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"
60 #include "stream-ssl.h"
64 #include "unaligned.h"
69 VLOG_DEFINE_THIS_MODULE(ofproto);
71 COVERAGE_DEFINE(facet_changed_rule);
72 COVERAGE_DEFINE(facet_revalidate);
73 COVERAGE_DEFINE(odp_overflow);
74 COVERAGE_DEFINE(ofproto_agg_request);
75 COVERAGE_DEFINE(ofproto_costly_flags);
76 COVERAGE_DEFINE(ofproto_ctlr_action);
77 COVERAGE_DEFINE(ofproto_del_rule);
78 COVERAGE_DEFINE(ofproto_error);
79 COVERAGE_DEFINE(ofproto_expiration);
80 COVERAGE_DEFINE(ofproto_expired);
81 COVERAGE_DEFINE(ofproto_flows_req);
82 COVERAGE_DEFINE(ofproto_flush);
83 COVERAGE_DEFINE(ofproto_invalidated);
84 COVERAGE_DEFINE(ofproto_no_packet_in);
85 COVERAGE_DEFINE(ofproto_ofp2odp);
86 COVERAGE_DEFINE(ofproto_packet_in);
87 COVERAGE_DEFINE(ofproto_packet_out);
88 COVERAGE_DEFINE(ofproto_queue_req);
89 COVERAGE_DEFINE(ofproto_recv_openflow);
90 COVERAGE_DEFINE(ofproto_reinit_ports);
91 COVERAGE_DEFINE(ofproto_unexpected_rule);
92 COVERAGE_DEFINE(ofproto_uninstallable);
93 COVERAGE_DEFINE(ofproto_update_port);
95 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
96 * flow translation. */
97 #define MAX_RESUBMIT_RECURSION 16
102 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
103 struct netdev *netdev;
104 struct ofp_phy_port opp;
106 struct cfm *cfm; /* Connectivity Fault Management, if any. */
109 static void ofport_free(struct ofport *);
110 static void ofport_run(struct ofproto *, struct ofport *);
111 static void ofport_wait(struct ofport *);
113 struct action_xlate_ctx {
114 /* action_xlate_ctx_init() initializes these members. */
117 struct ofproto *ofproto;
119 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
120 * this flow when actions change header fields. */
123 /* The packet corresponding to 'flow', or a null pointer if we are
124 * revalidating without a packet to refer to. */
125 const struct ofpbuf *packet;
127 /* If nonnull, called just before executing a resubmit action.
129 * This is normally null so the client has to set it manually after
130 * calling action_xlate_ctx_init(). */
131 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule *);
133 /* If true, the speciality of 'flow' should be checked before executing
134 * its actions. If special_cb returns false on 'flow' rendered
135 * uninstallable and no actions will be executed. */
138 /* xlate_actions() initializes and uses these members. The client might want
139 * to look at them after it returns. */
141 struct ofpbuf *odp_actions; /* Datapath actions. */
142 tag_type tags; /* Tags associated with OFPP_NORMAL actions. */
143 bool may_set_up_flow; /* True ordinarily; false if the actions must
144 * be reassessed for every packet. */
145 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
147 /* xlate_actions() initializes and uses these members, but the client has no
148 * reason to look at them. */
150 int recurse; /* Recursion level, via xlate_table_action. */
151 int last_pop_priority; /* Offset in 'odp_actions' just past most
152 * recent ODP_ACTION_ATTR_SET_PRIORITY. */
155 static void action_xlate_ctx_init(struct action_xlate_ctx *,
156 struct ofproto *, const struct flow *,
157 const struct ofpbuf *);
158 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
159 const union ofp_action *in, size_t n_in);
161 /* An OpenFlow flow. */
163 long long int used; /* Time last used; time created if not used. */
164 long long int created; /* Creation time. */
168 * - Do include packets and bytes from facets that have been deleted or
169 * whose own statistics have been folded into the rule.
171 * - Do include packets and bytes sent "by hand" that were accounted to
172 * the rule without any facet being involved (this is a rare corner
173 * case in rule_execute()).
175 * - Do not include packet or bytes that can be obtained from any facet's
176 * packet_count or byte_count member or that can be obtained from the
177 * datapath by, e.g., dpif_flow_get() for any facet.
179 uint64_t packet_count; /* Number of packets received. */
180 uint64_t byte_count; /* Number of bytes received. */
182 ovs_be64 flow_cookie; /* Controller-issued identifier. */
184 struct cls_rule cr; /* In owning ofproto's classifier. */
185 uint16_t idle_timeout; /* In seconds from time of last use. */
186 uint16_t hard_timeout; /* In seconds from time of creation. */
187 bool send_flow_removed; /* Send a flow removed message? */
188 int n_actions; /* Number of elements in actions[]. */
189 union ofp_action *actions; /* OpenFlow actions. */
190 struct list facets; /* List of "struct facet"s. */
193 static struct rule *rule_from_cls_rule(const struct cls_rule *);
194 static bool rule_is_hidden(const struct rule *);
196 static struct rule *rule_create(const struct cls_rule *,
197 const union ofp_action *, size_t n_actions,
198 uint16_t idle_timeout, uint16_t hard_timeout,
199 ovs_be64 flow_cookie, bool send_flow_removed);
200 static void rule_destroy(struct ofproto *, struct rule *);
201 static void rule_free(struct rule *);
203 static struct rule *rule_lookup(struct ofproto *, const struct flow *);
204 static void rule_insert(struct ofproto *, struct rule *);
205 static void rule_remove(struct ofproto *, struct rule *);
207 static void rule_send_removed(struct ofproto *, struct rule *, uint8_t reason);
208 static void rule_get_stats(const struct rule *, uint64_t *packets,
211 /* An exact-match instantiation of an OpenFlow flow. */
213 long long int used; /* Time last used; time created if not used. */
217 * - Do include packets and bytes sent "by hand", e.g. with
220 * - Do include packets and bytes that were obtained from the datapath
221 * when a flow was deleted (e.g. dpif_flow_del()) or when its
222 * statistics were reset (e.g. dpif_flow_put() with
223 * DPIF_FP_ZERO_STATS).
225 * - Do not include any packets or bytes that can currently be obtained
226 * from the datapath by, e.g., dpif_flow_get().
228 uint64_t packet_count; /* Number of packets received. */
229 uint64_t byte_count; /* Number of bytes received. */
231 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
232 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
234 uint64_t rs_packet_count; /* Packets pushed to resubmit children. */
235 uint64_t rs_byte_count; /* Bytes pushed to resubmit children. */
236 long long int rs_used; /* Used time pushed to resubmit children. */
238 /* Number of bytes passed to account_cb. This may include bytes that can
239 * currently obtained from the datapath (thus, it can be greater than
241 uint64_t accounted_bytes;
243 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
244 struct list list_node; /* In owning rule's 'facets' list. */
245 struct rule *rule; /* Owning rule. */
246 struct flow flow; /* Exact-match flow. */
247 bool installed; /* Installed in datapath? */
248 bool may_install; /* True ordinarily; false if actions must
249 * be reassessed for every packet. */
250 size_t actions_len; /* Number of bytes in actions[]. */
251 struct nlattr *actions; /* Datapath actions. */
252 tag_type tags; /* Tags (set only by hooks). */
253 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
256 static struct facet *facet_create(struct ofproto *, struct rule *,
258 const struct ofpbuf *packet);
259 static void facet_remove(struct ofproto *, struct facet *);
260 static void facet_free(struct facet *);
262 static struct facet *facet_lookup_valid(struct ofproto *, const struct flow *);
263 static bool facet_revalidate(struct ofproto *, struct facet *);
265 static void facet_install(struct ofproto *, struct facet *, bool zero_stats);
266 static void facet_uninstall(struct ofproto *, struct facet *);
267 static void facet_flush_stats(struct ofproto *, struct facet *);
269 static void facet_make_actions(struct ofproto *, struct facet *,
270 const struct ofpbuf *packet);
271 static void facet_update_stats(struct ofproto *, struct facet *,
272 const struct dpif_flow_stats *);
273 static void facet_push_stats(struct ofproto *, struct facet *);
275 static void send_packet_in(struct ofproto *, struct dpif_upcall *,
276 const struct flow *, bool clone);
280 uint64_t datapath_id; /* Datapath ID. */
281 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
282 char *mfr_desc; /* Manufacturer. */
283 char *hw_desc; /* Hardware. */
284 char *sw_desc; /* Software version. */
285 char *serial_desc; /* Serial number. */
286 char *dp_desc; /* Datapath description. */
290 struct netdev_monitor *netdev_monitor;
291 struct hmap ports; /* Contains "struct ofport"s. */
292 struct shash port_by_name;
296 struct netflow *netflow;
297 struct ofproto_sflow *sflow;
300 struct classifier cls;
301 struct timer next_expiration;
305 bool need_revalidate;
306 struct tag_set revalidate_set;
308 /* OpenFlow connections. */
309 struct connmgr *connmgr;
311 /* Hooks for ovs-vswitchd. */
312 const struct ofhooks *ofhooks;
315 /* Used by default ofhooks. */
316 struct mac_learning *ml;
319 /* Map from dpif name to struct ofproto, for use by unixctl commands. */
320 static struct shash all_ofprotos = SHASH_INITIALIZER(&all_ofprotos);
322 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
324 static const struct ofhooks default_ofhooks;
326 static uint64_t pick_datapath_id(const struct ofproto *);
327 static uint64_t pick_fallback_dpid(void);
329 static void ofproto_flush_flows__(struct ofproto *);
330 static int ofproto_expire(struct ofproto *);
331 static void flow_push_stats(struct ofproto *, const struct rule *,
332 struct flow *, uint64_t packets, uint64_t bytes,
335 static void handle_upcall(struct ofproto *, struct dpif_upcall *);
337 static void handle_openflow(struct ofconn *, struct ofpbuf *);
339 static struct ofport *get_port(const struct ofproto *, uint16_t odp_port);
340 static void update_port(struct ofproto *, const char *devname);
341 static int init_ports(struct ofproto *);
342 static void reinit_ports(struct ofproto *);
344 static void ofproto_unixctl_init(void);
347 ofproto_create(const char *datapath, const char *datapath_type,
348 const struct ofhooks *ofhooks, void *aux,
349 struct ofproto **ofprotop)
351 char local_name[IF_NAMESIZE];
358 ofproto_unixctl_init();
360 /* Connect to datapath and start listening for messages. */
361 error = dpif_create_and_open(datapath, datapath_type, &dpif);
363 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
366 error = dpif_recv_set_mask(dpif,
367 ((1u << DPIF_UC_MISS) |
368 (1u << DPIF_UC_ACTION) |
369 (1u << DPIF_UC_SAMPLE)));
371 VLOG_ERR("failed to listen on datapath %s: %s",
372 datapath, strerror(error));
376 dpif_flow_flush(dpif);
377 dpif_recv_purge(dpif);
379 error = dpif_port_get_name(dpif, ODPP_LOCAL,
380 local_name, sizeof local_name);
382 VLOG_ERR("%s: cannot get name of datapath local port (%s)",
383 datapath, strerror(error));
387 /* Initialize settings. */
388 p = xzalloc(sizeof *p);
389 p->fallback_dpid = pick_fallback_dpid();
390 p->datapath_id = p->fallback_dpid;
391 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
392 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
393 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
394 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
395 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
397 /* Initialize datapath. */
399 p->netdev_monitor = netdev_monitor_create();
400 hmap_init(&p->ports);
401 shash_init(&p->port_by_name);
402 p->max_ports = dpif_get_max_ports(dpif);
404 /* Initialize submodules. */
408 /* Initialize flow table. */
409 classifier_init(&p->cls);
410 timer_set_duration(&p->next_expiration, 1000);
412 /* Initialize facet table. */
413 hmap_init(&p->facets);
414 p->need_revalidate = false;
415 tag_set_init(&p->revalidate_set);
417 /* Initialize hooks. */
419 p->ofhooks = ofhooks;
423 p->ofhooks = &default_ofhooks;
425 p->ml = mac_learning_create();
428 /* Pick final datapath ID. */
429 p->datapath_id = pick_datapath_id(p);
430 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
432 shash_add_once(&all_ofprotos, dpif_name(p->dpif), p);
434 /* Initialize OpenFlow connections. */
435 p->connmgr = connmgr_create(p, datapath, local_name);
442 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
444 uint64_t old_dpid = p->datapath_id;
445 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
446 if (p->datapath_id != old_dpid) {
447 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
449 /* Force all active connections to reconnect, since there is no way to
450 * notify a controller that the datapath ID has changed. */
451 ofproto_reconnect_controllers(p);
456 ofproto_set_controllers(struct ofproto *p,
457 const struct ofproto_controller *controllers,
458 size_t n_controllers)
460 connmgr_set_controllers(p->connmgr, controllers, n_controllers);
464 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
466 connmgr_set_fail_mode(p->connmgr, fail_mode);
469 /* Drops the connections between 'ofproto' and all of its controllers, forcing
470 * them to reconnect. */
472 ofproto_reconnect_controllers(struct ofproto *ofproto)
474 connmgr_reconnect(ofproto->connmgr);
477 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
478 * in-band control should guarantee access, in the same way that in-band
479 * control guarantees access to OpenFlow controllers. */
481 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
482 const struct sockaddr_in *extras, size_t n)
484 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
487 /* Sets the OpenFlow queue used by flows set up by in-band control on
488 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
489 * flows will use the default queue. */
491 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
493 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
497 ofproto_set_desc(struct ofproto *p,
498 const char *mfr_desc, const char *hw_desc,
499 const char *sw_desc, const char *serial_desc,
502 struct ofp_desc_stats *ods;
505 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
506 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
507 sizeof ods->mfr_desc);
510 p->mfr_desc = xstrdup(mfr_desc);
513 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
514 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
515 sizeof ods->hw_desc);
518 p->hw_desc = xstrdup(hw_desc);
521 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
522 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
523 sizeof ods->sw_desc);
526 p->sw_desc = xstrdup(sw_desc);
529 if (strlen(serial_desc) >= sizeof ods->serial_num) {
530 VLOG_WARN("truncating serial_desc, must be less than %zu "
532 sizeof ods->serial_num);
534 free(p->serial_desc);
535 p->serial_desc = xstrdup(serial_desc);
538 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
539 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
540 sizeof ods->dp_desc);
543 p->dp_desc = xstrdup(dp_desc);
548 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
550 return connmgr_set_snoops(ofproto->connmgr, snoops);
554 ofproto_set_netflow(struct ofproto *ofproto,
555 const struct netflow_options *nf_options)
557 if (nf_options && !sset_is_empty(&nf_options->collectors)) {
558 if (!ofproto->netflow) {
559 ofproto->netflow = netflow_create();
561 return netflow_set_options(ofproto->netflow, nf_options);
563 netflow_destroy(ofproto->netflow);
564 ofproto->netflow = NULL;
570 ofproto_set_sflow(struct ofproto *ofproto,
571 const struct ofproto_sflow_options *oso)
573 struct ofproto_sflow *os = ofproto->sflow;
576 struct ofport *ofport;
578 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
579 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
580 ofproto_sflow_add_port(os, ofport->odp_port,
581 netdev_get_name(ofport->netdev));
584 ofproto_sflow_set_options(os, oso);
586 ofproto_sflow_destroy(os);
587 ofproto->sflow = NULL;
591 /* Connectivity Fault Management configuration. */
593 /* Clears the CFM configuration from 'port_no' on 'ofproto'. */
595 ofproto_iface_clear_cfm(struct ofproto *ofproto, uint32_t port_no)
597 struct ofport *ofport = get_port(ofproto, port_no);
598 if (ofport && ofport->cfm){
599 cfm_destroy(ofport->cfm);
604 /* Configures connectivity fault management on 'port_no' in 'ofproto'. Takes
605 * basic configuration from the configuration members in 'cfm', and the set of
606 * remote maintenance points from the 'n_remote_mps' elements in 'remote_mps'.
