already exists in the flow table.
3469 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3470 * ofp_actions, to the ofproto's flow table. Returns 0 on success or an
3471 * OpenFlow error code as encoded by ofp_mkerr() on failure.
3473 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3476 add_flow(struct ofconn *ofconn, struct flow_mod *fm)
3478 struct ofproto *p = ofconn_get_ofproto(ofconn);
3479 struct ofpbuf *packet;
3484 if (fm->flags & OFPFF_CHECK_OVERLAP
3485 && classifier_rule_overlaps(&p->cls, &fm->cr)) {
3486 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3490 if (fm->buffer_id != UINT32_MAX) {
3491 error = ofconn_pktbuf_retrieve(ofconn, fm->buffer_id,
3495 in_port = UINT16_MAX;
3498 rule = rule_create(&fm->cr, fm->actions, fm->n_actions,
3499 fm->idle_timeout, fm->hard_timeout, fm->cookie,
3500 fm->flags & OFPFF_SEND_FLOW_REM);
3501 rule_insert(p, rule);
3503 rule_execute(p, rule, in_port, packet);
3508 static struct rule *
3509 find_flow_strict(struct ofproto *p, const struct flow_mod *fm)
3511 return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr));
3515 send_buffered_packet(struct ofconn *ofconn,
3516 struct rule *rule, uint32_t buffer_id)
3518 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3519 struct ofpbuf *packet;
3523 if (buffer_id == UINT32_MAX) {
3527 error = ofconn_pktbuf_retrieve(ofconn, buffer_id, &packet, &in_port);
3532 rule_execute(ofproto, rule, in_port, packet);
3537 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3539 struct modify_flows_cbdata {
3540 struct ofproto *ofproto;
3541 const struct flow_mod *fm;
3545 static int modify_flow(struct ofproto *, const struct flow_mod *,
3548 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3549 * encoded by ofp_mkerr() on failure.
3551 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3554 modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm)
3556 struct ofproto *p = ofconn_get_ofproto(ofconn);
3557 struct rule *match = NULL;
3558 struct cls_cursor cursor;
3561 cls_cursor_init(&cursor, &p->cls, &fm->cr);
3562 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3563 if (!rule_is_hidden(rule)) {
3565 modify_flow(p, fm, rule);
3570 /* This credits the packet to whichever flow happened to match last.
3571 * That's weird. Maybe we should do a lookup for the flow that
3572 * actually matches the packet? Who knows. */
3573 send_buffered_packet(ofconn, match, fm->buffer_id);
3576 return add_flow(ofconn, fm);
3580 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3581 * code as encoded by ofp_mkerr() on failure.
3583 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3586 modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm)
3588 struct ofproto *p = ofconn_get_ofproto(ofconn);
3589 struct rule *rule = find_flow_strict(p, fm);
3590 if (rule && !rule_is_hidden(rule)) {
3591 modify_flow(p, fm, rule);
3592 return send_buffered_packet(ofconn, rule, fm->buffer_id);
3594 return add_flow(ofconn, fm);
3598 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3599 * been identified as a flow in 'p''s flow table to be modified, by changing
3600 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3601 * ofp_action[] structures). */
3603 modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule)
3605 size_t actions_len = fm->n_actions * sizeof *rule->actions;
3607 rule->flow_cookie = fm->cookie;
3609 /* If the actions are the same, do nothing. */
3610 if (fm->n_actions == rule->n_actions
3612 || !memcmp(fm->actions, rule->actions, actions_len))) {
3616 /* Replace actions. */
3617 free(rule->actions);
3618 rule->actions = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL;
3619 rule->n_actions = fm->n_actions;
