2 * Copyright (c) 2009, 2010, 2011 Nicira Networks.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
22 #include <sys/socket.h>
24 #include <netinet/in.h>
28 #include "byte-order.h"
30 #include "classifier.h"
34 #include "dynamic-string.h"
35 #include "fail-open.h"
39 #include "mac-learning.h"
40 #include "multipath.h"
46 #include "ofp-print.h"
48 #include "ofproto-sflow.h"
50 #include "openflow/nicira-ext.h"
51 #include "openflow/openflow.h"
52 #include "openvswitch/datapath-protocol.h"
56 #include "poll-loop.h"
61 #include "stream-ssl.h"
65 #include "unaligned.h"
70 VLOG_DEFINE_THIS_MODULE(ofproto);
72 COVERAGE_DEFINE(facet_changed_rule);
73 COVERAGE_DEFINE(facet_revalidate);
74 COVERAGE_DEFINE(odp_overflow);
75 COVERAGE_DEFINE(ofproto_agg_request);
76 COVERAGE_DEFINE(ofproto_costly_flags);
77 COVERAGE_DEFINE(ofproto_ctlr_action);
78 COVERAGE_DEFINE(ofproto_del_rule);
79 COVERAGE_DEFINE(ofproto_error);
80 COVERAGE_DEFINE(ofproto_expiration);
81 COVERAGE_DEFINE(ofproto_expired);
82 COVERAGE_DEFINE(ofproto_flows_req);
83 COVERAGE_DEFINE(ofproto_flush);
84 COVERAGE_DEFINE(ofproto_invalidated);
85 COVERAGE_DEFINE(ofproto_no_packet_in);
86 COVERAGE_DEFINE(ofproto_ofp2odp);
87 COVERAGE_DEFINE(ofproto_packet_in);
88 COVERAGE_DEFINE(ofproto_packet_out);
89 COVERAGE_DEFINE(ofproto_queue_req);
90 COVERAGE_DEFINE(ofproto_recv_openflow);
91 COVERAGE_DEFINE(ofproto_reinit_ports);
92 COVERAGE_DEFINE(ofproto_unexpected_rule);
93 COVERAGE_DEFINE(ofproto_uninstallable);
94 COVERAGE_DEFINE(ofproto_update_port);
96 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
97 * flow translation. */
98 #define MAX_RESUBMIT_RECURSION 16
103 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
104 struct netdev *netdev;
105 struct ofp_phy_port opp;
107 struct cfm *cfm; /* Connectivity Fault Management, if any. */
110 static void ofport_free(struct ofport *);
111 static void ofport_run(struct ofproto *, struct ofport *);
112 static void ofport_wait(struct ofport *);
114 struct action_xlate_ctx {
115 /* action_xlate_ctx_init() initializes these members. */
118 struct ofproto *ofproto;
120 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
121 * this flow when actions change header fields. */
124 /* The packet corresponding to 'flow', or a null pointer if we are
125 * revalidating without a packet to refer to. */
126 const struct ofpbuf *packet;
128 /* If nonnull, called just before executing a resubmit action.
130 * This is normally null so the client has to set it manually after
131 * calling action_xlate_ctx_init(). */
132 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule *);
134 /* If true, the speciality of 'flow' should be checked before executing
135 * its actions. If special_cb returns false on 'flow' rendered
136 * uninstallable and no actions will be executed. */
139 /* xlate_actions() initializes and uses these members. The client might want
140 * to look at them after it returns. */
142 struct ofpbuf *odp_actions; /* Datapath actions. */
143 tag_type tags; /* Tags associated with OFPP_NORMAL actions. */
144 bool may_set_up_flow; /* True ordinarily; false if the actions must
145 * be reassessed for every packet. */
146 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
148 /* xlate_actions() initializes and uses these members, but the client has no
149 * reason to look at them. */
151 int recurse; /* Recursion level, via xlate_table_action. */
152 int last_pop_priority; /* Offset in 'odp_actions' just past most
153 * recent ODP_ACTION_ATTR_SET_PRIORITY. */
156 static void action_xlate_ctx_init(struct action_xlate_ctx *,
157 struct ofproto *, const struct flow *,
158 const struct ofpbuf *);
159 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
160 const union ofp_action *in, size_t n_in);
162 /* An OpenFlow flow. */
164 long long int used; /* Time last used; time created if not used. */
165 long long int created; /* Creation time. */
169 * - Do include packets and bytes from facets that have been deleted or
170 * whose own statistics have been folded into the rule.
172 * - Do include packets and bytes sent "by hand" that were accounted to
173 * the rule without any facet being involved (this is a rare corner
174 * case in rule_execute()).
176 * - Do not include packet or bytes that can be obtained from any facet's
177 * packet_count or byte_count member or that can be obtained from the
178 * datapath by, e.g., dpif_flow_get() for any facet.
180 uint64_t packet_count; /* Number of packets received. */
181 uint64_t byte_count; /* Number of bytes received. */
183 ovs_be64 flow_cookie; /* Controller-issued identifier. */
185 struct cls_rule cr; /* In owning ofproto's classifier. */
186 uint16_t idle_timeout; /* In seconds from time of last use. */
187 uint16_t hard_timeout; /* In seconds from time of creation. */
188 bool send_flow_removed; /* Send a flow removed message? */
189 int n_actions; /* Number of elements in actions[]. */
190 union ofp_action *actions; /* OpenFlow actions. */
191 struct list facets; /* List of "struct facet"s. */
194 static struct rule *rule_from_cls_rule(const struct cls_rule *);
195 static bool rule_is_hidden(const struct rule *);
197 static struct rule *rule_create(const struct cls_rule *,
198 const union ofp_action *, size_t n_actions,
199 uint16_t idle_timeout, uint16_t hard_timeout,
200 ovs_be64 flow_cookie, bool send_flow_removed);
201 static void rule_destroy(struct ofproto *, struct rule *);
202 static void rule_free(struct rule *);
204 static struct rule *rule_lookup(struct ofproto *, const struct flow *);
205 static void rule_insert(struct ofproto *, struct rule *);
206 static void rule_remove(struct ofproto *, struct rule *);
208 static void rule_send_removed(struct ofproto *, struct rule *, uint8_t reason);
209 static void rule_get_stats(const struct rule *, uint64_t *packets,
212 /* An exact-match instantiation of an OpenFlow flow. */
214 long long int used; /* Time last used; time created if not used. */
218 * - Do include packets and bytes sent "by hand", e.g. with
221 * - Do include packets and bytes that were obtained from the datapath
222 * when a flow was deleted (e.g. dpif_flow_del()) or when its
223 * statistics were reset (e.g. dpif_flow_put() with
224 * DPIF_FP_ZERO_STATS).
226 * - Do not include any packets or bytes that can currently be obtained
227 * from the datapath by, e.g., dpif_flow_get().
229 uint64_t packet_count; /* Number of packets received. */
230 uint64_t byte_count; /* Number of bytes received. */
232 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
233 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
235 uint64_t rs_packet_count; /* Packets pushed to resubmit children. */
236 uint64_t rs_byte_count; /* Bytes pushed to resubmit children. */
237 long long int rs_used; /* Used time pushed to resubmit children. */
239 /* Number of bytes passed to account_cb. This may include bytes that can
240 * currently obtained from the datapath (thus, it can be greater than
242 uint64_t accounted_bytes;
244 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
245 struct list list_node; /* In owning rule's 'facets' list. */
246 struct rule *rule; /* Owning rule. */
247 struct flow flow; /* Exact-match flow. */
248 bool installed; /* Installed in datapath? */
249 bool may_install; /* True ordinarily; false if actions must
250 * be reassessed for every packet. */
251 size_t actions_len; /* Number of bytes in actions[]. */
252 struct nlattr *actions; /* Datapath actions. */
253 tag_type tags; /* Tags (set only by hooks). */
254 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
257 static struct facet *facet_create(struct ofproto *, struct rule *,
259 const struct ofpbuf *packet);
260 static void facet_remove(struct ofproto *, struct facet *);
261 static void facet_free(struct facet *);
263 static struct facet *facet_lookup_valid(struct ofproto *, const struct flow *);
264 static bool facet_revalidate(struct ofproto *, struct facet *);
266 static void facet_install(struct ofproto *, struct facet *, bool zero_stats);
267 static void facet_uninstall(struct ofproto *, struct facet *);
268 static void facet_flush_stats(struct ofproto *, struct facet *);
270 static void facet_make_actions(struct ofproto *, struct facet *,
271 const struct ofpbuf *packet);
272 static void facet_update_stats(struct ofproto *, struct facet *,
273 const struct dpif_flow_stats *);
274 static void facet_push_stats(struct ofproto *, struct facet *);
276 static void send_packet_in(struct ofproto *, struct dpif_upcall *,
277 const struct flow *, bool clone);
280 char *name; /* Datapath name. */
281 struct hmap_node hmap_node; /* In global 'all_ofprotos' hmap. */
284 uint64_t datapath_id; /* Datapath ID. */
285 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
286 char *mfr_desc; /* Manufacturer. */
287 char *hw_desc; /* Hardware. */
288 char *sw_desc; /* Software version. */
289 char *serial_desc; /* Serial number. */
290 char *dp_desc; /* Datapath description. */
294 struct netdev_monitor *netdev_monitor;
295 struct hmap ports; /* Contains "struct ofport"s. */
296 struct shash port_by_name;
300 struct netflow *netflow;
301 struct ofproto_sflow *sflow;
304 struct classifier cls;
305 struct timer next_expiration;
309 bool need_revalidate;
310 struct tag_set revalidate_set;
312 /* OpenFlow connections. */
313 struct connmgr *connmgr;
315 /* Hooks for ovs-vswitchd. */
316 const struct ofhooks *ofhooks;
319 /* Used by default ofhooks. */
320 struct mac_learning *ml;
323 /* Map from dpif name to struct ofproto, for use by unixctl commands. */
324 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
326 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
328 static const struct ofhooks default_ofhooks;
330 static uint64_t pick_datapath_id(const struct ofproto *);
331 static uint64_t pick_fallback_dpid(void);
333 static void ofproto_flush_flows__(struct ofproto *);
334 static int ofproto_expire(struct ofproto *);
335 static void flow_push_stats(struct ofproto *, const struct rule *,
336 struct flow *, uint64_t packets, uint64_t bytes,
339 static void handle_upcall(struct ofproto *, struct dpif_upcall *);
341 static void handle_openflow(struct ofconn *, struct ofpbuf *);
343 static struct ofport *get_port(const struct ofproto *, uint16_t odp_port);
344 static void update_port(struct ofproto *, const char *devname);
345 static int init_ports(struct ofproto *);
346 static void reinit_ports(struct ofproto *);
348 static void ofproto_unixctl_init(void);
351 ofproto_create(const char *datapath, const char *datapath_type,
352 const struct ofhooks *ofhooks, void *aux,
353 struct ofproto **ofprotop)
355 char local_name[IF_NAMESIZE];
362 ofproto_unixctl_init();
364 /* Connect to datapath and start listening for messages. */
365 error = dpif_create_and_open(datapath, datapath_type, &dpif);
367 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
370 error = dpif_recv_set_mask(dpif,
371 ((1u << DPIF_UC_MISS) |
372 (1u << DPIF_UC_ACTION) |
373 (1u << DPIF_UC_SAMPLE)));
375 VLOG_ERR("failed to listen on datapath %s: %s",
376 datapath, strerror(error));
380 dpif_flow_flush(dpif);
381 dpif_recv_purge(dpif);
383 error = dpif_port_get_name(dpif, ODPP_LOCAL,
384 local_name, sizeof local_name);
386 VLOG_ERR("%s: cannot get name of datapath local port (%s)",
387 datapath, strerror(error));
391 /* Initialize settings. */
392 p = xzalloc(sizeof *p);
393 p->name = xstrdup(dpif_name(dpif));
394 hmap_insert(&all_ofprotos, &p->hmap_node, hash_string(p->name, 0));
395 p->fallback_dpid = pick_fallback_dpid();
396 p->datapath_id = p->fallback_dpid;
397 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
398 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
399 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
400 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
401 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
403 /* Initialize datapath. */
405 p->netdev_monitor = netdev_monitor_create();
406 hmap_init(&p->ports);
407 shash_init(&p->port_by_name);
408 p->max_ports = dpif_get_max_ports(dpif);
410 /* Initialize submodules. */
414 /* Initialize flow table. */
415 classifier_init(&p->cls);
416 timer_set_duration(&p->next_expiration, 1000);
418 /* Initialize facet table. */
419 hmap_init(&p->facets);
420 p->need_revalidate = false;
421 tag_set_init(&p->revalidate_set);
423 /* Initialize hooks. */
425 p->ofhooks = ofhooks;
429 p->ofhooks = &default_ofhooks;
431 p->ml = mac_learning_create();
434 /* Pick final datapath ID. */
435 p->datapath_id = pick_datapath_id(p);
436 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
438 /* Initialize OpenFlow connections. */
439 p->connmgr = connmgr_create(p, datapath, local_name);
446 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
448 uint64_t old_dpid = p->datapath_id;
449 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
450 if (p->datapath_id != old_dpid) {
451 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
453 /* Force all active connections to reconnect, since there is no way to
454 * notify a controller that the datapath ID has changed. */
455 ofproto_reconnect_controllers(p);
460 ofproto_set_controllers(struct ofproto *p,
461 const struct ofproto_controller *controllers,
462 size_t n_controllers)
464 connmgr_set_controllers(p->connmgr, controllers, n_controllers);
468 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
470 connmgr_set_fail_mode(p->connmgr, fail_mode);
473 /* Drops the connections between 'ofproto' and all of its controllers, forcing
474 * them to reconnect. */
476 ofproto_reconnect_controllers(struct ofproto *ofproto)
478 connmgr_reconnect(ofproto->connmgr);
481 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
482 * in-band control should guarantee access, in the same way that in-band
483 * control guarantees access to OpenFlow controllers. */
485 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
486 const struct sockaddr_in *extras, size_t n)
488 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
491 /* Sets the OpenFlow queue used by flows set up by in-band control on
492 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
493 * flows will use the default queue. */
495 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
497 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
501 ofproto_set_desc(struct ofproto *p,
502 const char *mfr_desc, const char *hw_desc,
503 const char *sw_desc, const char *serial_desc,
506 struct ofp_desc_stats *ods;
509 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
510 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
511 sizeof ods->mfr_desc);
514 p->mfr_desc = xstrdup(mfr_desc);
517 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
518 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
519 sizeof ods->hw_desc);
522 p->hw_desc = xstrdup(hw_desc);
525 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
526 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
527 sizeof ods->sw_desc);
530 p->sw_desc = xstrdup(sw_desc);
533 if (strlen(serial_desc) >= sizeof ods->serial_num) {
534 VLOG_WARN("truncating serial_desc, must be less than %zu "
536 sizeof ods->serial_num);
538 free(p->serial_desc);
539 p->serial_desc = xstrdup(serial_desc);
542 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
543 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
544 sizeof ods->dp_desc);
547 p->dp_desc = xstrdup(dp_desc);
552 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
554 return connmgr_set_snoops(ofproto->connmgr, snoops);
558 ofproto_set_netflow(struct ofproto *ofproto,
559 const struct netflow_options *nf_options)
561 if (nf_options && !sset_is_empty(&nf_options->collectors)) {
562 if (!ofproto->netflow) {
563 ofproto->netflow = netflow_create();
565 return netflow_set_options(ofproto->netflow, nf_options);
567 netflow_destroy(ofproto->netflow);
568 ofproto->netflow = NULL;
574 ofproto_set_sflow(struct ofproto *ofproto,
575 const struct ofproto_sflow_options *oso)
577 struct ofproto_sflow *os = ofproto->sflow;
580 struct ofport *ofport;
582 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
583 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
584 ofproto_sflow_add_port(os, ofport->odp_port,
585 netdev_get_name(ofport->netdev));
588 ofproto_sflow_set_options(os, oso);
590 ofproto_sflow_destroy(os);
591 ofproto->sflow = NULL;
595 /* Connectivity Fault Management configuration. */
597 /* Clears the CFM configuration from 'port_no' on 'ofproto'. */
599 ofproto_port_clear_cfm(struct ofproto *ofproto, uint32_t port_no)
601 struct ofport *ofport = get_port(ofproto, port_no);
602 if (ofport && ofport->cfm){
603 cfm_destroy(ofport->cfm);
608 /* Configures connectivity fault management on 'port_no' in 'ofproto'. Takes
609 * basic configuration from the configuration members in 'cfm', and the set of
610 * remote maintenance points from the 'n_remote_mps' elements in 'remote_mps'.