607 * Ignores the statistics members of 'cfm'.
609 * This function has no effect if 'ofproto' does not have a port 'port_no'. */
611 ofproto_iface_set_cfm(struct ofproto *ofproto, uint32_t port_no,
612 const struct cfm *cfm,
613 const uint16_t *remote_mps, size_t n_remote_mps)
615 struct ofport *ofport;
617 ofport = get_port(ofproto, port_no);
619 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu32,
620 dpif_name(ofproto->dpif), port_no);
625 ofport->cfm = cfm_create();
628 ofport->cfm->mpid = cfm->mpid;
629 ofport->cfm->interval = cfm->interval;
630 memcpy(ofport->cfm->maid, cfm->maid, CCM_MAID_LEN);
632 cfm_update_remote_mps(ofport->cfm, remote_mps, n_remote_mps);
634 if (!cfm_configure(ofport->cfm)) {
635 VLOG_WARN("%s: CFM configuration on port %"PRIu32" (%s) failed",
636 dpif_name(ofproto->dpif), port_no,
637 netdev_get_name(ofport->netdev));
638 cfm_destroy(ofport->cfm);
643 /* Returns the connectivity fault management object associated with 'port_no'
644 * within 'ofproto', or a null pointer if 'ofproto' does not have a port
645 * 'port_no' or if that port does not have CFM configured. The caller must not
646 * modify or destroy the returned object. */
648 ofproto_iface_get_cfm(struct ofproto *ofproto, uint32_t port_no)
650 struct ofport *ofport = get_port(ofproto, port_no);
651 return ofport ? ofport->cfm : NULL;
655 ofproto_get_datapath_id(const struct ofproto *ofproto)
657 return ofproto->datapath_id;
660 enum ofproto_fail_mode
661 ofproto_get_fail_mode(const struct ofproto *p)
663 return connmgr_get_fail_mode(p->connmgr);
667 ofproto_has_snoops(const struct ofproto *ofproto)
669 return connmgr_has_snoops(ofproto->connmgr);
673 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
675 connmgr_get_snoops(ofproto->connmgr, snoops);
679 ofproto_destroy__(struct ofproto *p, bool delete)
681 struct ofport *ofport, *next_ofport;
687 shash_find_and_delete(&all_ofprotos, dpif_name(p->dpif));
689 ofproto_flush_flows__(p);
690 connmgr_destroy(p->connmgr);
691 classifier_destroy(&p->cls);
692 hmap_destroy(&p->facets);
695 int error = dpif_delete(p->dpif);
696 if (error && error != ENOENT) {
697 VLOG_ERR("bridge %s: failed to destroy (%s)",
698 dpif_name(p->dpif), strerror(error));
703 netdev_monitor_destroy(p->netdev_monitor);
704 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
705 hmap_remove(&p->ports, &ofport->hmap_node);
708 shash_destroy(&p->port_by_name);
710 netflow_destroy(p->netflow);
711 ofproto_sflow_destroy(p->sflow);
713 mac_learning_destroy(p->ml);
718 free(p->serial_desc);
721 hmap_destroy(&p->ports);
727 ofproto_destroy(struct ofproto *p)
729 ofproto_destroy__(p, false);
733 ofproto_destroy_and_delete(struct ofproto *p)
735 ofproto_destroy__(p, true);
739 ofproto_run(struct ofproto *p)
741 int error = ofproto_run1(p);
743 error = ofproto_run2(p, false);
749 process_port_change(struct ofproto *ofproto, int error, char *devname)
751 if (error == ENOBUFS) {
752 reinit_ports(ofproto);
754 update_port(ofproto, devname);
760 ofproto_run1(struct ofproto *p)
762 struct ofport *ofport;
767 if (shash_is_empty(&p->port_by_name)) {
771 for (i = 0; i < 50; i++) {
772 struct dpif_upcall packet;
774 error = dpif_recv(p->dpif, &packet);
776 if (error == ENODEV) {
777 /* Someone destroyed the datapath behind our back. The caller
778 * better destroy us and give up, because we're just going to
779 * spin from here on out. */
780 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
781 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
788 handle_upcall(p, &packet);
791 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
792 process_port_change(p, error, devname);
794 while ((error = netdev_monitor_poll(p->netdev_monitor,
795 &devname)) != EAGAIN) {
796 process_port_change(p, error, devname);
799 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
800 ofport_run(p, ofport);
803 connmgr_run(p->connmgr, handle_openflow);
805 if (timer_expired(&p->next_expiration)) {
806 int delay = ofproto_expire(p);
807 timer_set_duration(&p->next_expiration, delay);
808 COVERAGE_INC(ofproto_expiration);
812 netflow_run(p->netflow);
815 ofproto_sflow_run(p->sflow);
822 ofproto_run2(struct ofproto *p, bool revalidate_all)
824 /* Figure out what we need to revalidate now, if anything. */
825 struct tag_set revalidate_set = p->revalidate_set;
826 if (p->need_revalidate) {
827 revalidate_all = true;
830 /* Clear the revalidation flags. */
831 tag_set_init(&p->revalidate_set);
832 p->need_revalidate = false;
834 /* Now revalidate if there's anything to do. */
835 if (revalidate_all || !tag_set_is_empty(&revalidate_set)) {
836 struct facet *facet, *next;
838 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &p->facets) {
840 || tag_set_intersects(&revalidate_set, facet->tags)) {
841 facet_revalidate(p, facet);
850 ofproto_wait(struct ofproto *p)
852 struct ofport *ofport;
854 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
857 dpif_recv_wait(p->dpif);
858 dpif_port_poll_wait(p->dpif);
859 netdev_monitor_poll_wait(p->netdev_monitor);
861 ofproto_sflow_wait(p->sflow);
863 if (!tag_set_is_empty(&p->revalidate_set)) {
864 poll_immediate_wake();
866 if (p->need_revalidate) {
867 /* Shouldn't happen, but if it does just go around again. */
868 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
869 poll_immediate_wake();
871 timer_wait(&p->next_expiration);
873 connmgr_wait(p->connmgr);
877 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
879 tag_set_add(&ofproto->revalidate_set, tag);
883 ofproto_get_revalidate_set(struct ofproto *ofproto)
885 return &ofproto->revalidate_set;
889 ofproto_is_alive(const struct ofproto *p)
891 return connmgr_has_controllers(p->connmgr);
895 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
898 connmgr_get_controller_info(ofproto->connmgr, info);
902 ofproto_free_ofproto_controller_info(struct shash *info)
904 struct shash_node *node;
906 SHASH_FOR_EACH (node, info) {
907 struct ofproto_controller_info *cinfo = node->data;
908 while (cinfo->pairs.n) {
909 free((char *) cinfo->pairs.values[--cinfo->pairs.n]);
916 /* Makes a deep copy of 'old' into 'port'. */
918 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
920 port->name = xstrdup(old->name);
921 port->type = xstrdup(old->type);
922 port->ofp_port = old->ofp_port;
925 /* Frees memory allocated to members of 'ofproto_port'.
927 * Do not call this function on an ofproto_port obtained from
928 * ofproto_port_dump_next(): that function retains ownership of the data in the
931 ofproto_port_destroy(struct ofproto_port *ofproto_port)
933 free(ofproto_port->name);
934 free(ofproto_port->type);
937 /* Converts a dpif_port into an ofproto_port.
939 * This only makes a shallow copy, so make sure that the dpif_port doesn't get
940 * freed while the ofproto_port is still in use. You can choose to free the
941 * ofproto_port instead of the dpif_port. */
943 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
944 struct dpif_port *dpif_port)
946 ofproto_port->name = dpif_port->name;
947 ofproto_port->type = dpif_port->type;
948 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
951 /* Initializes 'dump' to begin dumping the ports in an ofproto.
953 * This function provides no status indication. An error status for the entire
954 * dump operation is provided when it is completed by calling
955 * ofproto_port_dump_done().
958 ofproto_port_dump_start(struct ofproto_port_dump *dump,
959 const struct ofproto *ofproto)
961 struct dpif_port_dump *dpif_dump;
963 dump->state = dpif_dump = xmalloc(sizeof *dpif_dump);
964 dpif_port_dump_start(dpif_dump, ofproto->dpif);
967 /* Attempts to retrieve another port from 'dump', which must have been created
968 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
969 * 'port' and returns true. On failure, returns false.
971 * Failure might indicate an actual error or merely that the last port has been
972 * dumped. An error status for the entire dump operation is provided when it
973 * is completed by calling ofproto_port_dump_done().
975 * The ofproto owns the data stored in 'port'. It will remain valid until at
976 * least the next time 'dump' is passed to ofproto_port_dump_next() or
977 * ofproto_port_dump_done(). */
979 ofproto_port_dump_next(struct ofproto_port_dump *dump,
980 struct ofproto_port *port)
982 struct dpif_port_dump *dpif_dump = dump->state;
983 struct dpif_port dpif_port;
986 ok = dpif_port_dump_next(dpif_dump, &dpif_port);
988 ofproto_port_from_dpif_port(port, &dpif_port);
993 /* Completes port table dump operation 'dump', which must have been created
994 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
995 * error-free, otherwise a positive errno value describing the problem. */
997 ofproto_port_dump_done(struct ofproto_port_dump *dump)
999 struct dpif_port_dump *dpif_dump = dump->state;
1000 int error = dpif_port_dump_done(dpif_dump);
1005 /* Attempts to add 'netdev' as a port on 'ofproto'. If successful, returns 0
1006 * and sets '*ofp_portp' to the new port's port OpenFlow number (if 'ofp_portp'
1007 * is non-null). On failure, returns a positive errno value and sets
1008 * '*ofp_portp' to OFPP_NONE (if 'ofp_portp' is non-null). */
1010 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1011 uint16_t *ofp_portp)
1016 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
1018 *ofp_portp = error ? OFPP_NONE : odp_port_to_ofp_port(odp_port);
1023 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1024 * initializes '*port' appropriately; on failure, returns a positive errno
1027 * The caller owns the data in 'port' and must free it with
1028 * ofproto_port_destroy() when it is no longer needed. */
1030 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1031 struct ofproto_port *port)
1033 struct dpif_port dpif_port;
1036 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
1038 ofproto_port_from_dpif_port(port, &dpif_port);
1043 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1045 * This is almost the same as calling dpif_port_del() directly on the
1046 * datapath, but it also makes 'ofproto' close its open netdev for the port
1047 * (if any). This makes it possible to create a new netdev of a different
1048 * type under the same name, which otherwise the netdev library would refuse
1049 * to do because of the conflict. (The netdev would eventually get closed on
1050 * the next trip through ofproto_run(), but this interface is more direct.)
1052 * Returns 0 if successful, otherwise a positive errno. */
1054 ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port)
1056 uint32_t odp_port = ofp_port_to_odp_port(ofp_port);
1057 struct ofport *ofport = get_port(ofproto, odp_port);
1058 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1061 error = dpif_port_del(ofproto->dpif, odp_port);
1063 VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)",
1064 dpif_name(ofproto->dpif), odp_port, name, strerror(error));
1065 } else if (ofport) {
1066 /* 'name' is the netdev's name and update_port() is going to close the
1067 * netdev. Just in case update_port() refers to 'name' after it
1068 * destroys 'ofport', make a copy of it around the update_port()
1070 char *devname = xstrdup(name);
1071 update_port(ofproto, devname);
1077 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
1078 * true if 'odp_port' exists and should be included, false otherwise. */
1080 ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port)
1082 struct ofport *ofport = get_port(ofproto, odp_port);
1083 return ofport && !(ofport->opp.config & htonl(OFPPC_NO_FLOOD));
1086 /* Sends 'packet' out of port 'port_no' within 'p'. If 'vlan_tci' is zero the
1087 * packet will not have any 802.1Q hader; if it is nonzero, then the packet
1088 * will be sent with the VLAN TCI specified by 'vlan_tci & ~VLAN_CFI'.
1090 * Returns 0 if successful, otherwise a positive errno value. */
1092 ofproto_send_packet(struct ofproto *ofproto,
1093 uint32_t port_no, uint16_t vlan_tci,
1094 const struct ofpbuf *packet)
1096 struct ofpbuf odp_actions;
1099 ofpbuf_init(&odp_actions, 32);
1100 if (vlan_tci != 0) {
1101 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
1102 ntohs(vlan_tci & ~VLAN_CFI));
1104 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, port_no);
1105 error = dpif_execute(ofproto->dpif, odp_actions.data, odp_actions.size,
1107 ofpbuf_uninit(&odp_actions);
1110 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
1111 dpif_name(ofproto->dpif), port_no, strerror(error));
1116 /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and
1117 * performs the 'n_actions' actions in 'actions'. The new flow will not
1120 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1121 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1122 * controllers; otherwise, it will be hidden.