3621 p->need_revalidate = true;
3626 /* OFPFC_DELETE implementation. */
3628 static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port);
3630 /* Implements OFPFC_DELETE. */
3632 delete_flows_loose(struct ofproto *p, const struct flow_mod *fm)
3634 struct rule *rule, *next_rule;
3635 struct cls_cursor cursor;
3637 cls_cursor_init(&cursor, &p->cls, &fm->cr);
3638 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
3639 delete_flow(p, rule, htons(fm->out_port));
3643 /* Implements OFPFC_DELETE_STRICT. */
3645 delete_flow_strict(struct ofproto *p, struct flow_mod *fm)
3647 struct rule *rule = find_flow_strict(p, fm);
3649 delete_flow(p, rule, htons(fm->out_port));
3653 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3654 * been identified as a flow to delete from 'p''s flow table, by deleting the
3655 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3658 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3659 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3660 * specified 'out_port'. */
3662 delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port)
3664 if (rule_is_hidden(rule)) {
3668 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
3672 rule_send_removed(p, rule, OFPRR_DELETE);
3673 rule_remove(p, rule);
3677 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3679 struct ofproto *p = ofconn_get_ofproto(ofconn);
3683 error = reject_slave_controller(ofconn, "flow_mod");
3688 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_flow_format(ofconn));
3693 /* We do not support the emergency flow cache. It will hopefully get
3694 * dropped from OpenFlow in the near future. */
3695 if (fm.flags & OFPFF_EMERG) {
3696 /* There isn't a good fit for an error code, so just state that the
3697 * flow table is full. */
3698 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3701 error = validate_actions(fm.actions, fm.n_actions,
3702 &fm.cr.flow, p->max_ports);
3707 switch (fm.command) {
3709 return add_flow(ofconn, &fm);
3712 return modify_flows_loose(ofconn, &fm);
3714 case OFPFC_MODIFY_STRICT:
3715 return modify_flow_strict(ofconn, &fm);
3718 delete_flows_loose(p, &fm);
3721 case OFPFC_DELETE_STRICT:
3722 delete_flow_strict(p, &fm);
3726 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3731 handle_tun_id_from_cookie(struct ofconn *ofconn, const struct ofp_header *oh)
3733 const struct nxt_tun_id_cookie *msg
3734 = (const struct nxt_tun_id_cookie *) oh;
3735 enum nx_flow_format flow_format;
3737 flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10;
3738 ofconn_set_flow_format(ofconn, flow_format);
3744 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
3746 struct nx_role_request *nrr = (struct nx_role_request *) oh;
3747 struct nx_role_request *reply;
3751 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY) {
3752 VLOG_WARN_RL(&rl, "ignoring role request on service connection");
3753 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3756 role = ntohl(nrr->role);
3757 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3758 && role != NX_ROLE_SLAVE) {
3759 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
3761 /* There's no good error code for this. */
3762 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
3765 ofconn_set_role(ofconn, role);
3767 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
3768 reply->role = htonl(role);
3769 ofconn_send_reply(ofconn, buf);
3775 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
3777 const struct nxt_set_flow_format *msg
3778 = (const struct nxt_set_flow_format *) oh;
3781 format = ntohl(msg->format);
3782 if (format == NXFF_OPENFLOW10
3783 || format == NXFF_TUN_ID_FROM_COOKIE
3784 || format == NXFF_NXM) {
