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 hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
104 struct netdev *netdev;
105 struct ofp_phy_port opp;
107 struct cfm *cfm; /* Connectivity Fault Management, if any. */
110 static void ofport_free(struct ofport *);
111 static void ofport_run(struct ofproto *, struct ofport *);
112 static void ofport_wait(struct ofport *);
114 struct action_xlate_ctx {
115 /* action_xlate_ctx_init() initializes these members. */
118 struct ofproto *ofproto;
120 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
121 * this flow when actions change header fields. */
124 /* The packet corresponding to 'flow', or a null pointer if we are
125 * revalidating without a packet to refer to. */
126 const struct ofpbuf *packet;
128 /* If nonnull, called just before executing a resubmit action.
130 * This is normally null so the client has to set it manually after
131 * calling action_xlate_ctx_init(). */
132 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule *);
134 /* If true, the speciality of 'flow' should be checked before executing
135 * its actions. If special_cb returns false on 'flow' rendered
136 * uninstallable and no actions will be executed. */
139 /* xlate_actions() initializes and uses these members. The client might want
140 * to look at them after it returns. */
142 struct ofpbuf *odp_actions; /* Datapath actions. */
143 tag_type tags; /* Tags associated with OFPP_NORMAL actions. */
144 bool may_set_up_flow; /* True ordinarily; false if the actions must
145 * be reassessed for every packet. */
146 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
148 /* xlate_actions() initializes and uses these members, but the client has no
149 * reason to look at them. */
151 int recurse; /* Recursion level, via xlate_table_action. */
152 int last_pop_priority; /* Offset in 'odp_actions' just past most
153 * recent ODP_ACTION_ATTR_SET_PRIORITY. */
156 static void action_xlate_ctx_init(struct action_xlate_ctx *,
157 struct ofproto *, const struct flow *,
158 const struct ofpbuf *);
159 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
160 const union ofp_action *in, size_t n_in);
162 /* An OpenFlow flow. */
164 long long int used; /* Time last used; time created if not used. */
165 long long int created; /* Creation time. */
169 * - Do include packets and bytes from facets that have been deleted or
170 * whose own statistics have been folded into the rule.
172 * - Do include packets and bytes sent "by hand" that were accounted to
173 * the rule without any facet being involved (this is a rare corner
174 * case in rule_execute()).
176 * - Do not include packet or bytes that can be obtained from any facet's
177 * packet_count or byte_count member or that can be obtained from the
178 * datapath by, e.g., dpif_flow_get() for any facet.
180 uint64_t packet_count; /* Number of packets received. */
181 uint64_t byte_count; /* Number of bytes received. */
183 ovs_be64 flow_cookie; /* Controller-issued identifier. */
185 struct cls_rule cr; /* In owning ofproto's classifier. */
186 uint16_t idle_timeout; /* In seconds from time of last use. */
187 uint16_t hard_timeout; /* In seconds from time of creation. */
188 bool send_flow_removed; /* Send a flow removed message? */
189 int n_actions; /* Number of elements in actions[]. */
190 union ofp_action *actions; /* OpenFlow actions. */
191 struct list facets; /* List of "struct facet"s. */
194 static struct rule *rule_from_cls_rule(const struct cls_rule *);
195 static bool rule_is_hidden(const struct rule *);
197 static struct rule *rule_create(const struct cls_rule *,
198 const union ofp_action *, size_t n_actions,
199 uint16_t idle_timeout, uint16_t hard_timeout,
200 ovs_be64 flow_cookie, bool send_flow_removed);
201 static void rule_destroy(struct ofproto *, struct rule *);
202 static void rule_free(struct rule *);
204 static struct rule *rule_lookup(struct ofproto *, const struct flow *);
205 static void rule_insert(struct ofproto *, struct rule *);
206 static void rule_remove(struct ofproto *, struct rule *);
208 static void rule_send_removed(struct ofproto *, struct rule *, uint8_t reason);
209 static void rule_get_stats(const struct rule *, uint64_t *packets,
212 /* An exact-match instantiation of an OpenFlow flow. */
214 long long int used; /* Time last used; time created if not used. */
218 * - Do include packets and bytes sent "by hand", e.g. with
221 * - Do include packets and bytes that were obtained from the datapath
222 * when a flow was deleted (e.g. dpif_flow_del()) or when its
223 * statistics were reset (e.g. dpif_flow_put() with
224 * DPIF_FP_ZERO_STATS).
226 * - Do not include any packets or bytes that can currently be obtained
227 * from the datapath by, e.g., dpif_flow_get().
229 uint64_t packet_count; /* Number of packets received. */
230 uint64_t byte_count; /* Number of bytes received. */
232 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
233 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
235 uint64_t rs_packet_count; /* Packets pushed to resubmit children. */
236 uint64_t rs_byte_count; /* Bytes pushed to resubmit children. */
237 long long int rs_used; /* Used time pushed to resubmit children. */
239 /* Number of bytes passed to account_cb. This may include bytes that can
240 * currently obtained from the datapath (thus, it can be greater than
242 uint64_t accounted_bytes;
244 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
245 struct list list_node; /* In owning rule's 'facets' list. */
246 struct rule *rule; /* Owning rule. */
247 struct flow flow; /* Exact-match flow. */
248 bool installed; /* Installed in datapath? */
249 bool may_install; /* True ordinarily; false if actions must
250 * be reassessed for every packet. */
251 size_t actions_len; /* Number of bytes in actions[]. */
252 struct nlattr *actions; /* Datapath actions. */
253 tag_type tags; /* Tags (set only by hooks). */
254 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
257 static struct facet *facet_create(struct ofproto *, struct rule *,
259 const struct ofpbuf *packet);
260 static void facet_remove(struct ofproto *, struct facet *);
261 static void facet_free(struct facet *);
263 static struct facet *facet_lookup_valid(struct ofproto *, const struct flow *);
264 static bool facet_revalidate(struct ofproto *, struct facet *);
266 static void facet_install(struct ofproto *, struct facet *, bool zero_stats);
267 static void facet_uninstall(struct ofproto *, struct facet *);
268 static void facet_flush_stats(struct ofproto *, struct facet *);
270 static void facet_make_actions(struct ofproto *, struct facet *,
271 const struct ofpbuf *packet);
272 static void facet_update_stats(struct ofproto *, struct facet *,
273 const struct dpif_flow_stats *);
274 static void facet_push_stats(struct ofproto *, struct facet *);
276 static void send_packet_in(struct ofproto *, struct dpif_upcall *,
277 const struct flow *, bool clone);
280 char *name; /* Datapath name. */
281 struct hmap_node hmap_node; /* In global 'all_ofprotos' hmap. */
284 uint64_t datapath_id; /* Datapath ID. */
285 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
286 char *mfr_desc; /* Manufacturer. */
287 char *hw_desc; /* Hardware. */
288 char *sw_desc; /* Software version. */
289 char *serial_desc; /* Serial number. */
290 char *dp_desc; /* Datapath description. */
294 struct netdev_monitor *netdev_monitor;
295 struct hmap ports; /* Contains "struct ofport"s. */
296 struct shash port_by_name;
300 struct netflow *netflow;
301 struct ofproto_sflow *sflow;
304 struct classifier cls;
305 struct timer next_expiration;
309 bool need_revalidate;
310 struct tag_set revalidate_set;
312 /* OpenFlow connections. */
313 struct connmgr *connmgr;
315 /* Hooks for ovs-vswitchd. */
316 const struct ofhooks *ofhooks;
319 /* Used by default ofhooks. */
320 struct mac_learning *ml;
323 /* Map from dpif name to struct ofproto, for use by unixctl commands. */
324 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
326 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
328 static const struct ofhooks default_ofhooks;
330 static uint64_t pick_datapath_id(const struct ofproto *);
331 static uint64_t pick_fallback_dpid(void);
333 static void ofproto_flush_flows__(struct ofproto *);
334 static int ofproto_expire(struct ofproto *);
335 static void flow_push_stats(struct ofproto *, const struct rule *,
336 struct flow *, uint64_t packets, uint64_t bytes,
339 static void handle_upcall(struct ofproto *, struct dpif_upcall *);
341 static void handle_openflow(struct ofconn *, struct ofpbuf *);
343 static struct ofport *get_port(const struct ofproto *, uint16_t odp_port);
344 static void update_port(struct ofproto *, const char *devname);
345 static int init_ports(struct ofproto *);
346 static void reinit_ports(struct ofproto *);
348 static void ofproto_unixctl_init(void);
351 ofproto_create(const char *datapath, const char *datapath_type,
352 const struct ofhooks *ofhooks, void *aux,
353 struct ofproto **ofprotop)
355 char local_name[IF_NAMESIZE];
362 ofproto_unixctl_init();
364 /* Connect to datapath and start listening for messages. */
365 error = dpif_create_and_open(datapath, datapath_type, &dpif);
367 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
370 error = dpif_recv_set_mask(dpif,
371 ((1u << DPIF_UC_MISS) |
372 (1u << DPIF_UC_ACTION) |
373 (1u << DPIF_UC_SAMPLE)));
375 VLOG_ERR("failed to listen on datapath %s: %s",
376 datapath, strerror(error));
380 dpif_flow_flush(dpif);
381 dpif_recv_purge(dpif);
383 error = dpif_port_get_name(dpif, ODPP_LOCAL,
384 local_name, sizeof local_name);
386 VLOG_ERR("%s: cannot get name of datapath local port (%s)",
387 datapath, strerror(error));
391 /* Initialize settings. */
392 p = xzalloc(sizeof *p);
393 p->name = xstrdup(dpif_name(dpif));
394 hmap_insert(&all_ofprotos, &p->hmap_node, hash_string(p->name, 0));
395 p->fallback_dpid = pick_fallback_dpid();
396 p->datapath_id = p->fallback_dpid;
397 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
398 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
399 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
400 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
401 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
403 /* Initialize datapath. */
405 p->netdev_monitor = netdev_monitor_create();
406 hmap_init(&p->ports);
407 shash_init(&p->port_by_name);
408 p->max_ports = dpif_get_max_ports(dpif);
410 /* Initialize submodules. */
414 /* Initialize flow table. */
415 classifier_init(&p->cls);
416 timer_set_duration(&p->next_expiration, 1000);
418 /* Initialize facet table. */
419 hmap_init(&p->facets);
420 p->need_revalidate = false;
421 tag_set_init(&p->revalidate_set);
423 /* Initialize hooks. */
425 p->ofhooks = ofhooks;
429 p->ofhooks = &default_ofhooks;
431 p->ml = mac_learning_create();
434 /* Pick final datapath ID. */
435 p->datapath_id = pick_datapath_id(p);
436 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
438 /* Initialize OpenFlow connections. */
439 p->connmgr = connmgr_create(p, datapath, local_name);
448 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
450 uint64_t old_dpid = p->datapath_id;
451 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
452 if (p->datapath_id != old_dpid) {
453 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
455 /* Force all active connections to reconnect, since there is no way to
456 * notify a controller that the datapath ID has changed. */
457 ofproto_reconnect_controllers(p);
462 ofproto_set_controllers(struct ofproto *p,
463 const struct ofproto_controller *controllers,
464 size_t n_controllers)
466 connmgr_set_controllers(p->connmgr, controllers, n_controllers);
470 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
472 connmgr_set_fail_mode(p->connmgr, fail_mode);
475 /* Drops the connections between 'ofproto' and all of its controllers, forcing
476 * them to reconnect. */
478 ofproto_reconnect_controllers(struct ofproto *ofproto)
480 connmgr_reconnect(ofproto->connmgr);
483 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
484 * in-band control should guarantee access, in the same way that in-band
485 * control guarantees access to OpenFlow controllers. */
487 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
488 const struct sockaddr_in *extras, size_t n)
490 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
493 /* Sets the OpenFlow queue used by flows set up by in-band control on
494 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
495 * flows will use the default queue. */
497 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
499 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
503 ofproto_set_desc(struct ofproto *p,
504 const char *mfr_desc, const char *hw_desc,
505 const char *sw_desc, const char *serial_desc,
508 struct ofp_desc_stats *ods;
511 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
512 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
513 sizeof ods->mfr_desc);
516 p->mfr_desc = xstrdup(mfr_desc);
519 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
520 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
521 sizeof ods->hw_desc);
524 p->hw_desc = xstrdup(hw_desc);
527 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
528 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
529 sizeof ods->sw_desc);
532 p->sw_desc = xstrdup(sw_desc);
535 if (strlen(serial_desc) >= sizeof ods->serial_num) {
536 VLOG_WARN("truncating serial_desc, must be less than %zu "
538 sizeof ods->serial_num);
540 free(p->serial_desc);
541 p->serial_desc = xstrdup(serial_desc);
544 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
545 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
546 sizeof ods->dp_desc);
549 p->dp_desc = xstrdup(dp_desc);
554 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
556 return connmgr_set_snoops(ofproto->connmgr, snoops);
560 ofproto_set_netflow(struct ofproto *ofproto,
561 const struct netflow_options *nf_options)
563 if (nf_options && !sset_is_empty(&nf_options->collectors)) {
564 if (!ofproto->netflow) {
565 ofproto->netflow = netflow_create();
567 return netflow_set_options(ofproto->netflow, nf_options);
569 netflow_destroy(ofproto->netflow);
570 ofproto->netflow = NULL;
576 ofproto_set_sflow(struct ofproto *ofproto,
577 const struct ofproto_sflow_options *oso)
579 struct ofproto_sflow *os = ofproto->sflow;
582 struct ofport *ofport;
584 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
585 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
586 ofproto_sflow_add_port(os, ofport->odp_port,
587 netdev_get_name(ofport->netdev));
590 ofproto_sflow_set_options(os, oso);
592 ofproto_sflow_destroy(os);
593 ofproto->sflow = NULL;
597 /* Connectivity Fault Management configuration. */
599 /* Clears the CFM configuration from 'port_no' on 'ofproto'. */
601 ofproto_port_clear_cfm(struct ofproto *ofproto, uint32_t port_no)
603 struct ofport *ofport = get_port(ofproto, port_no);
604 if (ofport && ofport->cfm){
605 cfm_destroy(ofport->cfm);
610 /* Configures connectivity fault management on 'port_no' in 'ofproto'. Takes
611 * basic configuration from the configuration members in 'cfm', and the set of
612 * remote maintenance points from the 'n_remote_mps' elements in 'remote_mps'.
613 * Ignores the statistics members of 'cfm'.