1124 * The caller retains ownership of 'cls_rule' and 'actions'. */
1126 ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule,
1127 const union ofp_action *actions, size_t n_actions)
1130 rule = rule_create(cls_rule, actions, n_actions, 0, 0, 0, false);
1131 rule_insert(p, rule);
1135 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1139 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1142 rule_remove(ofproto, rule);
1147 ofproto_flush_flows__(struct ofproto *ofproto)
1149 struct facet *facet, *next_facet;
1150 struct rule *rule, *next_rule;
1151 struct cls_cursor cursor;
1153 COVERAGE_INC(ofproto_flush);
1155 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1156 /* Mark the facet as not installed so that facet_remove() doesn't
1157 * bother trying to uninstall it. There is no point in uninstalling it
1158 * individually since we are about to blow away all the facets with
1159 * dpif_flow_flush(). */
1160 facet->installed = false;
1161 facet->dp_packet_count = 0;
1162 facet->dp_byte_count = 0;
1163 facet_remove(ofproto, facet);
1166 cls_cursor_init(&cursor, &ofproto->cls, NULL);
1167 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1168 rule_remove(ofproto, rule);
1171 dpif_flow_flush(ofproto->dpif);
1175 ofproto_flush_flows(struct ofproto *ofproto)
1177 ofproto_flush_flows__(ofproto);
1178 connmgr_flushed(ofproto->connmgr);
1182 reinit_ports(struct ofproto *p)
1184 struct dpif_port_dump dump;
1185 struct sset devnames;
1186 struct ofport *ofport;
1187 struct dpif_port dpif_port;
1188 const char *devname;
1190 COVERAGE_INC(ofproto_reinit_ports);
1192 sset_init(&devnames);
1193 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1194 sset_add(&devnames, netdev_get_name(ofport->netdev));
1196 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1197 sset_add(&devnames, dpif_port.name);
1200 SSET_FOR_EACH (devname, &devnames) {
1201 update_port(p, devname);
1203 sset_destroy(&devnames);
1206 /* Opens and returns a netdev for 'dpif_port', or a null pointer if the netdev
1207 * cannot be opened. On success, also fills in 'opp'. */
1208 static struct netdev *
1209 ofport_open(const struct dpif_port *dpif_port, struct ofp_phy_port *opp)
1211 uint32_t curr, advertised, supported, peer;
1212 struct netdev_options netdev_options;
1213 enum netdev_flags flags;
1214 struct netdev *netdev;
1217 memset(&netdev_options, 0, sizeof netdev_options);
1218 netdev_options.name = dpif_port->name;
1219 netdev_options.type = dpif_port->type;
1220 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1222 error = netdev_open(&netdev_options, &netdev);
1224 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1225 "cannot be opened (%s)",
1226 dpif_port->name, dpif_port->port_no,
1227 dpif_port->name, strerror(error));
1231 netdev_get_flags(netdev, &flags);
1232 netdev_get_features(netdev, &curr, &advertised, &supported, &peer);
1234 opp->port_no = htons(odp_port_to_ofp_port(dpif_port->port_no));
1235 netdev_get_etheraddr(netdev, opp->hw_addr);
1236 ovs_strzcpy(opp->name, dpif_port->name, sizeof opp->name);
1237 opp->config = flags & NETDEV_UP ? 0 : htonl(OFPPC_PORT_DOWN);
1238 opp->state = netdev_get_carrier(netdev) ? 0 : htonl(OFPPS_LINK_DOWN);
1239 opp->curr = htonl(curr);
1240 opp->advertised = htonl(advertised);
1241 opp->supported = htonl(supported);
1242 opp->peer = htonl(peer);
1248 ofport_conflicts(const struct ofproto *p, const struct dpif_port *dpif_port)
1250 if (get_port(p, dpif_port->port_no)) {
1251 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1252 dpif_port->port_no);
1254 } else if (shash_find(&p->port_by_name, dpif_port->name)) {
1255 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1263 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1264 * port number, and 'config' bits other than OFPPC_PORT_DOWN are
1267 ofport_equal(const struct ofp_phy_port *a, const struct ofp_phy_port *b)
1269 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1270 return (!memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1271 && a->state == b->state
1272 && !((a->config ^ b->config) & htonl(OFPPC_PORT_DOWN))
1273 && a->curr == b->curr
1274 && a->advertised == b->advertised
1275 && a->supported == b->supported
1276 && a->peer == b->peer);
1279 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1280 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1281 * one with the same name or port number). */
1283 ofport_install(struct ofproto *p,
1284 struct netdev *netdev, const struct ofp_phy_port *opp)
1286 const char *netdev_name = netdev_get_name(netdev);
1287 struct ofport *ofport;
1289 connmgr_send_port_status(p->connmgr, opp, OFPPR_ADD);
1291 /* Create ofport. */
1292 ofport = xmalloc(sizeof *ofport);
1293 ofport->netdev = netdev;
1295 ofport->odp_port = ofp_port_to_odp_port(ntohs(opp->port_no));
1298 /* Add port to 'p'. */
1299 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1300 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
1301 shash_add(&p->port_by_name, netdev_name, ofport);
1303 ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
1307 /* Removes 'ofport' from 'p' and destroys it. */
1309 ofport_remove(struct ofproto *p, struct ofport *ofport)
1311 connmgr_send_port_status(p->connmgr, &ofport->opp, OFPPR_DELETE);
1313 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1314 hmap_remove(&p->ports, &ofport->hmap_node);
1315 shash_delete(&p->port_by_name,
1316 shash_find(&p->port_by_name,
1317 netdev_get_name(ofport->netdev)));
1319 ofproto_sflow_del_port(p->sflow, ofport->odp_port);
1322 ofport_free(ofport);
1325 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1328 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1330 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1332 ofport_remove(ofproto, port);
1336 /* Updates 'port' within 'ofproto' with the new 'netdev' and 'opp'.
1338 * Does not handle a name or port number change. The caller must implement
1339 * such a change as a delete followed by an add. */
1341 ofport_modified(struct ofproto *ofproto, struct ofport *port,
1342 struct netdev *netdev, struct ofp_phy_port *opp)
1344 memcpy(port->opp.hw_addr, opp->hw_addr, ETH_ADDR_LEN);
1345 port->opp.config = ((port->opp.config & ~htonl(OFPPC_PORT_DOWN))
1346 | (opp->config & htonl(OFPPC_PORT_DOWN)));
1347 port->opp.state = opp->state;
1348 port->opp.curr = opp->curr;
1349 port->opp.advertised = opp->advertised;
1350 port->opp.supported = opp->supported;
1351 port->opp.peer = opp->peer;
1353 netdev_monitor_remove(ofproto->netdev_monitor, port->netdev);
1354 netdev_monitor_add(ofproto->netdev_monitor, netdev);
1356 netdev_close(port->netdev);
1357 port->netdev = netdev;
1359 connmgr_send_port_status(ofproto->connmgr, &port->opp, OFPPR_MODIFY);
1363 ofport_run(struct ofproto *ofproto, struct ofport *ofport)
1366 cfm_run(ofport->cfm);
1368 if (cfm_should_send_ccm(ofport->cfm)) {
1369 struct ofpbuf packet;
1372 ofpbuf_init(&packet, 0);
1373 ccm = eth_compose(&packet, eth_addr_ccm, ofport->opp.hw_addr,
1374 ETH_TYPE_CFM, sizeof *ccm);
1375 cfm_compose_ccm(ofport->cfm, ccm);
1376 ofproto_send_packet(ofproto, ofport->odp_port, 0, &packet);
1377 ofpbuf_uninit(&packet);
1383 ofport_wait(struct ofport *ofport)
1386 cfm_wait(ofport->cfm);
1391 ofport_free(struct ofport *ofport)
1394 cfm_destroy(ofport->cfm);
1395 netdev_close(ofport->netdev);
1400 static struct ofport *
1401 get_port(const struct ofproto *ofproto, uint16_t odp_port)
1403 struct ofport *port;
1405 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1406 hash_int(odp_port, 0), &ofproto->ports) {
1407 if (port->odp_port == odp_port) {
1415 update_port(struct ofproto *ofproto, const char *name)
1417 struct dpif_port dpif_port;
1418 struct ofp_phy_port opp;
1419 struct netdev *netdev;
1420 struct ofport *port;
1422 COVERAGE_INC(ofproto_update_port);
1424 /* Fetch 'name''s location and properties from the datapath. */
1425 netdev = (!dpif_port_query_by_name(ofproto->dpif, name, &dpif_port)
1426 ? ofport_open(&dpif_port, &opp)
1429 port = get_port(ofproto, dpif_port.port_no);
1430 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
1431 /* 'name' hasn't changed location. Any properties changed? */
1432 if (!ofport_equal(&port->opp, &opp)) {
1433 ofport_modified(ofproto, port, netdev, &opp);
1435 netdev_close(netdev);
1438 /* If 'port' is nonnull then its name differs from 'name' and thus
1439 * we should delete it. If we think there's a port named 'name'
1440 * then its port number must be wrong now so delete it too. */
1442 ofport_remove(ofproto, port);
1444 ofport_remove_with_name(ofproto, name);
1445 ofport_install(ofproto, netdev, &opp);
1448 /* Any port named 'name' is gone now. */
1449 ofport_remove_with_name(ofproto, name);
1451 dpif_port_destroy(&dpif_port);
1455 init_ports(struct ofproto *p)
1457 struct dpif_port_dump dump;
1458 struct dpif_port dpif_port;
1460 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1461 if (!ofport_conflicts(p, &dpif_port)) {
1462 struct ofp_phy_port opp;
1463 struct netdev *netdev;
1465 netdev = ofport_open(&dpif_port, &opp);
1467 ofport_install(p, netdev, &opp);
1475 /* Returns true if 'rule' should be hidden from the controller.
1477 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1478 * (e.g. by in-band control) and are intentionally hidden from the
1481 rule_is_hidden(const struct rule *rule)
1483 return rule->cr.priority > UINT16_MAX;
1486 /* Creates and returns a new rule initialized as specified.
1488 * The caller is responsible for inserting the rule into the classifier (with
1489 * rule_insert()). */
1490 static struct rule *
1491 rule_create(const struct cls_rule *cls_rule,
1492 const union ofp_action *actions, size_t n_actions,
1493 uint16_t idle_timeout, uint16_t hard_timeout,
1494 ovs_be64 flow_cookie, bool send_flow_removed)
1496 struct rule *rule = xzalloc(sizeof *rule);
1497 rule->cr = *cls_rule;
1498 rule->idle_timeout = idle_timeout;
1499 rule->hard_timeout = hard_timeout;
1500 rule->flow_cookie = flow_cookie;
1501 rule->used = rule->created = time_msec();
1502 rule->send_flow_removed = send_flow_removed;
1503 list_init(&rule->facets);
1504 if (n_actions > 0) {
1505 rule->n_actions = n_actions;
1506 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1512 static struct rule *
1513 rule_from_cls_rule(const struct cls_rule *cls_rule)
1515 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1519 rule_free(struct rule *rule)
1521 free(rule->actions);
1525 /* Destroys 'rule' and iterates through all of its facets and revalidates them,
1526 * destroying any that no longer has a rule (which is probably all of them).
1528 * The caller must have already removed 'rule' from the classifier. */
1530 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1532 struct facet *facet, *next_facet;
1533 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
1534 facet_revalidate(ofproto, facet);
1539 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
1540 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
1543 rule_has_out_port(const struct rule *rule, ovs_be16 out_port)
1545 const union ofp_action *oa;
1546 struct actions_iterator i;
1548 if (out_port == htons(OFPP_NONE)) {
1551 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1552 oa = actions_next(&i)) {
1553 if (action_outputs_to_port(oa, out_port)) {
1560 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
1561 * 'packet', which arrived on 'in_port'.
1563 * Takes ownership of 'packet'. */
1565 execute_odp_actions(struct ofproto *ofproto, const struct flow *flow,
1566 const struct nlattr *odp_actions, size_t actions_len,
1567 struct ofpbuf *packet)
1569 if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t))
1570 && odp_actions->nla_type == ODP_ACTION_ATTR_CONTROLLER) {
1571 /* As an optimization, avoid a round-trip from userspace to kernel to
1572 * userspace. This also avoids possibly filling up kernel packet
1573 * buffers along the way. */
1574 struct dpif_upcall upcall;
1576 upcall.type = DPIF_UC_ACTION;
1577 upcall.packet = packet;
1580 upcall.userdata = nl_attr_get_u64(odp_actions);
1581 upcall.sample_pool = 0;
1582 upcall.actions = NULL;
1583 upcall.actions_len = 0;
1585 send_packet_in(ofproto, &upcall, flow, false);
1591 error = dpif_execute(ofproto->dpif, odp_actions, actions_len, packet);
1592 ofpbuf_delete(packet);
1597 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
1598 * statistics appropriately. 'packet' must have at least sizeof(struct
1599 * ofp_packet_in) bytes of headroom.
1601 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
1602 * applying flow_extract() to 'packet' would yield the same flow as
1605 * 'facet' must have accurately composed ODP actions; that is, it must not be
1606 * in need of revalidation.
1608 * Takes ownership of 'packet'. */
1610 facet_execute(struct ofproto *ofproto, struct facet *facet,
1611 struct ofpbuf *packet)
1613 struct dpif_flow_stats stats;
1615 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1617 flow_extract_stats(&facet->flow, packet, &stats);
1618 stats.used = time_msec();
1619 if (execute_odp_actions(ofproto, &facet->flow,
1620 facet->actions, facet->actions_len, packet)) {
1621 facet_update_stats(ofproto, facet, &stats);
1625 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
1626 * statistics (or the statistics for one of its facets) appropriately.
1627 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom.
1629 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
1630 * with statistics for 'packet' either way.
1632 * Takes ownership of 'packet'. */
1634 rule_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port,
1635 struct ofpbuf *packet)
1637 struct action_xlate_ctx ctx;
1638 struct ofpbuf *odp_actions;
1639 struct facet *facet;
1643 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1645 flow_extract(packet, 0, in_port, &flow);
1647 /* First look for a related facet. If we find one, account it to that. */
1648 facet = facet_lookup_valid(ofproto, &flow);
1649 if (facet && facet->rule == rule) {
1650 facet_execute(ofproto, facet, packet);
1654 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
1655 * create a new facet for it and use that. */
1656 if (rule_lookup(ofproto, &flow) == rule) {
1657 facet = facet_create(ofproto, rule, &flow, packet);
1658 facet_execute(ofproto, facet, packet);
1659 facet_install(ofproto, facet, true);
1663 /* We can't account anything to a facet. If we were to try, then that
1664 * facet would have a non-matching rule, busting our invariants. */
1665 action_xlate_ctx_init(&ctx, ofproto, &flow, packet);
1666 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
1667 size = packet->size;
1668 if (execute_odp_actions(ofproto, &flow, odp_actions->data,
1669 odp_actions->size, packet)) {
1670 rule->used = time_msec();
1671 rule->packet_count++;
1672 rule->byte_count += size;
1673 flow_push_stats(ofproto, rule, &flow, 1, size, rule->used);
1675 ofpbuf_delete(odp_actions);
1678 /* Inserts 'rule' into 'p''s flow table. */
1680 rule_insert(struct ofproto *p, struct rule *rule)
1682 struct rule *displaced_rule;
1684 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1685 if (displaced_rule) {
1686 rule_destroy(p, displaced_rule);
1688 p->need_revalidate = true;
1691 /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a
1692 * 'flow' and an example 'packet' within that flow.
1694 * The caller must already have determined that no facet with an identical
1695 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
1696 * 'ofproto''s classifier table. */
1697 static struct facet *
1698 facet_create(struct ofproto *ofproto, struct rule *rule,
1699 const struct flow *flow, const struct ofpbuf *packet)
1701 struct facet *facet;
1703 facet = xzalloc(sizeof *facet);
1704 facet->used = time_msec();
1705 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
1706 list_push_back(&rule->facets, &facet->list_node);
1708 facet->flow = *flow;
1709 netflow_flow_init(&facet->nf_flow);
1710 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
1712 facet_make_actions(ofproto, facet, packet);
1718 facet_free(struct facet *facet)
1720 free(facet->actions);
1724 /* Remove 'rule' from 'ofproto' and free up the associated memory:
1726 * - Removes 'rule' from the classifier.
1728 * - If 'rule' has facets, revalidates them (and possibly uninstalls and
1729 * destroys them), via rule_destroy().
1732 rule_remove(struct ofproto *ofproto, struct rule *rule)
1734 COVERAGE_INC(ofproto_del_rule);
1735 ofproto->need_revalidate = true;
1736 classifier_remove(&ofproto->cls, &rule->cr);
1737 rule_destroy(ofproto, rule);
1740 /* Remove 'facet' from 'ofproto' and free up the associated memory:
1742 * - If 'facet' was installed in the datapath, uninstalls it and updates its
1743 * rule's statistics, via facet_uninstall().
1745 * - Removes 'facet' from its rule and from ofproto->facets.