3785 ofconn_set_flow_format(ofconn, format);
3788 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3793 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
3795 struct ofp_header *ob;
3798 /* Currently, everything executes synchronously, so we can just
3799 * immediately send the barrier reply. */
3800 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
3801 ofconn_send_reply(ofconn, buf);
3806 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
3808 const struct ofp_header *oh = msg->data;
3809 const struct ofputil_msg_type *type;
3812 error = ofputil_decode_msg_type(oh, &type);
3817 switch (ofputil_msg_type_code(type)) {
3818 /* OpenFlow requests. */
3819 case OFPUTIL_OFPT_ECHO_REQUEST:
3820 return handle_echo_request(ofconn, oh);
3822 case OFPUTIL_OFPT_FEATURES_REQUEST:
3823 return handle_features_request(ofconn, oh);
3825 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
3826 return handle_get_config_request(ofconn, oh);
3828 case OFPUTIL_OFPT_SET_CONFIG:
3829 return handle_set_config(ofconn, msg->data);
3831 case OFPUTIL_OFPT_PACKET_OUT:
3832 return handle_packet_out(ofconn, oh);
3834 case OFPUTIL_OFPT_PORT_MOD:
3835 return handle_port_mod(ofconn, oh);
3837 case OFPUTIL_OFPT_FLOW_MOD:
3838 return handle_flow_mod(ofconn, oh);
3840 case OFPUTIL_OFPT_BARRIER_REQUEST:
3841 return handle_barrier_request(ofconn, oh);
3843 /* OpenFlow replies. */
3844 case OFPUTIL_OFPT_ECHO_REPLY:
3847 /* Nicira extension requests. */
3848 case OFPUTIL_NXT_TUN_ID_FROM_COOKIE:
3849 return handle_tun_id_from_cookie(ofconn, oh);
3851 case OFPUTIL_NXT_ROLE_REQUEST:
3852 return handle_role_request(ofconn, oh);
3854 case OFPUTIL_NXT_SET_FLOW_FORMAT:
3855 return handle_nxt_set_flow_format(ofconn, oh);
3857 case OFPUTIL_NXT_FLOW_MOD:
3858 return handle_flow_mod(ofconn, oh);
3860 /* OpenFlow statistics requests. */
3861 case OFPUTIL_OFPST_DESC_REQUEST:
3862 return handle_desc_stats_request(ofconn, oh);
3864 case OFPUTIL_OFPST_FLOW_REQUEST:
3865 return handle_flow_stats_request(ofconn, oh);
3867 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
3868 return handle_aggregate_stats_request(ofconn, oh);
3870 case OFPUTIL_OFPST_TABLE_REQUEST:
3871 return handle_table_stats_request(ofconn, oh);
3873 case OFPUTIL_OFPST_PORT_REQUEST:
3874 return handle_port_stats_request(ofconn, oh);
3876 case OFPUTIL_OFPST_QUEUE_REQUEST:
3877 return handle_queue_stats_request(ofconn, oh);
3879 /* Nicira extension statistics requests. */
3880 case OFPUTIL_NXST_FLOW_REQUEST:
3881 return handle_nxst_flow(ofconn, oh);
3883 case OFPUTIL_NXST_AGGREGATE_REQUEST:
3884 return handle_nxst_aggregate(ofconn, oh);
3886 case OFPUTIL_INVALID:
3887 case OFPUTIL_OFPT_HELLO:
3888 case OFPUTIL_OFPT_ERROR:
3889 case OFPUTIL_OFPT_FEATURES_REPLY:
3890 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
3891 case OFPUTIL_OFPT_PACKET_IN:
3892 case OFPUTIL_OFPT_FLOW_REMOVED:
3893 case OFPUTIL_OFPT_PORT_STATUS:
3894 case OFPUTIL_OFPT_BARRIER_REPLY:
3895 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
3896 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
3897 case OFPUTIL_OFPST_DESC_REPLY:
3898 case OFPUTIL_OFPST_FLOW_REPLY:
3899 case OFPUTIL_OFPST_QUEUE_REPLY:
3900 case OFPUTIL_OFPST_PORT_REPLY:
3901 case OFPUTIL_OFPST_TABLE_REPLY:
3902 case OFPUTIL_OFPST_AGGREGATE_REPLY:
3903 case OFPUTIL_NXT_ROLE_REPLY:
3904 case OFPUTIL_NXT_FLOW_REMOVED:
3905 case OFPUTIL_NXST_FLOW_REPLY:
3906 case OFPUTIL_NXST_AGGREGATE_REPLY:
3908 if (VLOG_IS_WARN_ENABLED()) {
3909 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3910 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3913 if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) {