615 * This function has no effect if 'ofproto' does not have a port 'port_no'. */
617 ofproto_port_set_cfm(struct ofproto *ofproto, uint32_t port_no,
618 const struct cfm *cfm,
619 const uint16_t *remote_mps, size_t n_remote_mps)
621 struct ofport *ofport;
623 ofport = get_port(ofproto, port_no);
625 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu32,
626 ofproto->name, port_no);
631 ofport->cfm = cfm_create();
634 ofport->cfm->mpid = cfm->mpid;
635 ofport->cfm->interval = cfm->interval;
636 memcpy(ofport->cfm->maid, cfm->maid, CCM_MAID_LEN);
638 cfm_update_remote_mps(ofport->cfm, remote_mps, n_remote_mps);
640 if (!cfm_configure(ofport->cfm)) {
641 VLOG_WARN("%s: CFM configuration on port %"PRIu32" (%s) failed",
642 ofproto->name, port_no,
643 netdev_get_name(ofport->netdev));
644 cfm_destroy(ofport->cfm);
649 /* Returns the connectivity fault management object associated with 'port_no'
650 * within 'ofproto', or a null pointer if 'ofproto' does not have a port
651 * 'port_no' or if that port does not have CFM configured. The caller must not
652 * modify or destroy the returned object. */
654 ofproto_port_get_cfm(struct ofproto *ofproto, uint32_t port_no)
656 struct ofport *ofport = get_port(ofproto, port_no);
657 return ofport ? ofport->cfm : NULL;
661 ofproto_get_datapath_id(const struct ofproto *ofproto)
663 return ofproto->datapath_id;
666 enum ofproto_fail_mode
667 ofproto_get_fail_mode(const struct ofproto *p)
669 return connmgr_get_fail_mode(p->connmgr);
673 ofproto_has_snoops(const struct ofproto *ofproto)
675 return connmgr_has_snoops(ofproto->connmgr);
679 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
681 connmgr_get_snoops(ofproto->connmgr, snoops);
685 ofproto_destroy__(struct ofproto *p, bool delete)
687 struct ofport *ofport, *next_ofport;
693 hmap_remove(&all_ofprotos, &p->hmap_node);
695 ofproto_flush_flows__(p);
696 connmgr_destroy(p->connmgr);
697 classifier_destroy(&p->cls);
698 hmap_destroy(&p->facets);
701 int error = dpif_delete(p->dpif);
702 if (error && error != ENOENT) {
703 VLOG_ERR("bridge %s: failed to destroy (%s)",
704 p->name, strerror(error));
709 netdev_monitor_destroy(p->netdev_monitor);
710 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
711 hmap_remove(&p->ports, &ofport->hmap_node);
714 shash_destroy(&p->port_by_name);
716 netflow_destroy(p->netflow);
717 ofproto_sflow_destroy(p->sflow);
719 mac_learning_destroy(p->ml);
724 free(p->serial_desc);
727 hmap_destroy(&p->ports);
734 ofproto_destroy(struct ofproto *p)
736 ofproto_destroy__(p, false);
740 ofproto_destroy_and_delete(struct ofproto *p)
742 ofproto_destroy__(p, true);
746 ofproto_run(struct ofproto *p)
748 int error = ofproto_run1(p);
750 error = ofproto_run2(p, false);
756 process_port_change(struct ofproto *ofproto, int error, char *devname)
758 if (error == ENOBUFS) {
759 reinit_ports(ofproto);
761 update_port(ofproto, devname);
767 ofproto_run1(struct ofproto *p)
769 struct ofport *ofport;
774 for (i = 0; i < 50; i++) {
775 struct dpif_upcall packet;
777 error = dpif_recv(p->dpif, &packet);
779 if (error == ENODEV) {
780 /* Someone destroyed the datapath behind our back. The caller
781 * better destroy us and give up, because we're just going to
782 * spin from here on out. */
783 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
784 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
791 handle_upcall(p, &packet);
794 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
795 process_port_change(p, error, devname);
797 while ((error = netdev_monitor_poll(p->netdev_monitor,
798 &devname)) != EAGAIN) {
799 process_port_change(p, error, devname);
802 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
803 ofport_run(p, ofport);
806 connmgr_run(p->connmgr, handle_openflow);
808 if (timer_expired(&p->next_expiration)) {
809 int delay = ofproto_expire(p);
810 timer_set_duration(&p->next_expiration, delay);
811 COVERAGE_INC(ofproto_expiration);
815 netflow_run(p->netflow);
818 ofproto_sflow_run(p->sflow);
825 ofproto_run2(struct ofproto *p, bool revalidate_all)
827 /* Figure out what we need to revalidate now, if anything. */
828 struct tag_set revalidate_set = p->revalidate_set;
829 if (p->need_revalidate) {
830 revalidate_all = true;
833 /* Clear the revalidation flags. */
834 tag_set_init(&p->revalidate_set);
835 p->need_revalidate = false;
837 /* Now revalidate if there's anything to do. */
838 if (revalidate_all || !tag_set_is_empty(&revalidate_set)) {
839 struct facet *facet, *next;
841 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &p->facets) {
843 || tag_set_intersects(&revalidate_set, facet->tags)) {
844 facet_revalidate(p, facet);
853 ofproto_wait(struct ofproto *p)
855 struct ofport *ofport;
857 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
860 dpif_recv_wait(p->dpif);
861 dpif_port_poll_wait(p->dpif);
862 netdev_monitor_poll_wait(p->netdev_monitor);
864 ofproto_sflow_wait(p->sflow);
866 if (!tag_set_is_empty(&p->revalidate_set)) {
867 poll_immediate_wake();
869 if (p->need_revalidate) {
870 /* Shouldn't happen, but if it does just go around again. */
871 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
872 poll_immediate_wake();
874 timer_wait(&p->next_expiration);
876 connmgr_wait(p->connmgr);
880 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
882 tag_set_add(&ofproto->revalidate_set, tag);
886 ofproto_get_revalidate_set(struct ofproto *ofproto)
888 return &ofproto->revalidate_set;
892 ofproto_is_alive(const struct ofproto *p)
894 return connmgr_has_controllers(p->connmgr);
898 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
901 connmgr_get_controller_info(ofproto->connmgr, info);
905 ofproto_free_ofproto_controller_info(struct shash *info)
907 struct shash_node *node;
909 SHASH_FOR_EACH (node, info) {
910 struct ofproto_controller_info *cinfo = node->data;
911 while (cinfo->pairs.n) {
912 free((char *) cinfo->pairs.values[--cinfo->pairs.n]);
919 /* Makes a deep copy of 'old' into 'port'. */
921 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
923 port->name = xstrdup(old->name);
924 port->type = xstrdup(old->type);
925 port->ofp_port = old->ofp_port;
928 /* Frees memory allocated to members of 'ofproto_port'.
930 * Do not call this function on an ofproto_port obtained from
931 * ofproto_port_dump_next(): that function retains ownership of the data in the
934 ofproto_port_destroy(struct ofproto_port *ofproto_port)
936 free(ofproto_port->name);
937 free(ofproto_port->type);
940 /* Converts a dpif_port into an ofproto_port.
942 * This only makes a shallow copy, so make sure that the dpif_port doesn't get
943 * freed while the ofproto_port is still in use. You can choose to free the
944 * ofproto_port instead of the dpif_port. */
946 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
947 struct dpif_port *dpif_port)
949 ofproto_port->name = dpif_port->name;
950 ofproto_port->type = dpif_port->type;
951 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
954 /* Initializes 'dump' to begin dumping the ports in an ofproto.
956 * This function provides no status indication. An error status for the entire
957 * dump operation is provided when it is completed by calling
958 * ofproto_port_dump_done().
961 ofproto_port_dump_start(struct ofproto_port_dump *dump,
962 const struct ofproto *ofproto)
964 struct dpif_port_dump *dpif_dump;
966 dump->state = dpif_dump = xmalloc(sizeof *dpif_dump);
967 dpif_port_dump_start(dpif_dump, ofproto->dpif);
970 /* Attempts to retrieve another port from 'dump', which must have been created
971 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
972 * 'port' and returns true. On failure, returns false.
974 * Failure might indicate an actual error or merely that the last port has been
975 * dumped. An error status for the entire dump operation is provided when it
976 * is completed by calling ofproto_port_dump_done().
978 * The ofproto owns the data stored in 'port'. It will remain valid until at
979 * least the next time 'dump' is passed to ofproto_port_dump_next() or
980 * ofproto_port_dump_done(). */
982 ofproto_port_dump_next(struct ofproto_port_dump *dump,
983 struct ofproto_port *port)
985 struct dpif_port_dump *dpif_dump = dump->state;
986 struct dpif_port dpif_port;
989 ok = dpif_port_dump_next(dpif_dump, &dpif_port);
991 ofproto_port_from_dpif_port(port, &dpif_port);
996 /* Completes port table dump operation 'dump', which must have been created
997 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
998 * error-free, otherwise a positive errno value describing the problem. */
1000 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1002 struct dpif_port_dump *dpif_dump = dump->state;
1003 int error = dpif_port_dump_done(dpif_dump);
1008 /* Attempts to add 'netdev' as a port on 'ofproto'. If successful, returns 0
1009 * and sets '*ofp_portp' to the new port's port OpenFlow number (if 'ofp_portp'
1010 * is non-null). On failure, returns a positive errno value and sets
1011 * '*ofp_portp' to OFPP_NONE (if 'ofp_portp' is non-null). */
1013 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1014 uint16_t *ofp_portp)
1019 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
1021 *ofp_portp = error ? OFPP_NONE : odp_port_to_ofp_port(odp_port);
1026 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1027 * initializes '*port' appropriately; on failure, returns a positive errno
1030 * The caller owns the data in 'port' and must free it with
1031 * ofproto_port_destroy() when it is no longer needed. */
1033 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1034 struct ofproto_port *port)
1036 struct dpif_port dpif_port;
1039 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
1041 ofproto_port_from_dpif_port(port, &dpif_port);
1046 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1048 * This is almost the same as calling dpif_port_del() directly on the
1049 * datapath, but it also makes 'ofproto' close its open netdev for the port
1050 * (if any). This makes it possible to create a new netdev of a different
1051 * type under the same name, which otherwise the netdev library would refuse
1052 * to do because of the conflict. (The netdev would eventually get closed on
1053 * the next trip through ofproto_run(), but this interface is more direct.)
1055 * Returns 0 if successful, otherwise a positive errno. */
1057 ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port)
1059 uint32_t odp_port = ofp_port_to_odp_port(ofp_port);
1060 struct ofport *ofport = get_port(ofproto, odp_port);
1061 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1064 error = dpif_port_del(ofproto->dpif, odp_port);
1066 VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)",
1067 ofproto->name, odp_port, name, strerror(error));
1068 } else if (ofport) {
1069 /* 'name' is the netdev's name and update_port() is going to close the
1070 * netdev. Just in case update_port() refers to 'name' after it
1071 * destroys 'ofport', make a copy of it around the update_port()
1073 char *devname = xstrdup(name);
1074 update_port(ofproto, devname);
1080 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
1081 * true if 'odp_port' exists and should be included, false otherwise. */
1083 ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port)
1085 struct ofport *ofport = get_port(ofproto, odp_port);
1086 return ofport && !(ofport->opp.config & htonl(OFPPC_NO_FLOOD));
1089 /* Sends 'packet' out of port 'port_no' within 'p'. If 'vlan_tci' is zero the
1090 * packet will not have any 802.1Q hader; if it is nonzero, then the packet
1091 * will be sent with the VLAN TCI specified by 'vlan_tci & ~VLAN_CFI'.
1093 * Returns 0 if successful, otherwise a positive errno value. */
1095 ofproto_send_packet(struct ofproto *ofproto,
1096 uint32_t port_no, uint16_t vlan_tci,
1097 const struct ofpbuf *packet)
1099 struct ofpbuf odp_actions;
1102 ofpbuf_init(&odp_actions, 32);
1103 if (vlan_tci != 0) {
1104 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
1105 ntohs(vlan_tci & ~VLAN_CFI));
1107 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, port_no);
1108 error = dpif_execute(ofproto->dpif, odp_actions.data, odp_actions.size,
1110 ofpbuf_uninit(&odp_actions);
1113 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
1114 ofproto->name, port_no, strerror(error));
1119 /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and
1120 * performs the 'n_actions' actions in 'actions'. The new flow will not
1123 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1124 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1125 * controllers; otherwise, it will be hidden.
1127 * The caller retains ownership of 'cls_rule' and 'actions'. */
1129 ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule,
1130 const union ofp_action *actions, size_t n_actions)
1133 rule = rule_create(cls_rule, actions, n_actions, 0, 0, 0, false);
1134 rule_insert(p, rule);
1138 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1142 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1145 rule_remove(ofproto, rule);
1150 ofproto_flush_flows__(struct ofproto *ofproto)
1152 struct facet *facet, *next_facet;
1153 struct rule *rule, *next_rule;
1154 struct cls_cursor cursor;
1156 COVERAGE_INC(ofproto_flush);
1158 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1159 /* Mark the facet as not installed so that facet_remove() doesn't
1160 * bother trying to uninstall it. There is no point in uninstalling it
1161 * individually since we are about to blow away all the facets with
1162 * dpif_flow_flush(). */
1163 facet->installed = false;
1164 facet->dp_packet_count = 0;
1165 facet->dp_byte_count = 0;
1166 facet_remove(ofproto, facet);
1169 cls_cursor_init(&cursor, &ofproto->cls, NULL);
1170 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1171 rule_remove(ofproto, rule);
1174 dpif_flow_flush(ofproto->dpif);
1178 ofproto_flush_flows(struct ofproto *ofproto)
1180 ofproto_flush_flows__(ofproto);
1181 connmgr_flushed(ofproto->connmgr);
1185 reinit_ports(struct ofproto *p)
1187 struct dpif_port_dump dump;
1188 struct sset devnames;
1189 struct ofport *ofport;
1190 struct dpif_port dpif_port;
1191 const char *devname;
1193 COVERAGE_INC(ofproto_reinit_ports);
1195 sset_init(&devnames);
1196 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1197 sset_add(&devnames, netdev_get_name(ofport->netdev));
1199 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1200 sset_add(&devnames, dpif_port.name);
1203 SSET_FOR_EACH (devname, &devnames) {
1204 update_port(p, devname);
1206 sset_destroy(&devnames);
1209 /* Opens and returns a netdev for 'dpif_port', or a null pointer if the netdev
1210 * cannot be opened. On success, also fills in 'opp'. */
1211 static struct netdev *
1212 ofport_open(const struct dpif_port *dpif_port, struct ofp_phy_port *opp)
1214 uint32_t curr, advertised, supported, peer;
1215 struct netdev_options netdev_options;
1216 enum netdev_flags flags;
1217 struct netdev *netdev;
1220 memset(&netdev_options, 0, sizeof netdev_options);
1221 netdev_options.name = dpif_port->name;
1222 netdev_options.type = dpif_port->type;
1223 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1225 error = netdev_open(&netdev_options, &netdev);
1227 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1228 "cannot be opened (%s)",
1229 dpif_port->name, dpif_port->port_no,
1230 dpif_port->name, strerror(error));
1234 netdev_get_flags(netdev, &flags);
1235 netdev_get_features(netdev, &curr, &advertised, &supported, &peer);
1237 opp->port_no = htons(odp_port_to_ofp_port(dpif_port->port_no));
1238 netdev_get_etheraddr(netdev, opp->hw_addr);
1239 ovs_strzcpy(opp->name, dpif_port->name, sizeof opp->name);
1240 opp->config = flags & NETDEV_UP ? 0 : htonl(OFPPC_PORT_DOWN);
1241 opp->state = netdev_get_carrier(netdev) ? 0 : htonl(OFPPS_LINK_DOWN);
1242 opp->curr = htonl(curr);
1243 opp->advertised = htonl(advertised);
1244 opp->supported = htonl(supported);
1245 opp->peer = htonl(peer);
1251 ofport_conflicts(const struct ofproto *p, const struct dpif_port *dpif_port)
1253 if (get_port(p, dpif_port->port_no)) {
1254 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1255 dpif_port->port_no);
1257 } else if (shash_find(&p->port_by_name, dpif_port->name)) {
1258 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1266 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1267 * port number, and 'config' bits other than OFPPC_PORT_DOWN are
1270 ofport_equal(const struct ofp_phy_port *a, const struct ofp_phy_port *b)
1272 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1273 return (!memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1274 && a->state == b->state
1275 && !((a->config ^ b->config) & htonl(OFPPC_PORT_DOWN))
1276 && a->curr == b->curr
1277 && a->advertised == b->advertised
1278 && a->supported == b->supported
1279 && a->peer == b->peer);
1282 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1283 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1284 * one with the same name or port number). */
1286 ofport_install(struct ofproto *p,
1287 struct netdev *netdev, const struct ofp_phy_port *opp)
1289 const char *netdev_name = netdev_get_name(netdev);
1290 struct ofport *ofport;
1292 connmgr_send_port_status(p->connmgr, opp, OFPPR_ADD);
1294 /* Create ofport. */
1295 ofport = xmalloc(sizeof *ofport);
1296 ofport->netdev = netdev;
1298 ofport->odp_port = ofp_port_to_odp_port(ntohs(opp->port_no));
1301 /* Add port to 'p'. */
1302 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1303 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
1304 shash_add(&p->port_by_name, netdev_name, ofport);
1306 ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
1310 /* Removes 'ofport' from 'p' and destroys it. */
1312 ofport_remove(struct ofproto *p, struct ofport *ofport)
1314 connmgr_send_port_status(p->connmgr, &ofport->opp, OFPPR_DELETE);
1316 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1317 hmap_remove(&p->ports, &ofport->hmap_node);
1318 shash_delete(&p->port_by_name,
1319 shash_find(&p->port_by_name,
1320 netdev_get_name(ofport->netdev)));
1322 ofproto_sflow_del_port(p->sflow, ofport->odp_port);
1325 ofport_free(ofport);
1328 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1331 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1333 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1335 ofport_remove(ofproto, port);
1339 /* Updates 'port' within 'ofproto' with the new 'netdev' and 'opp'.