1748 facet_remove(struct ofproto *ofproto, struct facet *facet)
1750 facet_uninstall(ofproto, facet);
1751 facet_flush_stats(ofproto, facet);
1752 hmap_remove(&ofproto->facets, &facet->hmap_node);
1753 list_remove(&facet->list_node);
1757 /* Composes the ODP actions for 'facet' based on its rule's actions. */
1759 facet_make_actions(struct ofproto *p, struct facet *facet,
1760 const struct ofpbuf *packet)
1762 const struct rule *rule = facet->rule;
1763 struct ofpbuf *odp_actions;
1764 struct action_xlate_ctx ctx;
1766 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
1767 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
1768 facet->tags = ctx.tags;
1769 facet->may_install = ctx.may_set_up_flow;
1770 facet->nf_flow.output_iface = ctx.nf_output_iface;
1772 if (facet->actions_len != odp_actions->size
1773 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
1774 free(facet->actions);
1775 facet->actions_len = odp_actions->size;
1776 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
1779 ofpbuf_delete(odp_actions);
1783 facet_put__(struct ofproto *ofproto, struct facet *facet,
1784 const struct nlattr *actions, size_t actions_len,
1785 struct dpif_flow_stats *stats)
1787 struct odputil_keybuf keybuf;
1788 enum dpif_flow_put_flags flags;
1791 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
1793 flags |= DPIF_FP_ZERO_STATS;
1794 facet->dp_packet_count = 0;
1795 facet->dp_byte_count = 0;
1798 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1799 odp_flow_key_from_flow(&key, &facet->flow);
1801 return dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
1802 actions, actions_len, stats);
1805 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
1806 * 'zero_stats' is true, clears any existing statistics from the datapath for
1809 facet_install(struct ofproto *p, struct facet *facet, bool zero_stats)
1811 struct dpif_flow_stats stats;
1813 if (facet->may_install
1814 && !facet_put__(p, facet, facet->actions, facet->actions_len,
1815 zero_stats ? &stats : NULL)) {
1816 facet->installed = true;
1820 /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up
1821 * to the accounting hook function in the ofhooks structure. */
1823 facet_account(struct ofproto *ofproto,
1824 struct facet *facet, uint64_t extra_bytes)
1826 uint64_t total_bytes = facet->byte_count + extra_bytes;
1828 if (ofproto->ofhooks->account_flow_cb
1829 && total_bytes > facet->accounted_bytes)
1831 ofproto->ofhooks->account_flow_cb(
1832 &facet->flow, facet->tags, facet->actions, facet->actions_len,
1833 total_bytes - facet->accounted_bytes, ofproto->aux);
1834 facet->accounted_bytes = total_bytes;
1838 /* If 'rule' is installed in the datapath, uninstalls it. */
1840 facet_uninstall(struct ofproto *p, struct facet *facet)
1842 if (facet->installed) {
1843 struct odputil_keybuf keybuf;
1844 struct dpif_flow_stats stats;
1847 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1848 odp_flow_key_from_flow(&key, &facet->flow);
1850 if (!dpif_flow_del(p->dpif, key.data, key.size, &stats)) {
1851 facet_update_stats(p, facet, &stats);
1853 facet->installed = false;
1854 facet->dp_packet_count = 0;
1855 facet->dp_byte_count = 0;
1857 assert(facet->dp_packet_count == 0);
1858 assert(facet->dp_byte_count == 0);
1862 /* Returns true if the only action for 'facet' is to send to the controller.
1863 * (We don't report NetFlow expiration messages for such facets because they
1864 * are just part of the control logic for the network, not real traffic). */
1866 facet_is_controller_flow(struct facet *facet)
1869 && facet->rule->n_actions == 1
1870 && action_outputs_to_port(&facet->rule->actions[0],
1871 htons(OFPP_CONTROLLER)));
1874 /* Folds all of 'facet''s statistics into its rule. Also updates the
1875 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
1876 * 'facet''s statistics in the datapath should have been zeroed and folded into
1877 * its packet and byte counts before this function is called. */
1879 facet_flush_stats(struct ofproto *ofproto, struct facet *facet)
1881 assert(!facet->dp_byte_count);
1882 assert(!facet->dp_packet_count);
1884 facet_push_stats(ofproto, facet);
1885 facet_account(ofproto, facet, 0);
1887 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
1888 struct ofexpired expired;
1889 expired.flow = facet->flow;
1890 expired.packet_count = facet->packet_count;
1891 expired.byte_count = facet->byte_count;
1892 expired.used = facet->used;
1893 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
1896 facet->rule->packet_count += facet->packet_count;
1897 facet->rule->byte_count += facet->byte_count;
1899 /* Reset counters to prevent double counting if 'facet' ever gets
1901 facet->packet_count = 0;
1902 facet->byte_count = 0;
1903 facet->rs_packet_count = 0;
1904 facet->rs_byte_count = 0;
1905 facet->accounted_bytes = 0;
1907 netflow_flow_clear(&facet->nf_flow);
1910 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
1911 * Returns it if found, otherwise a null pointer.
1913 * The returned facet might need revalidation; use facet_lookup_valid()
1914 * instead if that is important. */
1915 static struct facet *
1916 facet_find(struct ofproto *ofproto, const struct flow *flow)
1918 struct facet *facet;
1920 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
1922 if (flow_equal(flow, &facet->flow)) {
1930 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
1931 * Returns it if found, otherwise a null pointer.
1933 * The returned facet is guaranteed to be valid. */
1934 static struct facet *
1935 facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow)
1937 struct facet *facet = facet_find(ofproto, flow);
1939 /* The facet we found might not be valid, since we could be in need of
1940 * revalidation. If it is not valid, don't return it. */
1942 && ofproto->need_revalidate
1943 && !facet_revalidate(ofproto, facet)) {
1944 COVERAGE_INC(ofproto_invalidated);
1951 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
1953 * - If the rule found is different from 'facet''s current rule, moves
1954 * 'facet' to the new rule and recompiles its actions.
1956 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
1957 * where it is and recompiles its actions anyway.
1959 * - If there is none, destroys 'facet'.
1961 * Returns true if 'facet' still exists, false if it has been destroyed. */
1963 facet_revalidate(struct ofproto *ofproto, struct facet *facet)
1965 struct action_xlate_ctx ctx;
1966 struct ofpbuf *odp_actions;
1967 struct rule *new_rule;
1968 bool actions_changed;
1970 COVERAGE_INC(facet_revalidate);
1972 /* Determine the new rule. */
1973 new_rule = rule_lookup(ofproto, &facet->flow);
1975 /* No new rule, so delete the facet. */
1976 facet_remove(ofproto, facet);
1980 /* Calculate new ODP actions.
1982 * We do not modify any 'facet' state yet, because we might need to, e.g.,
1983 * emit a NetFlow expiration and, if so, we need to have the old state
1984 * around to properly compose it. */
1985 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
1986 odp_actions = xlate_actions(&ctx, new_rule->actions, new_rule->n_actions);
1987 actions_changed = (facet->actions_len != odp_actions->size
1988 || memcmp(facet->actions, odp_actions->data,
1989 facet->actions_len));
1991 /* If the ODP actions changed or the installability changed, then we need
1992 * to talk to the datapath. */
1993 if (actions_changed || ctx.may_set_up_flow != facet->installed) {
1994 if (ctx.may_set_up_flow) {
1995 struct dpif_flow_stats stats;
1997 facet_put__(ofproto, facet,
1998 odp_actions->data, odp_actions->size, &stats);
1999 facet_update_stats(ofproto, facet, &stats);
2001 facet_uninstall(ofproto, facet);
2004 /* The datapath flow is gone or has zeroed stats, so push stats out of
2005 * 'facet' into 'rule'. */
2006 facet_flush_stats(ofproto, facet);
2009 /* Update 'facet' now that we've taken care of all the old state. */
2010 facet->tags = ctx.tags;
2011 facet->nf_flow.output_iface = ctx.nf_output_iface;
2012 facet->may_install = ctx.may_set_up_flow;
2013 if (actions_changed) {
2014 free(facet->actions);
2015 facet->actions_len = odp_actions->size;
2016 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2018 if (facet->rule != new_rule) {
2019 COVERAGE_INC(facet_changed_rule);
2020 list_remove(&facet->list_node);
2021 list_push_back(&new_rule->facets, &facet->list_node);
2022 facet->rule = new_rule;
2023 facet->used = new_rule->created;
2024 facet->rs_used = facet->used;
2027 ofpbuf_delete(odp_actions);
2033 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2036 struct ofpbuf *buf = ofputil_encode_error_msg(error, oh);
2038 COVERAGE_INC(ofproto_error);
2039 ofconn_send_reply(ofconn, buf);
2044 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2046 ofconn_send_reply(ofconn, make_echo_reply(oh));
2051 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2053 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2054 struct ofp_switch_features *osf;
2056 struct ofport *port;
2058 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2059 osf->datapath_id = htonll(ofproto->datapath_id);
2060 osf->n_buffers = htonl(pktbuf_capacity());
2062 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2063 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2064 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2065 (1u << OFPAT_SET_VLAN_VID) |
2066 (1u << OFPAT_SET_VLAN_PCP) |
2067 (1u << OFPAT_STRIP_VLAN) |
2068 (1u << OFPAT_SET_DL_SRC) |
2069 (1u << OFPAT_SET_DL_DST) |
2070 (1u << OFPAT_SET_NW_SRC) |
2071 (1u << OFPAT_SET_NW_DST) |
2072 (1u << OFPAT_SET_NW_TOS) |
2073 (1u << OFPAT_SET_TP_SRC) |
2074 (1u << OFPAT_SET_TP_DST) |
2075 (1u << OFPAT_ENQUEUE));
2077 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2078 ofpbuf_put(buf, &port->opp, sizeof port->opp);
2081 ofconn_send_reply(ofconn, buf);
2086 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2088 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2090 struct ofp_switch_config *osc;
2094 /* Figure out flags. */
2095 dpif_get_drop_frags(ofproto->dpif, &drop_frags);
2096 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2099 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2100 osc->flags = htons(flags);
2101 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2102 ofconn_send_reply(ofconn, buf);
2108 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
2110 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2111 uint16_t flags = ntohs(osc->flags);
2113 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2114 && ofconn_get_role(ofconn) != NX_ROLE_SLAVE) {
2115 switch (flags & OFPC_FRAG_MASK) {
2116 case OFPC_FRAG_NORMAL:
2117 dpif_set_drop_frags(ofproto->dpif, false);
2119 case OFPC_FRAG_DROP:
2120 dpif_set_drop_frags(ofproto->dpif, true);
2123 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2129 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2134 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2135 struct action_xlate_ctx *ctx);
2138 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2140 const struct ofport *ofport = get_port(ctx->ofproto, port);
2143 if (ofport->opp.config & htonl(OFPPC_NO_FWD)) {
2144 /* Forwarding disabled on port. */
2149 * We don't have an ofport record for this port, but it doesn't hurt to
2150 * allow forwarding to it anyhow. Maybe such a port will appear later
2151 * and we're pre-populating the flow table.