611 * Ignores the statistics members of 'cfm'.
613 * This function has no effect if 'ofproto' does not have a port 'port_no'. */
615 ofproto_port_set_cfm(struct ofproto *ofproto, uint32_t port_no,
616 const struct cfm *cfm,
617 const uint16_t *remote_mps, size_t n_remote_mps)
619 struct ofport *ofport;
621 ofport = get_port(ofproto, port_no);
623 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu32,
624 ofproto->name, port_no);
629 ofport->cfm = cfm_create();
632 ofport->cfm->mpid = cfm->mpid;
633 ofport->cfm->interval = cfm->interval;
634 memcpy(ofport->cfm->maid, cfm->maid, CCM_MAID_LEN);
636 cfm_update_remote_mps(ofport->cfm, remote_mps, n_remote_mps);
638 if (!cfm_configure(ofport->cfm)) {
639 VLOG_WARN("%s: CFM configuration on port %"PRIu32" (%s) failed",
640 ofproto->name, port_no,
641 netdev_get_name(ofport->netdev));
642 cfm_destroy(ofport->cfm);
647 /* Returns the connectivity fault management object associated with 'port_no'
648 * within 'ofproto', or a null pointer if 'ofproto' does not have a port
649 * 'port_no' or if that port does not have CFM configured. The caller must not
650 * modify or destroy the returned object. */
652 ofproto_port_get_cfm(struct ofproto *ofproto, uint32_t port_no)
654 struct ofport *ofport = get_port(ofproto, port_no);
655 return ofport ? ofport->cfm : NULL;
659 ofproto_get_datapath_id(const struct ofproto *ofproto)
661 return ofproto->datapath_id;
664 enum ofproto_fail_mode
665 ofproto_get_fail_mode(const struct ofproto *p)
667 return connmgr_get_fail_mode(p->connmgr);
671 ofproto_has_snoops(const struct ofproto *ofproto)
673 return connmgr_has_snoops(ofproto->connmgr);
677 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
679 connmgr_get_snoops(ofproto->connmgr, snoops);
683 ofproto_destroy__(struct ofproto *p, bool delete)
685 struct ofport *ofport, *next_ofport;
691 hmap_remove(&all_ofprotos, &p->hmap_node);
693 ofproto_flush_flows__(p);
694 connmgr_destroy(p->connmgr);
695 classifier_destroy(&p->cls);
696 hmap_destroy(&p->facets);
699 int error = dpif_delete(p->dpif);
700 if (error && error != ENOENT) {
701 VLOG_ERR("bridge %s: failed to destroy (%s)",
702 p->name, strerror(error));
707 netdev_monitor_destroy(p->netdev_monitor);
708 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
709 hmap_remove(&p->ports, &ofport->hmap_node);
712 shash_destroy(&p->port_by_name);
714 netflow_destroy(p->netflow);
715 ofproto_sflow_destroy(p->sflow);
717 mac_learning_destroy(p->ml);
722 free(p->serial_desc);
725 hmap_destroy(&p->ports);
732 ofproto_destroy(struct ofproto *p)
734 ofproto_destroy__(p, false);
738 ofproto_destroy_and_delete(struct ofproto *p)
740 ofproto_destroy__(p, true);
744 ofproto_run(struct ofproto *p)
746 int error = ofproto_run1(p);
748 error = ofproto_run2(p, false);
754 process_port_change(struct ofproto *ofproto, int error, char *devname)
756 if (error == ENOBUFS) {
757 reinit_ports(ofproto);
759 update_port(ofproto, devname);
765 ofproto_run1(struct ofproto *p)
767 struct ofport *ofport;
772 if (shash_is_empty(&p->port_by_name)) {
776 for (i = 0; i < 50; i++) {
777 struct dpif_upcall packet;
779 error = dpif_recv(p->dpif, &packet);
781 if (error == ENODEV) {
782 /* Someone destroyed the datapath behind our back. The caller
783 * better destroy us and give up, because we're just going to
784 * spin from here on out. */
785 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
786 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
793 handle_upcall(p, &packet);
796 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
797 process_port_change(p, error, devname);
799 while ((error = netdev_monitor_poll(p->netdev_monitor,
800 &devname)) != EAGAIN) {
801 process_port_change(p, error, devname);
804 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
805 ofport_run(p, ofport);
808 connmgr_run(p->connmgr, handle_openflow);
810 if (timer_expired(&p->next_expiration)) {
811 int delay = ofproto_expire(p);
812 timer_set_duration(&p->next_expiration, delay);
813 COVERAGE_INC(ofproto_expiration);
817 netflow_run(p->netflow);
820 ofproto_sflow_run(p->sflow);
827 ofproto_run2(struct ofproto *p, bool revalidate_all)
829 /* Figure out what we need to revalidate now, if anything. */
830 struct tag_set revalidate_set = p->revalidate_set;
831 if (p->need_revalidate) {
832 revalidate_all = true;
835 /* Clear the revalidation flags. */
836 tag_set_init(&p->revalidate_set);
837 p->need_revalidate = false;
839 /* Now revalidate if there's anything to do. */
840 if (revalidate_all || !tag_set_is_empty(&revalidate_set)) {
841 struct facet *facet, *next;
843 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &p->facets) {
845 || tag_set_intersects(&revalidate_set, facet->tags)) {
846 facet_revalidate(p, facet);
855 ofproto_wait(struct ofproto *p)
857 struct ofport *ofport;
859 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
862 dpif_recv_wait(p->dpif);
863 dpif_port_poll_wait(p->dpif);
864 netdev_monitor_poll_wait(p->netdev_monitor);
866 ofproto_sflow_wait(p->sflow);
868 if (!tag_set_is_empty(&p->revalidate_set)) {
869 poll_immediate_wake();
871 if (p->need_revalidate) {
872 /* Shouldn't happen, but if it does just go around again. */
873 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
874 poll_immediate_wake();
876 timer_wait(&p->next_expiration);
878 connmgr_wait(p->connmgr);
882 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
884 tag_set_add(&ofproto->revalidate_set, tag);
888 ofproto_get_revalidate_set(struct ofproto *ofproto)
890 return &ofproto->revalidate_set;
894 ofproto_is_alive(const struct ofproto *p)
896 return connmgr_has_controllers(p->connmgr);
900 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
903 connmgr_get_controller_info(ofproto->connmgr, info);
907 ofproto_free_ofproto_controller_info(struct shash *info)
909 struct shash_node *node;
911 SHASH_FOR_EACH (node, info) {
912 struct ofproto_controller_info *cinfo = node->data;
913 while (cinfo->pairs.n) {
914 free((char *) cinfo->pairs.values[--cinfo->pairs.n]);
921 /* Makes a deep copy of 'old' into 'port'. */
923 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
925 port->name = xstrdup(old->name);
926 port->type = xstrdup(old->type);
927 port->ofp_port = old->ofp_port;
930 /* Frees memory allocated to members of 'ofproto_port'.
932 * Do not call this function on an ofproto_port obtained from
933 * ofproto_port_dump_next(): that function retains ownership of the data in the
936 ofproto_port_destroy(struct ofproto_port *ofproto_port)
938 free(ofproto_port->name);
939 free(ofproto_port->type);
942 /* Converts a dpif_port into an ofproto_port.
944 * This only makes a shallow copy, so make sure that the dpif_port doesn't get
945 * freed while the ofproto_port is still in use. You can choose to free the
946 * ofproto_port instead of the dpif_port. */
948 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
949 struct dpif_port *dpif_port)
951 ofproto_port->name = dpif_port->name;
952 ofproto_port->type = dpif_port->type;
953 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
956 /* Initializes 'dump' to begin dumping the ports in an ofproto.
958 * This function provides no status indication. An error status for the entire
959 * dump operation is provided when it is completed by calling
960 * ofproto_port_dump_done().
963 ofproto_port_dump_start(struct ofproto_port_dump *dump,
964 const struct ofproto *ofproto)
966 struct dpif_port_dump *dpif_dump;
968 dump->state = dpif_dump = xmalloc(sizeof *dpif_dump);
969 dpif_port_dump_start(dpif_dump, ofproto->dpif);
972 /* Attempts to retrieve another port from 'dump', which must have been created
973 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
974 * 'port' and returns true. On failure, returns false.
976 * Failure might indicate an actual error or merely that the last port has been
977 * dumped. An error status for the entire dump operation is provided when it
978 * is completed by calling ofproto_port_dump_done().
980 * The ofproto owns the data stored in 'port'. It will remain valid until at
981 * least the next time 'dump' is passed to ofproto_port_dump_next() or
982 * ofproto_port_dump_done(). */
984 ofproto_port_dump_next(struct ofproto_port_dump *dump,
985 struct ofproto_port *port)
987 struct dpif_port_dump *dpif_dump = dump->state;
988 struct dpif_port dpif_port;
991 ok = dpif_port_dump_next(dpif_dump, &dpif_port);
993 ofproto_port_from_dpif_port(port, &dpif_port);
998 /* Completes port table dump operation 'dump', which must have been created
999 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1000 * error-free, otherwise a positive errno value describing the problem. */
1002 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1004 struct dpif_port_dump *dpif_dump = dump->state;
1005 int error = dpif_port_dump_done(dpif_dump);
1010 /* Attempts to add 'netdev' as a port on 'ofproto'. If successful, returns 0
1011 * and sets '*ofp_portp' to the new port's port OpenFlow number (if 'ofp_portp'
1012 * is non-null). On failure, returns a positive errno value and sets
1013 * '*ofp_portp' to OFPP_NONE (if 'ofp_portp' is non-null). */
1015 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1016 uint16_t *ofp_portp)
1021 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
1023 *ofp_portp = error ? OFPP_NONE : odp_port_to_ofp_port(odp_port);
1028 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1029 * initializes '*port' appropriately; on failure, returns a positive errno
1032 * The caller owns the data in 'port' and must free it with
1033 * ofproto_port_destroy() when it is no longer needed. */
1035 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1036 struct ofproto_port *port)
1038 struct dpif_port dpif_port;
1041 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
1043 ofproto_port_from_dpif_port(port, &dpif_port);
1048 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1050 * This is almost the same as calling dpif_port_del() directly on the
1051 * datapath, but it also makes 'ofproto' close its open netdev for the port
1052 * (if any). This makes it possible to create a new netdev of a different
1053 * type under the same name, which otherwise the netdev library would refuse
1054 * to do because of the conflict. (The netdev would eventually get closed on
1055 * the next trip through ofproto_run(), but this interface is more direct.)
1057 * Returns 0 if successful, otherwise a positive errno. */
1059 ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port)
1061 uint32_t odp_port = ofp_port_to_odp_port(ofp_port);
1062 struct ofport *ofport = get_port(ofproto, odp_port);
1063 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1066 error = dpif_port_del(ofproto->dpif, odp_port);
1068 VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)",
1069 ofproto->name, odp_port, name, strerror(error));
1070 } else if (ofport) {
1071 /* 'name' is the netdev's name and update_port() is going to close the
1072 * netdev. Just in case update_port() refers to 'name' after it
1073 * destroys 'ofport', make a copy of it around the update_port()
1075 char *devname = xstrdup(name);
1076 update_port(ofproto, devname);
1082 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
1083 * true if 'odp_port' exists and should be included, false otherwise. */
1085 ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port)
1087 struct ofport *ofport = get_port(ofproto, odp_port);
1088 return ofport && !(ofport->opp.config & htonl(OFPPC_NO_FLOOD));
1091 /* Sends 'packet' out of port 'port_no' within 'p'. If 'vlan_tci' is zero the
1092 * packet will not have any 802.1Q hader; if it is nonzero, then the packet
1093 * will be sent with the VLAN TCI specified by 'vlan_tci & ~VLAN_CFI'.
1095 * Returns 0 if successful, otherwise a positive errno value. */
1097 ofproto_send_packet(struct ofproto *ofproto,
1098 uint32_t port_no, uint16_t vlan_tci,
1099 const struct ofpbuf *packet)
1101 struct ofpbuf odp_actions;
1104 ofpbuf_init(&odp_actions, 32);
1105 if (vlan_tci != 0) {
1106 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
1107 ntohs(vlan_tci & ~VLAN_CFI));
1109 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, port_no);
1110 error = dpif_execute(ofproto->dpif, odp_actions.data, odp_actions.size,
1112 ofpbuf_uninit(&odp_actions);
1115 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
1116 ofproto->name, port_no, strerror(error));
1121 /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and
1122 * performs the 'n_actions' actions in 'actions'. The new flow will not
1125 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1126 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1127 * controllers; otherwise, it will be hidden.