3914 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
3916 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3922 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
3924 int error = handle_openflow__(ofconn, ofp_msg);
3926 send_error_oh(ofconn, ofp_msg->data, error);
3928 COVERAGE_INC(ofproto_recv_openflow);
3932 handle_miss_upcall(struct ofproto *p, struct dpif_upcall *upcall)
3934 struct facet *facet;
3937 /* Obtain in_port and tun_id, at least. */
3938 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
3940 /* Set header pointers in 'flow'. */
3941 flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow);
3943 if (cfm_should_process_flow(&flow)) {
3944 ofproto_process_cfm(p, &flow, upcall->packet);
3945 ofpbuf_delete(upcall->packet);
3947 } else if (p->ofhooks->special_cb
3948 && !p->ofhooks->special_cb(&flow, upcall->packet, p->aux)) {
3949 ofpbuf_delete(upcall->packet);
3953 /* Check with in-band control to see if this packet should be sent
3954 * to the local port regardless of the flow table. */
3955 if (connmgr_msg_in_hook(p->connmgr, &flow, upcall->packet)) {
3956 ofproto_send_packet(p, ODPP_LOCAL, 0, upcall->packet);
3959 facet = facet_lookup_valid(p, &flow);
3961 struct rule *rule = rule_lookup(p, &flow);
3963 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3964 struct ofport *port = get_port(p, flow.in_port);
3966 if (port->opp.config & htonl(OFPPC_NO_PACKET_IN)) {
3967 COVERAGE_INC(ofproto_no_packet_in);
3968 /* XXX install 'drop' flow entry */
3969 ofpbuf_delete(upcall->packet);
3973 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
3977 COVERAGE_INC(ofproto_packet_in);
3978 send_packet_in(p, upcall, &flow, false);
3982 facet = facet_create(p, rule, &flow, upcall->packet);
3983 } else if (!facet->may_install) {
3984 /* The facet is not installable, that is, we need to process every
3985 * packet, so process the current packet's actions into 'facet'. */
3986 facet_make_actions(p, facet, upcall->packet);
3989 if (facet->rule->cr.priority == FAIL_OPEN_PRIORITY) {
3991 * Extra-special case for fail-open mode.
3993 * We are in fail-open mode and the packet matched the fail-open rule,
3994 * but we are connected to a controller too. We should send the packet
3995 * up to the controller in the hope that it will try to set up a flow
3996 * and thereby allow us to exit fail-open.
3998 * See the top-level comment in fail-open.c for more information.
4000 send_packet_in(p, upcall, &flow, true);
4003 facet_execute(p, facet, upcall->packet);
4004 facet_install(p, facet, false);
4008 handle_upcall(struct ofproto *p, struct dpif_upcall *upcall)
4012 switch (upcall->type) {
4013 case DPIF_UC_ACTION:
4014 COVERAGE_INC(ofproto_ctlr_action);
4015 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4016 send_packet_in(p, upcall, &flow, false);
4019 case DPIF_UC_SAMPLE:
4021 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4022 ofproto_sflow_received(p->sflow, upcall, &flow);
4024 ofpbuf_delete(upcall->packet);
4028 handle_miss_upcall(p, upcall);
4031 case DPIF_N_UC_TYPES:
4033 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4038 /* Flow expiration. */
4040 static int ofproto_dp_max_idle(const struct ofproto *);
4041 static void ofproto_update_stats(struct ofproto *);
4042 static void rule_expire(struct ofproto *, struct rule *);
4043 static void ofproto_expire_facets(struct ofproto *, int dp_max_idle);
4045 /* This function is called periodically by ofproto_run(). Its job is to
4046 * collect updates for the flows that have been installed into the datapath,
4047 * most importantly when they last were used, and then use that information to
4048 * expire flows that have not been used recently.