2155 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, port);
2156 ctx->nf_output_iface = port;
2159 static struct rule *
2160 rule_lookup(struct ofproto *ofproto, const struct flow *flow)
2162 return rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2166 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2168 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2169 uint16_t old_in_port;
2172 /* Look up a flow with 'in_port' as the input port. Then restore the
2173 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2174 * have surprising behavior). */
2175 old_in_port = ctx->flow.in_port;
2176 ctx->flow.in_port = in_port;
2177 rule = rule_lookup(ctx->ofproto, &ctx->flow);
2178 ctx->flow.in_port = old_in_port;
2180 if (ctx->resubmit_hook) {
2181 ctx->resubmit_hook(ctx, rule);
2186 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2190 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2192 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2193 MAX_RESUBMIT_RECURSION);
2198 flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, ovs_be32 mask,
2199 uint16_t *nf_output_iface, struct ofpbuf *odp_actions)
2201 struct ofport *ofport;
2203 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
2204 uint16_t odp_port = ofport->odp_port;
2205 if (odp_port != odp_in_port && !(ofport->opp.config & mask)) {
2206 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port);
2209 *nf_output_iface = NF_OUT_FLOOD;
2213 xlate_output_action__(struct action_xlate_ctx *ctx,
2214 uint16_t port, uint16_t max_len)
2217 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2219 ctx->nf_output_iface = NF_OUT_DROP;
2223 add_output_action(ctx, ctx->flow.in_port);
2226 xlate_table_action(ctx, ctx->flow.in_port);
2229 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2230 ctx->odp_actions, &ctx->tags,
2231 &ctx->nf_output_iface,
2232 ctx->ofproto->aux)) {
2233 COVERAGE_INC(ofproto_uninstallable);
2234 ctx->may_set_up_flow = false;
2238 flood_packets(ctx->ofproto, ctx->flow.in_port, htonl(OFPPC_NO_FLOOD),
2239 &ctx->nf_output_iface, ctx->odp_actions);
2242 flood_packets(ctx->ofproto, ctx->flow.in_port, htonl(0),
2243 &ctx->nf_output_iface, ctx->odp_actions);
2245 case OFPP_CONTROLLER:
2246 nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_CONTROLLER, max_len);
2249 add_output_action(ctx, ODPP_LOCAL);
2252 odp_port = ofp_port_to_odp_port(port);
2253 if (odp_port != ctx->flow.in_port) {
2254 add_output_action(ctx, odp_port);
2259 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2260 ctx->nf_output_iface = NF_OUT_FLOOD;
2261 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2262 ctx->nf_output_iface = prev_nf_output_iface;
2263 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2264 ctx->nf_output_iface != NF_OUT_FLOOD) {
2265 ctx->nf_output_iface = NF_OUT_MULTI;
2270 xlate_output_action(struct action_xlate_ctx *ctx,
2271 const struct ofp_action_output *oao)
2273 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2276 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2277 * optimization, because we're going to add another action that sets the
2278 * priority immediately after, or because there are no actions following the
2281 remove_pop_action(struct action_xlate_ctx *ctx)
2283 if (ctx->odp_actions->size == ctx->last_pop_priority) {
2284 ctx->odp_actions->size -= NLA_ALIGN(NLA_HDRLEN);
2285 ctx->last_pop_priority = -1;
2290 add_pop_action(struct action_xlate_ctx *ctx)
2292 if (ctx->odp_actions->size != ctx->last_pop_priority) {
2293 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_POP_PRIORITY);
2294 ctx->last_pop_priority = ctx->odp_actions->size;
2299 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2300 const struct ofp_action_enqueue *oae)
2302 uint16_t ofp_port, odp_port;
2306 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2309 /* Fall back to ordinary output action. */
2310 xlate_output_action__(ctx, ntohs(oae->port), 0);
2314 /* Figure out ODP output port. */
2315 ofp_port = ntohs(oae->port);
2316 if (ofp_port != OFPP_IN_PORT) {
2317 odp_port = ofp_port_to_odp_port(ofp_port);
2319 odp_port = ctx->flow.in_port;
2322 /* Add ODP actions. */
2323 remove_pop_action(ctx);
2324 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2325 add_output_action(ctx, odp_port);
2326 add_pop_action(ctx);
2328 /* Update NetFlow output port. */
2329 if (ctx->nf_output_iface == NF_OUT_DROP) {
2330 ctx->nf_output_iface = odp_port;
2331 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2332 ctx->nf_output_iface = NF_OUT_MULTI;
2337 xlate_set_queue_action(struct action_xlate_ctx *ctx,
2338 const struct nx_action_set_queue *nasq)
2343 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
2346 /* Couldn't translate queue to a priority, so ignore. A warning
2347 * has already been logged. */
2351 remove_pop_action(ctx);
2352 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2356 xlate_set_dl_tci(struct action_xlate_ctx *ctx)
2358 ovs_be16 tci = ctx->flow.vlan_tci;
2359 if (!(tci & htons(VLAN_CFI))) {
2360 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN);
2362 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
2363 tci & ~htons(VLAN_CFI));
2367 struct xlate_reg_state {
2373 save_reg_state(const struct action_xlate_ctx *ctx,
2374 struct xlate_reg_state *state)
2376 state->vlan_tci = ctx->flow.vlan_tci;
2377 state->tun_id = ctx->flow.tun_id;
2381 update_reg_state(struct action_xlate_ctx *ctx,
2382 const struct xlate_reg_state *state)
2384 if (ctx->flow.vlan_tci != state->vlan_tci) {
2385 xlate_set_dl_tci(ctx);
2387 if (ctx->flow.tun_id != state->tun_id) {
2388 nl_msg_put_be64(ctx->odp_actions,
2389 ODP_ACTION_ATTR_SET_TUNNEL, ctx->flow.tun_id);
2394 xlate_nicira_action(struct action_xlate_ctx *ctx,
2395 const struct nx_action_header *nah)
2397 const struct nx_action_resubmit *nar;
2398 const struct nx_action_set_tunnel *nast;
2399 const struct nx_action_set_queue *nasq;
2400 const struct nx_action_multipath *nam;
2401 const struct nx_action_autopath *naa;
2402 enum nx_action_subtype subtype = ntohs(nah->subtype);
2403 const struct ofhooks *ofhooks = ctx->ofproto->ofhooks;
2404 struct xlate_reg_state state;
2405 uint16_t autopath_port;
2408 assert(nah->vendor == htonl(NX_VENDOR_ID));
2410 case NXAST_RESUBMIT:
2411 nar = (const struct nx_action_resubmit *) nah;
2412 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2415 case NXAST_SET_TUNNEL:
2416 nast = (const struct nx_action_set_tunnel *) nah;
2417 tun_id = htonll(ntohl(nast->tun_id));
2418 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
2419 ctx->flow.tun_id = tun_id;
2422 case NXAST_DROP_SPOOFED_ARP:
2423 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
2424 nl_msg_put_flag(ctx->odp_actions,
2425 ODP_ACTION_ATTR_DROP_SPOOFED_ARP);
2429 case NXAST_SET_QUEUE:
2430 nasq = (const struct nx_action_set_queue *) nah;
2431 xlate_set_queue_action(ctx, nasq);
2434 case NXAST_POP_QUEUE:
2435 add_pop_action(ctx);
2438 case NXAST_REG_MOVE:
2439 save_reg_state(ctx, &state);
2440 nxm_execute_reg_move((const struct nx_action_reg_move *) nah,
2442 update_reg_state(ctx, &state);
2445 case NXAST_REG_LOAD:
2446 save_reg_state(ctx, &state);
2447 nxm_execute_reg_load((const struct nx_action_reg_load *) nah,
2449 update_reg_state(ctx, &state);
2453 /* Nothing to do. */
2456 case NXAST_SET_TUNNEL64:
2457 tun_id = ((const struct nx_action_set_tunnel64 *) nah)->tun_id;
2458 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
2459 ctx->flow.tun_id = tun_id;
2462 case NXAST_MULTIPATH:
2463 nam = (const struct nx_action_multipath *) nah;
2464 multipath_execute(nam, &ctx->flow);
2467 case NXAST_AUTOPATH:
2468 naa = (const struct nx_action_autopath *) nah;
2469 autopath_port = (ofhooks->autopath_cb
2470 ? ofhooks->autopath_cb(&ctx->flow, ntohl(naa->id),
2471 &ctx->tags, ctx->ofproto->aux)
2473 autopath_execute(naa, &ctx->flow, autopath_port);
2476 /* If you add a new action here that modifies flow data, don't forget to
2477 * update the flow key in ctx->flow at the same time. */
2479 case NXAST_SNAT__OBSOLETE:
2481 VLOG_DBG_RL(&rl, "unknown Nicira action type %d", (int) subtype);
2487 do_xlate_actions(const union ofp_action *in, size_t n_in,
2488 struct action_xlate_ctx *ctx)
2490 struct actions_iterator iter;
2491 const union ofp_action *ia;
2492 const struct ofport *port;
2494 port = get_port(ctx->ofproto, ctx->flow.in_port);
2495 if (port && port->opp.config & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2496 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
2497 ? htonl(OFPPC_NO_RECV_STP)
2498 : htonl(OFPPC_NO_RECV))) {
2499 /* Drop this flow. */
2503 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2504 enum ofp_action_type type = ntohs(ia->type);
2505 const struct ofp_action_dl_addr *oada;
2509 xlate_output_action(ctx, &ia->output);
2512 case OFPAT_SET_VLAN_VID:
2513 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
2514 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
2515 xlate_set_dl_tci(ctx);
2518 case OFPAT_SET_VLAN_PCP:
2519 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
2520 ctx->flow.vlan_tci |= htons(
2521 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
2522 xlate_set_dl_tci(ctx);
2525 case OFPAT_STRIP_VLAN:
2526 ctx->flow.vlan_tci = htons(0);
2527 xlate_set_dl_tci(ctx);
2530 case OFPAT_SET_DL_SRC:
2531 oada = ((struct ofp_action_dl_addr *) ia);
2532 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_SRC,
2533 oada->dl_addr, ETH_ADDR_LEN);
2534 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
2537 case OFPAT_SET_DL_DST:
2538 oada = ((struct ofp_action_dl_addr *) ia);
2539 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_DST,
2540 oada->dl_addr, ETH_ADDR_LEN);
2541 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
2544 case OFPAT_SET_NW_SRC:
2545 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_SRC,
2546 ia->nw_addr.nw_addr);
2547 ctx->flow.nw_src = ia->nw_addr.nw_addr;
2550 case OFPAT_SET_NW_DST:
2551 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_DST,
2552 ia->nw_addr.nw_addr);
2553 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
2556 case OFPAT_SET_NW_TOS:
2557 nl_msg_put_u8(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_TOS,
2559 ctx->flow.nw_tos = ia->nw_tos.nw_tos;
2562 case OFPAT_SET_TP_SRC:
2563 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_SRC,
2564 ia->tp_port.tp_port);
2565 ctx->flow.tp_src = ia->tp_port.tp_port;
2568 case OFPAT_SET_TP_DST:
2569 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_DST,
2570 ia->tp_port.tp_port);
2571 ctx->flow.tp_dst = ia->tp_port.tp_port;
2575 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2579 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
2583 VLOG_DBG_RL(&rl, "unknown action type %d", (int) type);
2590 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
2591 struct ofproto *ofproto, const struct flow *flow,
2592 const struct ofpbuf *packet)
2594 ctx->ofproto = ofproto;
2596 ctx->packet = packet;
2597 ctx->resubmit_hook = NULL;
2598 ctx->check_special = true;
2602 ofproto_process_cfm(struct ofproto *ofproto, const struct flow *flow,
2603 const struct ofpbuf *packet)
2605 struct ofport *ofport;
2607 ofport = get_port(ofproto, flow->in_port);
2608 if (ofport && ofport->cfm) {
2609 cfm_process_heartbeat(ofport->cfm, packet);
2613 static struct ofpbuf *
2614 xlate_actions(struct action_xlate_ctx *ctx,
2615 const union ofp_action *in, size_t n_in)
2617 COVERAGE_INC(ofproto_ofp2odp);
2619 ctx->odp_actions = ofpbuf_new(512);
2621 ctx->may_set_up_flow = true;
2622 ctx->nf_output_iface = NF_OUT_DROP;
2624 ctx->last_pop_priority = -1;
2626 if (ctx->check_special && cfm_should_process_flow(&ctx->flow)) {
2628 ofproto_process_cfm(ctx->ofproto, &ctx->flow, ctx->packet);
2630 ctx->may_set_up_flow = false;
2631 } else if (ctx->check_special
2632 && ctx->ofproto->ofhooks->special_cb
2633 && !ctx->ofproto->ofhooks->special_cb(&ctx->flow, ctx->packet,
2634 ctx->ofproto->aux)) {
2635 ctx->may_set_up_flow = false;
2637 do_xlate_actions(in, n_in, ctx);
2640 remove_pop_action(ctx);
2642 /* Check with in-band control to see if we're allowed to set up this
2644 if (!connmgr_may_set_up_flow(ctx->ofproto->connmgr, &ctx->flow,
2645 ctx->odp_actions->data,
2646 ctx->odp_actions->size)) {
2647 ctx->may_set_up_flow = false;
2650 return ctx->odp_actions;
2653 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2654 * error message code (composed with ofp_mkerr()) for the caller to propagate
2655 * upward. Otherwise, returns 0.
2657 * The log message mentions 'msg_type'. */
2659 reject_slave_controller(struct ofconn *ofconn, const const char *msg_type)
2661 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2662 && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
2663 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
2664 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
2667 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2674 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2676 struct ofproto *p = ofconn_get_ofproto(ofconn);
2677 struct ofp_packet_out *opo;
2678 struct ofpbuf payload, *buffer;
2679 union ofp_action *ofp_actions;
2680 struct action_xlate_ctx ctx;
2681 struct ofpbuf *odp_actions;
2682 struct ofpbuf request;
2684 size_t n_ofp_actions;
2688 COVERAGE_INC(ofproto_packet_out);
2690 error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT");
2695 /* Get ofp_packet_out. */
2696 ofpbuf_use_const(&request, oh, ntohs(oh->length));
2697 opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions));
2700 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
2701 &ofp_actions, &n_ofp_actions);
2707 if (opo->buffer_id != htonl(UINT32_MAX)) {
2708 error = ofconn_pktbuf_retrieve(ofconn, ntohl(opo->buffer_id),
2710 if (error || !buffer) {
2719 /* Extract flow, check actions. */
2720 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)),
2722 error = validate_actions(ofp_actions, n_ofp_actions, &flow, p->max_ports);
2728 action_xlate_ctx_init(&ctx, p, &flow, &payload);
2729 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
2730 dpif_execute(p->dpif, odp_actions->data, odp_actions->size, &payload);
2731 ofpbuf_delete(odp_actions);
2734 ofpbuf_delete(buffer);
2739 update_port_config(struct ofproto *p, struct ofport *port,
2740 ovs_be32 config, ovs_be32 mask)
2742 mask &= config ^ port->opp.config;
2743 if (mask & htonl(OFPPC_PORT_DOWN)) {
2744 if (config & htonl(OFPPC_PORT_DOWN)) {
2745 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2747 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2750 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
2751 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
2752 if (mask & htonl(REVALIDATE_BITS)) {
2753 COVERAGE_INC(ofproto_costly_flags);
2754 port->opp.config ^= mask & htonl(REVALIDATE_BITS);
2755 p->need_revalidate = true;
2757 #undef REVALIDATE_BITS
2758 if (mask & htonl(OFPPC_NO_PACKET_IN)) {
2759 port->opp.config ^= htonl(OFPPC_NO_PACKET_IN);
2764 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2766 struct ofproto *p = ofconn_get_ofproto(ofconn);
2767 const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh;
2768 struct ofport *port;
2771 error = reject_slave_controller(ofconn, "OFPT_PORT_MOD");
2776 port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no)));
2778 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2779 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2780 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2782 update_port_config(p, port, opm->config, opm->mask);
2783 if (opm->advertise) {
2784 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2790 static struct ofpbuf *
2791 make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len)
2793 struct ofp_stats_reply *osr;
2796 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2797 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2799 osr->flags = htons(0);
2803 static struct ofpbuf *
2804 start_ofp_stats_reply(const struct ofp_header *request, size_t body_len)
2806 const struct ofp_stats_request *osr
2807 = (const struct ofp_stats_request *) request;
2808 return make_ofp_stats_reply(osr->header.xid, osr->type, body_len);
2812 append_ofp_stats_reply(size_t nbytes, struct ofconn *ofconn,
2813 struct ofpbuf **msgp)
2815 struct ofpbuf *msg = *msgp;
2816 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2817 if (nbytes + msg->size > UINT16_MAX) {
2818 struct ofp_stats_reply *reply = msg->data;
2819 reply->flags = htons(OFPSF_REPLY_MORE);
2820 *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes);
2821 ofconn_send_reply(ofconn, msg);
2823 return ofpbuf_put_uninit(*msgp, nbytes);
2826 static struct ofpbuf *
2827 make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len)
2829 struct nicira_stats_msg *nsm;
2832 msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX));
2833 nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg);
2834 nsm->type = htons(OFPST_VENDOR);
2835 nsm->flags = htons(0);
2836 nsm->vendor = htonl(NX_VENDOR_ID);
2837 nsm->subtype = subtype;
2841 static struct ofpbuf *
2842 start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len)
2844 return make_nxstats_reply(request->header.xid, request->subtype, body_len);
2848 append_nxstats_reply(size_t nbytes, struct ofconn *ofconn,