1129 * The caller retains ownership of 'cls_rule' and 'actions'. */
1131 ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule,
1132 const union ofp_action *actions, size_t n_actions)
1135 rule = rule_create(cls_rule, actions, n_actions, 0, 0, 0, false);
1136 rule_insert(p, rule);
1140 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1144 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1147 rule_remove(ofproto, rule);
1152 ofproto_flush_flows__(struct ofproto *ofproto)
1154 struct facet *facet, *next_facet;
1155 struct rule *rule, *next_rule;
1156 struct cls_cursor cursor;
1158 COVERAGE_INC(ofproto_flush);
1160 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1161 /* Mark the facet as not installed so that facet_remove() doesn't
1162 * bother trying to uninstall it. There is no point in uninstalling it
1163 * individually since we are about to blow away all the facets with
1164 * dpif_flow_flush(). */
1165 facet->installed = false;
1166 facet->dp_packet_count = 0;
1167 facet->dp_byte_count = 0;
1168 facet_remove(ofproto, facet);
1171 cls_cursor_init(&cursor, &ofproto->cls, NULL);
1172 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1173 rule_remove(ofproto, rule);
1176 dpif_flow_flush(ofproto->dpif);
1180 ofproto_flush_flows(struct ofproto *ofproto)
1182 ofproto_flush_flows__(ofproto);
1183 connmgr_flushed(ofproto->connmgr);
1187 reinit_ports(struct ofproto *p)
1189 struct dpif_port_dump dump;
1190 struct sset devnames;
1191 struct ofport *ofport;
1192 struct dpif_port dpif_port;
1193 const char *devname;
1195 COVERAGE_INC(ofproto_reinit_ports);
1197 sset_init(&devnames);
1198 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1199 sset_add(&devnames, netdev_get_name(ofport->netdev));
1201 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1202 sset_add(&devnames, dpif_port.name);
1205 SSET_FOR_EACH (devname, &devnames) {
1206 update_port(p, devname);
1208 sset_destroy(&devnames);
1211 /* Opens and returns a netdev for 'dpif_port', or a null pointer if the netdev
1212 * cannot be opened. On success, also fills in 'opp'. */
1213 static struct netdev *
1214 ofport_open(const struct dpif_port *dpif_port, struct ofp_phy_port *opp)
1216 uint32_t curr, advertised, supported, peer;
1217 struct netdev_options netdev_options;
1218 enum netdev_flags flags;
1219 struct netdev *netdev;
1222 memset(&netdev_options, 0, sizeof netdev_options);
1223 netdev_options.name = dpif_port->name;
1224 netdev_options.type = dpif_port->type;
1225 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1227 error = netdev_open(&netdev_options, &netdev);
1229 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1230 "cannot be opened (%s)",
1231 dpif_port->name, dpif_port->port_no,
1232 dpif_port->name, strerror(error));
1236 netdev_get_flags(netdev, &flags);
1237 netdev_get_features(netdev, &curr, &advertised, &supported, &peer);
1239 opp->port_no = htons(odp_port_to_ofp_port(dpif_port->port_no));
1240 netdev_get_etheraddr(netdev, opp->hw_addr);
1241 ovs_strzcpy(opp->name, dpif_port->name, sizeof opp->name);
1242 opp->config = flags & NETDEV_UP ? 0 : htonl(OFPPC_PORT_DOWN);
1243 opp->state = netdev_get_carrier(netdev) ? 0 : htonl(OFPPS_LINK_DOWN);
1244 opp->curr = htonl(curr);
1245 opp->advertised = htonl(advertised);
1246 opp->supported = htonl(supported);
1247 opp->peer = htonl(peer);
1253 ofport_conflicts(const struct ofproto *p, const struct dpif_port *dpif_port)
1255 if (get_port(p, dpif_port->port_no)) {
1256 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1257 dpif_port->port_no);
1259 } else if (shash_find(&p->port_by_name, dpif_port->name)) {
1260 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1268 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1269 * port number, and 'config' bits other than OFPPC_PORT_DOWN are
1272 ofport_equal(const struct ofp_phy_port *a, const struct ofp_phy_port *b)
1274 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1275 return (!memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1276 && a->state == b->state
1277 && !((a->config ^ b->config) & htonl(OFPPC_PORT_DOWN))
1278 && a->curr == b->curr
1279 && a->advertised == b->advertised
1280 && a->supported == b->supported
1281 && a->peer == b->peer);
1284 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1285 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1286 * one with the same name or port number). */
1288 ofport_install(struct ofproto *p,
1289 struct netdev *netdev, const struct ofp_phy_port *opp)
1291 const char *netdev_name = netdev_get_name(netdev);
1292 struct ofport *ofport;
1294 connmgr_send_port_status(p->connmgr, opp, OFPPR_ADD);
1296 /* Create ofport. */
1297 ofport = xmalloc(sizeof *ofport);
1298 ofport->netdev = netdev;
1300 ofport->odp_port = ofp_port_to_odp_port(ntohs(opp->port_no));
1303 /* Add port to 'p'. */
1304 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1305 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
1306 shash_add(&p->port_by_name, netdev_name, ofport);
1308 ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
1312 /* Removes 'ofport' from 'p' and destroys it. */
1314 ofport_remove(struct ofproto *p, struct ofport *ofport)
1316 connmgr_send_port_status(p->connmgr, &ofport->opp, OFPPR_DELETE);
1318 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1319 hmap_remove(&p->ports, &ofport->hmap_node);
1320 shash_delete(&p->port_by_name,
1321 shash_find(&p->port_by_name,
1322 netdev_get_name(ofport->netdev)));
1324 ofproto_sflow_del_port(p->sflow, ofport->odp_port);
1327 ofport_free(ofport);
1330 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1333 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1335 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1337 ofport_remove(ofproto, port);
1341 /* Updates 'port' within 'ofproto' with the new 'netdev' and 'opp'.
1343 * Does not handle a name or port number change. The caller must implement
1344 * such a change as a delete followed by an add. */
1346 ofport_modified(struct ofproto *ofproto, struct ofport *port,
1347 struct netdev *netdev, struct ofp_phy_port *opp)
1349 memcpy(port->opp.hw_addr, opp->hw_addr, ETH_ADDR_LEN);
1350 port->opp.config = ((port->opp.config & ~htonl(OFPPC_PORT_DOWN))
1351 | (opp->config & htonl(OFPPC_PORT_DOWN)));
1352 port->opp.state = opp->state;
1353 port->opp.curr = opp->curr;
1354 port->opp.advertised = opp->advertised;
1355 port->opp.supported = opp->supported;
1356 port->opp.peer = opp->peer;
1358 netdev_monitor_remove(ofproto->netdev_monitor, port->netdev);
1359 netdev_monitor_add(ofproto->netdev_monitor, netdev);
1361 netdev_close(port->netdev);
1362 port->netdev = netdev;
1364 connmgr_send_port_status(ofproto->connmgr, &port->opp, OFPPR_MODIFY);
1368 ofport_run(struct ofproto *ofproto, struct ofport *ofport)
1371 cfm_run(ofport->cfm);
1373 if (cfm_should_send_ccm(ofport->cfm)) {
1374 struct ofpbuf packet;
1377 ofpbuf_init(&packet, 0);
1378 ccm = eth_compose(&packet, eth_addr_ccm, ofport->opp.hw_addr,
1379 ETH_TYPE_CFM, sizeof *ccm);
1380 cfm_compose_ccm(ofport->cfm, ccm);
1381 ofproto_send_packet(ofproto, ofport->odp_port, 0, &packet);
1382 ofpbuf_uninit(&packet);
1388 ofport_wait(struct ofport *ofport)
1391 cfm_wait(ofport->cfm);
1396 ofport_free(struct ofport *ofport)
1399 cfm_destroy(ofport->cfm);
1400 netdev_close(ofport->netdev);
1405 static struct ofport *
1406 get_port(const struct ofproto *ofproto, uint16_t odp_port)
1408 struct ofport *port;
1410 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1411 hash_int(odp_port, 0), &ofproto->ports) {
1412 if (port->odp_port == odp_port) {
1420 update_port(struct ofproto *ofproto, const char *name)
1422 struct dpif_port dpif_port;
1423 struct ofp_phy_port opp;
1424 struct netdev *netdev;
1425 struct ofport *port;
1427 COVERAGE_INC(ofproto_update_port);
1429 /* Fetch 'name''s location and properties from the datapath. */
1430 netdev = (!dpif_port_query_by_name(ofproto->dpif, name, &dpif_port)
1431 ? ofport_open(&dpif_port, &opp)
1434 port = get_port(ofproto, dpif_port.port_no);
1435 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
1436 /* 'name' hasn't changed location. Any properties changed? */
1437 if (!ofport_equal(&port->opp, &opp)) {
1438 ofport_modified(ofproto, port, netdev, &opp);
1440 netdev_close(netdev);
1443 /* If 'port' is nonnull then its name differs from 'name' and thus
1444 * we should delete it. If we think there's a port named 'name'
1445 * then its port number must be wrong now so delete it too. */
1447 ofport_remove(ofproto, port);
1449 ofport_remove_with_name(ofproto, name);
1450 ofport_install(ofproto, netdev, &opp);
1453 /* Any port named 'name' is gone now. */
1454 ofport_remove_with_name(ofproto, name);
1456 dpif_port_destroy(&dpif_port);
1460 init_ports(struct ofproto *p)
1462 struct dpif_port_dump dump;
1463 struct dpif_port dpif_port;
1465 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1466 if (!ofport_conflicts(p, &dpif_port)) {
1467 struct ofp_phy_port opp;
1468 struct netdev *netdev;
1470 netdev = ofport_open(&dpif_port, &opp);
1472 ofport_install(p, netdev, &opp);
1480 /* Returns true if 'rule' should be hidden from the controller.
1482 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1483 * (e.g. by in-band control) and are intentionally hidden from the
1486 rule_is_hidden(const struct rule *rule)
1488 return rule->cr.priority > UINT16_MAX;
1491 /* Creates and returns a new rule initialized as specified.
1493 * The caller is responsible for inserting the rule into the classifier (with
1494 * rule_insert()). */
1495 static struct rule *
1496 rule_create(const struct cls_rule *cls_rule,
1497 const union ofp_action *actions, size_t n_actions,
1498 uint16_t idle_timeout, uint16_t hard_timeout,
1499 ovs_be64 flow_cookie, bool send_flow_removed)
1501 struct rule *rule = xzalloc(sizeof *rule);
1502 rule->cr = *cls_rule;
1503 rule->idle_timeout = idle_timeout;
1504 rule->hard_timeout = hard_timeout;
1505 rule->flow_cookie = flow_cookie;
1506 rule->used = rule->created = time_msec();
1507 rule->send_flow_removed = send_flow_removed;
1508 list_init(&rule->facets);
1509 if (n_actions > 0) {
1510 rule->n_actions = n_actions;
1511 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1517 static struct rule *
1518 rule_from_cls_rule(const struct cls_rule *cls_rule)
1520 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1524 rule_free(struct rule *rule)
1526 free(rule->actions);
1530 /* Destroys 'rule' and iterates through all of its facets and revalidates them,
1531 * destroying any that no longer has a rule (which is probably all of them).
1533 * The caller must have already removed 'rule' from the classifier. */
1535 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1537 struct facet *facet, *next_facet;
1538 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
1539 facet_revalidate(ofproto, facet);
1544 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
1545 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
1548 rule_has_out_port(const struct rule *rule, ovs_be16 out_port)
1550 const union ofp_action *oa;
1551 struct actions_iterator i;
1553 if (out_port == htons(OFPP_NONE)) {
1556 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1557 oa = actions_next(&i)) {
1558 if (action_outputs_to_port(oa, out_port)) {
1565 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
1566 * 'packet', which arrived on 'in_port'.
1568 * Takes ownership of 'packet'. */
1570 execute_odp_actions(struct ofproto *ofproto, const struct flow *flow,
1571 const struct nlattr *odp_actions, size_t actions_len,
1572 struct ofpbuf *packet)
1574 if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t))
1575 && odp_actions->nla_type == ODP_ACTION_ATTR_CONTROLLER) {
1576 /* As an optimization, avoid a round-trip from userspace to kernel to
1577 * userspace. This also avoids possibly filling up kernel packet
1578 * buffers along the way. */
1579 struct dpif_upcall upcall;
1581 upcall.type = DPIF_UC_ACTION;
1582 upcall.packet = packet;
1585 upcall.userdata = nl_attr_get_u64(odp_actions);
1586 upcall.sample_pool = 0;
1587 upcall.actions = NULL;
1588 upcall.actions_len = 0;
1590 send_packet_in(ofproto, &upcall, flow, false);
1596 error = dpif_execute(ofproto->dpif, odp_actions, actions_len, packet);
1597 ofpbuf_delete(packet);
1602 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
1603 * statistics appropriately. 'packet' must have at least sizeof(struct
1604 * ofp_packet_in) bytes of headroom.
1606 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
1607 * applying flow_extract() to 'packet' would yield the same flow as
1610 * 'facet' must have accurately composed ODP actions; that is, it must not be
1611 * in need of revalidation.
1613 * Takes ownership of 'packet'. */
1615 facet_execute(struct ofproto *ofproto, struct facet *facet,
1616 struct ofpbuf *packet)
1618 struct dpif_flow_stats stats;
1620 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1622 flow_extract_stats(&facet->flow, packet, &stats);
1623 stats.used = time_msec();
1624 if (execute_odp_actions(ofproto, &facet->flow,
1625 facet->actions, facet->actions_len, packet)) {
1626 facet_update_stats(ofproto, facet, &stats);
1630 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
1631 * statistics (or the statistics for one of its facets) appropriately.
1632 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom.
1634 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
1635 * with statistics for 'packet' either way.
1637 * Takes ownership of 'packet'. */
1639 rule_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port,
1640 struct ofpbuf *packet)
1642 struct action_xlate_ctx ctx;
1643 struct ofpbuf *odp_actions;
1644 struct facet *facet;
1648 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1650 flow_extract(packet, 0, in_port, &flow);
1652 /* First look for a related facet. If we find one, account it to that. */
1653 facet = facet_lookup_valid(ofproto, &flow);
1654 if (facet && facet->rule == rule) {
1655 facet_execute(ofproto, facet, packet);
1659 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
1660 * create a new facet for it and use that. */
1661 if (rule_lookup(ofproto, &flow) == rule) {
1662 facet = facet_create(ofproto, rule, &flow, packet);
1663 facet_execute(ofproto, facet, packet);
1664 facet_install(ofproto, facet, true);
1668 /* We can't account anything to a facet. If we were to try, then that
1669 * facet would have a non-matching rule, busting our invariants. */
1670 action_xlate_ctx_init(&ctx, ofproto, &flow, packet);
1671 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
1672 size = packet->size;
1673 if (execute_odp_actions(ofproto, &flow, odp_actions->data,
1674 odp_actions->size, packet)) {
1675 rule->used = time_msec();
1676 rule->packet_count++;
1677 rule->byte_count += size;
1678 flow_push_stats(ofproto, rule, &flow, 1, size, rule->used);
1680 ofpbuf_delete(odp_actions);
1683 /* Inserts 'rule' into 'p''s flow table. */
1685 rule_insert(struct ofproto *p, struct rule *rule)
1687 struct rule *displaced_rule;
1689 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1690 if (displaced_rule) {
1691 rule_destroy(p, displaced_rule);
1693 p->need_revalidate = true;
1696 /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a
1697 * 'flow' and an example 'packet' within that flow.
1699 * The caller must already have determined that no facet with an identical
1700 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
1701 * 'ofproto''s classifier table. */
1702 static struct facet *
1703 facet_create(struct ofproto *ofproto, struct rule *rule,
1704 const struct flow *flow, const struct ofpbuf *packet)
1706 struct facet *facet;
1708 facet = xzalloc(sizeof *facet);
1709 facet->used = time_msec();
1710 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
1711 list_push_back(&rule->facets, &facet->list_node);
1713 facet->flow = *flow;
1714 netflow_flow_init(&facet->nf_flow);
1715 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
1717 facet_make_actions(ofproto, facet, packet);
1723 facet_free(struct facet *facet)
1725 free(facet->actions);
1729 /* Remove 'rule' from 'ofproto' and free up the associated memory:
1731 * - Removes 'rule' from the classifier.
1733 * - If 'rule' has facets, revalidates them (and possibly uninstalls and
1734 * destroys them), via rule_destroy().
1737 rule_remove(struct ofproto *ofproto, struct rule *rule)
1739 COVERAGE_INC(ofproto_del_rule);
1740 ofproto->need_revalidate = true;
1741 classifier_remove(&ofproto->cls, &rule->cr);
1742 rule_destroy(ofproto, rule);
1745 /* Remove 'facet' from 'ofproto' and free up the associated memory:
1747 * - If 'facet' was installed in the datapath, uninstalls it and updates its
1748 * rule's statistics, via facet_uninstall().