4050 * Returns the number of milliseconds after which it should be called again. */
4052 ofproto_expire(struct ofproto *ofproto)
4054 struct rule *rule, *next_rule;
4055 struct cls_cursor cursor;
4058 /* Update stats for each flow in the datapath. */
4059 ofproto_update_stats(ofproto);
4061 /* Expire facets that have been idle too long. */
4062 dp_max_idle = ofproto_dp_max_idle(ofproto);
4063 ofproto_expire_facets(ofproto, dp_max_idle);
4065 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
4066 cls_cursor_init(&cursor, &ofproto->cls, NULL);
4067 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4068 rule_expire(ofproto, rule);
4071 /* Let the hook know that we're at a stable point: all outstanding data
4072 * in existing flows has been accounted to the account_cb. Thus, the
4073 * hook can now reasonably do operations that depend on having accurate
4074 * flow volume accounting (currently, that's just bond rebalancing). */
4075 if (ofproto->ofhooks->account_checkpoint_cb) {
4076 ofproto->ofhooks->account_checkpoint_cb(ofproto->aux);
4079 return MIN(dp_max_idle, 1000);
4082 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4084 * This function also pushes statistics updates to rules which each facet
4085 * resubmits into. Generally these statistics will be accurate. However, if a
4086 * facet changes the rule it resubmits into at some time in between
4087 * ofproto_update_stats() runs, it is possible that statistics accrued to the
4088 * old rule will be incorrectly attributed to the new rule. This could be
4089 * avoided by calling ofproto_update_stats() whenever rules are created or
4090 * deleted. However, the performance impact of making so many calls to the
4091 * datapath do not justify the benefit of having perfectly accurate statistics.
4094 ofproto_update_stats(struct ofproto *p)
4096 const struct dpif_flow_stats *stats;
4097 struct dpif_flow_dump dump;
4098 const struct nlattr *key;
4101 dpif_flow_dump_start(&dump, p->dpif);
4102 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4103 struct facet *facet;
4106 if (odp_flow_key_to_flow(key, key_len, &flow)) {
4110 odp_flow_key_format(key, key_len, &s);
4111 VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s",
4117 facet = facet_find(p, &flow);
4119 if (facet && facet->installed) {
4121 if (stats->n_packets >= facet->dp_packet_count) {
4122 facet->packet_count += stats->n_packets - facet->dp_packet_count;
4124 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4127 if (stats->n_bytes >= facet->dp_byte_count) {
4128 facet->byte_count += stats->n_bytes - facet->dp_byte_count;
4130 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4133 facet->dp_packet_count = stats->n_packets;
4134 facet->dp_byte_count = stats->n_bytes;
4136 facet_update_time(p, facet, stats->used);
4137 facet_account(p, facet, stats->n_bytes);
4138 facet_push_stats(p, facet);
4140 /* There's a flow in the datapath that we know nothing about.
4142 COVERAGE_INC(ofproto_unexpected_rule);
4143 dpif_flow_del(p->dpif, key, key_len, NULL);
4146 dpif_flow_dump_done(&dump);
4149 /* Calculates and returns the number of milliseconds of idle time after which
4150 * facets should expire from the datapath and we should fold their statistics
4151 * into their parent rules in userspace. */
4153 ofproto_dp_max_idle(const struct ofproto *ofproto)
4156 * Idle time histogram.
4158 * Most of the time a switch has a relatively small number of facets. When
4159 * this is the case we might as well keep statistics for all of them in
4160 * userspace and to cache them in the kernel datapath for performance as
4163 * As the number of facets increases, the memory required to maintain
4164 * statistics about them in userspace and in the kernel becomes
4165 * significant. However, with a large number of facets it is likely that
4166 * only a few of them are "heavy hitters" that consume a large amount of
4167 * bandwidth. At this point, only heavy hitters are worth caching in the
4168 * kernel and maintaining in userspaces; other facets we can discard.
4170 * The technique used to compute the idle time is to build a histogram with
4171 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
4172 * that is installed in the kernel gets dropped in the appropriate bucket.
4173 * After the histogram has been built, we compute the cutoff so that only
4174 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
4175 * cached. At least the most-recently-used bucket of facets is kept, so
4176 * actually an arbitrary number of facets can be kept in any given
4177 * expiration run (though the next run will delete most of those unless
4178 * they receive additional data).
4180 * This requires a second pass through the facets, in addition to the pass
4181 * made by ofproto_update_stats(), because the former function never looks
4182 * at uninstallable facets.