2849 struct ofpbuf **msgp)
2851 struct ofpbuf *msg = *msgp;
2852 assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg));
2853 if (nbytes + msg->size > UINT16_MAX) {
2854 struct nicira_stats_msg *reply = msg->data;
2855 reply->flags = htons(OFPSF_REPLY_MORE);
2856 *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes);
2857 ofconn_send_reply(ofconn, msg);
2859 ofpbuf_prealloc_tailroom(*msgp, nbytes);
2863 handle_desc_stats_request(struct ofconn *ofconn,
2864 const struct ofp_header *request)
2866 struct ofproto *p = ofconn_get_ofproto(ofconn);
2867 struct ofp_desc_stats *ods;
2870 msg = start_ofp_stats_reply(request, sizeof *ods);
2871 ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg);
2872 memset(ods, 0, sizeof *ods);
2873 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2874 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2875 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
2876 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
2877 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2878 ofconn_send_reply(ofconn, msg);
2884 handle_table_stats_request(struct ofconn *ofconn,
2885 const struct ofp_header *request)
2887 struct ofproto *p = ofconn_get_ofproto(ofconn);
2888 struct ofp_table_stats *ots;
2891 msg = start_ofp_stats_reply(request, sizeof *ots * 2);
2893 /* Classifier table. */
2894 ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg);
2895 memset(ots, 0, sizeof *ots);
2896 strcpy(ots->name, "classifier");
2897 ots->wildcards = (ofconn_get_flow_format(ofconn) == NXFF_OPENFLOW10
2898 ? htonl(OFPFW_ALL) : htonl(OVSFW_ALL));
2899 ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */
2900 ots->active_count = htonl(classifier_count(&p->cls));
2901 put_32aligned_be64(&ots->lookup_count, htonll(0)); /* XXX */
2902 put_32aligned_be64(&ots->matched_count, htonll(0)); /* XXX */
2904 ofconn_send_reply(ofconn, msg);
2909 append_port_stat(struct ofport *port, struct ofconn *ofconn,
2910 struct ofpbuf **msgp)
2912 struct netdev_stats stats;
2913 struct ofp_port_stats *ops;
2915 /* Intentionally ignore return value, since errors will set
2916 * 'stats' to all-1s, which is correct for OpenFlow, and
2917 * netdev_get_stats() will log errors. */
2918 netdev_get_stats(port->netdev, &stats);
2920 ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp);
2921 ops->port_no = port->opp.port_no;
2922 memset(ops->pad, 0, sizeof ops->pad);
2923 put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
2924 put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
2925 put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
2926 put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
2927 put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
2928 put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
2929 put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
2930 put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
2931 put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
2932 put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
2933 put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
2934 put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
2938 handle_port_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
2940 struct ofproto *p = ofconn_get_ofproto(ofconn);
2941 const struct ofp_port_stats_request *psr = ofputil_stats_body(oh);
2942 struct ofp_port_stats *ops;
2944 struct ofport *port;
2946 msg = start_ofp_stats_reply(oh, sizeof *ops * 16);
2947 if (psr->port_no != htons(OFPP_NONE)) {
2948 port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no)));
2950 append_port_stat(port, ofconn, &msg);
2953 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2954 append_port_stat(port, ofconn, &msg);
2958 ofconn_send_reply(ofconn, msg);
2963 calc_flow_duration__(long long int start, uint32_t *sec, uint32_t *nsec)
2965 long long int msecs = time_msec() - start;
2966 *sec = msecs / 1000;
2967 *nsec = (msecs % 1000) * (1000 * 1000);
2971 calc_flow_duration(long long int start, ovs_be32 *sec_be, ovs_be32 *nsec_be)
2975 calc_flow_duration__(start, &sec, &nsec);
2976 *sec_be = htonl(sec);
2977 *nsec_be = htonl(nsec);
2981 put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule,
2982 ovs_be16 out_port, struct ofpbuf **replyp)
2984 struct ofp_flow_stats *ofs;
2985 uint64_t packet_count, byte_count;
2987 size_t act_len, len;
2989 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
2993 act_len = sizeof *rule->actions * rule->n_actions;
2994 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2996 rule_get_stats(rule, &packet_count, &byte_count);
2998 ofs = append_ofp_stats_reply(len, ofconn, replyp);
2999 ofs->length = htons(len);
3002 ofputil_cls_rule_to_match(&rule->cr, ofconn_get_flow_format(ofconn),
3003 &ofs->match, rule->flow_cookie, &cookie);
3004 put_32aligned_be64(&ofs->cookie, cookie);
3005 calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec);
3006 ofs->priority = htons(rule->cr.priority);
3007 ofs->idle_timeout = htons(rule->idle_timeout);
3008 ofs->hard_timeout = htons(rule->hard_timeout);
3009 memset(ofs->pad2, 0, sizeof ofs->pad2);
3010 put_32aligned_be64(&ofs->packet_count, htonll(packet_count));
3011 put_32aligned_be64(&ofs->byte_count, htonll(byte_count));
3012 if (rule->n_actions > 0) {
3013 memcpy(ofs->actions, rule->actions, act_len);
3018 is_valid_table(uint8_t table_id)
3020 if (table_id == 0 || table_id == 0xff) {
3023 /* It would probably be better to reply with an error but there doesn't
3024 * seem to be any appropriate value, so that might just be
3026 VLOG_WARN_RL(&rl, "controller asked for invalid table %"PRIu8,
3033 handle_flow_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3035 const struct ofp_flow_stats_request *fsr = ofputil_stats_body(oh);
3036 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3037 struct ofpbuf *reply;
3039 COVERAGE_INC(ofproto_flows_req);
3040 reply = start_ofp_stats_reply(oh, 1024);
3041 if (is_valid_table(fsr->table_id)) {
3042 struct cls_cursor cursor;
3043 struct cls_rule target;
3046 ofputil_cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0,
3048 cls_cursor_init(&cursor, &ofproto->cls, &target);
3049 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3050 put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply);
3053 ofconn_send_reply(ofconn, reply);
3059 put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule,
3060 ovs_be16 out_port, struct ofpbuf **replyp)
3062 struct nx_flow_stats *nfs;
3063 uint64_t packet_count, byte_count;
3064 size_t act_len, start_len;
3065 struct ofpbuf *reply;
3067 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3071 rule_get_stats(rule, &packet_count, &byte_count);
3073 act_len = sizeof *rule->actions * rule->n_actions;
3075 append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, ofconn, replyp);
3076 start_len = (*replyp)->size;
3079 nfs = ofpbuf_put_uninit(reply, sizeof *nfs);
3082 calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec);
3083 nfs->cookie = rule->flow_cookie;
3084 nfs->priority = htons(rule->cr.priority);
3085 nfs->idle_timeout = htons(rule->idle_timeout);
3086 nfs->hard_timeout = htons(rule->hard_timeout);
3087 nfs->match_len = htons(nx_put_match(reply, &rule->cr));
3088 memset(nfs->pad2, 0, sizeof nfs->pad2);
3089 nfs->packet_count = htonll(packet_count);
3090 nfs->byte_count = htonll(byte_count);
3091 if (rule->n_actions > 0) {
3092 ofpbuf_put(reply, rule->actions, act_len);
3094 nfs->length = htons(reply->size - start_len);
3098 handle_nxst_flow(struct ofconn *ofconn, const struct ofp_header *oh)
3100 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3101 struct nx_flow_stats_request *nfsr;
3102 struct cls_rule target;
3103 struct ofpbuf *reply;
3107 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3109 /* Dissect the message. */
3110 nfsr = ofpbuf_pull(&b, sizeof *nfsr);
3111 error = nx_pull_match(&b, ntohs(nfsr->match_len), 0, &target);
3116 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3119 COVERAGE_INC(ofproto_flows_req);
3120 reply = start_nxstats_reply(&nfsr->nsm, 1024);
3121 if (is_valid_table(nfsr->table_id)) {
3122 struct cls_cursor cursor;
3125 cls_cursor_init(&cursor, &ofproto->cls, &target);
3126 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3127 put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply);
3130 ofconn_send_reply(ofconn, reply);
3136 flow_stats_ds(struct rule *rule, struct ds *results)
3138 uint64_t packet_count, byte_count;
3139 size_t act_len = sizeof *rule->actions * rule->n_actions;
3141 rule_get_stats(rule, &packet_count, &byte_count);
3143 ds_put_format(results, "duration=%llds, ",
3144 (time_msec() - rule->created) / 1000);
3145 ds_put_format(results, "idle=%.3fs, ", (time_msec() - rule->used) / 1000.0);
3146 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3147 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3148 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3149 cls_rule_format(&rule->cr, results);
3150 ds_put_char(results, ',');
3152 ofp_print_actions(results, &rule->actions->header, act_len);
3154 ds_put_cstr(results, "drop");
3156 ds_put_cstr(results, "\n");
3159 /* Adds a pretty-printed description of all flows to 'results', including
3160 * hidden flows (e.g., set up by in-band control). */
3162 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3164 struct cls_cursor cursor;
3167 cls_cursor_init(&cursor, &p->cls, NULL);
3168 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3169 flow_stats_ds(rule, results);
3173 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
3174 * '*engine_type' and '*engine_id', respectively. */
3176 ofproto_get_netflow_ids(const struct ofproto *ofproto,
3177 uint8_t *engine_type, uint8_t *engine_id)
3179 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
3183 query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target,
3184 ovs_be16 out_port, uint8_t table_id,
3185 struct ofp_aggregate_stats_reply *oasr)
3187 uint64_t total_packets = 0;
3188 uint64_t total_bytes = 0;
3191 COVERAGE_INC(ofproto_agg_request);
3193 if (is_valid_table(table_id)) {
3194 struct cls_cursor cursor;
3197 cls_cursor_init(&cursor, &ofproto->cls, target);
3198 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3199 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
3200 uint64_t packet_count;
3201 uint64_t byte_count;
3203 rule_get_stats(rule, &packet_count, &byte_count);
3205 total_packets += packet_count;
3206 total_bytes += byte_count;
3212 oasr->flow_count = htonl(n_flows);
3213 put_32aligned_be64(&oasr->packet_count, htonll(total_packets));
3214 put_32aligned_be64(&oasr->byte_count, htonll(total_bytes));
3215 memset(oasr->pad, 0, sizeof oasr->pad);
3219 handle_aggregate_stats_request(struct ofconn *ofconn,
3220 const struct ofp_header *oh)
3222 const struct ofp_aggregate_stats_request *request = ofputil_stats_body(oh);
3223 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3224 struct ofp_aggregate_stats_reply *reply;
3225 struct cls_rule target;
3228 ofputil_cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0,
3231 msg = start_ofp_stats_reply(oh, sizeof *reply);
3232 reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg);
3233 query_aggregate_stats(ofproto, &target, request->out_port,
3234 request->table_id, reply);
3235 ofconn_send_reply(ofconn, msg);
3240 handle_nxst_aggregate(struct ofconn *ofconn, const struct ofp_header *oh)
3242 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3243 struct nx_aggregate_stats_request *request;
3244 struct ofp_aggregate_stats_reply *reply;
3245 struct cls_rule target;
3250 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3252 /* Dissect the message. */
3253 request = ofpbuf_pull(&b, sizeof *request);
3254 error = nx_pull_match(&b, ntohs(request->match_len), 0, &target);
3259 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3263 COVERAGE_INC(ofproto_flows_req);
3264 buf = start_nxstats_reply(&request->nsm, sizeof *reply);
3265 reply = ofpbuf_put_uninit(buf, sizeof *reply);
3266 query_aggregate_stats(ofproto, &target, request->out_port,
3267 request->table_id, reply);
3268 ofconn_send_reply(ofconn, buf);
3273 struct queue_stats_cbdata {
3274 struct ofconn *ofconn;
3275 struct ofport *ofport;
3280 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3281 const struct netdev_queue_stats *stats)
3283 struct ofp_queue_stats *reply;
3285 reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3286 reply->port_no = cbdata->ofport->opp.port_no;
3287 memset(reply->pad, 0, sizeof reply->pad);
3288 reply->queue_id = htonl(queue_id);
3289 put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
3290 put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
3291 put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
3295 handle_queue_stats_dump_cb(uint32_t queue_id,
3296 struct netdev_queue_stats *stats,
3299 struct queue_stats_cbdata *cbdata = cbdata_;
3301 put_queue_stats(cbdata, queue_id, stats);
3305 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3306 struct queue_stats_cbdata *cbdata)
3308 cbdata->ofport = port;
3309 if (queue_id == OFPQ_ALL) {
3310 netdev_dump_queue_stats(port->netdev,
3311 handle_queue_stats_dump_cb, cbdata);
3313 struct netdev_queue_stats stats;
3315 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3316 put_queue_stats(cbdata, queue_id, &stats);
3322 handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3324 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3325 const struct ofp_queue_stats_request *qsr;
3326 struct queue_stats_cbdata cbdata;
3327 struct ofport *port;
3328 unsigned int port_no;
3331 qsr = ofputil_stats_body(oh);
3333 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3336 COVERAGE_INC(ofproto_queue_req);
3338 cbdata.ofconn = ofconn;
3339 cbdata.msg = start_ofp_stats_reply(oh, 128);
3341 port_no = ntohs(qsr->port_no);
3342 queue_id = ntohl(qsr->queue_id);
3343 if (port_no == OFPP_ALL) {
3344 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3345 handle_queue_stats_for_port(port, queue_id, &cbdata);
3347 } else if (port_no < ofproto->max_ports) {
3348 port = get_port(ofproto, ofp_port_to_odp_port(port_no));
3350 handle_queue_stats_for_port(port, queue_id, &cbdata);
3353 ofpbuf_delete(cbdata.msg);
3354 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3356 ofconn_send_reply(ofconn, cbdata.msg);
3361 /* Updates 'facet''s used time. Caller is responsible for calling
3362 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3364 facet_update_time(struct ofproto *ofproto, struct facet *facet,
3367 if (used > facet->used) {
3369 if (used > facet->rule->used) {
3370 facet->rule->used = used;
3372 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3376 /* Folds the statistics from 'stats' into the counters in 'facet'.
3378 * Because of the meaning of a facet's counters, it only makes sense to do this
3379 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3380 * packet that was sent by hand or if it represents statistics that have been
3381 * cleared out of the datapath. */
3383 facet_update_stats(struct ofproto *ofproto, struct facet *facet,
3384 const struct dpif_flow_stats *stats)
3386 if (stats->n_packets || stats->used > facet->used) {
3387 facet_update_time(ofproto, facet, stats->used);
3388 facet->packet_count += stats->n_packets;
3389 facet->byte_count += stats->n_bytes;
3390 facet_push_stats(ofproto, facet);
3391 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3396 facet_push_stats(struct ofproto *ofproto, struct facet *facet)
3398 uint64_t rs_packets, rs_bytes;
3400 assert(facet->packet_count >= facet->rs_packet_count);
3401 assert(facet->byte_count >= facet->rs_byte_count);
3402 assert(facet->used >= facet->rs_used);
3404 rs_packets = facet->packet_count - facet->rs_packet_count;
3405 rs_bytes = facet->byte_count - facet->rs_byte_count;
3407 if (rs_packets || rs_bytes || facet->used > facet->rs_used) {
3408 facet->rs_packet_count = facet->packet_count;
3409 facet->rs_byte_count = facet->byte_count;
3410 facet->rs_used = facet->used;
3412 flow_push_stats(ofproto, facet->rule, &facet->flow,
3413 rs_packets, rs_bytes, facet->used);
3417 struct ofproto_push {
3418 struct action_xlate_ctx ctx;
3425 push_resubmit(struct action_xlate_ctx *ctx, struct rule *rule)
3427 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3430 rule->packet_count += push->packets;
3431 rule->byte_count += push->bytes;
3432 rule->used = MAX(push->used, rule->used);
3436 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3437 * 'rule''s actions. */
3439 flow_push_stats(struct ofproto *ofproto, const struct rule *rule,
3440 struct flow *flow, uint64_t packets, uint64_t bytes,
3443 struct ofproto_push push;
3445 push.packets = packets;
3449 action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL);
3450 push.ctx.resubmit_hook = push_resubmit;
3451 ofpbuf_delete(xlate_actions(&push.ctx, rule->actions, rule->n_actions));
3454 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3455 * in which no matching flow already exists in the flow table.
3457 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3458 * ofp_actions, to the ofproto's flow table. Returns 0 on success or an
3459 * OpenFlow error code as encoded by ofp_mkerr() on failure.