1750 * - Removes 'facet' from its rule and from ofproto->facets.
1753 facet_remove(struct ofproto *ofproto, struct facet *facet)
1755 facet_uninstall(ofproto, facet);
1756 facet_flush_stats(ofproto, facet);
1757 hmap_remove(&ofproto->facets, &facet->hmap_node);
1758 list_remove(&facet->list_node);
1762 /* Composes the ODP actions for 'facet' based on its rule's actions. */
1764 facet_make_actions(struct ofproto *p, struct facet *facet,
1765 const struct ofpbuf *packet)
1767 const struct rule *rule = facet->rule;
1768 struct ofpbuf *odp_actions;
1769 struct action_xlate_ctx ctx;
1771 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
1772 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
1773 facet->tags = ctx.tags;
1774 facet->may_install = ctx.may_set_up_flow;
1775 facet->nf_flow.output_iface = ctx.nf_output_iface;
1777 if (facet->actions_len != odp_actions->size
1778 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
1779 free(facet->actions);
1780 facet->actions_len = odp_actions->size;
1781 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
1784 ofpbuf_delete(odp_actions);
1788 facet_put__(struct ofproto *ofproto, struct facet *facet,
1789 const struct nlattr *actions, size_t actions_len,
1790 struct dpif_flow_stats *stats)
1792 struct odputil_keybuf keybuf;
1793 enum dpif_flow_put_flags flags;
1796 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
1798 flags |= DPIF_FP_ZERO_STATS;
1799 facet->dp_packet_count = 0;
1800 facet->dp_byte_count = 0;
1803 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1804 odp_flow_key_from_flow(&key, &facet->flow);
1806 return dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
1807 actions, actions_len, stats);
1810 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
1811 * 'zero_stats' is true, clears any existing statistics from the datapath for
1814 facet_install(struct ofproto *p, struct facet *facet, bool zero_stats)
1816 struct dpif_flow_stats stats;
1818 if (facet->may_install
1819 && !facet_put__(p, facet, facet->actions, facet->actions_len,
1820 zero_stats ? &stats : NULL)) {
1821 facet->installed = true;
1825 /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up
1826 * to the accounting hook function in the ofhooks structure. */
1828 facet_account(struct ofproto *ofproto,
1829 struct facet *facet, uint64_t extra_bytes)
1831 uint64_t total_bytes = facet->byte_count + extra_bytes;
1833 if (ofproto->ofhooks->account_flow_cb
1834 && total_bytes > facet->accounted_bytes)
1836 ofproto->ofhooks->account_flow_cb(
1837 &facet->flow, facet->tags, facet->actions, facet->actions_len,
1838 total_bytes - facet->accounted_bytes, ofproto->aux);
1839 facet->accounted_bytes = total_bytes;
1843 /* If 'rule' is installed in the datapath, uninstalls it. */
1845 facet_uninstall(struct ofproto *p, struct facet *facet)
1847 if (facet->installed) {
1848 struct odputil_keybuf keybuf;
1849 struct dpif_flow_stats stats;
1852 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1853 odp_flow_key_from_flow(&key, &facet->flow);
1855 if (!dpif_flow_del(p->dpif, key.data, key.size, &stats)) {
1856 facet_update_stats(p, facet, &stats);
1858 facet->installed = false;
1859 facet->dp_packet_count = 0;
1860 facet->dp_byte_count = 0;
1862 assert(facet->dp_packet_count == 0);
1863 assert(facet->dp_byte_count == 0);
1867 /* Returns true if the only action for 'facet' is to send to the controller.
1868 * (We don't report NetFlow expiration messages for such facets because they
1869 * are just part of the control logic for the network, not real traffic). */
1871 facet_is_controller_flow(struct facet *facet)
1874 && facet->rule->n_actions == 1
1875 && action_outputs_to_port(&facet->rule->actions[0],
1876 htons(OFPP_CONTROLLER)));
1879 /* Folds all of 'facet''s statistics into its rule. Also updates the
1880 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
1881 * 'facet''s statistics in the datapath should have been zeroed and folded into
1882 * its packet and byte counts before this function is called. */
1884 facet_flush_stats(struct ofproto *ofproto, struct facet *facet)
1886 assert(!facet->dp_byte_count);
1887 assert(!facet->dp_packet_count);
1889 facet_push_stats(ofproto, facet);
1890 facet_account(ofproto, facet, 0);
1892 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
1893 struct ofexpired expired;
1894 expired.flow = facet->flow;
1895 expired.packet_count = facet->packet_count;
1896 expired.byte_count = facet->byte_count;
1897 expired.used = facet->used;
1898 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
1901 facet->rule->packet_count += facet->packet_count;
1902 facet->rule->byte_count += facet->byte_count;
1904 /* Reset counters to prevent double counting if 'facet' ever gets
1906 facet->packet_count = 0;
1907 facet->byte_count = 0;
1908 facet->rs_packet_count = 0;
1909 facet->rs_byte_count = 0;
1910 facet->accounted_bytes = 0;
1912 netflow_flow_clear(&facet->nf_flow);
1915 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
1916 * Returns it if found, otherwise a null pointer.
1918 * The returned facet might need revalidation; use facet_lookup_valid()
1919 * instead if that is important. */
1920 static struct facet *
1921 facet_find(struct ofproto *ofproto, const struct flow *flow)
1923 struct facet *facet;
1925 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
1927 if (flow_equal(flow, &facet->flow)) {
1935 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
1936 * Returns it if found, otherwise a null pointer.
1938 * The returned facet is guaranteed to be valid. */
1939 static struct facet *
1940 facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow)
1942 struct facet *facet = facet_find(ofproto, flow);
1944 /* The facet we found might not be valid, since we could be in need of
1945 * revalidation. If it is not valid, don't return it. */
1947 && ofproto->need_revalidate
1948 && !facet_revalidate(ofproto, facet)) {
1949 COVERAGE_INC(ofproto_invalidated);
1956 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
1958 * - If the rule found is different from 'facet''s current rule, moves
1959 * 'facet' to the new rule and recompiles its actions.
1961 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
1962 * where it is and recompiles its actions anyway.
1964 * - If there is none, destroys 'facet'.
1966 * Returns true if 'facet' still exists, false if it has been destroyed. */
1968 facet_revalidate(struct ofproto *ofproto, struct facet *facet)
1970 struct action_xlate_ctx ctx;
1971 struct ofpbuf *odp_actions;
1972 struct rule *new_rule;
1973 bool actions_changed;
1975 COVERAGE_INC(facet_revalidate);
1977 /* Determine the new rule. */
1978 new_rule = rule_lookup(ofproto, &facet->flow);
1980 /* No new rule, so delete the facet. */
1981 facet_remove(ofproto, facet);
1985 /* Calculate new ODP actions.
1987 * We do not modify any 'facet' state yet, because we might need to, e.g.,
1988 * emit a NetFlow expiration and, if so, we need to have the old state
1989 * around to properly compose it. */
1990 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
1991 odp_actions = xlate_actions(&ctx, new_rule->actions, new_rule->n_actions);
1992 actions_changed = (facet->actions_len != odp_actions->size
1993 || memcmp(facet->actions, odp_actions->data,
1994 facet->actions_len));
1996 /* If the ODP actions changed or the installability changed, then we need
1997 * to talk to the datapath. */
1998 if (actions_changed || ctx.may_set_up_flow != facet->installed) {
1999 if (ctx.may_set_up_flow) {
2000 struct dpif_flow_stats stats;
2002 facet_put__(ofproto, facet,
2003 odp_actions->data, odp_actions->size, &stats);
2004 facet_update_stats(ofproto, facet, &stats);
2006 facet_uninstall(ofproto, facet);
2009 /* The datapath flow is gone or has zeroed stats, so push stats out of
2010 * 'facet' into 'rule'. */
2011 facet_flush_stats(ofproto, facet);
2014 /* Update 'facet' now that we've taken care of all the old state. */
2015 facet->tags = ctx.tags;
2016 facet->nf_flow.output_iface = ctx.nf_output_iface;
2017 facet->may_install = ctx.may_set_up_flow;
2018 if (actions_changed) {
2019 free(facet->actions);
2020 facet->actions_len = odp_actions->size;
2021 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2023 if (facet->rule != new_rule) {
2024 COVERAGE_INC(facet_changed_rule);
2025 list_remove(&facet->list_node);
2026 list_push_back(&new_rule->facets, &facet->list_node);
2027 facet->rule = new_rule;
2028 facet->used = new_rule->created;
2029 facet->rs_used = facet->used;
2032 ofpbuf_delete(odp_actions);
2038 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2041 struct ofpbuf *buf = ofputil_encode_error_msg(error, oh);
2043 COVERAGE_INC(ofproto_error);
2044 ofconn_send_reply(ofconn, buf);
2049 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2051 ofconn_send_reply(ofconn, make_echo_reply(oh));
2056 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2058 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2059 struct ofp_switch_features *osf;
2061 struct ofport *port;
2063 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2064 osf->datapath_id = htonll(ofproto->datapath_id);
2065 osf->n_buffers = htonl(pktbuf_capacity());
2067 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2068 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2069 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2070 (1u << OFPAT_SET_VLAN_VID) |
2071 (1u << OFPAT_SET_VLAN_PCP) |
2072 (1u << OFPAT_STRIP_VLAN) |
2073 (1u << OFPAT_SET_DL_SRC) |
2074 (1u << OFPAT_SET_DL_DST) |
2075 (1u << OFPAT_SET_NW_SRC) |
2076 (1u << OFPAT_SET_NW_DST) |
2077 (1u << OFPAT_SET_NW_TOS) |
2078 (1u << OFPAT_SET_TP_SRC) |
2079 (1u << OFPAT_SET_TP_DST) |
2080 (1u << OFPAT_ENQUEUE));
2082 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2083 ofpbuf_put(buf, &port->opp, sizeof port->opp);
2086 ofconn_send_reply(ofconn, buf);
2091 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2093 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2095 struct ofp_switch_config *osc;
2099 /* Figure out flags. */
2100 dpif_get_drop_frags(ofproto->dpif, &drop_frags);
2101 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2104 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2105 osc->flags = htons(flags);
2106 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2107 ofconn_send_reply(ofconn, buf);
2113 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
2115 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2116 uint16_t flags = ntohs(osc->flags);
2118 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2119 && ofconn_get_role(ofconn) != NX_ROLE_SLAVE) {
2120 switch (flags & OFPC_FRAG_MASK) {
2121 case OFPC_FRAG_NORMAL:
2122 dpif_set_drop_frags(ofproto->dpif, false);
2124 case OFPC_FRAG_DROP:
2125 dpif_set_drop_frags(ofproto->dpif, true);
2128 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2134 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2139 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2140 struct action_xlate_ctx *ctx);
2143 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2145 const struct ofport *ofport = get_port(ctx->ofproto, port);
2148 if (ofport->opp.config & htonl(OFPPC_NO_FWD)) {
2149 /* Forwarding disabled on port. */
2154 * We don't have an ofport record for this port, but it doesn't hurt to
2155 * allow forwarding to it anyhow. Maybe such a port will appear later
2156 * and we're pre-populating the flow table.