4184 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4185 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4186 int buckets[N_BUCKETS] = { 0 };
4187 struct facet *facet;
4192 total = hmap_count(&ofproto->facets);
4193 if (total <= 1000) {
4194 return N_BUCKETS * BUCKET_WIDTH;
4197 /* Build histogram. */
4199 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
4200 long long int idle = now - facet->used;
4201 int bucket = (idle <= 0 ? 0
4202 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4203 : (unsigned int) idle / BUCKET_WIDTH);
4207 /* Find the first bucket whose flows should be expired. */
4208 for (bucket = 0; bucket < N_BUCKETS; bucket++) {
4209 if (buckets[bucket]) {
4212 subtotal += buckets[bucket++];
4213 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
4218 if (VLOG_IS_DBG_ENABLED()) {
4222 ds_put_cstr(&s, "keep");
4223 for (i = 0; i < N_BUCKETS; i++) {
4225 ds_put_cstr(&s, ", drop");
4228 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4231 VLOG_INFO("%s: %s (msec:count)", ofproto->name, ds_cstr(&s));
4235 return bucket * BUCKET_WIDTH;
4239 facet_active_timeout(struct ofproto *ofproto, struct facet *facet)
4241 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
4242 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
4243 struct ofexpired expired;
4245 if (facet->installed) {
4246 struct dpif_flow_stats stats;
4248 facet_put__(ofproto, facet, facet->actions, facet->actions_len,
4250 facet_update_stats(ofproto, facet, &stats);
4253 expired.flow = facet->flow;
4254 expired.packet_count = facet->packet_count;
4255 expired.byte_count = facet->byte_count;
4256 expired.used = facet->used;
4257 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4262 ofproto_expire_facets(struct ofproto *ofproto, int dp_max_idle)
4264 long long int cutoff = time_msec() - dp_max_idle;
4265 struct facet *facet, *next_facet;
4267 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
4268 facet_active_timeout(ofproto, facet);
4269 if (facet->used < cutoff) {
4270 facet_remove(ofproto, facet);
4275 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4276 * then delete it entirely. */
4278 rule_expire(struct ofproto *ofproto, struct rule *rule)
4280 struct facet *facet, *next_facet;
4284 /* Has 'rule' expired? */
4286 if (rule->hard_timeout
4287 && now > rule->created + rule->hard_timeout * 1000) {
4288 reason = OFPRR_HARD_TIMEOUT;
4289 } else if (rule->idle_timeout && list_is_empty(&rule->facets)
4290 && now >rule->used + rule->idle_timeout * 1000) {
4291 reason = OFPRR_IDLE_TIMEOUT;
4296 COVERAGE_INC(ofproto_expired);
4298 /* Update stats. (This is a no-op if the rule expired due to an idle
4299 * timeout, because that only happens when the rule has no facets left.) */
4300 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4301 facet_remove(ofproto, facet);
4304 /* Get rid of the rule. */
4305 if (!rule_is_hidden(rule)) {
4306 rule_send_removed(ofproto, rule, reason);
4308 rule_remove(ofproto, rule);
4312 rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason)
4314 struct ofputil_flow_removed fr;
4316 if (!rule->send_flow_removed) {
4321 fr.cookie = rule->flow_cookie;
4323 calc_flow_duration__(rule->created, &fr.duration_sec, &fr.duration_nsec);
4324 fr.idle_timeout = rule->idle_timeout;
4325 fr.packet_count = rule->packet_count;
4326 fr.byte_count = rule->byte_count;
4328 connmgr_send_flow_removed(p->connmgr, &fr);
4331 /* Obtains statistics for 'rule' and stores them in '*packets' and '*bytes'.
4332 * The returned statistics include statistics for all of 'rule''s facets. */
4334 rule_get_stats(const struct rule *rule, uint64_t *packets, uint64_t *bytes)
4337 struct facet *facet;
4339 /* Start from historical data for 'rule' itself that are no longer tracked
4340 * in facets. This counts, for example, facets that have expired. */
4341 p = rule->packet_count;
4342 b = rule->byte_count;
4344 /* Add any statistics that are tracked by facets. This includes
4345 * statistical data recently updated by ofproto_update_stats() as well as
4346 * stats for packets that were executed "by hand" via dpif_execute(). */
4347 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4348 p += facet->packet_count;
4349 b += facet->byte_count;
4356 /* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an
4357 * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to
4358 * their individual configurations.