3461 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3464 add_flow(struct ofconn *ofconn, struct flow_mod *fm)
3466 struct ofproto *p = ofconn_get_ofproto(ofconn);
3467 struct ofpbuf *packet;
3472 if (fm->flags & OFPFF_CHECK_OVERLAP
3473 && classifier_rule_overlaps(&p->cls, &fm->cr)) {
3474 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3478 if (fm->buffer_id != UINT32_MAX) {
3479 error = ofconn_pktbuf_retrieve(ofconn, fm->buffer_id,
3483 in_port = UINT16_MAX;
3486 rule = rule_create(&fm->cr, fm->actions, fm->n_actions,
3487 fm->idle_timeout, fm->hard_timeout, fm->cookie,
3488 fm->flags & OFPFF_SEND_FLOW_REM);
3489 rule_insert(p, rule);
3491 rule_execute(p, rule, in_port, packet);
3496 static struct rule *
3497 find_flow_strict(struct ofproto *p, const struct flow_mod *fm)
3499 return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr));
3503 send_buffered_packet(struct ofconn *ofconn,
3504 struct rule *rule, uint32_t buffer_id)
3506 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3507 struct ofpbuf *packet;
3511 if (buffer_id == UINT32_MAX) {
3515 error = ofconn_pktbuf_retrieve(ofconn, buffer_id, &packet, &in_port);
3520 rule_execute(ofproto, rule, in_port, packet);
3525 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3527 struct modify_flows_cbdata {
3528 struct ofproto *ofproto;
3529 const struct flow_mod *fm;
3533 static int modify_flow(struct ofproto *, const struct flow_mod *,
3536 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3537 * encoded by ofp_mkerr() on failure.
3539 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3542 modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm)
3544 struct ofproto *p = ofconn_get_ofproto(ofconn);
3545 struct rule *match = NULL;
3546 struct cls_cursor cursor;
3549 cls_cursor_init(&cursor, &p->cls, &fm->cr);
3550 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3551 if (!rule_is_hidden(rule)) {
3553 modify_flow(p, fm, rule);
3558 /* This credits the packet to whichever flow happened to match last.
3559 * That's weird. Maybe we should do a lookup for the flow that
3560 * actually matches the packet? Who knows. */
3561 send_buffered_packet(ofconn, match, fm->buffer_id);
3564 return add_flow(ofconn, fm);
3568 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3569 * code as encoded by ofp_mkerr() on failure.
3571 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3574 modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm)
3576 struct ofproto *p = ofconn_get_ofproto(ofconn);
3577 struct rule *rule = find_flow_strict(p, fm);
3578 if (rule && !rule_is_hidden(rule)) {
3579 modify_flow(p, fm, rule);
3580 return send_buffered_packet(ofconn, rule, fm->buffer_id);
3582 return add_flow(ofconn, fm);
3586 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3587 * been identified as a flow in 'p''s flow table to be modified, by changing
3588 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3589 * ofp_action[] structures). */
3591 modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule)
3593 size_t actions_len = fm->n_actions * sizeof *rule->actions;
3595 rule->flow_cookie = fm->cookie;
3597 /* If the actions are the same, do nothing. */
3598 if (fm->n_actions == rule->n_actions
3600 || !memcmp(fm->actions, rule->actions, actions_len))) {
3604 /* Replace actions. */
3605 free(rule->actions);
3606 rule->actions = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL;
3607 rule->n_actions = fm->n_actions;
3609 p->need_revalidate = true;
3614 /* OFPFC_DELETE implementation. */
3616 static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port);
3618 /* Implements OFPFC_DELETE. */
3620 delete_flows_loose(struct ofproto *p, const struct flow_mod *fm)
3622 struct rule *rule, *next_rule;
3623 struct cls_cursor cursor;
3625 cls_cursor_init(&cursor, &p->cls, &fm->cr);
3626 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
3627 delete_flow(p, rule, htons(fm->out_port));
3631 /* Implements OFPFC_DELETE_STRICT. */
3633 delete_flow_strict(struct ofproto *p, struct flow_mod *fm)
3635 struct rule *rule = find_flow_strict(p, fm);
3637 delete_flow(p, rule, htons(fm->out_port));
3641 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3642 * been identified as a flow to delete from 'p''s flow table, by deleting the
3643 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3646 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3647 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3648 * specified 'out_port'. */
3650 delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port)
3652 if (rule_is_hidden(rule)) {
3656 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
3660 rule_send_removed(p, rule, OFPRR_DELETE);
3661 rule_remove(p, rule);
3665 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3667 struct ofproto *p = ofconn_get_ofproto(ofconn);
3671 error = reject_slave_controller(ofconn, "flow_mod");
3676 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_flow_format(ofconn));
3681 /* We do not support the emergency flow cache. It will hopefully get
3682 * dropped from OpenFlow in the near future. */
3683 if (fm.flags & OFPFF_EMERG) {
3684 /* There isn't a good fit for an error code, so just state that the
3685 * flow table is full. */
3686 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3689 error = validate_actions(fm.actions, fm.n_actions,
3690 &fm.cr.flow, p->max_ports);
3695 switch (fm.command) {
3697 return add_flow(ofconn, &fm);
3700 return modify_flows_loose(ofconn, &fm);
3702 case OFPFC_MODIFY_STRICT:
3703 return modify_flow_strict(ofconn, &fm);
3706 delete_flows_loose(p, &fm);
3709 case OFPFC_DELETE_STRICT:
3710 delete_flow_strict(p, &fm);
3714 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3719 handle_tun_id_from_cookie(struct ofconn *ofconn, const struct ofp_header *oh)
3721 const struct nxt_tun_id_cookie *msg
3722 = (const struct nxt_tun_id_cookie *) oh;
3723 enum nx_flow_format flow_format;
3725 flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10;
3726 ofconn_set_flow_format(ofconn, flow_format);
3732 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
3734 struct nx_role_request *nrr = (struct nx_role_request *) oh;
3735 struct nx_role_request *reply;
3739 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY) {
3740 VLOG_WARN_RL(&rl, "ignoring role request on service connection");
3741 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3744 role = ntohl(nrr->role);
3745 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3746 && role != NX_ROLE_SLAVE) {
3747 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
3749 /* There's no good error code for this. */
3750 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
3753 ofconn_set_role(ofconn, role);
3755 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
3756 reply->role = htonl(role);
3757 ofconn_send_reply(ofconn, buf);
3763 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
3765 const struct nxt_set_flow_format *msg
3766 = (const struct nxt_set_flow_format *) oh;
3769 format = ntohl(msg->format);
3770 if (format == NXFF_OPENFLOW10
3771 || format == NXFF_TUN_ID_FROM_COOKIE
3772 || format == NXFF_NXM) {
3773 ofconn_set_flow_format(ofconn, format);
3776 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3781 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
3783 struct ofp_header *ob;
3786 /* Currently, everything executes synchronously, so we can just
3787 * immediately send the barrier reply. */
3788 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
3789 ofconn_send_reply(ofconn, buf);
3794 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
3796 const struct ofp_header *oh = msg->data;
3797 const struct ofputil_msg_type *type;
3800 error = ofputil_decode_msg_type(oh, &type);
3805 switch (ofputil_msg_type_code(type)) {
3806 /* OpenFlow requests. */
3807 case OFPUTIL_OFPT_ECHO_REQUEST:
3808 return handle_echo_request(ofconn, oh);
3810 case OFPUTIL_OFPT_FEATURES_REQUEST:
3811 return handle_features_request(ofconn, oh);
3813 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
3814 return handle_get_config_request(ofconn, oh);
3816 case OFPUTIL_OFPT_SET_CONFIG:
3817 return handle_set_config(ofconn, msg->data);
3819 case OFPUTIL_OFPT_PACKET_OUT:
3820 return handle_packet_out(ofconn, oh);
3822 case OFPUTIL_OFPT_PORT_MOD:
3823 return handle_port_mod(ofconn, oh);
3825 case OFPUTIL_OFPT_FLOW_MOD:
3826 return handle_flow_mod(ofconn, oh);
3828 case OFPUTIL_OFPT_BARRIER_REQUEST:
3829 return handle_barrier_request(ofconn, oh);
3831 /* OpenFlow replies. */
3832 case OFPUTIL_OFPT_ECHO_REPLY:
3835 /* Nicira extension requests. */
3836 case OFPUTIL_NXT_TUN_ID_FROM_COOKIE:
3837 return handle_tun_id_from_cookie(ofconn, oh);
3839 case OFPUTIL_NXT_ROLE_REQUEST:
3840 return handle_role_request(ofconn, oh);
3842 case OFPUTIL_NXT_SET_FLOW_FORMAT:
3843 return handle_nxt_set_flow_format(ofconn, oh);
3845 case OFPUTIL_NXT_FLOW_MOD:
3846 return handle_flow_mod(ofconn, oh);
3848 /* OpenFlow statistics requests. */
3849 case OFPUTIL_OFPST_DESC_REQUEST:
3850 return handle_desc_stats_request(ofconn, oh);
3852 case OFPUTIL_OFPST_FLOW_REQUEST:
3853 return handle_flow_stats_request(ofconn, oh);
3855 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
3856 return handle_aggregate_stats_request(ofconn, oh);
3858 case OFPUTIL_OFPST_TABLE_REQUEST:
3859 return handle_table_stats_request(ofconn, oh);
3861 case OFPUTIL_OFPST_PORT_REQUEST:
3862 return handle_port_stats_request(ofconn, oh);
3864 case OFPUTIL_OFPST_QUEUE_REQUEST:
3865 return handle_queue_stats_request(ofconn, oh);
3867 /* Nicira extension statistics requests. */
3868 case OFPUTIL_NXST_FLOW_REQUEST:
3869 return handle_nxst_flow(ofconn, oh);
3871 case OFPUTIL_NXST_AGGREGATE_REQUEST:
3872 return handle_nxst_aggregate(ofconn, oh);
3874 case OFPUTIL_INVALID:
3875 case OFPUTIL_OFPT_HELLO:
3876 case OFPUTIL_OFPT_ERROR:
3877 case OFPUTIL_OFPT_FEATURES_REPLY:
3878 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
3879 case OFPUTIL_OFPT_PACKET_IN:
3880 case OFPUTIL_OFPT_FLOW_REMOVED:
3881 case OFPUTIL_OFPT_PORT_STATUS:
3882 case OFPUTIL_OFPT_BARRIER_REPLY:
3883 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
3884 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
3885 case OFPUTIL_OFPST_DESC_REPLY:
3886 case OFPUTIL_OFPST_FLOW_REPLY:
3887 case OFPUTIL_OFPST_QUEUE_REPLY:
3888 case OFPUTIL_OFPST_PORT_REPLY:
3889 case OFPUTIL_OFPST_TABLE_REPLY:
3890 case OFPUTIL_OFPST_AGGREGATE_REPLY:
3891 case OFPUTIL_NXT_ROLE_REPLY:
3892 case OFPUTIL_NXT_FLOW_REMOVED:
3893 case OFPUTIL_NXST_FLOW_REPLY:
3894 case OFPUTIL_NXST_AGGREGATE_REPLY:
3896 if (VLOG_IS_WARN_ENABLED()) {
3897 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3898 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3901 if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) {
3902 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
3904 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3910 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
3912 int error = handle_openflow__(ofconn, ofp_msg);
3914 send_error_oh(ofconn, ofp_msg->data, error);
3916 COVERAGE_INC(ofproto_recv_openflow);
3920 handle_miss_upcall(struct ofproto *p, struct dpif_upcall *upcall)
3922 struct facet *facet;
3925 /* Obtain in_port and tun_id, at least. */
3926 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
3928 /* Set header pointers in 'flow'. */
3929 flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow);
3931 if (cfm_should_process_flow(&flow)) {
3932 ofproto_process_cfm(p, &flow, upcall->packet);
3933 ofpbuf_delete(upcall->packet);
3935 } else if (p->ofhooks->special_cb
3936 && !p->ofhooks->special_cb(&flow, upcall->packet, p->aux)) {
3937 ofpbuf_delete(upcall->packet);
3941 /* Check with in-band control to see if this packet should be sent
3942 * to the local port regardless of the flow table. */
3943 if (connmgr_msg_in_hook(p->connmgr, &flow, upcall->packet)) {
3944 ofproto_send_packet(p, ODPP_LOCAL, 0, upcall->packet);
3947 facet = facet_lookup_valid(p, &flow);
3949 struct rule *rule = rule_lookup(p, &flow);
3951 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3952 struct ofport *port = get_port(p, flow.in_port);
3954 if (port->opp.config & htonl(OFPPC_NO_PACKET_IN)) {
3955 COVERAGE_INC(ofproto_no_packet_in);
3956 /* XXX install 'drop' flow entry */
3957 ofpbuf_delete(upcall->packet);
3961 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
3965 COVERAGE_INC(ofproto_packet_in);
3966 send_packet_in(p, upcall, &flow, false);
3970 facet = facet_create(p, rule, &flow, upcall->packet);
3971 } else if (!facet->may_install) {
3972 /* The facet is not installable, that is, we need to process every
3973 * packet, so process the current packet's actions into 'facet'. */
3974 facet_make_actions(p, facet, upcall->packet);
3977 if (facet->rule->cr.priority == FAIL_OPEN_PRIORITY) {
3979 * Extra-special case for fail-open mode.
3981 * We are in fail-open mode and the packet matched the fail-open rule,
3982 * but we are connected to a controller too. We should send the packet
3983 * up to the controller in the hope that it will try to set up a flow
3984 * and thereby allow us to exit fail-open.
3986 * See the top-level comment in fail-open.c for more information.
3988 send_packet_in(p, upcall, &flow, true);
3991 facet_execute(p, facet, upcall->packet);
3992 facet_install(p, facet, false);
3996 handle_upcall(struct ofproto *p, struct dpif_upcall *upcall)
4000 switch (upcall->type) {
4001 case DPIF_UC_ACTION:
4002 COVERAGE_INC(ofproto_ctlr_action);
4003 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4004 send_packet_in(p, upcall, &flow, false);
4007 case DPIF_UC_SAMPLE:
4009 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4010 ofproto_sflow_received(p->sflow, upcall, &flow);
4012 ofpbuf_delete(upcall->packet);
4016 handle_miss_upcall(p, upcall);
4019 case DPIF_N_UC_TYPES:
4021 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4026 /* Flow expiration. */
4028 static int ofproto_dp_max_idle(const struct ofproto *);
4029 static void ofproto_update_stats(struct ofproto *);
4030 static void rule_expire(struct ofproto *, struct rule *);
4031 static void ofproto_expire_facets(struct ofproto *, int dp_max_idle);
4033 /* This function is called periodically by ofproto_run(). Its job is to
4034 * collect updates for the flows that have been installed into the datapath,
4035 * most importantly when they last were used, and then use that information to
4036 * expire flows that have not been used recently.
4038 * Returns the number of milliseconds after which it should be called again. */
4040 ofproto_expire(struct ofproto *ofproto)
4042 struct rule *rule, *next_rule;
4043 struct cls_cursor cursor;
4046 /* Update stats for each flow in the datapath. */
4047 ofproto_update_stats(ofproto);
4049 /* Expire facets that have been idle too long. */
4050 dp_max_idle = ofproto_dp_max_idle(ofproto);
4051 ofproto_expire_facets(ofproto, dp_max_idle);
4053 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
4054 cls_cursor_init(&cursor, &ofproto->cls, NULL);
4055 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4056 rule_expire(ofproto, rule);
4059 /* Let the hook know that we're at a stable point: all outstanding data
4060 * in existing flows has been accounted to the account_cb. Thus, the
4061 * hook can now reasonably do operations that depend on having accurate
4062 * flow volume accounting (currently, that's just bond rebalancing). */
4063 if (ofproto->ofhooks->account_checkpoint_cb) {
4064 ofproto->ofhooks->account_checkpoint_cb(ofproto->aux);
4067 return MIN(dp_max_idle, 1000);
4070 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4072 * This function also pushes statistics updates to rules which each facet
4073 * resubmits into. Generally these statistics will be accurate. However, if a
4074 * facet changes the rule it resubmits into at some time in between
4075 * ofproto_update_stats() runs, it is possible that statistics accrued to the
4076 * old rule will be incorrectly attributed to the new rule. This could be
4077 * avoided by calling ofproto_update_stats() whenever rules are created or
4078 * deleted. However, the performance impact of making so many calls to the
4079 * datapath do not justify the benefit of having perfectly accurate statistics.