2160 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, port);
2161 ctx->nf_output_iface = port;
2164 static struct rule *
2165 rule_lookup(struct ofproto *ofproto, const struct flow *flow)
2167 return rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2171 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2173 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2174 uint16_t old_in_port;
2177 /* Look up a flow with 'in_port' as the input port. Then restore the
2178 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2179 * have surprising behavior). */
2180 old_in_port = ctx->flow.in_port;
2181 ctx->flow.in_port = in_port;
2182 rule = rule_lookup(ctx->ofproto, &ctx->flow);
2183 ctx->flow.in_port = old_in_port;
2185 if (ctx->resubmit_hook) {
2186 ctx->resubmit_hook(ctx, rule);
2191 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2195 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2197 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2198 MAX_RESUBMIT_RECURSION);
2203 flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, ovs_be32 mask,
2204 uint16_t *nf_output_iface, struct ofpbuf *odp_actions)
2206 struct ofport *ofport;
2208 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
2209 uint16_t odp_port = ofport->odp_port;
2210 if (odp_port != odp_in_port && !(ofport->opp.config & mask)) {
2211 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port);
2214 *nf_output_iface = NF_OUT_FLOOD;
2218 xlate_output_action__(struct action_xlate_ctx *ctx,
2219 uint16_t port, uint16_t max_len)
2222 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2224 ctx->nf_output_iface = NF_OUT_DROP;
2228 add_output_action(ctx, ctx->flow.in_port);
2231 xlate_table_action(ctx, ctx->flow.in_port);
2234 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2235 ctx->odp_actions, &ctx->tags,
2236 &ctx->nf_output_iface,
2237 ctx->ofproto->aux)) {
2238 COVERAGE_INC(ofproto_uninstallable);
2239 ctx->may_set_up_flow = false;
2243 flood_packets(ctx->ofproto, ctx->flow.in_port, htonl(OFPPC_NO_FLOOD),
2244 &ctx->nf_output_iface, ctx->odp_actions);
2247 flood_packets(ctx->ofproto, ctx->flow.in_port, htonl(0),
2248 &ctx->nf_output_iface, ctx->odp_actions);
2250 case OFPP_CONTROLLER:
2251 nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_CONTROLLER, max_len);
2254 add_output_action(ctx, ODPP_LOCAL);
2257 odp_port = ofp_port_to_odp_port(port);
2258 if (odp_port != ctx->flow.in_port) {
2259 add_output_action(ctx, odp_port);
2264 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2265 ctx->nf_output_iface = NF_OUT_FLOOD;
2266 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2267 ctx->nf_output_iface = prev_nf_output_iface;
2268 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2269 ctx->nf_output_iface != NF_OUT_FLOOD) {
2270 ctx->nf_output_iface = NF_OUT_MULTI;
2275 xlate_output_action(struct action_xlate_ctx *ctx,
2276 const struct ofp_action_output *oao)
2278 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2281 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2282 * optimization, because we're going to add another action that sets the
2283 * priority immediately after, or because there are no actions following the
2286 remove_pop_action(struct action_xlate_ctx *ctx)
2288 if (ctx->odp_actions->size == ctx->last_pop_priority) {
2289 ctx->odp_actions->size -= NLA_ALIGN(NLA_HDRLEN);
2290 ctx->last_pop_priority = -1;
2295 add_pop_action(struct action_xlate_ctx *ctx)
2297 if (ctx->odp_actions->size != ctx->last_pop_priority) {
2298 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_POP_PRIORITY);
2299 ctx->last_pop_priority = ctx->odp_actions->size;
2304 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2305 const struct ofp_action_enqueue *oae)
2307 uint16_t ofp_port, odp_port;
2311 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2314 /* Fall back to ordinary output action. */
2315 xlate_output_action__(ctx, ntohs(oae->port), 0);
2319 /* Figure out ODP output port. */
2320 ofp_port = ntohs(oae->port);
2321 if (ofp_port != OFPP_IN_PORT) {
2322 odp_port = ofp_port_to_odp_port(ofp_port);
2324 odp_port = ctx->flow.in_port;
2327 /* Add ODP actions. */
2328 remove_pop_action(ctx);
2329 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2330 add_output_action(ctx, odp_port);
2331 add_pop_action(ctx);
2333 /* Update NetFlow output port. */
2334 if (ctx->nf_output_iface == NF_OUT_DROP) {
2335 ctx->nf_output_iface = odp_port;
2336 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2337 ctx->nf_output_iface = NF_OUT_MULTI;
2342 xlate_set_queue_action(struct action_xlate_ctx *ctx,
2343 const struct nx_action_set_queue *nasq)
2348 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
2351 /* Couldn't translate queue to a priority, so ignore. A warning
2352 * has already been logged. */
2356 remove_pop_action(ctx);
2357 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2361 xlate_set_dl_tci(struct action_xlate_ctx *ctx)
2363 ovs_be16 tci = ctx->flow.vlan_tci;
2364 if (!(tci & htons(VLAN_CFI))) {
2365 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN);
2367 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
2368 tci & ~htons(VLAN_CFI));
2372 struct xlate_reg_state {
2378 save_reg_state(const struct action_xlate_ctx *ctx,
2379 struct xlate_reg_state *state)
2381 state->vlan_tci = ctx->flow.vlan_tci;
2382 state->tun_id = ctx->flow.tun_id;
2386 update_reg_state(struct action_xlate_ctx *ctx,
2387 const struct xlate_reg_state *state)
2389 if (ctx->flow.vlan_tci != state->vlan_tci) {
2390 xlate_set_dl_tci(ctx);
2392 if (ctx->flow.tun_id != state->tun_id) {
2393 nl_msg_put_be64(ctx->odp_actions,
2394 ODP_ACTION_ATTR_SET_TUNNEL, ctx->flow.tun_id);
2399 xlate_nicira_action(struct action_xlate_ctx *ctx,
2400 const struct nx_action_header *nah)
2402 const struct nx_action_resubmit *nar;
2403 const struct nx_action_set_tunnel *nast;
2404 const struct nx_action_set_queue *nasq;
2405 const struct nx_action_multipath *nam;
2406 const struct nx_action_autopath *naa;
2407 enum nx_action_subtype subtype = ntohs(nah->subtype);
2408 const struct ofhooks *ofhooks = ctx->ofproto->ofhooks;
2409 struct xlate_reg_state state;
2410 uint16_t autopath_port;
2413 assert(nah->vendor == htonl(NX_VENDOR_ID));
2415 case NXAST_RESUBMIT:
2416 nar = (const struct nx_action_resubmit *) nah;
2417 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2420 case NXAST_SET_TUNNEL:
2421 nast = (const struct nx_action_set_tunnel *) nah;
2422 tun_id = htonll(ntohl(nast->tun_id));
2423 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
2424 ctx->flow.tun_id = tun_id;
2427 case NXAST_DROP_SPOOFED_ARP:
2428 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
2429 nl_msg_put_flag(ctx->odp_actions,
2430 ODP_ACTION_ATTR_DROP_SPOOFED_ARP);
2434 case NXAST_SET_QUEUE:
2435 nasq = (const struct nx_action_set_queue *) nah;
2436 xlate_set_queue_action(ctx, nasq);
2439 case NXAST_POP_QUEUE:
2440 add_pop_action(ctx);
2443 case NXAST_REG_MOVE:
2444 save_reg_state(ctx, &state);
2445 nxm_execute_reg_move((const struct nx_action_reg_move *) nah,
2447 update_reg_state(ctx, &state);
2450 case NXAST_REG_LOAD:
2451 save_reg_state(ctx, &state);
2452 nxm_execute_reg_load((const struct nx_action_reg_load *) nah,
2454 update_reg_state(ctx, &state);
2458 /* Nothing to do. */
2461 case NXAST_SET_TUNNEL64:
2462 tun_id = ((const struct nx_action_set_tunnel64 *) nah)->tun_id;
2463 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
2464 ctx->flow.tun_id = tun_id;
2467 case NXAST_MULTIPATH:
2468 nam = (const struct nx_action_multipath *) nah;
2469 multipath_execute(nam, &ctx->flow);
2472 case NXAST_AUTOPATH:
2473 naa = (const struct nx_action_autopath *) nah;
2474 autopath_port = (ofhooks->autopath_cb
2475 ? ofhooks->autopath_cb(&ctx->flow, ntohl(naa->id),
2476 &ctx->tags, ctx->ofproto->aux)
2478 autopath_execute(naa, &ctx->flow, autopath_port);
2481 /* If you add a new action here that modifies flow data, don't forget to
2482 * update the flow key in ctx->flow at the same time. */
2484 case NXAST_SNAT__OBSOLETE:
2486 VLOG_DBG_RL(&rl, "unknown Nicira action type %d", (int) subtype);
2492 do_xlate_actions(const union ofp_action *in, size_t n_in,
2493 struct action_xlate_ctx *ctx)
2495 struct actions_iterator iter;
2496 const union ofp_action *ia;
2497 const struct ofport *port;
2499 port = get_port(ctx->ofproto, ctx->flow.in_port);
2500 if (port && port->opp.config & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2501 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
2502 ? htonl(OFPPC_NO_RECV_STP)
2503 : htonl(OFPPC_NO_RECV))) {
2504 /* Drop this flow. */
2508 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2509 enum ofp_action_type type = ntohs(ia->type);
2510 const struct ofp_action_dl_addr *oada;
2514 xlate_output_action(ctx, &ia->output);
2517 case OFPAT_SET_VLAN_VID:
2518 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
2519 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
2520 xlate_set_dl_tci(ctx);
2523 case OFPAT_SET_VLAN_PCP:
2524 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
2525 ctx->flow.vlan_tci |= htons(
2526 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
2527 xlate_set_dl_tci(ctx);
2530 case OFPAT_STRIP_VLAN:
2531 ctx->flow.vlan_tci = htons(0);
2532 xlate_set_dl_tci(ctx);
2535 case OFPAT_SET_DL_SRC:
2536 oada = ((struct ofp_action_dl_addr *) ia);
2537 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_SRC,
2538 oada->dl_addr, ETH_ADDR_LEN);
2539 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
2542 case OFPAT_SET_DL_DST:
2543 oada = ((struct ofp_action_dl_addr *) ia);
2544 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_DST,
2545 oada->dl_addr, ETH_ADDR_LEN);
2546 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
2549 case OFPAT_SET_NW_SRC:
2550 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_SRC,
2551 ia->nw_addr.nw_addr);
2552 ctx->flow.nw_src = ia->nw_addr.nw_addr;
2555 case OFPAT_SET_NW_DST:
2556 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_DST,
2557 ia->nw_addr.nw_addr);
2558 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
2561 case OFPAT_SET_NW_TOS:
2562 nl_msg_put_u8(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_TOS,
2564 ctx->flow.nw_tos = ia->nw_tos.nw_tos;
2567 case OFPAT_SET_TP_SRC:
2568 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_SRC,
2569 ia->tp_port.tp_port);
2570 ctx->flow.tp_src = ia->tp_port.tp_port;
2573 case OFPAT_SET_TP_DST:
2574 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_DST,
2575 ia->tp_port.tp_port);
2576 ctx->flow.tp_dst = ia->tp_port.tp_port;
2580 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2584 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
2588 VLOG_DBG_RL(&rl, "unknown action type %d", (int) type);
2595 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
2596 struct ofproto *ofproto, const struct flow *flow,
2597 const struct ofpbuf *packet)
2599 ctx->ofproto = ofproto;
2601 ctx->packet = packet;
2602 ctx->resubmit_hook = NULL;
2603 ctx->check_special = true;
2607 ofproto_process_cfm(struct ofproto *ofproto, const struct flow *flow,
2608 const struct ofpbuf *packet)
2610 struct ofport *ofport;
2612 ofport = get_port(ofproto, flow->in_port);
2613 if (ofport && ofport->cfm) {
2614 cfm_process_heartbeat(ofport->cfm, packet);
2618 static struct ofpbuf *
2619 xlate_actions(struct action_xlate_ctx *ctx,
2620 const union ofp_action *in, size_t n_in)
2622 COVERAGE_INC(ofproto_ofp2odp);
2624 ctx->odp_actions = ofpbuf_new(512);
2626 ctx->may_set_up_flow = true;
2627 ctx->nf_output_iface = NF_OUT_DROP;
2629 ctx->last_pop_priority = -1;
2631 if (ctx->check_special && cfm_should_process_flow(&ctx->flow)) {
2633 ofproto_process_cfm(ctx->ofproto, &ctx->flow, ctx->packet);
2635 ctx->may_set_up_flow = false;
2636 } else if (ctx->check_special
2637 && ctx->ofproto->ofhooks->special_cb
2638 && !ctx->ofproto->ofhooks->special_cb(&ctx->flow, ctx->packet,
2639 ctx->ofproto->aux)) {
2640 ctx->may_set_up_flow = false;
2642 do_xlate_actions(in, n_in, ctx);
2645 remove_pop_action(ctx);
2647 /* Check with in-band control to see if we're allowed to set up this
2649 if (!connmgr_may_set_up_flow(ctx->ofproto->connmgr, &ctx->flow,
2650 ctx->odp_actions->data,
2651 ctx->odp_actions->size)) {
2652 ctx->may_set_up_flow = false;
2655 return ctx->odp_actions;
2658 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2659 * error message code (composed with ofp_mkerr()) for the caller to propagate
2660 * upward. Otherwise, returns 0.
2662 * The log message mentions 'msg_type'. */
2664 reject_slave_controller(struct ofconn *ofconn, const const char *msg_type)
2666 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2667 && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
2668 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
2669 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
2672 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2679 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2681 struct ofproto *p = ofconn_get_ofproto(ofconn);
2682 struct ofp_packet_out *opo;
2683 struct ofpbuf payload, *buffer;
2684 union ofp_action *ofp_actions;
2685 struct action_xlate_ctx ctx;
2686 struct ofpbuf *odp_actions;
2687 struct ofpbuf request;
2689 size_t n_ofp_actions;
2693 COVERAGE_INC(ofproto_packet_out);
2695 error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT");
2700 /* Get ofp_packet_out. */
2701 ofpbuf_use_const(&request, oh, ntohs(oh->length));
2702 opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions));
2705 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
2706 &ofp_actions, &n_ofp_actions);
2712 if (opo->buffer_id != htonl(UINT32_MAX)) {
2713 error = ofconn_pktbuf_retrieve(ofconn, ntohl(opo->buffer_id),
2715 if (error || !buffer) {
2724 /* Extract flow, check actions. */
2725 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)),
2727 error = validate_actions(ofp_actions, n_ofp_actions, &flow, p->max_ports);
2733 action_xlate_ctx_init(&ctx, p, &flow, &payload);
2734 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
2735 dpif_execute(p->dpif, odp_actions->data, odp_actions->size, &payload);
2736 ofpbuf_delete(odp_actions);
2739 ofpbuf_delete(buffer);
2744 update_port_config(struct ofproto *p, struct ofport *port,
2745 ovs_be32 config, ovs_be32 mask)
2747 mask &= config ^ port->opp.config;
2748 if (mask & htonl(OFPPC_PORT_DOWN)) {
2749 if (config & htonl(OFPPC_PORT_DOWN)) {
2750 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2752 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2755 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
2756 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
2757 if (mask & htonl(REVALIDATE_BITS)) {
2758 COVERAGE_INC(ofproto_costly_flags);
2759 port->opp.config ^= mask & htonl(REVALIDATE_BITS);
2760 p->need_revalidate = true;
2762 #undef REVALIDATE_BITS
2763 if (mask & htonl(OFPPC_NO_PACKET_IN)) {
2764 port->opp.config ^= htonl(OFPPC_NO_PACKET_IN);
2769 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2771 struct ofproto *p = ofconn_get_ofproto(ofconn);
2772 const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh;
2773 struct ofport *port;
2776 error = reject_slave_controller(ofconn, "OFPT_PORT_MOD");
2781 port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no)));
2783 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2784 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2785 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2787 update_port_config(p, port, opm->config, opm->mask);
2788 if (opm->advertise) {
2789 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2795 static struct ofpbuf *
2796 make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len)
2798 struct ofp_stats_reply *osr;
2801 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2802 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2804 osr->flags = htons(0);
2808 static struct ofpbuf *
2809 start_ofp_stats_reply(const struct ofp_header *request, size_t body_len)
2811 const struct ofp_stats_request *osr
2812 = (const struct ofp_stats_request *) request;
2813 return make_ofp_stats_reply(osr->header.xid, osr->type, body_len);
2817 append_ofp_stats_reply(size_t nbytes, struct ofconn *ofconn,
2818 struct ofpbuf **msgp)
2820 struct ofpbuf *msg = *msgp;
2821 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2822 if (nbytes + msg->size > UINT16_MAX) {
2823 struct ofp_stats_reply *reply = msg->data;
2824 reply->flags = htons(OFPSF_REPLY_MORE);
2825 *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes);
2826 ofconn_send_reply(ofconn, msg);
2828 return ofpbuf_put_uninit(*msgp, nbytes);
2831 static struct ofpbuf *
2832 make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len)
2834 struct nicira_stats_msg *nsm;
2837 msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX));
2838 nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg);
2839 nsm->type = htons(OFPST_VENDOR);
2840 nsm->flags = htons(0);
2841 nsm->vendor = htonl(NX_VENDOR_ID);
2842 nsm->subtype = subtype;
2846 static struct ofpbuf *
2847 start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len)
2849 return make_nxstats_reply(request->header.xid, request->subtype, body_len);
2853 append_nxstats_reply(size_t nbytes, struct ofconn *ofconn,