4360 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
4361 * Otherwise, ownership is transferred to this function. */
4363 send_packet_in(struct ofproto *ofproto, struct dpif_upcall *upcall,
4364 const struct flow *flow, bool clone)
4366 struct ofputil_packet_in pin;
4368 pin.packet = upcall->packet;
4369 pin.in_port = odp_port_to_ofp_port(flow->in_port);
4370 pin.reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION;
4371 pin.buffer_id = 0; /* not yet known */
4372 pin.send_len = upcall->userdata;
4373 connmgr_send_packet_in(ofproto->connmgr, upcall, flow,
4374 clone ? NULL : upcall->packet);
4378 pick_datapath_id(const struct ofproto *ofproto)
4380 const struct ofport *port;
4382 port = get_port(ofproto, ODPP_LOCAL);
4384 uint8_t ea[ETH_ADDR_LEN];
4387 error = netdev_get_etheraddr(port->netdev, ea);
4389 return eth_addr_to_uint64(ea);
4391 VLOG_WARN("could not get MAC address for %s (%s)",
4392 netdev_get_name(port->netdev), strerror(error));
4394 return ofproto->fallback_dpid;
4398 pick_fallback_dpid(void)
4400 uint8_t ea[ETH_ADDR_LEN];
4401 eth_addr_nicira_random(ea);
4402 return eth_addr_to_uint64(ea);
4405 static struct ofproto *
4406 ofproto_lookup(const char *name)
4408 struct ofproto *ofproto;
4410 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
4412 if (!strcmp(ofproto->name, name)) {
4420 ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED,
4421 void *aux OVS_UNUSED)
4423 struct ofproto *ofproto;
4427 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
4428 ds_put_format(&results, "%s\n", ofproto->name);
4430 unixctl_command_reply(conn, 200, ds_cstr(&results));
4431 ds_destroy(&results);
4434 struct ofproto_trace {
4435 struct action_xlate_ctx ctx;
4441 trace_format_rule(struct ds *result, int level, const struct rule *rule)
4443 ds_put_char_multiple(result, '\t', level);
4445 ds_put_cstr(result, "No match\n");
4449 ds_put_format(result, "Rule: cookie=%#"PRIx64" ",
4450 ntohll(rule->flow_cookie));
4451 cls_rule_format(&rule->cr, result);
4452 ds_put_char(result, '\n');
4454 ds_put_char_multiple(result, '\t', level);
4455 ds_put_cstr(result, "OpenFlow ");
4456 ofp_print_actions(result, (const struct ofp_action_header *) rule->actions,
4457 rule->n_actions * sizeof *rule->actions);
4458 ds_put_char(result, '\n');
4462 trace_format_flow(struct ds *result, int level, const char *title,
4463 struct ofproto_trace *trace)
4465 ds_put_char_multiple(result, '\t', level);
4466 ds_put_format(result, "%s: ", title);
4467 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
4468 ds_put_cstr(result, "unchanged");
4470 flow_format(result, &trace->ctx.flow);
4471 trace->flow = trace->ctx.flow;
4473 ds_put_char(result, '\n');
4477 trace_resubmit(struct action_xlate_ctx *ctx, struct rule *rule)
4479 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
4480 struct ds *result = trace->result;
4482 ds_put_char(result, '\n');
4483 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
4484 trace_format_rule(result, ctx->recurse + 1, rule);
4488 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
4489 void *aux OVS_UNUSED)
4491 char *dpname, *in_port_s, *tun_id_s, *packet_s;
4492 char *args = xstrdup(args_);
4493 char *save_ptr = NULL;
4494 struct ofproto *ofproto;
4495 struct ofpbuf packet;
4503 ofpbuf_init(&packet, strlen(args) / 2);
4506 dpname = strtok_r(args, " ", &save_ptr);
4507 tun_id_s = strtok_r(NULL, " ", &save_ptr);
4508 in_port_s = strtok_r(NULL, " ", &save_ptr);