1341 * Does not handle a name or port number change. The caller must implement
1342 * such a change as a delete followed by an add. */
1344 ofport_modified(struct ofproto *ofproto, struct ofport *port,
1345 struct netdev *netdev, struct ofp_phy_port *opp)
1347 memcpy(port->opp.hw_addr, opp->hw_addr, ETH_ADDR_LEN);
1348 port->opp.config = ((port->opp.config & ~htonl(OFPPC_PORT_DOWN))
1349 | (opp->config & htonl(OFPPC_PORT_DOWN)));
1350 port->opp.state = opp->state;
1351 port->opp.curr = opp->curr;
1352 port->opp.advertised = opp->advertised;
1353 port->opp.supported = opp->supported;
1354 port->opp.peer = opp->peer;
1356 netdev_monitor_remove(ofproto->netdev_monitor, port->netdev);
1357 netdev_monitor_add(ofproto->netdev_monitor, netdev);
1359 netdev_close(port->netdev);
1360 port->netdev = netdev;
1362 connmgr_send_port_status(ofproto->connmgr, &port->opp, OFPPR_MODIFY);
1366 ofport_run(struct ofproto *ofproto, struct ofport *ofport)
1369 cfm_run(ofport->cfm);
1371 if (cfm_should_send_ccm(ofport->cfm)) {
1372 struct ofpbuf packet;
1375 ofpbuf_init(&packet, 0);
1376 ccm = eth_compose(&packet, eth_addr_ccm, ofport->opp.hw_addr,
1377 ETH_TYPE_CFM, sizeof *ccm);
1378 cfm_compose_ccm(ofport->cfm, ccm);
1379 ofproto_send_packet(ofproto, ofport->odp_port, 0, &packet);
1380 ofpbuf_uninit(&packet);
1386 ofport_wait(struct ofport *ofport)
1389 cfm_wait(ofport->cfm);
1394 ofport_free(struct ofport *ofport)
1397 cfm_destroy(ofport->cfm);
1398 netdev_close(ofport->netdev);
1403 static struct ofport *
1404 get_port(const struct ofproto *ofproto, uint16_t odp_port)
1406 struct ofport *port;
1408 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1409 hash_int(odp_port, 0), &ofproto->ports) {
1410 if (port->odp_port == odp_port) {
1418 update_port(struct ofproto *ofproto, const char *name)
1420 struct dpif_port dpif_port;
1421 struct ofp_phy_port opp;
1422 struct netdev *netdev;
1423 struct ofport *port;
1425 COVERAGE_INC(ofproto_update_port);
1427 /* Fetch 'name''s location and properties from the datapath. */
1428 netdev = (!dpif_port_query_by_name(ofproto->dpif, name, &dpif_port)
1429 ? ofport_open(&dpif_port, &opp)
1432 port = get_port(ofproto, dpif_port.port_no);
1433 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
1434 /* 'name' hasn't changed location. Any properties changed? */
1435 if (!ofport_equal(&port->opp, &opp)) {
1436 ofport_modified(ofproto, port, netdev, &opp);
1438 netdev_close(netdev);
1441 /* If 'port' is nonnull then its name differs from 'name' and thus
1442 * we should delete it. If we think there's a port named 'name'
1443 * then its port number must be wrong now so delete it too. */
1445 ofport_remove(ofproto, port);
1447 ofport_remove_with_name(ofproto, name);
1448 ofport_install(ofproto, netdev, &opp);
1451 /* Any port named 'name' is gone now. */
1452 ofport_remove_with_name(ofproto, name);
1454 dpif_port_destroy(&dpif_port);
1458 init_ports(struct ofproto *p)
1460 struct dpif_port_dump dump;
1461 struct dpif_port dpif_port;
1463 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1464 if (!ofport_conflicts(p, &dpif_port)) {
1465 struct ofp_phy_port opp;
1466 struct netdev *netdev;
1468 netdev = ofport_open(&dpif_port, &opp);
1470 ofport_install(p, netdev, &opp);
1478 /* Returns true if 'rule' should be hidden from the controller.
1480 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1481 * (e.g. by in-band control) and are intentionally hidden from the
1484 rule_is_hidden(const struct rule *rule)
1486 return rule->cr.priority > UINT16_MAX;
1489 /* Creates and returns a new rule initialized as specified.
1491 * The caller is responsible for inserting the rule into the classifier (with
1492 * rule_insert()). */
1493 static struct rule *
1494 rule_create(const struct cls_rule *cls_rule,
1495 const union ofp_action *actions, size_t n_actions,
1496 uint16_t idle_timeout, uint16_t hard_timeout,
1497 ovs_be64 flow_cookie, bool send_flow_removed)
1499 struct rule *rule = xzalloc(sizeof *rule);
1500 rule->cr = *cls_rule;
1501 rule->idle_timeout = idle_timeout;
1502 rule->hard_timeout = hard_timeout;
1503 rule->flow_cookie = flow_cookie;
1504 rule->used = rule->created = time_msec();
1505 rule->send_flow_removed = send_flow_removed;
1506 list_init(&rule->facets);
1507 if (n_actions > 0) {
1508 rule->n_actions = n_actions;
1509 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1515 static struct rule *
1516 rule_from_cls_rule(const struct cls_rule *cls_rule)
1518 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1522 rule_free(struct rule *rule)
1524 free(rule->actions);
1528 /* Destroys 'rule' and iterates through all of its facets and revalidates them,
1529 * destroying any that no longer has a rule (which is probably all of them).
1531 * The caller must have already removed 'rule' from the classifier. */
1533 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1535 struct facet *facet, *next_facet;
1536 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
1537 facet_revalidate(ofproto, facet);
1542 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
1543 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
1546 rule_has_out_port(const struct rule *rule, ovs_be16 out_port)
1548 const union ofp_action *oa;
1549 struct actions_iterator i;
1551 if (out_port == htons(OFPP_NONE)) {
1554 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1555 oa = actions_next(&i)) {
1556 if (action_outputs_to_port(oa, out_port)) {
1563 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
1564 * 'packet', which arrived on 'in_port'.
1566 * Takes ownership of 'packet'. */
1568 execute_odp_actions(struct ofproto *ofproto, const struct flow *flow,
1569 const struct nlattr *odp_actions, size_t actions_len,
1570 struct ofpbuf *packet)
1572 if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t))
1573 && odp_actions->nla_type == ODP_ACTION_ATTR_CONTROLLER) {
1574 /* As an optimization, avoid a round-trip from userspace to kernel to
1575 * userspace. This also avoids possibly filling up kernel packet
1576 * buffers along the way. */
1577 struct dpif_upcall upcall;
1579 upcall.type = DPIF_UC_ACTION;
1580 upcall.packet = packet;
1583 upcall.userdata = nl_attr_get_u64(odp_actions);
1584 upcall.sample_pool = 0;
1585 upcall.actions = NULL;
1586 upcall.actions_len = 0;
1588 send_packet_in(ofproto, &upcall, flow, false);
1594 error = dpif_execute(ofproto->dpif, odp_actions, actions_len, packet);
1595 ofpbuf_delete(packet);
1600 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
1601 * statistics appropriately. 'packet' must have at least sizeof(struct
1602 * ofp_packet_in) bytes of headroom.
1604 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
1605 * applying flow_extract() to 'packet' would yield the same flow as
1608 * 'facet' must have accurately composed ODP actions; that is, it must not be
1609 * in need of revalidation.
1611 * Takes ownership of 'packet'. */
1613 facet_execute(struct ofproto *ofproto, struct facet *facet,
1614 struct ofpbuf *packet)
1616 struct dpif_flow_stats stats;
1618 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1620 flow_extract_stats(&facet->flow, packet, &stats);
1621 stats.used = time_msec();
1622 if (execute_odp_actions(ofproto, &facet->flow,
1623 facet->actions, facet->actions_len, packet)) {
1624 facet_update_stats(ofproto, facet, &stats);
1628 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
1629 * statistics (or the statistics for one of its facets) appropriately.
1630 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom.
1632 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
1633 * with statistics for 'packet' either way.
1635 * Takes ownership of 'packet'. */
1637 rule_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port,
1638 struct ofpbuf *packet)
1640 struct action_xlate_ctx ctx;
1641 struct ofpbuf *odp_actions;
1642 struct facet *facet;
1646 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1648 flow_extract(packet, 0, in_port, &flow);
1650 /* First look for a related facet. If we find one, account it to that. */
1651 facet = facet_lookup_valid(ofproto, &flow);
1652 if (facet && facet->rule == rule) {
1653 facet_execute(ofproto, facet, packet);
1657 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
1658 * create a new facet for it and use that. */
1659 if (rule_lookup(ofproto, &flow) == rule) {
1660 facet = facet_create(ofproto, rule, &flow, packet);
1661 facet_execute(ofproto, facet, packet);
1662 facet_install(ofproto, facet, true);
1666 /* We can't account anything to a facet. If we were to try, then that
1667 * facet would have a non-matching rule, busting our invariants. */
1668 action_xlate_ctx_init(&ctx, ofproto, &flow, packet);
1669 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
1670 size = packet->size;
1671 if (execute_odp_actions(ofproto, &flow, odp_actions->data,
1672 odp_actions->size, packet)) {
1673 rule->used = time_msec();
1674 rule->packet_count++;
1675 rule->byte_count += size;
1676 flow_push_stats(ofproto, rule, &flow, 1, size, rule->used);
1678 ofpbuf_delete(odp_actions);
1681 /* Inserts 'rule' into 'p''s flow table. */
1683 rule_insert(struct ofproto *p, struct rule *rule)
1685 struct rule *displaced_rule;
1687 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1688 if (displaced_rule) {
1689 rule_destroy(p, displaced_rule);
1691 p->need_revalidate = true;
1694 /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a
1695 * 'flow' and an example 'packet' within that flow.
1697 * The caller must already have determined that no facet with an identical
1698 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
1699 * 'ofproto''s classifier table. */
1700 static struct facet *
1701 facet_create(struct ofproto *ofproto, struct rule *rule,
1702 const struct flow *flow, const struct ofpbuf *packet)
1704 struct facet *facet;
1706 facet = xzalloc(sizeof *facet);
1707 facet->used = time_msec();
1708 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
1709 list_push_back(&rule->facets, &facet->list_node);
1711 facet->flow = *flow;
1712 netflow_flow_init(&facet->nf_flow);
1713 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
1715 facet_make_actions(ofproto, facet, packet);
1721 facet_free(struct facet *facet)
1723 free(facet->actions);
1727 /* Remove 'rule' from 'ofproto' and free up the associated memory:
1729 * - Removes 'rule' from the classifier.
1731 * - If 'rule' has facets, revalidates them (and possibly uninstalls and
1732 * destroys them), via rule_destroy().
1735 rule_remove(struct ofproto *ofproto, struct rule *rule)
1737 COVERAGE_INC(ofproto_del_rule);
1738 ofproto->need_revalidate = true;
1739 classifier_remove(&ofproto->cls, &rule->cr);
1740 rule_destroy(ofproto, rule);
1743 /* Remove 'facet' from 'ofproto' and free up the associated memory:
1745 * - If 'facet' was installed in the datapath, uninstalls it and updates its
1746 * rule's statistics, via facet_uninstall().
1748 * - Removes 'facet' from its rule and from ofproto->facets.