2854 struct ofpbuf **msgp)
2856 struct ofpbuf *msg = *msgp;
2857 assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg));
2858 if (nbytes + msg->size > UINT16_MAX) {
2859 struct nicira_stats_msg *reply = msg->data;
2860 reply->flags = htons(OFPSF_REPLY_MORE);
2861 *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes);
2862 ofconn_send_reply(ofconn, msg);
2864 ofpbuf_prealloc_tailroom(*msgp, nbytes);
2868 handle_desc_stats_request(struct ofconn *ofconn,
2869 const struct ofp_header *request)
2871 struct ofproto *p = ofconn_get_ofproto(ofconn);
2872 struct ofp_desc_stats *ods;
2875 msg = start_ofp_stats_reply(request, sizeof *ods);
2876 ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg);
2877 memset(ods, 0, sizeof *ods);
2878 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2879 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2880 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
2881 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
2882 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2883 ofconn_send_reply(ofconn, msg);
2889 handle_table_stats_request(struct ofconn *ofconn,
2890 const struct ofp_header *request)
2892 struct ofproto *p = ofconn_get_ofproto(ofconn);
2893 struct ofp_table_stats *ots;
2896 msg = start_ofp_stats_reply(request, sizeof *ots * 2);
2898 /* Classifier table. */
2899 ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg);
2900 memset(ots, 0, sizeof *ots);
2901 strcpy(ots->name, "classifier");
2902 ots->wildcards = (ofconn_get_flow_format(ofconn) == NXFF_OPENFLOW10
2903 ? htonl(OFPFW_ALL) : htonl(OVSFW_ALL));
2904 ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */
2905 ots->active_count = htonl(classifier_count(&p->cls));
2906 put_32aligned_be64(&ots->lookup_count, htonll(0)); /* XXX */
2907 put_32aligned_be64(&ots->matched_count, htonll(0)); /* XXX */
2909 ofconn_send_reply(ofconn, msg);
2914 append_port_stat(struct ofport *port, struct ofconn *ofconn,
2915 struct ofpbuf **msgp)
2917 struct netdev_stats stats;
2918 struct ofp_port_stats *ops;
2920 /* Intentionally ignore return value, since errors will set
2921 * 'stats' to all-1s, which is correct for OpenFlow, and
2922 * netdev_get_stats() will log errors. */
2923 netdev_get_stats(port->netdev, &stats);
2925 ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp);
2926 ops->port_no = port->opp.port_no;
2927 memset(ops->pad, 0, sizeof ops->pad);
2928 put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
2929 put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
2930 put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
2931 put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
2932 put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
2933 put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
2934 put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
2935 put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
2936 put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
2937 put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
2938 put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
2939 put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
2943 handle_port_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
2945 struct ofproto *p = ofconn_get_ofproto(ofconn);
2946 const struct ofp_port_stats_request *psr = ofputil_stats_body(oh);
2947 struct ofp_port_stats *ops;
2949 struct ofport *port;
2951 msg = start_ofp_stats_reply(oh, sizeof *ops * 16);
2952 if (psr->port_no != htons(OFPP_NONE)) {
2953 port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no)));
2955 append_port_stat(port, ofconn, &msg);
2958 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2959 append_port_stat(port, ofconn, &msg);
2963 ofconn_send_reply(ofconn, msg);
2968 calc_flow_duration__(long long int start, uint32_t *sec, uint32_t *nsec)
2970 long long int msecs = time_msec() - start;
2971 *sec = msecs / 1000;
2972 *nsec = (msecs % 1000) * (1000 * 1000);
2976 calc_flow_duration(long long int start, ovs_be32 *sec_be, ovs_be32 *nsec_be)
2980 calc_flow_duration__(start, &sec, &nsec);
2981 *sec_be = htonl(sec);
2982 *nsec_be = htonl(nsec);
2986 put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule,
2987 ovs_be16 out_port, struct ofpbuf **replyp)
2989 struct ofp_flow_stats *ofs;
2990 uint64_t packet_count, byte_count;
2992 size_t act_len, len;
2994 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
2998 act_len = sizeof *rule->actions * rule->n_actions;
2999 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3001 rule_get_stats(rule, &packet_count, &byte_count);
3003 ofs = append_ofp_stats_reply(len, ofconn, replyp);
3004 ofs->length = htons(len);
3007 ofputil_cls_rule_to_match(&rule->cr, ofconn_get_flow_format(ofconn),
3008 &ofs->match, rule->flow_cookie, &cookie);
3009 put_32aligned_be64(&ofs->cookie, cookie);
3010 calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec);
3011 ofs->priority = htons(rule->cr.priority);
3012 ofs->idle_timeout = htons(rule->idle_timeout);
3013 ofs->hard_timeout = htons(rule->hard_timeout);
3014 memset(ofs->pad2, 0, sizeof ofs->pad2);
3015 put_32aligned_be64(&ofs->packet_count, htonll(packet_count));
3016 put_32aligned_be64(&ofs->byte_count, htonll(byte_count));
3017 if (rule->n_actions > 0) {
3018 memcpy(ofs->actions, rule->actions, act_len);
3023 is_valid_table(uint8_t table_id)
3025 if (table_id == 0 || table_id == 0xff) {
3028 /* It would probably be better to reply with an error but there doesn't
3029 * seem to be any appropriate value, so that might just be
3031 VLOG_WARN_RL(&rl, "controller asked for invalid table %"PRIu8,
3038 handle_flow_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3040 const struct ofp_flow_stats_request *fsr = ofputil_stats_body(oh);
3041 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3042 struct ofpbuf *reply;
3044 COVERAGE_INC(ofproto_flows_req);
3045 reply = start_ofp_stats_reply(oh, 1024);
3046 if (is_valid_table(fsr->table_id)) {
3047 struct cls_cursor cursor;
3048 struct cls_rule target;
3051 ofputil_cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0,
3053 cls_cursor_init(&cursor, &ofproto->cls, &target);
3054 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3055 put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply);
3058 ofconn_send_reply(ofconn, reply);
3064 put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule,
3065 ovs_be16 out_port, struct ofpbuf **replyp)
3067 struct nx_flow_stats *nfs;
3068 uint64_t packet_count, byte_count;
3069 size_t act_len, start_len;
3070 struct ofpbuf *reply;
3072 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3076 rule_get_stats(rule, &packet_count, &byte_count);
3078 act_len = sizeof *rule->actions * rule->n_actions;
3080 append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, ofconn, replyp);
3081 start_len = (*replyp)->size;
3084 nfs = ofpbuf_put_uninit(reply, sizeof *nfs);
3087 calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec);
3088 nfs->cookie = rule->flow_cookie;
3089 nfs->priority = htons(rule->cr.priority);
3090 nfs->idle_timeout = htons(rule->idle_timeout);
3091 nfs->hard_timeout = htons(rule->hard_timeout);
3092 nfs->match_len = htons(nx_put_match(reply, &rule->cr));
3093 memset(nfs->pad2, 0, sizeof nfs->pad2);
3094 nfs->packet_count = htonll(packet_count);
3095 nfs->byte_count = htonll(byte_count);
3096 if (rule->n_actions > 0) {
3097 ofpbuf_put(reply, rule->actions, act_len);
3099 nfs->length = htons(reply->size - start_len);
3103 handle_nxst_flow(struct ofconn *ofconn, const struct ofp_header *oh)
3105 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3106 struct nx_flow_stats_request *nfsr;
3107 struct cls_rule target;
3108 struct ofpbuf *reply;
3112 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3114 /* Dissect the message. */
3115 nfsr = ofpbuf_pull(&b, sizeof *nfsr);
3116 error = nx_pull_match(&b, ntohs(nfsr->match_len), 0, &target);
3121 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3124 COVERAGE_INC(ofproto_flows_req);
3125 reply = start_nxstats_reply(&nfsr->nsm, 1024);
3126 if (is_valid_table(nfsr->table_id)) {
3127 struct cls_cursor cursor;
3130 cls_cursor_init(&cursor, &ofproto->cls, &target);
3131 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3132 put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply);
3135 ofconn_send_reply(ofconn, reply);
3141 flow_stats_ds(struct rule *rule, struct ds *results)
3143 uint64_t packet_count, byte_count;
3144 size_t act_len = sizeof *rule->actions * rule->n_actions;
3146 rule_get_stats(rule, &packet_count, &byte_count);
3148 ds_put_format(results, "duration=%llds, ",
3149 (time_msec() - rule->created) / 1000);
3150 ds_put_format(results, "idle=%.3fs, ", (time_msec() - rule->used) / 1000.0);
3151 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3152 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3153 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3154 cls_rule_format(&rule->cr, results);
3155 ds_put_char(results, ',');
3157 ofp_print_actions(results, &rule->actions->header, act_len);
3159 ds_put_cstr(results, "drop");
3161 ds_put_cstr(results, "\n");
3164 /* Adds a pretty-printed description of all flows to 'results', including
3165 * hidden flows (e.g., set up by in-band control). */
3167 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3169 struct cls_cursor cursor;
3172 cls_cursor_init(&cursor, &p->cls, NULL);
3173 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3174 flow_stats_ds(rule, results);
3178 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
3179 * '*engine_type' and '*engine_id', respectively. */
3181 ofproto_get_netflow_ids(const struct ofproto *ofproto,
3182 uint8_t *engine_type, uint8_t *engine_id)
3184 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
3188 query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target,
3189 ovs_be16 out_port, uint8_t table_id,
3190 struct ofp_aggregate_stats_reply *oasr)
3192 uint64_t total_packets = 0;
3193 uint64_t total_bytes = 0;
3196 COVERAGE_INC(ofproto_agg_request);
3198 if (is_valid_table(table_id)) {
3199 struct cls_cursor cursor;
3202 cls_cursor_init(&cursor, &ofproto->cls, target);
3203 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3204 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
3205 uint64_t packet_count;
3206 uint64_t byte_count;
3208 rule_get_stats(rule, &packet_count, &byte_count);
3210 total_packets += packet_count;
3211 total_bytes += byte_count;
3217 oasr->flow_count = htonl(n_flows);
3218 put_32aligned_be64(&oasr->packet_count, htonll(total_packets));
3219 put_32aligned_be64(&oasr->byte_count, htonll(total_bytes));
3220 memset(oasr->pad, 0, sizeof oasr->pad);
3224 handle_aggregate_stats_request(struct ofconn *ofconn,
3225 const struct ofp_header *oh)
3227 const struct ofp_aggregate_stats_request *request = ofputil_stats_body(oh);
3228 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3229 struct ofp_aggregate_stats_reply *reply;
3230 struct cls_rule target;
3233 ofputil_cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0,
3236 msg = start_ofp_stats_reply(oh, sizeof *reply);
3237 reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg);
3238 query_aggregate_stats(ofproto, &target, request->out_port,
3239 request->table_id, reply);
3240 ofconn_send_reply(ofconn, msg);
3245 handle_nxst_aggregate(struct ofconn *ofconn, const struct ofp_header *oh)
3247 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3248 struct nx_aggregate_stats_request *request;
3249 struct ofp_aggregate_stats_reply *reply;
3250 struct cls_rule target;
3255 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3257 /* Dissect the message. */
3258 request = ofpbuf_pull(&b, sizeof *request);
3259 error = nx_pull_match(&b, ntohs(request->match_len), 0, &target);
3264 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3268 COVERAGE_INC(ofproto_flows_req);
3269 buf = start_nxstats_reply(&request->nsm, sizeof *reply);
3270 reply = ofpbuf_put_uninit(buf, sizeof *reply);
3271 query_aggregate_stats(ofproto, &target, request->out_port,
3272 request->table_id, reply);
3273 ofconn_send_reply(ofconn, buf);
3278 struct queue_stats_cbdata {
3279 struct ofconn *ofconn;
3280 struct ofport *ofport;
3285 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3286 const struct netdev_queue_stats *stats)
3288 struct ofp_queue_stats *reply;
3290 reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3291 reply->port_no = cbdata->ofport->opp.port_no;
3292 memset(reply->pad, 0, sizeof reply->pad);
3293 reply->queue_id = htonl(queue_id);
3294 put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
3295 put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
3296 put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
3300 handle_queue_stats_dump_cb(uint32_t queue_id,
3301 struct netdev_queue_stats *stats,
3304 struct queue_stats_cbdata *cbdata = cbdata_;
3306 put_queue_stats(cbdata, queue_id, stats);
3310 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3311 struct queue_stats_cbdata *cbdata)
3313 cbdata->ofport = port;
3314 if (queue_id == OFPQ_ALL) {
3315 netdev_dump_queue_stats(port->netdev,
3316 handle_queue_stats_dump_cb, cbdata);
3318 struct netdev_queue_stats stats;
3320 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3321 put_queue_stats(cbdata, queue_id, &stats);
3327 handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3329 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3330 const struct ofp_queue_stats_request *qsr;
3331 struct queue_stats_cbdata cbdata;
3332 struct ofport *port;
3333 unsigned int port_no;
3336 qsr = ofputil_stats_body(oh);
3338 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3341 COVERAGE_INC(ofproto_queue_req);
3343 cbdata.ofconn = ofconn;
3344 cbdata.msg = start_ofp_stats_reply(oh, 128);
3346 port_no = ntohs(qsr->port_no);
3347 queue_id = ntohl(qsr->queue_id);
3348 if (port_no == OFPP_ALL) {
3349 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3350 handle_queue_stats_for_port(port, queue_id, &cbdata);
3352 } else if (port_no < ofproto->max_ports) {
3353 port = get_port(ofproto, ofp_port_to_odp_port(port_no));
3355 handle_queue_stats_for_port(port, queue_id, &cbdata);
3358 ofpbuf_delete(cbdata.msg);
3359 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3361 ofconn_send_reply(ofconn, cbdata.msg);
3366 /* Updates 'facet''s used time. Caller is responsible for calling
3367 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3369 facet_update_time(struct ofproto *ofproto, struct facet *facet,
3372 if (used > facet->used) {
3374 if (used > facet->rule->used) {
3375 facet->rule->used = used;
3377 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3381 /* Folds the statistics from 'stats' into the counters in 'facet'.
3383 * Because of the meaning of a facet's counters, it only makes sense to do this
3384 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3385 * packet that was sent by hand or if it represents statistics that have been
3386 * cleared out of the datapath. */
3388 facet_update_stats(struct ofproto *ofproto, struct facet *facet,
3389 const struct dpif_flow_stats *stats)
3391 if (stats->n_packets || stats->used > facet->used) {
3392 facet_update_time(ofproto, facet, stats->used);
3393 facet->packet_count += stats->n_packets;
3394 facet->byte_count += stats->n_bytes;
3395 facet_push_stats(ofproto, facet);
3396 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3401 facet_push_stats(struct ofproto *ofproto, struct facet *facet)
3403 uint64_t rs_packets, rs_bytes;
3405 assert(facet->packet_count >= facet->rs_packet_count);
3406 assert(facet->byte_count >= facet->rs_byte_count);
3407 assert(facet->used >= facet->rs_used);
3409 rs_packets = facet->packet_count - facet->rs_packet_count;
3410 rs_bytes = facet->byte_count - facet->rs_byte_count;
3412 if (rs_packets || rs_bytes || facet->used > facet->rs_used) {
3413 facet->rs_packet_count = facet->packet_count;
3414 facet->rs_byte_count = facet->byte_count;
3415 facet->rs_used = facet->used;
3417 flow_push_stats(ofproto, facet->rule, &facet->flow,
3418 rs_packets, rs_bytes, facet->used);
3422 struct ofproto_push {
3423 struct action_xlate_ctx ctx;
3430 push_resubmit(struct action_xlate_ctx *ctx, struct rule *rule)
3432 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3435 rule->packet_count += push->packets;
3436 rule->byte_count += push->bytes;
3437 rule->used = MAX(push->used, rule->used);
3441 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3442 * 'rule''s actions. */
3444 flow_push_stats(struct ofproto *ofproto, const struct rule *rule,
3445 struct flow *flow, uint64_t packets, uint64_t bytes,
3448 struct ofproto_push push;
3450 push.packets = packets;
3454 action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL);
3455 push.ctx.resubmit_hook = push_resubmit;
3456 ofpbuf_delete(xlate_actions(&push.ctx, rule->actions, rule->n_actions));
3459 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3460 * in which no matching flow already exists in the flow table.
3462 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3463 * ofp_actions, to the ofproto's flow table. Returns 0 on success or an
3464 * OpenFlow error code as encoded by ofp_mkerr() on failure.