4509 packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
4510 if (!dpname || !in_port_s || !packet_s) {
4511 unixctl_command_reply(conn, 501, "Bad command syntax");
4515 ofproto = ofproto_lookup(dpname);
4517 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
4522 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
4523 in_port = ofp_port_to_odp_port(atoi(in_port_s));
4525 packet_s = ofpbuf_put_hex(&packet, packet_s, NULL);
4526 packet_s += strspn(packet_s, " ");
4527 if (*packet_s != '\0') {
4528 unixctl_command_reply(conn, 501, "Trailing garbage in command");
4531 if (packet.size < ETH_HEADER_LEN) {
4532 unixctl_command_reply(conn, 501, "Packet data too short for Ethernet");
4536 ds_put_cstr(&result, "Packet: ");
4537 s = ofp_packet_to_string(packet.data, packet.size, packet.size);
4538 ds_put_cstr(&result, s);
4541 flow_extract(&packet, tun_id, in_port, &flow);
4542 ds_put_cstr(&result, "Flow: ");
4543 flow_format(&result, &flow);
4544 ds_put_char(&result, '\n');
4546 rule = rule_lookup(ofproto, &flow);
4547 trace_format_rule(&result, 0, rule);
4549 struct ofproto_trace trace;
4550 struct ofpbuf *odp_actions;
4552 trace.result = &result;
4554 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet);
4555 trace.ctx.resubmit_hook = trace_resubmit;
4556 odp_actions = xlate_actions(&trace.ctx,
4557 rule->actions, rule->n_actions);
4559 ds_put_char(&result, '\n');
4560 trace_format_flow(&result, 0, "Final flow", &trace);
4561 ds_put_cstr(&result, "Datapath actions: ");
4562 format_odp_actions(&result, odp_actions->data, odp_actions->size);
4563 ofpbuf_delete(odp_actions);
4566 unixctl_command_reply(conn, 200, ds_cstr(&result));
4569 ds_destroy(&result);
4570 ofpbuf_uninit(&packet);
4575 ofproto_unixctl_init(void)
4577 static bool registered;
4583 unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL);
4584 unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL);
4588 default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
4589 struct ofpbuf *odp_actions, tag_type *tags,
4590 uint16_t *nf_output_iface, void *ofproto_)
4592 struct ofproto *ofproto = ofproto_;
4593 struct mac_entry *dst_mac;
4595 /* Drop frames for reserved multicast addresses. */
4596 if (eth_addr_is_reserved(flow->dl_dst)) {
4600 /* Learn source MAC (but don't try to learn from revalidation). */
4602 && mac_learning_may_learn(ofproto->ml, flow->dl_src, 0)) {
4603 struct mac_entry *src_mac;
4605 src_mac = mac_learning_insert(ofproto->ml, flow->dl_src, 0);
4606 if (mac_entry_is_new(src_mac) || src_mac->port.i != flow->in_port) {
4607 /* The log messages here could actually be useful in debugging,
4608 * so keep the rate limit relatively high. */
4609 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4610 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
4611 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
4613 ofproto_revalidate(ofproto,
4614 mac_learning_changed(ofproto->ml, src_mac));
4615 src_mac->port.i = flow->in_port;
4619 /* Determine output port. */
4620 dst_mac = mac_learning_lookup(ofproto->ml, flow->dl_dst, 0, tags);
4622 flood_packets(ofproto, flow->in_port, htonl(OFPPC_NO_FLOOD),
4623 nf_output_iface, odp_actions);
4625 int out_port = dst_mac->port.i;
4626 if (out_port != flow->in_port) {
4627 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, out_port);
4628 *nf_output_iface = out_port;
4637 static const struct ofhooks default_ofhooks = {
4638 default_normal_ofhook_cb,