1751 facet_remove(struct ofproto *ofproto, struct facet *facet)
1753 facet_uninstall(ofproto, facet);
1754 facet_flush_stats(ofproto, facet);
1755 hmap_remove(&ofproto->facets, &facet->hmap_node);
1756 list_remove(&facet->list_node);
1760 /* Composes the ODP actions for 'facet' based on its rule's actions. */
1762 facet_make_actions(struct ofproto *p, struct facet *facet,
1763 const struct ofpbuf *packet)
1765 const struct rule *rule = facet->rule;
1766 struct ofpbuf *odp_actions;
1767 struct action_xlate_ctx ctx;
1769 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
1770 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
1771 facet->tags = ctx.tags;
1772 facet->may_install = ctx.may_set_up_flow;
1773 facet->nf_flow.output_iface = ctx.nf_output_iface;
1775 if (facet->actions_len != odp_actions->size
1776 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
1777 free(facet->actions);
1778 facet->actions_len = odp_actions->size;
1779 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
1782 ofpbuf_delete(odp_actions);
1786 facet_put__(struct ofproto *ofproto, struct facet *facet,
1787 const struct nlattr *actions, size_t actions_len,
1788 struct dpif_flow_stats *stats)
1790 struct odputil_keybuf keybuf;
1791 enum dpif_flow_put_flags flags;
1794 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
1796 flags |= DPIF_FP_ZERO_STATS;
1797 facet->dp_packet_count = 0;
1798 facet->dp_byte_count = 0;
1801 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1802 odp_flow_key_from_flow(&key, &facet->flow);
1804 return dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
1805 actions, actions_len, stats);
1808 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
1809 * 'zero_stats' is true, clears any existing statistics from the datapath for
1812 facet_install(struct ofproto *p, struct facet *facet, bool zero_stats)
1814 struct dpif_flow_stats stats;
1816 if (facet->may_install
1817 && !facet_put__(p, facet, facet->actions, facet->actions_len,
1818 zero_stats ? &stats : NULL)) {
1819 facet->installed = true;
1823 /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up
1824 * to the accounting hook function in the ofhooks structure. */
1826 facet_account(struct ofproto *ofproto,
1827 struct facet *facet, uint64_t extra_bytes)
1829 uint64_t total_bytes = facet->byte_count + extra_bytes;
1831 if (ofproto->ofhooks->account_flow_cb
1832 && total_bytes > facet->accounted_bytes)
1834 ofproto->ofhooks->account_flow_cb(
1835 &facet->flow, facet->tags, facet->actions, facet->actions_len,
1836 total_bytes - facet->accounted_bytes, ofproto->aux);
1837 facet->accounted_bytes = total_bytes;
1841 /* If 'rule' is installed in the datapath, uninstalls it. */
1843 facet_uninstall(struct ofproto *p, struct facet *facet)
1845 if (facet->installed) {
1846 struct odputil_keybuf keybuf;
1847 struct dpif_flow_stats stats;
1850 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1851 odp_flow_key_from_flow(&key, &facet->flow);
1853 if (!dpif_flow_del(p->dpif, key.data, key.size, &stats)) {
1854 facet_update_stats(p, facet, &stats);
1856 facet->installed = false;
1857 facet->dp_packet_count = 0;
1858 facet->dp_byte_count = 0;
1860 assert(facet->dp_packet_count == 0);
1861 assert(facet->dp_byte_count == 0);
1865 /* Returns true if the only action for 'facet' is to send to the controller.
1866 * (We don't report NetFlow expiration messages for such facets because they
1867 * are just part of the control logic for the network, not real traffic). */
1869 facet_is_controller_flow(struct facet *facet)
1872 && facet->rule->n_actions == 1
1873 && action_outputs_to_port(&facet->rule->actions[0],
1874 htons(OFPP_CONTROLLER)));
1877 /* Folds all of 'facet''s statistics into its rule. Also updates the
1878 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
1879 * 'facet''s statistics in the datapath should have been zeroed and folded into
1880 * its packet and byte counts before this function is called. */
1882 facet_flush_stats(struct ofproto *ofproto, struct facet *facet)
1884 assert(!facet->dp_byte_count);
1885 assert(!facet->dp_packet_count);
1887 facet_push_stats(ofproto, facet);
1888 facet_account(ofproto, facet, 0);
1890 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
1891 struct ofexpired expired;
1892 expired.flow = facet->flow;
1893 expired.packet_count = facet->packet_count;
1894 expired.byte_count = facet->byte_count;
1895 expired.used = facet->used;
1896 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
1899 facet->rule->packet_count += facet->packet_count;
1900 facet->rule->byte_count += facet->byte_count;
1902 /* Reset counters to prevent double counting if 'facet' ever gets
1904 facet->packet_count = 0;
1905 facet->byte_count = 0;
1906 facet->rs_packet_count = 0;
1907 facet->rs_byte_count = 0;
1908 facet->accounted_bytes = 0;
1910 netflow_flow_clear(&facet->nf_flow);
1913 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
1914 * Returns it if found, otherwise a null pointer.
1916 * The returned facet might need revalidation; use facet_lookup_valid()
1917 * instead if that is important. */
1918 static struct facet *
1919 facet_find(struct ofproto *ofproto, const struct flow *flow)
1921 struct facet *facet;
1923 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
1925 if (flow_equal(flow, &facet->flow)) {
1933 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
1934 * Returns it if found, otherwise a null pointer.
1936 * The returned facet is guaranteed to be valid. */
1937 static struct facet *
1938 facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow)
1940 struct facet *facet = facet_find(ofproto, flow);
1942 /* The facet we found might not be valid, since we could be in need of
1943 * revalidation. If it is not valid, don't return it. */
1945 && ofproto->need_revalidate
1946 && !facet_revalidate(ofproto, facet)) {
1947 COVERAGE_INC(ofproto_invalidated);
1954 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
1956 * - If the rule found is different from 'facet''s current rule, moves
1957 * 'facet' to the new rule and recompiles its actions.
1959 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
1960 * where it is and recompiles its actions anyway.
1962 * - If there is none, destroys 'facet'.
1964 * Returns true if 'facet' still exists, false if it has been destroyed. */
1966 facet_revalidate(struct ofproto *ofproto, struct facet *facet)
1968 struct action_xlate_ctx ctx;
1969 struct ofpbuf *odp_actions;
1970 struct rule *new_rule;
1971 bool actions_changed;
1973 COVERAGE_INC(facet_revalidate);
1975 /* Determine the new rule. */
1976 new_rule = rule_lookup(ofproto, &facet->flow);
1978 /* No new rule, so delete the facet. */
1979 facet_remove(ofproto, facet);
1983 /* Calculate new ODP actions.
1985 * We do not modify any 'facet' state yet, because we might need to, e.g.,
1986 * emit a NetFlow expiration and, if so, we need to have the old state
1987 * around to properly compose it. */
1988 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
1989 odp_actions = xlate_actions(&ctx, new_rule->actions, new_rule->n_actions);
1990 actions_changed = (facet->actions_len != odp_actions->size
1991 || memcmp(facet->actions, odp_actions->data,
1992 facet->actions_len));
1994 /* If the ODP actions changed or the installability changed, then we need
1995 * to talk to the datapath. */
1996 if (actions_changed || ctx.may_set_up_flow != facet->installed) {
1997 if (ctx.may_set_up_flow) {
1998 struct dpif_flow_stats stats;
2000 facet_put__(ofproto, facet,
2001 odp_actions->data, odp_actions->size, &stats);
2002 facet_update_stats(ofproto, facet, &stats);
2004 facet_uninstall(ofproto, facet);
2007 /* The datapath flow is gone or has zeroed stats, so push stats out of
2008 * 'facet' into 'rule'. */
2009 facet_flush_stats(ofproto, facet);
2012 /* Update 'facet' now that we've taken care of all the old state. */
2013 facet->tags = ctx.tags;
2014 facet->nf_flow.output_iface = ctx.nf_output_iface;
2015 facet->may_install = ctx.may_set_up_flow;
2016 if (actions_changed) {
2017 free(facet->actions);
2018 facet->actions_len = odp_actions->size;
2019 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2021 if (facet->rule != new_rule) {
2022 COVERAGE_INC(facet_changed_rule);
2023 list_remove(&facet->list_node);
2024 list_push_back(&new_rule->facets, &facet->list_node);
2025 facet->rule = new_rule;
2026 facet->used = new_rule->created;
2027 facet->rs_used = facet->used;
2030 ofpbuf_delete(odp_actions);
2036 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2039 struct ofpbuf *buf = ofputil_encode_error_msg(error, oh);
2041 COVERAGE_INC(ofproto_error);
2042 ofconn_send_reply(ofconn, buf);
2047 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2049 ofconn_send_reply(ofconn, make_echo_reply(oh));
2054 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2056 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2057 struct ofp_switch_features *osf;
2059 struct ofport *port;
2061 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2062 osf->datapath_id = htonll(ofproto->datapath_id);
2063 osf->n_buffers = htonl(pktbuf_capacity());
2065 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2066 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2067 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2068 (1u << OFPAT_SET_VLAN_VID) |
2069 (1u << OFPAT_SET_VLAN_PCP) |
2070 (1u << OFPAT_STRIP_VLAN) |
2071 (1u << OFPAT_SET_DL_SRC) |
2072 (1u << OFPAT_SET_DL_DST) |
2073 (1u << OFPAT_SET_NW_SRC) |
2074 (1u << OFPAT_SET_NW_DST) |
2075 (1u << OFPAT_SET_NW_TOS) |
2076 (1u << OFPAT_SET_TP_SRC) |
2077 (1u << OFPAT_SET_TP_DST) |
2078 (1u << OFPAT_ENQUEUE));
2080 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2081 ofpbuf_put(buf, &port->opp, sizeof port->opp);
2084 ofconn_send_reply(ofconn, buf);
2089 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2091 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2093 struct ofp_switch_config *osc;
2097 /* Figure out flags. */
2098 dpif_get_drop_frags(ofproto->dpif, &drop_frags);
2099 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2102 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2103 osc->flags = htons(flags);
2104 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2105 ofconn_send_reply(ofconn, buf);
2111 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
2113 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2114 uint16_t flags = ntohs(osc->flags);
2116 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2117 && ofconn_get_role(ofconn) != NX_ROLE_SLAVE) {
2118 switch (flags & OFPC_FRAG_MASK) {
2119 case OFPC_FRAG_NORMAL:
2120 dpif_set_drop_frags(ofproto->dpif, false);
2122 case OFPC_FRAG_DROP:
2123 dpif_set_drop_frags(ofproto->dpif, true);
2126 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2132 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2137 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2138 struct action_xlate_ctx *ctx);
2141 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2143 const struct ofport *ofport = get_port(ctx->ofproto, port);
2146 if (ofport->opp.config & htonl(OFPPC_NO_FWD)) {
2147 /* Forwarding disabled on port. */
2152 * We don't have an ofport record for this port, but it doesn't hurt to
2153 * allow forwarding to it anyhow. Maybe such a port will appear later
2154 * and we're pre-populating the flow table.
2158 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, port);
2159 ctx->nf_output_iface = port;
2162 static struct rule *
2163 rule_lookup(struct ofproto *ofproto, const struct flow *flow)
2165 return rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2169 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2171 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2172 uint16_t old_in_port;
2175 /* Look up a flow with 'in_port' as the input port. Then restore the
2176 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2177 * have surprising behavior). */
2178 old_in_port = ctx->flow.in_port;
2179 ctx->flow.in_port = in_port;
2180 rule = rule_lookup(ctx->ofproto, &ctx->flow);
2181 ctx->flow.in_port = old_in_port;
2183 if (ctx->resubmit_hook) {
2184 ctx->resubmit_hook(ctx, rule);
2189 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2193 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2195 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2196 MAX_RESUBMIT_RECURSION);
2201 flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, ovs_be32 mask,
2202 uint16_t *nf_output_iface, struct ofpbuf *odp_actions)
2204 struct ofport *ofport;
2206 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
2207 uint16_t odp_port = ofport->odp_port;
2208 if (odp_port != odp_in_port && !(ofport->opp.config & mask)) {
2209 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port);
2212 *nf_output_iface = NF_OUT_FLOOD;
2216 xlate_output_action__(struct action_xlate_ctx *ctx,
2217 uint16_t port, uint16_t max_len)
2220 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2222 ctx->nf_output_iface = NF_OUT_DROP;
2226 add_output_action(ctx, ctx->flow.in_port);
2229 xlate_table_action(ctx, ctx->flow.in_port);
2232 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2233 ctx->odp_actions, &ctx->tags,
2234 &ctx->nf_output_iface,
2235 ctx->ofproto->aux)) {
2236 COVERAGE_INC(ofproto_uninstallable);
2237 ctx->may_set_up_flow = false;
2241 flood_packets(ctx->ofproto, ctx->flow.in_port, htonl(OFPPC_NO_FLOOD),
2242 &ctx->nf_output_iface, ctx->odp_actions);
2245 flood_packets(ctx->ofproto, ctx->flow.in_port, htonl(0),
2246 &ctx->nf_output_iface, ctx->odp_actions);
2248 case OFPP_CONTROLLER:
2249 nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_CONTROLLER, max_len);
2252 add_output_action(ctx, ODPP_LOCAL);
2255 odp_port = ofp_port_to_odp_port(port);
2256 if (odp_port != ctx->flow.in_port) {
2257 add_output_action(ctx, odp_port);
2262 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2263 ctx->nf_output_iface = NF_OUT_FLOOD;
2264 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2265 ctx->nf_output_iface = prev_nf_output_iface;
2266 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2267 ctx->nf_output_iface != NF_OUT_FLOOD) {
2268 ctx->nf_output_iface = NF_OUT_MULTI;
2273 xlate_output_action(struct action_xlate_ctx *ctx,
2274 const struct ofp_action_output *oao)
2276 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2279 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2280 * optimization, because we're going to add another action that sets the
2281 * priority immediately after, or because there are no actions following the
2284 remove_pop_action(struct action_xlate_ctx *ctx)
2286 if (ctx->odp_actions->size == ctx->last_pop_priority) {
2287 ctx->odp_actions->size -= NLA_ALIGN(NLA_HDRLEN);
2288 ctx->last_pop_priority = -1;
2293 add_pop_action(struct action_xlate_ctx *ctx)
2295 if (ctx->odp_actions->size != ctx->last_pop_priority) {
2296 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_POP_PRIORITY);
2297 ctx->last_pop_priority = ctx->odp_actions->size;
2302 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2303 const struct ofp_action_enqueue *oae)
2305 uint16_t ofp_port, odp_port;
2309 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2312 /* Fall back to ordinary output action. */
2313 xlate_output_action__(ctx, ntohs(oae->port), 0);
2317 /* Figure out ODP output port. */
2318 ofp_port = ntohs(oae->port);
2319 if (ofp_port != OFPP_IN_PORT) {
2320 odp_port = ofp_port_to_odp_port(ofp_port);
2322 odp_port = ctx->flow.in_port;
2325 /* Add ODP actions. */
2326 remove_pop_action(ctx);
2327 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2328 add_output_action(ctx, odp_port);
2329 add_pop_action(ctx);
2331 /* Update NetFlow output port. */
2332 if (ctx->nf_output_iface == NF_OUT_DROP) {
2333 ctx->nf_output_iface = odp_port;