3466 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3469 add_flow(struct ofconn *ofconn, struct flow_mod *fm)
3471 struct ofproto *p = ofconn_get_ofproto(ofconn);
3472 struct ofpbuf *packet;
3477 if (fm->flags & OFPFF_CHECK_OVERLAP
3478 && classifier_rule_overlaps(&p->cls, &fm->cr)) {
3479 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3483 if (fm->buffer_id != UINT32_MAX) {
3484 error = ofconn_pktbuf_retrieve(ofconn, fm->buffer_id,
3488 in_port = UINT16_MAX;
3491 rule = rule_create(&fm->cr, fm->actions, fm->n_actions,
3492 fm->idle_timeout, fm->hard_timeout, fm->cookie,
3493 fm->flags & OFPFF_SEND_FLOW_REM);
3494 rule_insert(p, rule);
3496 rule_execute(p, rule, in_port, packet);
3501 static struct rule *
3502 find_flow_strict(struct ofproto *p, const struct flow_mod *fm)
3504 return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr));
3508 send_buffered_packet(struct ofconn *ofconn,
3509 struct rule *rule, uint32_t buffer_id)
3511 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3512 struct ofpbuf *packet;
3516 if (buffer_id == UINT32_MAX) {
3520 error = ofconn_pktbuf_retrieve(ofconn, buffer_id, &packet, &in_port);
3525 rule_execute(ofproto, rule, in_port, packet);
3530 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3532 struct modify_flows_cbdata {
3533 struct ofproto *ofproto;
3534 const struct flow_mod *fm;
3538 static int modify_flow(struct ofproto *, const struct flow_mod *,
3541 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3542 * encoded by ofp_mkerr() on failure.
3544 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3547 modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm)
3549 struct ofproto *p = ofconn_get_ofproto(ofconn);
3550 struct rule *match = NULL;
3551 struct cls_cursor cursor;
3554 cls_cursor_init(&cursor, &p->cls, &fm->cr);
3555 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3556 if (!rule_is_hidden(rule)) {
3558 modify_flow(p, fm, rule);
3563 /* This credits the packet to whichever flow happened to match last.
3564 * That's weird. Maybe we should do a lookup for the flow that
3565 * actually matches the packet? Who knows. */
3566 send_buffered_packet(ofconn, match, fm->buffer_id);
3569 return add_flow(ofconn, fm);
3573 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3574 * code as encoded by ofp_mkerr() on failure.
3576 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3579 modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm)
3581 struct ofproto *p = ofconn_get_ofproto(ofconn);
3582 struct rule *rule = find_flow_strict(p, fm);
3583 if (rule && !rule_is_hidden(rule)) {
3584 modify_flow(p, fm, rule);
3585 return send_buffered_packet(ofconn, rule, fm->buffer_id);
3587 return add_flow(ofconn, fm);
3591 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3592 * been identified as a flow in 'p''s flow table to be modified, by changing
3593 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3594 * ofp_action[] structures). */
3596 modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule)
3598 size_t actions_len = fm->n_actions * sizeof *rule->actions;
3600 rule->flow_cookie = fm->cookie;
3602 /* If the actions are the same, do nothing. */
3603 if (fm->n_actions == rule->n_actions
3605 || !memcmp(fm->actions, rule->actions, actions_len))) {
3609 /* Replace actions. */
3610 free(rule->actions);
3611 rule->actions = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL;
3612 rule->n_actions = fm->n_actions;
3614 p->need_revalidate = true;
3619 /* OFPFC_DELETE implementation. */
3621 static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port);
3623 /* Implements OFPFC_DELETE. */
3625 delete_flows_loose(struct ofproto *p, const struct flow_mod *fm)
3627 struct rule *rule, *next_rule;
3628 struct cls_cursor cursor;
3630 cls_cursor_init(&cursor, &p->cls, &fm->cr);
3631 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
3632 delete_flow(p, rule, htons(fm->out_port));
3636 /* Implements OFPFC_DELETE_STRICT. */
3638 delete_flow_strict(struct ofproto *p, struct flow_mod *fm)
3640 struct rule *rule = find_flow_strict(p, fm);
3642 delete_flow(p, rule, htons(fm->out_port));
3646 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3647 * been identified as a flow to delete from 'p''s flow table, by deleting the
3648 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3651 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3652 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3653 * specified 'out_port'. */
3655 delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port)
3657 if (rule_is_hidden(rule)) {
3661 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
3665 rule_send_removed(p, rule, OFPRR_DELETE);
3666 rule_remove(p, rule);
3670 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3672 struct ofproto *p = ofconn_get_ofproto(ofconn);
3676 error = reject_slave_controller(ofconn, "flow_mod");
3681 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_flow_format(ofconn));
3686 /* We do not support the emergency flow cache. It will hopefully get
3687 * dropped from OpenFlow in the near future. */
3688 if (fm.flags & OFPFF_EMERG) {
3689 /* There isn't a good fit for an error code, so just state that the
3690 * flow table is full. */
3691 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3694 error = validate_actions(fm.actions, fm.n_actions,
3695 &fm.cr.flow, p->max_ports);
3700 switch (fm.command) {
3702 return add_flow(ofconn, &fm);
3705 return modify_flows_loose(ofconn, &fm);
3707 case OFPFC_MODIFY_STRICT:
3708 return modify_flow_strict(ofconn, &fm);
3711 delete_flows_loose(p, &fm);
3714 case OFPFC_DELETE_STRICT:
3715 delete_flow_strict(p, &fm);
3719 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3724 handle_tun_id_from_cookie(struct ofconn *ofconn, const struct ofp_header *oh)
3726 const struct nxt_tun_id_cookie *msg
3727 = (const struct nxt_tun_id_cookie *) oh;
3728 enum nx_flow_format flow_format;
3730 flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10;
3731 ofconn_set_flow_format(ofconn, flow_format);
3737 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
3739 struct nx_role_request *nrr = (struct nx_role_request *) oh;
3740 struct nx_role_request *reply;
3744 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY) {
3745 VLOG_WARN_RL(&rl, "ignoring role request on service connection");
3746 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3749 role = ntohl(nrr->role);
3750 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3751 && role != NX_ROLE_SLAVE) {
3752 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
3754 /* There's no good error code for this. */
3755 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
3758 ofconn_set_role(ofconn, role);
3760 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
3761 reply->role = htonl(role);
3762 ofconn_send_reply(ofconn, buf);
3768 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
3770 const struct nxt_set_flow_format *msg
3771 = (const struct nxt_set_flow_format *) oh;
3774 format = ntohl(msg->format);
3775 if (format == NXFF_OPENFLOW10
3776 || format == NXFF_TUN_ID_FROM_COOKIE
3777 || format == NXFF_NXM) {
3778 ofconn_set_flow_format(ofconn, format);
3781 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3786 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
3788 struct ofp_header *ob;
3791 /* Currently, everything executes synchronously, so we can just
3792 * immediately send the barrier reply. */
3793 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
3794 ofconn_send_reply(ofconn, buf);
3799 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
3801 const struct ofp_header *oh = msg->data;
3802 const struct ofputil_msg_type *type;
3805 error = ofputil_decode_msg_type(oh, &type);
3810 switch (ofputil_msg_type_code(type)) {
3811 /* OpenFlow requests. */
3812 case OFPUTIL_OFPT_ECHO_REQUEST:
3813 return handle_echo_request(ofconn, oh);
3815 case OFPUTIL_OFPT_FEATURES_REQUEST:
3816 return handle_features_request(ofconn, oh);
3818 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
3819 return handle_get_config_request(ofconn, oh);
3821 case OFPUTIL_OFPT_SET_CONFIG:
3822 return handle_set_config(ofconn, msg->data);
3824 case OFPUTIL_OFPT_PACKET_OUT:
3825 return handle_packet_out(ofconn, oh);
3827 case OFPUTIL_OFPT_PORT_MOD:
3828 return handle_port_mod(ofconn, oh);
3830 case OFPUTIL_OFPT_FLOW_MOD:
3831 return handle_flow_mod(ofconn, oh);
3833 case OFPUTIL_OFPT_BARRIER_REQUEST:
3834 return handle_barrier_request(ofconn, oh);
3836 /* OpenFlow replies. */
3837 case OFPUTIL_OFPT_ECHO_REPLY:
3840 /* Nicira extension requests. */
3841 case OFPUTIL_NXT_TUN_ID_FROM_COOKIE:
3842 return handle_tun_id_from_cookie(ofconn, oh);
3844 case OFPUTIL_NXT_ROLE_REQUEST:
3845 return handle_role_request(ofconn, oh);
3847 case OFPUTIL_NXT_SET_FLOW_FORMAT:
3848 return handle_nxt_set_flow_format(ofconn, oh);
3850 case OFPUTIL_NXT_FLOW_MOD:
3851 return handle_flow_mod(ofconn, oh);
3853 /* OpenFlow statistics requests. */
3854 case OFPUTIL_OFPST_DESC_REQUEST:
3855 return handle_desc_stats_request(ofconn, oh);
3857 case OFPUTIL_OFPST_FLOW_REQUEST:
3858 return handle_flow_stats_request(ofconn, oh);
3860 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
3861 return handle_aggregate_stats_request(ofconn, oh);
3863 case OFPUTIL_OFPST_TABLE_REQUEST:
3864 return handle_table_stats_request(ofconn, oh);
3866 case OFPUTIL_OFPST_PORT_REQUEST:
3867 return handle_port_stats_request(ofconn, oh);
3869 case OFPUTIL_OFPST_QUEUE_REQUEST:
3870 return handle_queue_stats_request(ofconn, oh);
3872 /* Nicira extension statistics requests. */
3873 case OFPUTIL_NXST_FLOW_REQUEST:
3874 return handle_nxst_flow(ofconn, oh);
3876 case OFPUTIL_NXST_AGGREGATE_REQUEST:
3877 return handle_nxst_aggregate(ofconn, oh);
3879 case OFPUTIL_INVALID:
3880 case OFPUTIL_OFPT_HELLO:
3881 case OFPUTIL_OFPT_ERROR:
3882 case OFPUTIL_OFPT_FEATURES_REPLY:
3883 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
3884 case OFPUTIL_OFPT_PACKET_IN:
3885 case OFPUTIL_OFPT_FLOW_REMOVED:
3886 case OFPUTIL_OFPT_PORT_STATUS:
3887 case OFPUTIL_OFPT_BARRIER_REPLY:
3888 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
3889 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
3890 case OFPUTIL_OFPST_DESC_REPLY:
3891 case OFPUTIL_OFPST_FLOW_REPLY:
3892 case OFPUTIL_OFPST_QUEUE_REPLY:
3893 case OFPUTIL_OFPST_PORT_REPLY:
3894 case OFPUTIL_OFPST_TABLE_REPLY:
3895 case OFPUTIL_OFPST_AGGREGATE_REPLY:
3896 case OFPUTIL_NXT_ROLE_REPLY:
3897 case OFPUTIL_NXT_FLOW_REMOVED:
3898 case OFPUTIL_NXST_FLOW_REPLY:
3899 case OFPUTIL_NXST_AGGREGATE_REPLY:
3901 if (VLOG_IS_WARN_ENABLED()) {
3902 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3903 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3906 if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) {
3907 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
3909 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3915 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
3917 int error = handle_openflow__(ofconn, ofp_msg);
3919 send_error_oh(ofconn, ofp_msg->data, error);
3921 COVERAGE_INC(ofproto_recv_openflow);
3925 handle_miss_upcall(struct ofproto *p, struct dpif_upcall *upcall)
3927 struct facet *facet;
3930 /* Obtain in_port and tun_id, at least. */
3931 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
3933 /* Set header pointers in 'flow'. */
3934 flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow);
3936 if (cfm_should_process_flow(&flow)) {
3937 ofproto_process_cfm(p, &flow, upcall->packet);
3938 ofpbuf_delete(upcall->packet);
3940 } else if (p->ofhooks->special_cb
3941 && !p->ofhooks->special_cb(&flow, upcall->packet, p->aux)) {
3942 ofpbuf_delete(upcall->packet);
3946 /* Check with in-band control to see if this packet should be sent
3947 * to the local port regardless of the flow table. */
3948 if (connmgr_msg_in_hook(p->connmgr, &flow, upcall->packet)) {
3949 ofproto_send_packet(p, ODPP_LOCAL, 0, upcall->packet);
3952 facet = facet_lookup_valid(p, &flow);
3954 struct rule *rule = rule_lookup(p, &flow);
3956 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3957 struct ofport *port = get_port(p, flow.in_port);
3959 if (port->opp.config & htonl(OFPPC_NO_PACKET_IN)) {
3960 COVERAGE_INC(ofproto_no_packet_in);
3961 /* XXX install 'drop' flow entry */
3962 ofpbuf_delete(upcall->packet);
3966 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
3970 COVERAGE_INC(ofproto_packet_in);
3971 send_packet_in(p, upcall, &flow, false);
3975 facet = facet_create(p, rule, &flow, upcall->packet);
3976 } else if (!facet->may_install) {
3977 /* The facet is not installable, that is, we need to process every
3978 * packet, so process the current packet's actions into 'facet'. */
3979 facet_make_actions(p, facet, upcall->packet);
3982 if (facet->rule->cr.priority == FAIL_OPEN_PRIORITY) {
3984 * Extra-special case for fail-open mode.
3986 * We are in fail-open mode and the packet matched the fail-open rule,
3987 * but we are connected to a controller too. We should send the packet
3988 * up to the controller in the hope that it will try to set up a flow
3989 * and thereby allow us to exit fail-open.
3991 * See the top-level comment in fail-open.c for more information.
3993 send_packet_in(p, upcall, &flow, true);
3996 facet_execute(p, facet, upcall->packet);
3997 facet_install(p, facet, false);
4001 handle_upcall(struct ofproto *p, struct dpif_upcall *upcall)
4005 switch (upcall->type) {
4006 case DPIF_UC_ACTION:
4007 COVERAGE_INC(ofproto_ctlr_action);
4008 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4009 send_packet_in(p, upcall, &flow, false);
4012 case DPIF_UC_SAMPLE:
4014 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4015 ofproto_sflow_received(p->sflow, upcall, &flow);
4017 ofpbuf_delete(upcall->packet);
4021 handle_miss_upcall(p, upcall);
4024 case DPIF_N_UC_TYPES:
4026 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4031 /* Flow expiration. */
4033 static int ofproto_dp_max_idle(const struct ofproto *);
4034 static void ofproto_update_stats(struct ofproto *);
4035 static void rule_expire(struct ofproto *, struct rule *);
4036 static void ofproto_expire_facets(struct ofproto *, int dp_max_idle);
4038 /* This function is called periodically by ofproto_run(). Its job is to
4039 * collect updates for the flows that have been installed into the datapath,
4040 * most importantly when they last were used, and then use that information to
4041 * expire flows that have not been used recently.
4043 * Returns the number of milliseconds after which it should be called again. */
4045 ofproto_expire(struct ofproto *ofproto)
4047 struct rule *rule, *next_rule;
4048 struct cls_cursor cursor;
4051 /* Update stats for each flow in the datapath. */
4052 ofproto_update_stats(ofproto);
4054 /* Expire facets that have been idle too long. */
4055 dp_max_idle = ofproto_dp_max_idle(ofproto);
4056 ofproto_expire_facets(ofproto, dp_max_idle);
4058 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
4059 cls_cursor_init(&cursor, &ofproto->cls, NULL);
4060 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4061 rule_expire(ofproto, rule);
4064 /* Let the hook know that we're at a stable point: all outstanding data
4065 * in existing flows has been accounted to the account_cb. Thus, the
4066 * hook can now reasonably do operations that depend on having accurate
4067 * flow volume accounting (currently, that's just bond rebalancing). */
4068 if (ofproto->ofhooks->account_checkpoint_cb) {
4069 ofproto->ofhooks->account_checkpoint_cb(ofproto->aux);
4072 return MIN(dp_max_idle, 1000);
4075 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4077 * This function also pushes statistics updates to rules which each facet
4078 * resubmits into. Generally these statistics will be accurate. However, if a
4079 * facet changes the rule it resubmits into at some time in between
4080 * ofproto_update_stats() runs, it is possible that statistics accrued to the
4081 * old rule will be incorrectly attributed to the new rule. This could be
4082 * avoided by calling ofproto_update_stats() whenever rules are created or
4083 * deleted. However, the performance impact of making so many calls to the
4084 * datapath do not justify the benefit of having perfectly accurate statistics.