4082 ofproto_update_stats(struct ofproto *p)
4084 const struct dpif_flow_stats *stats;
4085 struct dpif_flow_dump dump;
4086 const struct nlattr *key;
4089 dpif_flow_dump_start(&dump, p->dpif);
4090 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4091 struct facet *facet;
4094 if (odp_flow_key_to_flow(key, key_len, &flow)) {
4098 odp_flow_key_format(key, key_len, &s);
4099 VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s",
4105 facet = facet_find(p, &flow);
4107 if (facet && facet->installed) {
4109 if (stats->n_packets >= facet->dp_packet_count) {
4110 facet->packet_count += stats->n_packets - facet->dp_packet_count;
4112 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4115 if (stats->n_bytes >= facet->dp_byte_count) {
4116 facet->byte_count += stats->n_bytes - facet->dp_byte_count;
4118 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4121 facet->dp_packet_count = stats->n_packets;
4122 facet->dp_byte_count = stats->n_bytes;
4124 facet_update_time(p, facet, stats->used);
4125 facet_account(p, facet, stats->n_bytes);
4126 facet_push_stats(p, facet);
4128 /* There's a flow in the datapath that we know nothing about.
4130 COVERAGE_INC(ofproto_unexpected_rule);
4131 dpif_flow_del(p->dpif, key, key_len, NULL);
4134 dpif_flow_dump_done(&dump);
4137 /* Calculates and returns the number of milliseconds of idle time after which
4138 * facets should expire from the datapath and we should fold their statistics
4139 * into their parent rules in userspace. */
4141 ofproto_dp_max_idle(const struct ofproto *ofproto)
4144 * Idle time histogram.
4146 * Most of the time a switch has a relatively small number of facets. When
4147 * this is the case we might as well keep statistics for all of them in
4148 * userspace and to cache them in the kernel datapath for performance as
4151 * As the number of facets increases, the memory required to maintain
4152 * statistics about them in userspace and in the kernel becomes
4153 * significant. However, with a large number of facets it is likely that
4154 * only a few of them are "heavy hitters" that consume a large amount of
4155 * bandwidth. At this point, only heavy hitters are worth caching in the
4156 * kernel and maintaining in userspaces; other facets we can discard.
4158 * The technique used to compute the idle time is to build a histogram with
4159 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
4160 * that is installed in the kernel gets dropped in the appropriate bucket.
4161 * After the histogram has been built, we compute the cutoff so that only
4162 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
4163 * cached. At least the most-recently-used bucket of facets is kept, so
4164 * actually an arbitrary number of facets can be kept in any given
4165 * expiration run (though the next run will delete most of those unless
4166 * they receive additional data).
4168 * This requires a second pass through the facets, in addition to the pass
4169 * made by ofproto_update_stats(), because the former function never looks
4170 * at uninstallable facets.
4172 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4173 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4174 int buckets[N_BUCKETS] = { 0 };
4175 struct facet *facet;
4180 total = hmap_count(&ofproto->facets);
4181 if (total <= 1000) {
4182 return N_BUCKETS * BUCKET_WIDTH;
4185 /* Build histogram. */
4187 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
4188 long long int idle = now - facet->used;
4189 int bucket = (idle <= 0 ? 0
4190 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4191 : (unsigned int) idle / BUCKET_WIDTH);
4195 /* Find the first bucket whose flows should be expired. */
4196 for (bucket = 0; bucket < N_BUCKETS; bucket++) {
4197 if (buckets[bucket]) {
4200 subtotal += buckets[bucket++];
4201 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
4206 if (VLOG_IS_DBG_ENABLED()) {
4210 ds_put_cstr(&s, "keep");
4211 for (i = 0; i < N_BUCKETS; i++) {
4213 ds_put_cstr(&s, ", drop");
4216 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4219 VLOG_INFO("%s: %s (msec:count)",
4220 dpif_name(ofproto->dpif), ds_cstr(&s));
4224 return bucket * BUCKET_WIDTH;
4228 facet_active_timeout(struct ofproto *ofproto, struct facet *facet)
4230 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
4231 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
4232 struct ofexpired expired;
4234 if (facet->installed) {
4235 struct dpif_flow_stats stats;
4237 facet_put__(ofproto, facet, facet->actions, facet->actions_len,
4239 facet_update_stats(ofproto, facet, &stats);
4242 expired.flow = facet->flow;
4243 expired.packet_count = facet->packet_count;
4244 expired.byte_count = facet->byte_count;
4245 expired.used = facet->used;
4246 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4251 ofproto_expire_facets(struct ofproto *ofproto, int dp_max_idle)
4253 long long int cutoff = time_msec() - dp_max_idle;
4254 struct facet *facet, *next_facet;
4256 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
4257 facet_active_timeout(ofproto, facet);
4258 if (facet->used < cutoff) {
4259 facet_remove(ofproto, facet);
4264 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4265 * then delete it entirely. */
4267 rule_expire(struct ofproto *ofproto, struct rule *rule)
4269 struct facet *facet, *next_facet;
4273 /* Has 'rule' expired? */
4275 if (rule->hard_timeout
4276 && now > rule->created + rule->hard_timeout * 1000) {
4277 reason = OFPRR_HARD_TIMEOUT;
4278 } else if (rule->idle_timeout && list_is_empty(&rule->facets)
4279 && now >rule->used + rule->idle_timeout * 1000) {
4280 reason = OFPRR_IDLE_TIMEOUT;
4285 COVERAGE_INC(ofproto_expired);
4287 /* Update stats. (This is a no-op if the rule expired due to an idle
4288 * timeout, because that only happens when the rule has no facets left.) */
4289 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4290 facet_remove(ofproto, facet);
4293 /* Get rid of the rule. */
4294 if (!rule_is_hidden(rule)) {
4295 rule_send_removed(ofproto, rule, reason);
4297 rule_remove(ofproto, rule);
4301 rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason)
4303 struct ofputil_flow_removed fr;
4305 if (!rule->send_flow_removed) {
4310 fr.cookie = rule->flow_cookie;
4312 calc_flow_duration__(rule->created, &fr.duration_sec, &fr.duration_nsec);
4313 fr.idle_timeout = rule->idle_timeout;
4314 fr.packet_count = rule->packet_count;
4315 fr.byte_count = rule->byte_count;
4317 connmgr_send_flow_removed(p->connmgr, &fr);
4320 /* Obtains statistics for 'rule' and stores them in '*packets' and '*bytes'.
4321 * The returned statistics include statistics for all of 'rule''s facets. */
4323 rule_get_stats(const struct rule *rule, uint64_t *packets, uint64_t *bytes)
4326 struct facet *facet;
4328 /* Start from historical data for 'rule' itself that are no longer tracked
4329 * in facets. This counts, for example, facets that have expired. */
4330 p = rule->packet_count;
4331 b = rule->byte_count;
4333 /* Add any statistics that are tracked by facets. This includes
4334 * statistical data recently updated by ofproto_update_stats() as well as
4335 * stats for packets that were executed "by hand" via dpif_execute(). */
4336 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4337 p += facet->packet_count;
4338 b += facet->byte_count;
4345 /* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an
4346 * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to
4347 * their individual configurations.
4349 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
4350 * Otherwise, ownership is transferred to this function. */
4352 send_packet_in(struct ofproto *ofproto, struct dpif_upcall *upcall,
4353 const struct flow *flow, bool clone)
4355 struct ofputil_packet_in pin;
4357 pin.packet = upcall->packet;
4358 pin.in_port = odp_port_to_ofp_port(flow->in_port);
4359 pin.reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION;
4360 pin.buffer_id = 0; /* not yet known */
4361 pin.send_len = upcall->userdata;
4362 connmgr_send_packet_in(ofproto->connmgr, upcall, flow,
4363 clone ? NULL : upcall->packet);
4367 pick_datapath_id(const struct ofproto *ofproto)
4369 const struct ofport *port;
4371 port = get_port(ofproto, ODPP_LOCAL);
4373 uint8_t ea[ETH_ADDR_LEN];
4376 error = netdev_get_etheraddr(port->netdev, ea);
4378 return eth_addr_to_uint64(ea);
4380 VLOG_WARN("could not get MAC address for %s (%s)",
4381 netdev_get_name(port->netdev), strerror(error));
4383 return ofproto->fallback_dpid;
4387 pick_fallback_dpid(void)
4389 uint8_t ea[ETH_ADDR_LEN];
4390 eth_addr_nicira_random(ea);
4391 return eth_addr_to_uint64(ea);
4395 ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED,
4396 void *aux OVS_UNUSED)
4398 const struct shash_node *node;
4402 SHASH_FOR_EACH (node, &all_ofprotos) {
4403 ds_put_format(&results, "%s\n", node->name);
4405 unixctl_command_reply(conn, 200, ds_cstr(&results));
4406 ds_destroy(&results);
4409 struct ofproto_trace {
4410 struct action_xlate_ctx ctx;
4416 trace_format_rule(struct ds *result, int level, const struct rule *rule)
4418 ds_put_char_multiple(result, '\t', level);
4420 ds_put_cstr(result, "No match\n");
4424 ds_put_format(result, "Rule: cookie=%#"PRIx64" ",
4425 ntohll(rule->flow_cookie));
4426 cls_rule_format(&rule->cr, result);
4427 ds_put_char(result, '\n');
4429 ds_put_char_multiple(result, '\t', level);
4430 ds_put_cstr(result, "OpenFlow ");
4431 ofp_print_actions(result, (const struct ofp_action_header *) rule->actions,
4432 rule->n_actions * sizeof *rule->actions);
4433 ds_put_char(result, '\n');
4437 trace_format_flow(struct ds *result, int level, const char *title,
4438 struct ofproto_trace *trace)
4440 ds_put_char_multiple(result, '\t', level);
4441 ds_put_format(result, "%s: ", title);
4442 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
4443 ds_put_cstr(result, "unchanged");
4445 flow_format(result, &trace->ctx.flow);
4446 trace->flow = trace->ctx.flow;
4448 ds_put_char(result, '\n');
4452 trace_resubmit(struct action_xlate_ctx *ctx, struct rule *rule)
4454 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
4455 struct ds *result = trace->result;
4457 ds_put_char(result, '\n');
4458 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
4459 trace_format_rule(result, ctx->recurse + 1, rule);
4463 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
4464 void *aux OVS_UNUSED)
4466 char *dpname, *in_port_s, *tun_id_s, *packet_s;
4467 char *args = xstrdup(args_);
4468 char *save_ptr = NULL;
4469 struct ofproto *ofproto;
4470 struct ofpbuf packet;
4478 ofpbuf_init(&packet, strlen(args) / 2);
4481 dpname = strtok_r(args, " ", &save_ptr);
4482 tun_id_s = strtok_r(NULL, " ", &save_ptr);
4483 in_port_s = strtok_r(NULL, " ", &save_ptr);
4484 packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
4485 if (!dpname || !in_port_s || !packet_s) {
4486 unixctl_command_reply(conn, 501, "Bad command syntax");
4490 ofproto = shash_find_data(&all_ofprotos, dpname);
4492 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
4497 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
4498 in_port = ofp_port_to_odp_port(atoi(in_port_s));
4500 packet_s = ofpbuf_put_hex(&packet, packet_s, NULL);
4501 packet_s += strspn(packet_s, " ");
4502 if (*packet_s != '\0') {
4503 unixctl_command_reply(conn, 501, "Trailing garbage in command");
4506 if (packet.size < ETH_HEADER_LEN) {
4507 unixctl_command_reply(conn, 501, "Packet data too short for Ethernet");
4511 ds_put_cstr(&result, "Packet: ");
4512 s = ofp_packet_to_string(packet.data, packet.size, packet.size);
4513 ds_put_cstr(&result, s);
4516 flow_extract(&packet, tun_id, in_port, &flow);
4517 ds_put_cstr(&result, "Flow: ");
4518 flow_format(&result, &flow);
4519 ds_put_char(&result, '\n');
4521 rule = rule_lookup(ofproto, &flow);
4522 trace_format_rule(&result, 0, rule);
4524 struct ofproto_trace trace;
4525 struct ofpbuf *odp_actions;
4527 trace.result = &result;
4529 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet);
4530 trace.ctx.resubmit_hook = trace_resubmit;
4531 odp_actions = xlate_actions(&trace.ctx,
4532 rule->actions, rule->n_actions);
4534 ds_put_char(&result, '\n');
4535 trace_format_flow(&result, 0, "Final flow", &trace);
4536 ds_put_cstr(&result, "Datapath actions: ");
4537 format_odp_actions(&result, odp_actions->data, odp_actions->size);
4538 ofpbuf_delete(odp_actions);
4541 unixctl_command_reply(conn, 200, ds_cstr(&result));
4544 ds_destroy(&result);
4545 ofpbuf_uninit(&packet);
4550 ofproto_unixctl_init(void)
4552 static bool registered;
4558 unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL);
4559 unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL);
4563 default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
4564 struct ofpbuf *odp_actions, tag_type *tags,
4565 uint16_t *nf_output_iface, void *ofproto_)
4567 struct ofproto *ofproto = ofproto_;
4568 struct mac_entry *dst_mac;
4570 /* Drop frames for reserved multicast addresses. */
4571 if (eth_addr_is_reserved(flow->dl_dst)) {
4575 /* Learn source MAC (but don't try to learn from revalidation). */
4577 && mac_learning_may_learn(ofproto->ml, flow->dl_src, 0)) {
4578 struct mac_entry *src_mac;
4580 src_mac = mac_learning_insert(ofproto->ml, flow->dl_src, 0);
4581 if (mac_entry_is_new(src_mac) || src_mac->port.i != flow->in_port) {
4582 /* The log messages here could actually be useful in debugging,
4583 * so keep the rate limit relatively high. */
4584 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4585 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
4586 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
4588 ofproto_revalidate(ofproto,
4589 mac_learning_changed(ofproto->ml, src_mac));
4590 src_mac->port.i = flow->in_port;
4594 /* Determine output port. */
4595 dst_mac = mac_learning_lookup(ofproto->ml, flow->dl_dst, 0, tags);
4597 flood_packets(ofproto, flow->in_port, htonl(OFPPC_NO_FLOOD),
4598 nf_output_iface, odp_actions);
4600 int out_port = dst_mac->port.i;
4601 if (out_port != flow->in_port) {
4602 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, out_port);
4603 *nf_output_iface = out_port;
4612 static const struct ofhooks default_ofhooks = {
4613 default_normal_ofhook_cb,