2334 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2335 ctx->nf_output_iface = NF_OUT_MULTI;
2340 xlate_set_queue_action(struct action_xlate_ctx *ctx,
2341 const struct nx_action_set_queue *nasq)
2346 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
2349 /* Couldn't translate queue to a priority, so ignore. A warning
2350 * has already been logged. */
2354 remove_pop_action(ctx);
2355 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2359 xlate_set_dl_tci(struct action_xlate_ctx *ctx)
2361 ovs_be16 tci = ctx->flow.vlan_tci;
2362 if (!(tci & htons(VLAN_CFI))) {
2363 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN);
2365 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
2366 tci & ~htons(VLAN_CFI));
2370 struct xlate_reg_state {
2376 save_reg_state(const struct action_xlate_ctx *ctx,
2377 struct xlate_reg_state *state)
2379 state->vlan_tci = ctx->flow.vlan_tci;
2380 state->tun_id = ctx->flow.tun_id;
2384 update_reg_state(struct action_xlate_ctx *ctx,
2385 const struct xlate_reg_state *state)
2387 if (ctx->flow.vlan_tci != state->vlan_tci) {
2388 xlate_set_dl_tci(ctx);
2390 if (ctx->flow.tun_id != state->tun_id) {
2391 nl_msg_put_be64(ctx->odp_actions,
2392 ODP_ACTION_ATTR_SET_TUNNEL, ctx->flow.tun_id);
2397 xlate_nicira_action(struct action_xlate_ctx *ctx,
2398 const struct nx_action_header *nah)
2400 const struct nx_action_resubmit *nar;
2401 const struct nx_action_set_tunnel *nast;
2402 const struct nx_action_set_queue *nasq;
2403 const struct nx_action_multipath *nam;
2404 const struct nx_action_autopath *naa;
2405 enum nx_action_subtype subtype = ntohs(nah->subtype);
2406 const struct ofhooks *ofhooks = ctx->ofproto->ofhooks;
2407 struct xlate_reg_state state;
2408 uint16_t autopath_port;
2411 assert(nah->vendor == htonl(NX_VENDOR_ID));
2413 case NXAST_RESUBMIT:
2414 nar = (const struct nx_action_resubmit *) nah;
2415 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2418 case NXAST_SET_TUNNEL:
2419 nast = (const struct nx_action_set_tunnel *) nah;
2420 tun_id = htonll(ntohl(nast->tun_id));
2421 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
2422 ctx->flow.tun_id = tun_id;
2425 case NXAST_DROP_SPOOFED_ARP:
2426 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
2427 nl_msg_put_flag(ctx->odp_actions,
2428 ODP_ACTION_ATTR_DROP_SPOOFED_ARP);
2432 case NXAST_SET_QUEUE:
2433 nasq = (const struct nx_action_set_queue *) nah;
2434 xlate_set_queue_action(ctx, nasq);
2437 case NXAST_POP_QUEUE:
2438 add_pop_action(ctx);
2441 case NXAST_REG_MOVE:
2442 save_reg_state(ctx, &state);
2443 nxm_execute_reg_move((const struct nx_action_reg_move *) nah,
2445 update_reg_state(ctx, &state);
2448 case NXAST_REG_LOAD:
2449 save_reg_state(ctx, &state);
2450 nxm_execute_reg_load((const struct nx_action_reg_load *) nah,
2452 update_reg_state(ctx, &state);
2456 /* Nothing to do. */
2459 case NXAST_SET_TUNNEL64:
2460 tun_id = ((const struct nx_action_set_tunnel64 *) nah)->tun_id;
2461 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
2462 ctx->flow.tun_id = tun_id;
2465 case NXAST_MULTIPATH:
2466 nam = (const struct nx_action_multipath *) nah;
2467 multipath_execute(nam, &ctx->flow);
2470 case NXAST_AUTOPATH:
2471 naa = (const struct nx_action_autopath *) nah;
2472 autopath_port = (ofhooks->autopath_cb
2473 ? ofhooks->autopath_cb(&ctx->flow, ntohl(naa->id),
2474 &ctx->tags, ctx->ofproto->aux)
2476 autopath_execute(naa, &ctx->flow, autopath_port);
2479 /* If you add a new action here that modifies flow data, don't forget to
2480 * update the flow key in ctx->flow at the same time. */
2482 case NXAST_SNAT__OBSOLETE:
2484 VLOG_DBG_RL(&rl, "unknown Nicira action type %d", (int) subtype);
2490 do_xlate_actions(const union ofp_action *in, size_t n_in,
2491 struct action_xlate_ctx *ctx)
2493 struct actions_iterator iter;
2494 const union ofp_action *ia;
2495 const struct ofport *port;
2497 port = get_port(ctx->ofproto, ctx->flow.in_port);
2498 if (port && port->opp.config & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2499 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
2500 ? htonl(OFPPC_NO_RECV_STP)
2501 : htonl(OFPPC_NO_RECV))) {
2502 /* Drop this flow. */
2506 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2507 enum ofp_action_type type = ntohs(ia->type);
2508 const struct ofp_action_dl_addr *oada;
2512 xlate_output_action(ctx, &ia->output);
2515 case OFPAT_SET_VLAN_VID:
2516 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
2517 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
2518 xlate_set_dl_tci(ctx);
2521 case OFPAT_SET_VLAN_PCP:
2522 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
2523 ctx->flow.vlan_tci |= htons(
2524 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
2525 xlate_set_dl_tci(ctx);
2528 case OFPAT_STRIP_VLAN:
2529 ctx->flow.vlan_tci = htons(0);
2530 xlate_set_dl_tci(ctx);
2533 case OFPAT_SET_DL_SRC:
2534 oada = ((struct ofp_action_dl_addr *) ia);
2535 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_SRC,
2536 oada->dl_addr, ETH_ADDR_LEN);
2537 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
2540 case OFPAT_SET_DL_DST:
2541 oada = ((struct ofp_action_dl_addr *) ia);
2542 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_DST,
2543 oada->dl_addr, ETH_ADDR_LEN);
2544 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
2547 case OFPAT_SET_NW_SRC:
2548 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_SRC,
2549 ia->nw_addr.nw_addr);
2550 ctx->flow.nw_src = ia->nw_addr.nw_addr;
2553 case OFPAT_SET_NW_DST:
2554 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_DST,
2555 ia->nw_addr.nw_addr);
2556 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
2559 case OFPAT_SET_NW_TOS:
2560 nl_msg_put_u8(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_TOS,
2562 ctx->flow.nw_tos = ia->nw_tos.nw_tos;
2565 case OFPAT_SET_TP_SRC:
2566 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_SRC,
2567 ia->tp_port.tp_port);
2568 ctx->flow.tp_src = ia->tp_port.tp_port;
2571 case OFPAT_SET_TP_DST:
2572 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_DST,
2573 ia->tp_port.tp_port);
2574 ctx->flow.tp_dst = ia->tp_port.tp_port;
2578 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2582 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
2586 VLOG_DBG_RL(&rl, "unknown action type %d", (int) type);
2593 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
2594 struct ofproto *ofproto, const struct flow *flow,
2595 const struct ofpbuf *packet)
2597 ctx->ofproto = ofproto;
2599 ctx->packet = packet;
2600 ctx->resubmit_hook = NULL;
2601 ctx->check_special = true;
2605 ofproto_process_cfm(struct ofproto *ofproto, const struct flow *flow,
2606 const struct ofpbuf *packet)
2608 struct ofport *ofport;
2610 ofport = get_port(ofproto, flow->in_port);
2611 if (ofport && ofport->cfm) {
2612 cfm_process_heartbeat(ofport->cfm, packet);
2616 static struct ofpbuf *
2617 xlate_actions(struct action_xlate_ctx *ctx,
2618 const union ofp_action *in, size_t n_in)
2620 COVERAGE_INC(ofproto_ofp2odp);
2622 ctx->odp_actions = ofpbuf_new(512);
2624 ctx->may_set_up_flow = true;
2625 ctx->nf_output_iface = NF_OUT_DROP;
2627 ctx->last_pop_priority = -1;
2629 if (ctx->check_special && cfm_should_process_flow(&ctx->flow)) {
2631 ofproto_process_cfm(ctx->ofproto, &ctx->flow, ctx->packet);
2633 ctx->may_set_up_flow = false;
2634 } else if (ctx->check_special
2635 && ctx->ofproto->ofhooks->special_cb
2636 && !ctx->ofproto->ofhooks->special_cb(&ctx->flow, ctx->packet,
2637 ctx->ofproto->aux)) {
2638 ctx->may_set_up_flow = false;
2640 do_xlate_actions(in, n_in, ctx);
2643 remove_pop_action(ctx);
2645 /* Check with in-band control to see if we're allowed to set up this
2647 if (!connmgr_may_set_up_flow(ctx->ofproto->connmgr, &ctx->flow,
2648 ctx->odp_actions->data,
2649 ctx->odp_actions->size)) {
2650 ctx->may_set_up_flow = false;
2653 return ctx->odp_actions;
2656 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2657 * error message code (composed with ofp_mkerr()) for the caller to propagate
2658 * upward. Otherwise, returns 0.
2660 * The log message mentions 'msg_type'. */
2662 reject_slave_controller(struct ofconn *ofconn, const const char *msg_type)
2664 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2665 && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
2666 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
2667 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
2670 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2677 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2679 struct ofproto *p = ofconn_get_ofproto(ofconn);
2680 struct ofp_packet_out *opo;
2681 struct ofpbuf payload, *buffer;
2682 union ofp_action *ofp_actions;
2683 struct action_xlate_ctx ctx;
2684 struct ofpbuf *odp_actions;
2685 struct ofpbuf request;
2687 size_t n_ofp_actions;
2691 COVERAGE_INC(ofproto_packet_out);
2693 error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT");
2698 /* Get ofp_packet_out. */
2699 ofpbuf_use_const(&request, oh, ntohs(oh->length));
2700 opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions));
2703 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
2704 &ofp_actions, &n_ofp_actions);
2710 if (opo->buffer_id != htonl(UINT32_MAX)) {
2711 error = ofconn_pktbuf_retrieve(ofconn, ntohl(opo->buffer_id),
2713 if (error || !buffer) {
2722 /* Extract flow, check actions. */
2723 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)),
2725 error = validate_actions(ofp_actions, n_ofp_actions, &flow, p->max_ports);
2731 action_xlate_ctx_init(&ctx, p, &flow, &payload);
2732 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
2733 dpif_execute(p->dpif, odp_actions->data, odp_actions->size, &payload);
2734 ofpbuf_delete(odp_actions);
2737 ofpbuf_delete(buffer);
2742 update_port_config(struct ofproto *p, struct ofport *port,
2743 ovs_be32 config, ovs_be32 mask)
2745 mask &= config ^ port->opp.config;
2746 if (mask & htonl(OFPPC_PORT_DOWN)) {
2747 if (config & htonl(OFPPC_PORT_DOWN)) {
2748 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2750 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2753 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
2754 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
2755 if (mask & htonl(REVALIDATE_BITS)) {
2756 COVERAGE_INC(ofproto_costly_flags);
2757 port->opp.config ^= mask & htonl(REVALIDATE_BITS);
2758 p->need_revalidate = true;
2760 #undef REVALIDATE_BITS
2761 if (mask & htonl(OFPPC_NO_PACKET_IN)) {
2762 port->opp.config ^= htonl(OFPPC_NO_PACKET_IN);
2767 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2769 struct ofproto *p = ofconn_get_ofproto(ofconn);
2770 const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh;
2771 struct ofport *port;
2774 error = reject_slave_controller(ofconn, "OFPT_PORT_MOD");
2779 port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no)));
2781 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2782 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2783 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2785 update_port_config(p, port, opm->config, opm->mask);
2786 if (opm->advertise) {
2787 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2793 static struct ofpbuf *
2794 make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len)
2796 struct ofp_stats_reply *osr;
2799 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2800 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2802 osr->flags = htons(0);
2806 static struct ofpbuf *
2807 start_ofp_stats_reply(const struct ofp_header *request, size_t body_len)
2809 const struct ofp_stats_request *osr
2810 = (const struct ofp_stats_request *) request;
2811 return make_ofp_stats_reply(osr->header.xid, osr->type, body_len);
2815 append_ofp_stats_reply(size_t nbytes, struct ofconn *ofconn,
2816 struct ofpbuf **msgp)
2818 struct ofpbuf *msg = *msgp;
2819 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2820 if (nbytes + msg->size > UINT16_MAX) {
2821 struct ofp_stats_reply *reply = msg->data;
2822 reply->flags = htons(OFPSF_REPLY_MORE);
2823 *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes);
2824 ofconn_send_reply(ofconn, msg);
2826 return ofpbuf_put_uninit(*msgp, nbytes);
2829 static struct ofpbuf *
2830 make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len)
2832 struct nicira_stats_msg *nsm;
2835 msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX));
2836 nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg);
2837 nsm->type = htons(OFPST_VENDOR);
2838 nsm->flags = htons(0);
2839 nsm->vendor = htonl(NX_VENDOR_ID);
2840 nsm->subtype = subtype;
2844 static struct ofpbuf *
2845 start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len)
2847 return make_nxstats_reply(request->header.xid, request->subtype, body_len);
2851 append_nxstats_reply(size_t nbytes, struct ofconn *ofconn,
2852 struct ofpbuf **msgp)
2854 struct ofpbuf *msg = *msgp;
2855 assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg));
2856 if (nbytes + msg->size > UINT16_MAX) {
2857 struct nicira_stats_msg *reply = msg->data;
2858 reply->flags = htons(OFPSF_REPLY_MORE);
2859 *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes);
2860 ofconn_send_reply(ofconn, msg);
2862 ofpbuf_prealloc_tailroom(*msgp, nbytes);
2866 handle_desc_stats_request(struct ofconn *ofconn,
2867 const struct ofp_header *request)
2869 struct ofproto *p = ofconn_get_ofproto(ofconn);
2870 struct ofp_desc_stats *ods;
2873 msg = start_ofp_stats_reply(request, sizeof *ods);
2874 ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg);
2875 memset(ods, 0, sizeof *ods);
2876 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2877 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2878 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
2879 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
2880 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2881 ofconn_send_reply(ofconn, msg);
2887 handle_table_stats_request(struct ofconn *ofconn,
2888 const struct ofp_header *request)
2890 struct ofproto *p = ofconn_get_ofproto(ofconn);
2891 struct ofp_table_stats *ots;
2894 msg = start_ofp_stats_reply(request, sizeof *ots * 2);
2896 /* Classifier table. */
2897 ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg);
2898 memset(ots, 0, sizeof *ots);
2899 strcpy(ots->name, "classifier");
2900 ots->wildcards = (ofconn_get_flow_format(ofconn) == NXFF_OPENFLOW10
2901 ? htonl(OFPFW_ALL) : htonl(OVSFW_ALL));
2902 ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */
2903 ots->active_count = htonl(classifier_count(&p->cls));
2904 put_32aligned_be64(&ots->lookup_count, htonll(0)); /* XXX */
2905 put_32aligned_be64(&ots->matched_count, htonll(0)); /* XXX */
2907 ofconn_send_reply(ofconn, msg);
2912 append_port_stat(struct ofport *port, struct ofconn *ofconn,
2913 struct ofpbuf **msgp)
2915 struct netdev_stats stats;
2916 struct ofp_port_stats *ops;
2918 /* Intentionally ignore return value, since errors will set
2919 * 'stats' to all-1s, which is correct for OpenFlow, and
2920 * netdev_get_stats() will log errors. */
2921 netdev_get_stats(port->netdev, &stats);
2923 ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp);
2924 ops->port_no = port->opp.port_no;
2925 memset(ops->pad, 0, sizeof ops->pad);
2926 put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
2927 put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
2928 put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
2929 put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
2930 put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
2931 put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
2932 put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
2933 put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
2934 put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