4087 ofproto_update_stats(struct ofproto *p)
4089 const struct dpif_flow_stats *stats;
4090 struct dpif_flow_dump dump;
4091 const struct nlattr *key;
4094 dpif_flow_dump_start(&dump, p->dpif);
4095 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4096 struct facet *facet;
4099 if (odp_flow_key_to_flow(key, key_len, &flow)) {
4103 odp_flow_key_format(key, key_len, &s);
4104 VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s",
4110 facet = facet_find(p, &flow);
4112 if (facet && facet->installed) {
4114 if (stats->n_packets >= facet->dp_packet_count) {
4115 facet->packet_count += stats->n_packets - facet->dp_packet_count;
4117 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4120 if (stats->n_bytes >= facet->dp_byte_count) {
4121 facet->byte_count += stats->n_bytes - facet->dp_byte_count;
4123 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4126 facet->dp_packet_count = stats->n_packets;
4127 facet->dp_byte_count = stats->n_bytes;
4129 facet_update_time(p, facet, stats->used);
4130 facet_account(p, facet, stats->n_bytes);
4131 facet_push_stats(p, facet);
4133 /* There's a flow in the datapath that we know nothing about.
4135 COVERAGE_INC(ofproto_unexpected_rule);
4136 dpif_flow_del(p->dpif, key, key_len, NULL);
4139 dpif_flow_dump_done(&dump);
4142 /* Calculates and returns the number of milliseconds of idle time after which
4143 * facets should expire from the datapath and we should fold their statistics
4144 * into their parent rules in userspace. */
4146 ofproto_dp_max_idle(const struct ofproto *ofproto)
4149 * Idle time histogram.
4151 * Most of the time a switch has a relatively small number of facets. When
4152 * this is the case we might as well keep statistics for all of them in
4153 * userspace and to cache them in the kernel datapath for performance as
4156 * As the number of facets increases, the memory required to maintain
4157 * statistics about them in userspace and in the kernel becomes
4158 * significant. However, with a large number of facets it is likely that
4159 * only a few of them are "heavy hitters" that consume a large amount of
4160 * bandwidth. At this point, only heavy hitters are worth caching in the
4161 * kernel and maintaining in userspaces; other facets we can discard.
4163 * The technique used to compute the idle time is to build a histogram with
4164 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
4165 * that is installed in the kernel gets dropped in the appropriate bucket.
4166 * After the histogram has been built, we compute the cutoff so that only
4167 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
4168 * cached. At least the most-recently-used bucket of facets is kept, so
4169 * actually an arbitrary number of facets can be kept in any given
4170 * expiration run (though the next run will delete most of those unless
4171 * they receive additional data).
4173 * This requires a second pass through the facets, in addition to the pass
4174 * made by ofproto_update_stats(), because the former function never looks
4175 * at uninstallable facets.
4177 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4178 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4179 int buckets[N_BUCKETS] = { 0 };
4180 struct facet *facet;
4185 total = hmap_count(&ofproto->facets);
4186 if (total <= 1000) {
4187 return N_BUCKETS * BUCKET_WIDTH;
4190 /* Build histogram. */
4192 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
4193 long long int idle = now - facet->used;
4194 int bucket = (idle <= 0 ? 0
4195 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4196 : (unsigned int) idle / BUCKET_WIDTH);
4200 /* Find the first bucket whose flows should be expired. */
4201 for (bucket = 0; bucket < N_BUCKETS; bucket++) {
4202 if (buckets[bucket]) {
4205 subtotal += buckets[bucket++];
4206 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
4211 if (VLOG_IS_DBG_ENABLED()) {
4215 ds_put_cstr(&s, "keep");
4216 for (i = 0; i < N_BUCKETS; i++) {
4218 ds_put_cstr(&s, ", drop");
4221 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4224 VLOG_INFO("%s: %s (msec:count)", ofproto->name, ds_cstr(&s));
4228 return bucket * BUCKET_WIDTH;
4232 facet_active_timeout(struct ofproto *ofproto, struct facet *facet)
4234 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
4235 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
4236 struct ofexpired expired;
4238 if (facet->installed) {
4239 struct dpif_flow_stats stats;
4241 facet_put__(ofproto, facet, facet->actions, facet->actions_len,
4243 facet_update_stats(ofproto, facet, &stats);
4246 expired.flow = facet->flow;
4247 expired.packet_count = facet->packet_count;
4248 expired.byte_count = facet->byte_count;
4249 expired.used = facet->used;
4250 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4255 ofproto_expire_facets(struct ofproto *ofproto, int dp_max_idle)
4257 long long int cutoff = time_msec() - dp_max_idle;
4258 struct facet *facet, *next_facet;
4260 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
4261 facet_active_timeout(ofproto, facet);
4262 if (facet->used < cutoff) {
4263 facet_remove(ofproto, facet);
4268 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4269 * then delete it entirely. */
4271 rule_expire(struct ofproto *ofproto, struct rule *rule)
4273 struct facet *facet, *next_facet;
4277 /* Has 'rule' expired? */
4279 if (rule->hard_timeout
4280 && now > rule->created + rule->hard_timeout * 1000) {
4281 reason = OFPRR_HARD_TIMEOUT;
4282 } else if (rule->idle_timeout && list_is_empty(&rule->facets)
4283 && now >rule->used + rule->idle_timeout * 1000) {
4284 reason = OFPRR_IDLE_TIMEOUT;
4289 COVERAGE_INC(ofproto_expired);
4291 /* Update stats. (This is a no-op if the rule expired due to an idle
4292 * timeout, because that only happens when the rule has no facets left.) */
4293 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4294 facet_remove(ofproto, facet);
4297 /* Get rid of the rule. */
4298 if (!rule_is_hidden(rule)) {
4299 rule_send_removed(ofproto, rule, reason);
4301 rule_remove(ofproto, rule);
4305 rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason)
4307 struct ofputil_flow_removed fr;
4309 if (!rule->send_flow_removed) {
4314 fr.cookie = rule->flow_cookie;
4316 calc_flow_duration__(rule->created, &fr.duration_sec, &fr.duration_nsec);
4317 fr.idle_timeout = rule->idle_timeout;
4318 fr.packet_count = rule->packet_count;
4319 fr.byte_count = rule->byte_count;
4321 connmgr_send_flow_removed(p->connmgr, &fr);
4324 /* Obtains statistics for 'rule' and stores them in '*packets' and '*bytes'.
4325 * The returned statistics include statistics for all of 'rule''s facets. */
4327 rule_get_stats(const struct rule *rule, uint64_t *packets, uint64_t *bytes)
4330 struct facet *facet;
4332 /* Start from historical data for 'rule' itself that are no longer tracked
4333 * in facets. This counts, for example, facets that have expired. */
4334 p = rule->packet_count;
4335 b = rule->byte_count;
4337 /* Add any statistics that are tracked by facets. This includes
4338 * statistical data recently updated by ofproto_update_stats() as well as
4339 * stats for packets that were executed "by hand" via dpif_execute(). */
4340 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4341 p += facet->packet_count;
4342 b += facet->byte_count;
4349 /* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an
4350 * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to
4351 * their individual configurations.
4353 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
4354 * Otherwise, ownership is transferred to this function. */
4356 send_packet_in(struct ofproto *ofproto, struct dpif_upcall *upcall,
4357 const struct flow *flow, bool clone)
4359 struct ofputil_packet_in pin;
4361 pin.packet = upcall->packet;
4362 pin.in_port = odp_port_to_ofp_port(flow->in_port);
4363 pin.reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION;
4364 pin.buffer_id = 0; /* not yet known */
4365 pin.send_len = upcall->userdata;
4366 connmgr_send_packet_in(ofproto->connmgr, upcall, flow,
4367 clone ? NULL : upcall->packet);
4371 pick_datapath_id(const struct ofproto *ofproto)
4373 const struct ofport *port;
4375 port = get_port(ofproto, ODPP_LOCAL);
4377 uint8_t ea[ETH_ADDR_LEN];
4380 error = netdev_get_etheraddr(port->netdev, ea);
4382 return eth_addr_to_uint64(ea);
4384 VLOG_WARN("could not get MAC address for %s (%s)",
4385 netdev_get_name(port->netdev), strerror(error));
4387 return ofproto->fallback_dpid;
4391 pick_fallback_dpid(void)
4393 uint8_t ea[ETH_ADDR_LEN];
4394 eth_addr_nicira_random(ea);
4395 return eth_addr_to_uint64(ea);
4398 static struct ofproto *
4399 ofproto_lookup(const char *name)
4401 struct ofproto *ofproto;
4403 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
4405 if (!strcmp(ofproto->name, name)) {
4413 ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED,
4414 void *aux OVS_UNUSED)
4416 struct ofproto *ofproto;
4420 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
4421 ds_put_format(&results, "%s\n", ofproto->name);
4423 unixctl_command_reply(conn, 200, ds_cstr(&results));
4424 ds_destroy(&results);
4427 struct ofproto_trace {
4428 struct action_xlate_ctx ctx;
4434 trace_format_rule(struct ds *result, int level, const struct rule *rule)
4436 ds_put_char_multiple(result, '\t', level);
4438 ds_put_cstr(result, "No match\n");
4442 ds_put_format(result, "Rule: cookie=%#"PRIx64" ",
4443 ntohll(rule->flow_cookie));
4444 cls_rule_format(&rule->cr, result);
4445 ds_put_char(result, '\n');
4447 ds_put_char_multiple(result, '\t', level);
4448 ds_put_cstr(result, "OpenFlow ");
4449 ofp_print_actions(result, (const struct ofp_action_header *) rule->actions,
4450 rule->n_actions * sizeof *rule->actions);
4451 ds_put_char(result, '\n');
4455 trace_format_flow(struct ds *result, int level, const char *title,
4456 struct ofproto_trace *trace)
4458 ds_put_char_multiple(result, '\t', level);
4459 ds_put_format(result, "%s: ", title);
4460 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
4461 ds_put_cstr(result, "unchanged");
4463 flow_format(result, &trace->ctx.flow);
4464 trace->flow = trace->ctx.flow;
4466 ds_put_char(result, '\n');
4470 trace_resubmit(struct action_xlate_ctx *ctx, struct rule *rule)
4472 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
4473 struct ds *result = trace->result;
4475 ds_put_char(result, '\n');
4476 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
4477 trace_format_rule(result, ctx->recurse + 1, rule);
4481 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
4482 void *aux OVS_UNUSED)
4484 char *dpname, *in_port_s, *tun_id_s, *packet_s;
4485 char *args = xstrdup(args_);
4486 char *save_ptr = NULL;
4487 struct ofproto *ofproto;
4488 struct ofpbuf packet;
4496 ofpbuf_init(&packet, strlen(args) / 2);
4499 dpname = strtok_r(args, " ", &save_ptr);
4500 tun_id_s = strtok_r(NULL, " ", &save_ptr);
4501 in_port_s = strtok_r(NULL, " ", &save_ptr);
4502 packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
4503 if (!dpname || !in_port_s || !packet_s) {
4504 unixctl_command_reply(conn, 501, "Bad command syntax");
4508 ofproto = ofproto_lookup(dpname);
4510 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
4515 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
4516 in_port = ofp_port_to_odp_port(atoi(in_port_s));
4518 packet_s = ofpbuf_put_hex(&packet, packet_s, NULL);
4519 packet_s += strspn(packet_s, " ");
4520 if (*packet_s != '\0') {
4521 unixctl_command_reply(conn, 501, "Trailing garbage in command");
4524 if (packet.size < ETH_HEADER_LEN) {
4525 unixctl_command_reply(conn, 501, "Packet data too short for Ethernet");
4529 ds_put_cstr(&result, "Packet: ");
4530 s = ofp_packet_to_string(packet.data, packet.size, packet.size);
4531 ds_put_cstr(&result, s);
4534 flow_extract(&packet, tun_id, in_port, &flow);
4535 ds_put_cstr(&result, "Flow: ");
4536 flow_format(&result, &flow);
4537 ds_put_char(&result, '\n');
4539 rule = rule_lookup(ofproto, &flow);
4540 trace_format_rule(&result, 0, rule);
4542 struct ofproto_trace trace;
4543 struct ofpbuf *odp_actions;
4545 trace.result = &result;
4547 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet);
4548 trace.ctx.resubmit_hook = trace_resubmit;
4549 odp_actions = xlate_actions(&trace.ctx,
4550 rule->actions, rule->n_actions);
4552 ds_put_char(&result, '\n');
4553 trace_format_flow(&result, 0, "Final flow", &trace);
4554 ds_put_cstr(&result, "Datapath actions: ");
4555 format_odp_actions(&result, odp_actions->data, odp_actions->size);
4556 ofpbuf_delete(odp_actions);
4559 unixctl_command_reply(conn, 200, ds_cstr(&result));
4562 ds_destroy(&result);
4563 ofpbuf_uninit(&packet);
4568 ofproto_unixctl_init(void)
4570 static bool registered;
4576 unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL);
4577 unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL);
4581 default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
4582 struct ofpbuf *odp_actions, tag_type *tags,
4583 uint16_t *nf_output_iface, void *ofproto_)
4585 struct ofproto *ofproto = ofproto_;
4586 struct mac_entry *dst_mac;
4588 /* Drop frames for reserved multicast addresses. */
4589 if (eth_addr_is_reserved(flow->dl_dst)) {
4593 /* Learn source MAC (but don't try to learn from revalidation). */
4595 && mac_learning_may_learn(ofproto->ml, flow->dl_src, 0)) {
4596 struct mac_entry *src_mac;
4598 src_mac = mac_learning_insert(ofproto->ml, flow->dl_src, 0);
4599 if (mac_entry_is_new(src_mac) || src_mac->port.i != flow->in_port) {
4600 /* The log messages here could actually be useful in debugging,
4601 * so keep the rate limit relatively high. */
4602 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4603 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
4604 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
4606 ofproto_revalidate(ofproto,
4607 mac_learning_changed(ofproto->ml, src_mac));
4608 src_mac->port.i = flow->in_port;
4612 /* Determine output port. */
4613 dst_mac = mac_learning_lookup(ofproto->ml, flow->dl_dst, 0, tags);
4615 flood_packets(ofproto, flow->in_port, htonl(OFPPC_NO_FLOOD),
4616 nf_output_iface, odp_actions);
4618 int out_port = dst_mac->port.i;
4619 if (out_port != flow->in_port) {
4620 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, out_port);
4621 *nf_output_iface = out_port;
4630 static const struct ofhooks default_ofhooks = {
4631 default_normal_ofhook_cb,