2935 put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
2936 put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
2937 put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
2941 handle_port_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
2943 struct ofproto *p = ofconn_get_ofproto(ofconn);
2944 const struct ofp_port_stats_request *psr = ofputil_stats_body(oh);
2945 struct ofp_port_stats *ops;
2947 struct ofport *port;
2949 msg = start_ofp_stats_reply(oh, sizeof *ops * 16);
2950 if (psr->port_no != htons(OFPP_NONE)) {
2951 port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no)));
2953 append_port_stat(port, ofconn, &msg);
2956 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2957 append_port_stat(port, ofconn, &msg);
2961 ofconn_send_reply(ofconn, msg);
2966 calc_flow_duration__(long long int start, uint32_t *sec, uint32_t *nsec)
2968 long long int msecs = time_msec() - start;
2969 *sec = msecs / 1000;
2970 *nsec = (msecs % 1000) * (1000 * 1000);
2974 calc_flow_duration(long long int start, ovs_be32 *sec_be, ovs_be32 *nsec_be)
2978 calc_flow_duration__(start, &sec, &nsec);
2979 *sec_be = htonl(sec);
2980 *nsec_be = htonl(nsec);
2984 put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule,
2985 ovs_be16 out_port, struct ofpbuf **replyp)
2987 struct ofp_flow_stats *ofs;
2988 uint64_t packet_count, byte_count;
2990 size_t act_len, len;
2992 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
2996 act_len = sizeof *rule->actions * rule->n_actions;
2997 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2999 rule_get_stats(rule, &packet_count, &byte_count);
3001 ofs = append_ofp_stats_reply(len, ofconn, replyp);
3002 ofs->length = htons(len);
3005 ofputil_cls_rule_to_match(&rule->cr, ofconn_get_flow_format(ofconn),
3006 &ofs->match, rule->flow_cookie, &cookie);
3007 put_32aligned_be64(&ofs->cookie, cookie);
3008 calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec);
3009 ofs->priority = htons(rule->cr.priority);
3010 ofs->idle_timeout = htons(rule->idle_timeout);
3011 ofs->hard_timeout = htons(rule->hard_timeout);
3012 memset(ofs->pad2, 0, sizeof ofs->pad2);
3013 put_32aligned_be64(&ofs->packet_count, htonll(packet_count));
3014 put_32aligned_be64(&ofs->byte_count, htonll(byte_count));
3015 if (rule->n_actions > 0) {
3016 memcpy(ofs->actions, rule->actions, act_len);
3021 is_valid_table(uint8_t table_id)
3023 if (table_id == 0 || table_id == 0xff) {
3026 /* It would probably be better to reply with an error but there doesn't
3027 * seem to be any appropriate value, so that might just be
3029 VLOG_WARN_RL(&rl, "controller asked for invalid table %"PRIu8,
3036 handle_flow_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3038 const struct ofp_flow_stats_request *fsr = ofputil_stats_body(oh);
3039 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3040 struct ofpbuf *reply;
3042 COVERAGE_INC(ofproto_flows_req);
3043 reply = start_ofp_stats_reply(oh, 1024);
3044 if (is_valid_table(fsr->table_id)) {
3045 struct cls_cursor cursor;
3046 struct cls_rule target;
3049 ofputil_cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0,
3051 cls_cursor_init(&cursor, &ofproto->cls, &target);
3052 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3053 put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply);
3056 ofconn_send_reply(ofconn, reply);
3062 put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule,
3063 ovs_be16 out_port, struct ofpbuf **replyp)
3065 struct nx_flow_stats *nfs;
3066 uint64_t packet_count, byte_count;
3067 size_t act_len, start_len;
3068 struct ofpbuf *reply;
3070 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3074 rule_get_stats(rule, &packet_count, &byte_count);
3076 act_len = sizeof *rule->actions * rule->n_actions;
3078 append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, ofconn, replyp);
3079 start_len = (*replyp)->size;
3082 nfs = ofpbuf_put_uninit(reply, sizeof *nfs);
3085 calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec);
3086 nfs->cookie = rule->flow_cookie;
3087 nfs->priority = htons(rule->cr.priority);
3088 nfs->idle_timeout = htons(rule->idle_timeout);
3089 nfs->hard_timeout = htons(rule->hard_timeout);
3090 nfs->match_len = htons(nx_put_match(reply, &rule->cr));
3091 memset(nfs->pad2, 0, sizeof nfs->pad2);
3092 nfs->packet_count = htonll(packet_count);
3093 nfs->byte_count = htonll(byte_count);
3094 if (rule->n_actions > 0) {
3095 ofpbuf_put(reply, rule->actions, act_len);
3097 nfs->length = htons(reply->size - start_len);
3101 handle_nxst_flow(struct ofconn *ofconn, const struct ofp_header *oh)
3103 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3104 struct nx_flow_stats_request *nfsr;
3105 struct cls_rule target;
3106 struct ofpbuf *reply;
3110 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3112 /* Dissect the message. */
3113 nfsr = ofpbuf_pull(&b, sizeof *nfsr);
3114 error = nx_pull_match(&b, ntohs(nfsr->match_len), 0, &target);
3119 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3122 COVERAGE_INC(ofproto_flows_req);
3123 reply = start_nxstats_reply(&nfsr->nsm, 1024);
3124 if (is_valid_table(nfsr->table_id)) {
3125 struct cls_cursor cursor;
3128 cls_cursor_init(&cursor, &ofproto->cls, &target);
3129 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3130 put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply);
3133 ofconn_send_reply(ofconn, reply);
3139 flow_stats_ds(struct rule *rule, struct ds *results)
3141 uint64_t packet_count, byte_count;
3142 size_t act_len = sizeof *rule->actions * rule->n_actions;
3144 rule_get_stats(rule, &packet_count, &byte_count);
3146 ds_put_format(results, "duration=%llds, ",
3147 (time_msec() - rule->created) / 1000);
3148 ds_put_format(results, "idle=%.3fs, ", (time_msec() - rule->used) / 1000.0);
3149 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3150 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3151 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3152 cls_rule_format(&rule->cr, results);
3153 ds_put_char(results, ',');
3155 ofp_print_actions(results, &rule->actions->header, act_len);
3157 ds_put_cstr(results, "drop");
3159 ds_put_cstr(results, "\n");
3162 /* Adds a pretty-printed description of all flows to 'results', including
3163 * hidden flows (e.g., set up by in-band control). */
3165 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3167 struct cls_cursor cursor;
3170 cls_cursor_init(&cursor, &p->cls, NULL);
3171 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3172 flow_stats_ds(rule, results);
3176 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
3177 * '*engine_type' and '*engine_id', respectively. */
3179 ofproto_get_netflow_ids(const struct ofproto *ofproto,
3180 uint8_t *engine_type, uint8_t *engine_id)
3182 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
3186 query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target,
3187 ovs_be16 out_port, uint8_t table_id,
3188 struct ofp_aggregate_stats_reply *oasr)
3190 uint64_t total_packets = 0;
3191 uint64_t total_bytes = 0;
3194 COVERAGE_INC(ofproto_agg_request);
3196 if (is_valid_table(table_id)) {
3197 struct cls_cursor cursor;
3200 cls_cursor_init(&cursor, &ofproto->cls, target);
3201 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3202 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
3203 uint64_t packet_count;
3204 uint64_t byte_count;
3206 rule_get_stats(rule, &packet_count, &byte_count);
3208 total_packets += packet_count;
3209 total_bytes += byte_count;
3215 oasr->flow_count = htonl(n_flows);
3216 put_32aligned_be64(&oasr->packet_count, htonll(total_packets));
3217 put_32aligned_be64(&oasr->byte_count, htonll(total_bytes));
3218 memset(oasr->pad, 0, sizeof oasr->pad);
3222 handle_aggregate_stats_request(struct ofconn *ofconn,
3223 const struct ofp_header *oh)
3225 const struct ofp_aggregate_stats_request *request = ofputil_stats_body(oh);
3226 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3227 struct ofp_aggregate_stats_reply *reply;
3228 struct cls_rule target;
3231 ofputil_cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0,
3234 msg = start_ofp_stats_reply(oh, sizeof *reply);
3235 reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg);
3236 query_aggregate_stats(ofproto, &target, request->out_port,
3237 request->table_id, reply);
3238 ofconn_send_reply(ofconn, msg);
3243 handle_nxst_aggregate(struct ofconn *ofconn, const struct ofp_header *oh)
3245 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3246 struct nx_aggregate_stats_request *request;
3247 struct ofp_aggregate_stats_reply *reply;
3248 struct cls_rule target;
3253 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3255 /* Dissect the message. */
3256 request = ofpbuf_pull(&b, sizeof *request);
3257 error = nx_pull_match(&b, ntohs(request->match_len), 0, &target);
3262 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3266 COVERAGE_INC(ofproto_flows_req);
3267 buf = start_nxstats_reply(&request->nsm, sizeof *reply);
3268 reply = ofpbuf_put_uninit(buf, sizeof *reply);
3269 query_aggregate_stats(ofproto, &target, request->out_port,
3270 request->table_id, reply);
3271 ofconn_send_reply(ofconn, buf);
3276 struct queue_stats_cbdata {
3277 struct ofconn *ofconn;
3278 struct ofport *ofport;
3283 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3284 const struct netdev_queue_stats *stats)
3286 struct ofp_queue_stats *reply;
3288 reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3289 reply->port_no = cbdata->ofport->opp.port_no;
3290 memset(reply->pad, 0, sizeof reply->pad);
3291 reply->queue_id = htonl(queue_id);
3292 put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
3293 put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
3294 put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
3298 handle_queue_stats_dump_cb(uint32_t queue_id,
3299 struct netdev_queue_stats *stats,
3302 struct queue_stats_cbdata *cbdata = cbdata_;
3304 put_queue_stats(cbdata, queue_id, stats);
3308 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3309 struct queue_stats_cbdata *cbdata)
3311 cbdata->ofport = port;
3312 if (queue_id == OFPQ_ALL) {
3313 netdev_dump_queue_stats(port->netdev,
3314 handle_queue_stats_dump_cb, cbdata);
3316 struct netdev_queue_stats stats;
3318 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3319 put_queue_stats(cbdata, queue_id, &stats);
3325 handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3327 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3328 const struct ofp_queue_stats_request *qsr;
3329 struct queue_stats_cbdata cbdata;
3330 struct ofport *port;
3331 unsigned int port_no;
3334 qsr = ofputil_stats_body(oh);
3336 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3339 COVERAGE_INC(ofproto_queue_req);
3341 cbdata.ofconn = ofconn;
3342 cbdata.msg = start_ofp_stats_reply(oh, 128);
3344 port_no = ntohs(qsr->port_no);
3345 queue_id = ntohl(qsr->queue_id);
3346 if (port_no == OFPP_ALL) {
3347 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3348 handle_queue_stats_for_port(port, queue_id, &cbdata);
3350 } else if (port_no < ofproto->max_ports) {
3351 port = get_port(ofproto, ofp_port_to_odp_port(port_no));
3353 handle_queue_stats_for_port(port, queue_id, &cbdata);
3356 ofpbuf_delete(cbdata.msg);
3357 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3359 ofconn_send_reply(ofconn, cbdata.msg);
3364 /* Updates 'facet''s used time. Caller is responsible for calling
3365 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3367 facet_update_time(struct ofproto *ofproto, struct facet *facet,
3370 if (used > facet->used) {
3372 if (used > facet->rule->used) {
3373 facet->rule->used = used;
3375 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3379 /* Folds the statistics from 'stats' into the counters in 'facet'.
3381 * Because of the meaning of a facet's counters, it only makes sense to do this
3382 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3383 * packet that was sent by hand or if it represents statistics that have been
3384 * cleared out of the datapath. */
3386 facet_update_stats(struct ofproto *ofproto, struct facet *facet,
3387 const struct dpif_flow_stats *stats)
3389 if (stats->n_packets || stats->used > facet->used) {
3390 facet_update_time(ofproto, facet, stats->used);
3391 facet->packet_count += stats->n_packets;
3392 facet->byte_count += stats->n_bytes;
3393 facet_push_stats(ofproto, facet);
3394 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3399 facet_push_stats(struct ofproto *ofproto, struct facet *facet)
3401 uint64_t rs_packets, rs_bytes;
3403 assert(facet->packet_count >= facet->rs_packet_count);
3404 assert(facet->byte_count >= facet->rs_byte_count);
3405 assert(facet->used >= facet->rs_used);
3407 rs_packets = facet->packet_count - facet->rs_packet_count;
3408 rs_bytes = facet->byte_count - facet->rs_byte_count;
3410 if (rs_packets || rs_bytes || facet->used > facet->rs_used) {
3411 facet->rs_packet_count = facet->packet_count;
3412 facet->rs_byte_count = facet->byte_count;
3413 facet->rs_used = facet->used;
3415 flow_push_stats(ofproto, facet->rule, &facet->flow,
3416 rs_packets, rs_bytes, facet->used);
3420 struct ofproto_push {
3421 struct action_xlate_ctx ctx;
3428 push_resubmit(struct action_xlate_ctx *ctx, struct rule *rule)
3430 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3433 rule->packet_count += push->packets;
3434 rule->byte_count += push->bytes;
3435 rule->used = MAX(push->used, rule->used);
3439 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3440 * 'rule''s actions. */
3442 flow_push_stats(struct ofproto *ofproto, const struct rule *rule,
3443 struct flow *flow, uint64_t packets, uint64_t bytes,
3446 struct ofproto_push push;
3448 push.packets = packets;
3452 action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL);
3453 push.ctx.resubmit_hook = push_resubmit;
3454 ofpbuf_delete(xlate_actions(&push.ctx, rule->actions, rule->n_actions));
3457 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3458 * in which no matching flow already exists in the flow table.
3460 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3461 * ofp_actions, to the ofproto's flow table. Returns 0 on success or an
3462 * OpenFlow error code as encoded by ofp_mkerr() on failure.
3464 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3467 add_flow(struct ofconn *ofconn, struct flow_mod *fm)
3469 struct ofproto *p = ofconn_get_ofproto(ofconn);
3470 struct ofpbuf *packet;
3475 if (fm->flags & OFPFF_CHECK_OVERLAP
3476 && classifier_rule_overlaps(&p->cls, &fm->cr)) {
3477 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3481 if (fm->buffer_id != UINT32_MAX) {
3482 error = ofconn_pktbuf_retrieve(ofconn, fm->buffer_id,
3486 in_port = UINT16_MAX;
3489 rule = rule_create(&fm->cr, fm->actions, fm->n_actions,
3490 fm->idle_timeout, fm->hard_timeout, fm->cookie,
3491 fm->flags & OFPFF_SEND_FLOW_REM);
3492 rule_insert(p, rule);
3494 rule_execute(p, rule, in_port, packet);
3499 static struct rule *
3500 find_flow_strict(struct ofproto *p, const struct flow_mod *fm)
3502 return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr));
3506 send_buffered_packet(struct ofconn *ofconn,
3507 struct rule *rule, uint32_t buffer_id)
3509 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3510 struct ofpbuf *packet;
3514 if (buffer_id == UINT32_MAX) {
3518 error = ofconn_pktbuf_retrieve(ofconn, buffer_id, &packet, &in_port);
3523 rule_execute(ofproto, rule, in_port, packet);
3528 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3530 struct modify_flows_cbdata {
3531 struct ofproto *ofproto;
3532 const struct flow_mod *fm;
3536 static int modify_flow(struct ofproto *, const struct flow_mod *,
3539 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3540 * encoded by ofp_mkerr() on failure.
3542 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3545 modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm)
3547 struct ofproto *p = ofconn_get_ofproto(ofconn);
3548 struct rule *match = NULL;
3549 struct cls_cursor cursor;
3552 cls_cursor_init(&cursor, &p->cls, &fm->cr);
3553 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3554 if (!rule_is_hidden(rule)) {
3556 modify_flow(p, fm, rule);
3561 /* This credits the packet to whichever flow happened to match last.
3562 * That's weird. Maybe we should do a lookup for the flow that
3563 * actually matches the packet? Who knows. */
3564 send_buffered_packet(ofconn, match, fm->buffer_id);
3567 return add_flow(ofconn, fm);
3571 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3572 * code as encoded by ofp_mkerr() on failure.
3574 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3577 modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm)
3579 struct ofproto *p = ofconn_get_ofproto(ofconn);
3580 struct rule *rule = find_flow_strict(p, fm);
3581 if (rule && !rule_is_hidden(rule)) {
3582 modify_flow(p, fm, rule);
3583 return send_buffered_packet(ofconn, rule, fm->buffer_id);
3585 return add_flow(ofconn, fm);
3589 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3590 * been identified as a flow in 'p''s flow table to be modified, by changing
3591 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3592 * ofp_action[] structures). */
3594 modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule)
3596 size_t actions_len = fm->n_actions * sizeof *rule->actions;
3598 rule->flow_cookie = fm->cookie;
3600 /* If the actions are the same, do nothing. */
3601 if (fm->n_actions == rule->n_actions
3603 || !memcmp(fm->actions, rule->actions, actions_len))) {
3607 /* Replace actions. */
3608 free(rule->actions);
3609 rule->actions = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL;
3610 rule->n_actions = fm->n_actions;
3612 p->need_revalidate = true;
3617 /* OFPFC_DELETE implementation. */
3619 static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port);
3621 /* Implements OFPFC_DELETE. */
3623 delete_flows_loose(struct ofproto *p, const struct flow_mod *fm)
3625 struct rule *rule, *next_rule;
3626 struct cls_cursor cursor;
3628 cls_cursor_init(&cursor, &p->cls, &fm->cr);
3629 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
3630 delete_flow(p, rule, htons(fm->out_port));
3634 /* Implements OFPFC_DELETE_STRICT. */
3636 delete_flow_strict(struct ofproto *p, struct flow_mod *fm)
3638 struct rule *rule = find_flow_strict(p, fm);
3640 delete_flow(p, rule, htons(fm->out_port));
3644 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3645 * been identified as a flow to delete from 'p''s flow table, by deleting the
3646 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3649 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3650 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3651 * specified 'out_port'. */
3653 delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port)
3655 if (rule_is_hidden(rule)) {
3659 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
3663 rule_send_removed(p, rule, OFPRR_DELETE);
3664 rule_remove(p, rule);
3668 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3670 struct ofproto *p = ofconn_get_ofproto(ofconn);
3674 error = reject_slave_controller(ofconn, "flow_mod");
3679 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_flow_format(ofconn));
3684 /* We do not support the emergency flow cache. It will hopefully get
3685 * dropped from OpenFlow in the near future. */
3686 if (fm.flags & OFPFF_EMERG) {
3687 /* There isn't a good fit for an error code, so just state that the
3688 * flow table is full. */
3689 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3692 error = validate_actions(fm.actions, fm.n_actions,
3693 &fm.cr.flow, p->max_ports);
3698 switch (fm.command) {
3700 return add_flow(ofconn, &fm);
3703 return modify_flows_loose(ofconn, &fm);
3705 case OFPFC_MODIFY_STRICT:
3706 return modify_flow_strict(ofconn, &fm);
3709 delete_flows_loose(p, &fm);
3712 case OFPFC_DELETE_STRICT:
3713 delete_flow_strict(p, &fm);
3717 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3722 handle_tun_id_from_cookie(struct ofconn *ofconn, const struct ofp_header *oh)
3724 const struct nxt_tun_id_cookie *msg
3725 = (const struct nxt_tun_id_cookie *) oh;
3726 enum nx_flow_format flow_format;
3728 flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10;
3729 ofconn_set_flow_format(ofconn, flow_format);
3735 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
3737 struct nx_role_request *nrr = (struct nx_role_request *) oh;
3738 struct nx_role_request *reply;
3742 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY) {
3743 VLOG_WARN_RL(&rl, "ignoring role request on service connection");
3744 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3747 role = ntohl(nrr->role);
3748 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3749 && role != NX_ROLE_SLAVE) {
3750 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
3752 /* There's no good error code for this. */
3753 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
3756 ofconn_set_role(ofconn, role);
3758 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
3759 reply->role = htonl(role);
3760 ofconn_send_reply(ofconn, buf);
3766 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
3768 const struct nxt_set_flow_format *msg
3769 = (const struct nxt_set_flow_format *) oh;
3772 format = ntohl(msg->format);
3773 if (format == NXFF_OPENFLOW10
3774 || format == NXFF_TUN_ID_FROM_COOKIE
3775 || format == NXFF_NXM) {
3776 ofconn_set_flow_format(ofconn, format);
3779 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3784 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
3786 struct ofp_header *ob;
3789 /* Currently, everything executes synchronously, so we can just
3790 * immediately send the barrier reply. */
3791 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
3792 ofconn_send_reply(ofconn, buf);
3797 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
3799 const struct ofp_header *oh = msg->data;
3800 const struct ofputil_msg_type *type;
3803 error = ofputil_decode_msg_type(oh, &type);
3808 switch (ofputil_msg_type_code(type)) {
3809 /* OpenFlow requests. */
3810 case OFPUTIL_OFPT_ECHO_REQUEST:
3811 return handle_echo_request(ofconn, oh);
3813 case OFPUTIL_OFPT_FEATURES_REQUEST:
3814 return handle_features_request(ofconn, oh);
3816 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
3817 return handle_get_config_request(ofconn, oh);
3819 case OFPUTIL_OFPT_SET_CONFIG:
3820 return handle_set_config(ofconn, msg->data);
3822 case OFPUTIL_OFPT_PACKET_OUT:
3823 return handle_packet_out(ofconn, oh);
3825 case OFPUTIL_OFPT_PORT_MOD:
3826 return handle_port_mod(ofconn, oh);
3828 case OFPUTIL_OFPT_FLOW_MOD:
3829 return handle_flow_mod(ofconn, oh);
3831 case OFPUTIL_OFPT_BARRIER_REQUEST:
3832 return handle_barrier_request(ofconn, oh);
3834 /* OpenFlow replies. */
3835 case OFPUTIL_OFPT_ECHO_REPLY:
3838 /* Nicira extension requests. */
3839 case OFPUTIL_NXT_TUN_ID_FROM_COOKIE:
3840 return handle_tun_id_from_cookie(ofconn, oh);
3842 case OFPUTIL_NXT_ROLE_REQUEST:
3843 return handle_role_request(ofconn, oh);
3845 case OFPUTIL_NXT_SET_FLOW_FORMAT:
3846 return handle_nxt_set_flow_format(ofconn, oh);
3848 case OFPUTIL_NXT_FLOW_MOD:
3849 return handle_flow_mod(ofconn, oh);
3851 /* OpenFlow statistics requests. */
3852 case OFPUTIL_OFPST_DESC_REQUEST:
3853 return handle_desc_stats_request(ofconn, oh);
3855 case OFPUTIL_OFPST_FLOW_REQUEST:
3856 return handle_flow_stats_request(ofconn, oh);
3858 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
3859 return handle_aggregate_stats_request(ofconn, oh);
3861 case OFPUTIL_OFPST_TABLE_REQUEST:
3862 return handle_table_stats_request(ofconn, oh);
3864 case OFPUTIL_OFPST_PORT_REQUEST:
3865 return handle_port_stats_request(ofconn, oh);
3867 case OFPUTIL_OFPST_QUEUE_REQUEST:
3868 return handle_queue_stats_request(ofconn, oh);
3870 /* Nicira extension statistics requests. */
3871 case OFPUTIL_NXST_FLOW_REQUEST:
3872 return handle_nxst_flow(ofconn, oh);
3874 case OFPUTIL_NXST_AGGREGATE_REQUEST:
3875 return handle_nxst_aggregate(ofconn, oh);
3877 case OFPUTIL_INVALID:
3878 case OFPUTIL_OFPT_HELLO:
3879 case OFPUTIL_OFPT_ERROR:
3880 case OFPUTIL_OFPT_FEATURES_REPLY:
3881 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
3882 case OFPUTIL_OFPT_PACKET_IN:
3883 case OFPUTIL_OFPT_FLOW_REMOVED:
3884 case OFPUTIL_OFPT_PORT_STATUS:
3885 case OFPUTIL_OFPT_BARRIER_REPLY:
3886 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
3887 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
3888 case OFPUTIL_OFPST_DESC_REPLY:
3889 case OFPUTIL_OFPST_FLOW_REPLY:
3890 case OFPUTIL_OFPST_QUEUE_REPLY:
3891 case OFPUTIL_OFPST_PORT_REPLY:
3892 case OFPUTIL_OFPST_TABLE_REPLY:
3893 case OFPUTIL_OFPST_AGGREGATE_REPLY:
3894 case OFPUTIL_NXT_ROLE_REPLY:
3895 case OFPUTIL_NXT_FLOW_REMOVED:
3896 case OFPUTIL_NXST_FLOW_REPLY:
3897 case OFPUTIL_NXST_AGGREGATE_REPLY:
3899 if (VLOG_IS_WARN_ENABLED()) {
3900 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3901 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3904 if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) {
3905 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
3907 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3913 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
3915 int error = handle_openflow__(ofconn, ofp_msg);
3917 send_error_oh(ofconn, ofp_msg->data, error);
3919 COVERAGE_INC(ofproto_recv_openflow);
3923 handle_miss_upcall(struct ofproto *p, struct dpif_upcall *upcall)
3925 struct facet *facet;
3928 /* Obtain in_port and tun_id, at least. */
3929 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
3931 /* Set header pointers in 'flow'. */
3932 flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow);
3934 if (cfm_should_process_flow(&flow)) {
3935 ofproto_process_cfm(p, &flow, upcall->packet);
3936 ofpbuf_delete(upcall->packet);
3938 } else if (p->ofhooks->special_cb
3939 && !p->ofhooks->special_cb(&flow, upcall->packet, p->aux)) {
3940 ofpbuf_delete(upcall->packet);
3944 /* Check with in-band control to see if this packet should be sent
3945 * to the local port regardless of the flow table. */
3946 if (connmgr_msg_in_hook(p->connmgr, &flow, upcall->packet)) {
3947 ofproto_send_packet(p, ODPP_LOCAL, 0, upcall->packet);
3950 facet = facet_lookup_valid(p, &flow);
3952 struct rule *rule = rule_lookup(p, &flow);
3954 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3955 struct ofport *port = get_port(p, flow.in_port);
3957 if (port->opp.config & htonl(OFPPC_NO_PACKET_IN)) {
3958 COVERAGE_INC(ofproto_no_packet_in);
3959 /* XXX install 'drop' flow entry */
3960 ofpbuf_delete(upcall->packet);
3964 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
3968 COVERAGE_INC(ofproto_packet_in);
3969 send_packet_in(p, upcall, &flow, false);
3973 facet = facet_create(p, rule, &flow, upcall->packet);
3974 } else if (!facet->may_install) {
3975 /* The facet is not installable, that is, we need to process every
3976 * packet, so process the current packet's actions into 'facet'. */
3977 facet_make_actions(p, facet, upcall->packet);
3980 if (facet->rule->cr.priority == FAIL_OPEN_PRIORITY) {
3982 * Extra-special case for fail-open mode.
3984 * We are in fail-open mode and the packet matched the fail-open rule,
3985 * but we are connected to a controller too. We should send the packet
3986 * up to the controller in the hope that it will try to set up a flow
3987 * and thereby allow us to exit fail-open.
3989 * See the top-level comment in fail-open.c for more information.
3991 send_packet_in(p, upcall, &flow, true);
3994 facet_execute(p, facet, upcall->packet);
3995 facet_install(p, facet, false);
3999 handle_upcall(struct ofproto *p, struct dpif_upcall *upcall)
4003 switch (upcall->type) {
4004 case DPIF_UC_ACTION:
4005 COVERAGE_INC(ofproto_ctlr_action);
4006 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4007 send_packet_in(p, upcall, &flow, false);
4010 case DPIF_UC_SAMPLE:
4012 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4013 ofproto_sflow_received(p->sflow, upcall, &flow);
4015 ofpbuf_delete(upcall->packet);
4019 handle_miss_upcall(p, upcall);
4022 case DPIF_N_UC_TYPES:
4024 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4029 /* Flow expiration. */
4031 static int ofproto_dp_max_idle(const struct ofproto *);
4032 static void ofproto_update_stats(struct ofproto *);
4033 static void rule_expire(struct ofproto *, struct rule *);
4034 static void ofproto_expire_facets(struct ofproto *, int dp_max_idle);
4036 /* This function is called periodically by ofproto_run(). Its job is to
4037 * collect updates for the flows that have been installed into the datapath,
4038 * most importantly when they last were used, and then use that information to
4039 * expire flows that have not been used recently.
4041 * Returns the number of milliseconds after which it should be called again. */
4043 ofproto_expire(struct ofproto *ofproto)
4045 struct rule *rule, *next_rule;
4046 struct cls_cursor cursor;
4049 /* Update stats for each flow in the datapath. */
4050 ofproto_update_stats(ofproto);
4052 /* Expire facets that have been idle too long. */
4053 dp_max_idle = ofproto_dp_max_idle(ofproto);
4054 ofproto_expire_facets(ofproto, dp_max_idle);
4056 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
4057 cls_cursor_init(&cursor, &ofproto->cls, NULL);
4058 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4059 rule_expire(ofproto, rule);
4062 /* Let the hook know that we're at a stable point: all outstanding data
4063 * in existing flows has been accounted to the account_cb. Thus, the
4064 * hook can now reasonably do operations that depend on having accurate
4065 * flow volume accounting (currently, that's just bond rebalancing). */
4066 if (ofproto->ofhooks->account_checkpoint_cb) {
4067 ofproto->ofhooks->account_checkpoint_cb(ofproto->aux);
4070 return MIN(dp_max_idle, 1000);
4073 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4075 * This function also pushes statistics updates to rules which each facet
4076 * resubmits into. Generally these statistics will be accurate. However, if a
4077 * facet changes the rule it resubmits into at some time in between
4078 * ofproto_update_stats() runs, it is possible that statistics accrued to the
4079 * old rule will be incorrectly attributed to the new rule. This could be
4080 * avoided by calling ofproto_update_stats() whenever rules are created or
4081 * deleted. However, the performance impact of making so many calls to the
4082 * datapath do not justify the benefit of having perfectly accurate statistics.
4085 ofproto_update_stats(struct ofproto *p)
4087 const struct dpif_flow_stats *stats;
4088 struct dpif_flow_dump dump;
4089 const struct nlattr *key;
4092 dpif_flow_dump_start(&dump, p->dpif);
4093 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4094 struct facet *facet;
4097 if (odp_flow_key_to_flow(key, key_len, &flow)) {
4101 odp_flow_key_format(key, key_len, &s);
4102 VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s",
4108 facet = facet_find(p, &flow);
4110 if (facet && facet->installed) {
4112 if (stats->n_packets >= facet->dp_packet_count) {
4113 facet->packet_count += stats->n_packets - facet->dp_packet_count;
4115 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4118 if (stats->n_bytes >= facet->dp_byte_count) {
4119 facet->byte_count += stats->n_bytes - facet->dp_byte_count;
4121 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4124 facet->dp_packet_count = stats->n_packets;
4125 facet->dp_byte_count = stats->n_bytes;
4127 facet_update_time(p, facet, stats->used);
4128 facet_account(p, facet, stats->n_bytes);
4129 facet_push_stats(p, facet);
4131 /* There's a flow in the datapath that we know nothing about.
4133 COVERAGE_INC(ofproto_unexpected_rule);
4134 dpif_flow_del(p->dpif, key, key_len, NULL);
4137 dpif_flow_dump_done(&dump);
4140 /* Calculates and returns the number of milliseconds of idle time after which
4141 * facets should expire from the datapath and we should fold their statistics
4142 * into their parent rules in userspace. */
4144 ofproto_dp_max_idle(const struct ofproto *ofproto)
4147 * Idle time histogram.
4149 * Most of the time a switch has a relatively small number of facets. When
4150 * this is the case we might as well keep statistics for all of them in
4151 * userspace and to cache them in the kernel datapath for performance as
4154 * As the number of facets increases, the memory required to maintain
4155 * statistics about them in userspace and in the kernel becomes
4156 * significant. However, with a large number of facets it is likely that
4157 * only a few of them are "heavy hitters" that consume a large amount of
4158 * bandwidth. At this point, only heavy hitters are worth caching in the
4159 * kernel and maintaining in userspaces; other facets we can discard.
4161 * The technique used to compute the idle time is to build a histogram with
4162 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
4163 * that is installed in the kernel gets dropped in the appropriate bucket.
4164 * After the histogram has been built, we compute the cutoff so that only
4165 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
4166 * cached. At least the most-recently-used bucket of facets is kept, so
4167 * actually an arbitrary number of facets can be kept in any given
4168 * expiration run (though the next run will delete most of those unless
4169 * they receive additional data).
4171 * This requires a second pass through the facets, in addition to the pass
4172 * made by ofproto_update_stats(), because the former function never looks
4173 * at uninstallable facets.
4175 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4176 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4177 int buckets[N_BUCKETS] = { 0 };
4178 struct facet *facet;
4183 total = hmap_count(&ofproto->facets);
4184 if (total <= 1000) {
4185 return N_BUCKETS * BUCKET_WIDTH;
4188 /* Build histogram. */
4190 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
4191 long long int idle = now - facet->used;
4192 int bucket = (idle <= 0 ? 0
4193 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4194 : (unsigned int) idle / BUCKET_WIDTH);
4198 /* Find the first bucket whose flows should be expired. */
4199 for (bucket = 0; bucket < N_BUCKETS; bucket++) {
4200 if (buckets[bucket]) {
4203 subtotal += buckets[bucket++];
4204 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
4209 if (VLOG_IS_DBG_ENABLED()) {
4213 ds_put_cstr(&s, "keep");
4214 for (i = 0; i < N_BUCKETS; i++) {
4216 ds_put_cstr(&s, ", drop");
4219 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4222 VLOG_INFO("%s: %s (msec:count)", ofproto->name, ds_cstr(&s));
4226 return bucket * BUCKET_WIDTH;
4230 facet_active_timeout(struct ofproto *ofproto, struct facet *facet)
4232 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
4233 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
4234 struct ofexpired expired;
4236 if (facet->installed) {
4237 struct dpif_flow_stats stats;
4239 facet_put__(ofproto, facet, facet->actions, facet->actions_len,
4241 facet_update_stats(ofproto, facet, &stats);
4244 expired.flow = facet->flow;
4245 expired.packet_count = facet->packet_count;
4246 expired.byte_count = facet->byte_count;
4247 expired.used = facet->used;
4248 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4253 ofproto_expire_facets(struct ofproto *ofproto, int dp_max_idle)
4255 long long int cutoff = time_msec() - dp_max_idle;
4256 struct facet *facet, *next_facet;
4258 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
4259 facet_active_timeout(ofproto, facet);
4260 if (facet->used < cutoff) {
4261 facet_remove(ofproto, facet);
4266 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4267 * then delete it entirely. */
4269 rule_expire(struct ofproto *ofproto, struct rule *rule)
4271 struct facet *facet, *next_facet;
4275 /* Has 'rule' expired? */
4277 if (rule->hard_timeout
4278 && now > rule->created + rule->hard_timeout * 1000) {
4279 reason = OFPRR_HARD_TIMEOUT;
4280 } else if (rule->idle_timeout && list_is_empty(&rule->facets)
4281 && now >rule->used + rule->idle_timeout * 1000) {
4282 reason = OFPRR_IDLE_TIMEOUT;
4287 COVERAGE_INC(ofproto_expired);
4289 /* Update stats. (This is a no-op if the rule expired due to an idle
4290 * timeout, because that only happens when the rule has no facets left.) */
4291 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4292 facet_remove(ofproto, facet);
4295 /* Get rid of the rule. */
4296 if (!rule_is_hidden(rule)) {
4297 rule_send_removed(ofproto, rule, reason);
4299 rule_remove(ofproto, rule);
4303 rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason)
4305 struct ofputil_flow_removed fr;
4307 if (!rule->send_flow_removed) {
4312 fr.cookie = rule->flow_cookie;
4314 calc_flow_duration__(rule->created, &fr.duration_sec, &fr.duration_nsec);
4315 fr.idle_timeout = rule->idle_timeout;
4316 fr.packet_count = rule->packet_count;
4317 fr.byte_count = rule->byte_count;
4319 connmgr_send_flow_removed(p->connmgr, &fr);
4322 /* Obtains statistics for 'rule' and stores them in '*packets' and '*bytes'.
4323 * The returned statistics include statistics for all of 'rule''s facets. */
4325 rule_get_stats(const struct rule *rule, uint64_t *packets, uint64_t *bytes)
4328 struct facet *facet;
4330 /* Start from historical data for 'rule' itself that are no longer tracked
4331 * in facets. This counts, for example, facets that have expired. */
4332 p = rule->packet_count;
4333 b = rule->byte_count;
4335 /* Add any statistics that are tracked by facets. This includes
4336 * statistical data recently updated by ofproto_update_stats() as well as
4337 * stats for packets that were executed "by hand" via dpif_execute(). */
4338 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4339 p += facet->packet_count;
4340 b += facet->byte_count;
4347 /* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an
4348 * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to
4349 * their individual configurations.
4351 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
4352 * Otherwise, ownership is transferred to this function. */
4354 send_packet_in(struct ofproto *ofproto, struct dpif_upcall *upcall,
4355 const struct flow *flow, bool clone)
4357 struct ofputil_packet_in pin;
4359 pin.packet = upcall->packet;
4360 pin.in_port = odp_port_to_ofp_port(flow->in_port);
4361 pin.reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION;
4362 pin.buffer_id = 0; /* not yet known */
4363 pin.send_len = upcall->userdata;
4364 connmgr_send_packet_in(ofproto->connmgr, upcall, flow,
4365 clone ? NULL : upcall->packet);
4369 pick_datapath_id(const struct ofproto *ofproto)
4371 const struct ofport *port;
4373 port = get_port(ofproto, ODPP_LOCAL);
4375 uint8_t ea[ETH_ADDR_LEN];
4378 error = netdev_get_etheraddr(port->netdev, ea);
4380 return eth_addr_to_uint64(ea);
4382 VLOG_WARN("could not get MAC address for %s (%s)",
4383 netdev_get_name(port->netdev), strerror(error));
4385 return ofproto->fallback_dpid;
4389 pick_fallback_dpid(void)
4391 uint8_t ea[ETH_ADDR_LEN];
4392 eth_addr_nicira_random(ea);
4393 return eth_addr_to_uint64(ea);
4396 static struct ofproto *
4397 ofproto_lookup(const char *name)
4399 struct ofproto *ofproto;
4401 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
4403 if (!strcmp(ofproto->name, name)) {
4411 ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED,
4412 void *aux OVS_UNUSED)
4414 struct ofproto *ofproto;
4418 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
4419 ds_put_format(&results, "%s\n", ofproto->name);
4421 unixctl_command_reply(conn, 200, ds_cstr(&results));
4422 ds_destroy(&results);
4425 struct ofproto_trace {
4426 struct action_xlate_ctx ctx;
4432 trace_format_rule(struct ds *result, int level, const struct rule *rule)
4434 ds_put_char_multiple(result, '\t', level);
4436 ds_put_cstr(result, "No match\n");
4440 ds_put_format(result, "Rule: cookie=%#"PRIx64" ",
4441 ntohll(rule->flow_cookie));
4442 cls_rule_format(&rule->cr, result);
4443 ds_put_char(result, '\n');
4445 ds_put_char_multiple(result, '\t', level);
4446 ds_put_cstr(result, "OpenFlow ");
4447 ofp_print_actions(result, (const struct ofp_action_header *) rule->actions,
4448 rule->n_actions * sizeof *rule->actions);
4449 ds_put_char(result, '\n');
4453 trace_format_flow(struct ds *result, int level, const char *title,
4454 struct ofproto_trace *trace)
4456 ds_put_char_multiple(result, '\t', level);
4457 ds_put_format(result, "%s: ", title);
4458 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
4459 ds_put_cstr(result, "unchanged");
4461 flow_format(result, &trace->ctx.flow);
4462 trace->flow = trace->ctx.flow;
4464 ds_put_char(result, '\n');
4468 trace_resubmit(struct action_xlate_ctx *ctx, struct rule *rule)
4470 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
4471 struct ds *result = trace->result;
4473 ds_put_char(result, '\n');
4474 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
4475 trace_format_rule(result, ctx->recurse + 1, rule);
4479 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
4480 void *aux OVS_UNUSED)
4482 char *dpname, *in_port_s, *tun_id_s, *packet_s;
4483 char *args = xstrdup(args_);
4484 char *save_ptr = NULL;
4485 struct ofproto *ofproto;
4486 struct ofpbuf packet;
4494 ofpbuf_init(&packet, strlen(args) / 2);
4497 dpname = strtok_r(args, " ", &save_ptr);
4498 tun_id_s = strtok_r(NULL, " ", &save_ptr);
4499 in_port_s = strtok_r(NULL, " ", &save_ptr);
4500 packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
4501 if (!dpname || !in_port_s || !packet_s) {
4502 unixctl_command_reply(conn, 501, "Bad command syntax");
4506 ofproto = ofproto_lookup(dpname);
4508 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
4513 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
4514 in_port = ofp_port_to_odp_port(atoi(in_port_s));
4516 packet_s = ofpbuf_put_hex(&packet, packet_s, NULL);
4517 packet_s += strspn(packet_s, " ");
4518 if (*packet_s != '\0') {
4519 unixctl_command_reply(conn, 501, "Trailing garbage in command");
4522 if (packet.size < ETH_HEADER_LEN) {
4523 unixctl_command_reply(conn, 501, "Packet data too short for Ethernet");
4527 ds_put_cstr(&result, "Packet: ");
4528 s = ofp_packet_to_string(packet.data, packet.size, packet.size);
4529 ds_put_cstr(&result, s);
4532 flow_extract(&packet, tun_id, in_port, &flow);
4533 ds_put_cstr(&result, "Flow: ");
4534 flow_format(&result, &flow);
4535 ds_put_char(&result, '\n');
4537 rule = rule_lookup(ofproto, &flow);
4538 trace_format_rule(&result, 0, rule);
4540 struct ofproto_trace trace;
4541 struct ofpbuf *odp_actions;
4543 trace.result = &result;
4545 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet);
4546 trace.ctx.resubmit_hook = trace_resubmit;
4547 odp_actions = xlate_actions(&trace.ctx,
4548 rule->actions, rule->n_actions);
4550 ds_put_char(&result, '\n');
4551 trace_format_flow(&result, 0, "Final flow", &trace);
4552 ds_put_cstr(&result, "Datapath actions: ");
4553 format_odp_actions(&result, odp_actions->data, odp_actions->size);
4554 ofpbuf_delete(odp_actions);
4557 unixctl_command_reply(conn, 200, ds_cstr(&result));
4560 ds_destroy(&result);
4561 ofpbuf_uninit(&packet);
4566 ofproto_unixctl_init(void)
4568 static bool registered;
4574 unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL);
4575 unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL);
4579 default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
4580 struct ofpbuf *odp_actions, tag_type *tags,
4581 uint16_t *nf_output_iface, void *ofproto_)
4583 struct ofproto *ofproto = ofproto_;
4584 struct mac_entry *dst_mac;
4586 /* Drop frames for reserved multicast addresses. */
4587 if (eth_addr_is_reserved(flow->dl_dst)) {
4591 /* Learn source MAC (but don't try to learn from revalidation). */
4593 && mac_learning_may_learn(ofproto->ml, flow->dl_src, 0)) {
4594 struct mac_entry *src_mac;
4596 src_mac = mac_learning_insert(ofproto->ml, flow->dl_src, 0);
4597 if (mac_entry_is_new(src_mac) || src_mac->port.i != flow->in_port) {
4598 /* The log messages here could actually be useful in debugging,
4599 * so keep the rate limit relatively high. */
4600 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4601 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
4602 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
4604 ofproto_revalidate(ofproto,
4605 mac_learning_changed(ofproto->ml, src_mac));
4606 src_mac->port.i = flow->in_port;
4610 /* Determine output port. */
4611 dst_mac = mac_learning_lookup(ofproto->ml, flow->dl_dst, 0, tags);
4613 flood_packets(ofproto, flow->in_port, htonl(OFPPC_NO_FLOOD),
4614 nf_output_iface, odp_actions);
4616 int out_port = dst_mac->port.i;
4617 if (out_port != flow->in_port) {
4618 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, out_port);
4619 *nf_output_iface = out_port;
4628 static const struct ofhooks default_ofhooks = {
4629 default_normal_ofhook_cb,