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>
27 #include "byte-order.h"
29 #include "classifier.h"
33 #include "dynamic-string.h"
34 #include "fail-open.h"
38 #include "mac-learning.h"
39 #include "multipath.h"
45 #include "ofp-print.h"
47 #include "ofproto-sflow.h"
49 #include "openflow/nicira-ext.h"
50 #include "openflow/openflow.h"
51 #include "openvswitch/datapath-protocol.h"
55 #include "poll-loop.h"
58 #include "stream-ssl.h"
62 #include "unaligned.h"
67 VLOG_DEFINE_THIS_MODULE(ofproto);
69 COVERAGE_DEFINE(facet_changed_rule);
70 COVERAGE_DEFINE(facet_revalidate);
71 COVERAGE_DEFINE(odp_overflow);
72 COVERAGE_DEFINE(ofproto_agg_request);
73 COVERAGE_DEFINE(ofproto_costly_flags);
74 COVERAGE_DEFINE(ofproto_ctlr_action);
75 COVERAGE_DEFINE(ofproto_del_rule);
76 COVERAGE_DEFINE(ofproto_error);
77 COVERAGE_DEFINE(ofproto_expiration);
78 COVERAGE_DEFINE(ofproto_expired);
79 COVERAGE_DEFINE(ofproto_flows_req);
80 COVERAGE_DEFINE(ofproto_flush);
81 COVERAGE_DEFINE(ofproto_invalidated);
82 COVERAGE_DEFINE(ofproto_no_packet_in);
83 COVERAGE_DEFINE(ofproto_ofp2odp);
84 COVERAGE_DEFINE(ofproto_packet_in);
85 COVERAGE_DEFINE(ofproto_packet_out);
86 COVERAGE_DEFINE(ofproto_queue_req);
87 COVERAGE_DEFINE(ofproto_recv_openflow);
88 COVERAGE_DEFINE(ofproto_reinit_ports);
89 COVERAGE_DEFINE(ofproto_unexpected_rule);
90 COVERAGE_DEFINE(ofproto_uninstallable);
91 COVERAGE_DEFINE(ofproto_update_port);
93 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
94 * flow translation. */
95 #define MAX_RESUBMIT_RECURSION 16
100 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
101 struct netdev *netdev;
102 struct ofp_phy_port opp; /* In host byte order. */
104 struct cfm *cfm; /* Connectivity Fault Management, if any. */
107 static void ofport_free(struct ofport *);
108 static void ofport_run(struct ofproto *, struct ofport *);
109 static void ofport_wait(struct ofport *);
111 struct action_xlate_ctx {
112 /* action_xlate_ctx_init() initializes these members. */
115 struct ofproto *ofproto;
117 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
118 * this flow when actions change header fields. */
121 /* The packet corresponding to 'flow', or a null pointer if we are
122 * revalidating without a packet to refer to. */
123 const struct ofpbuf *packet;
125 /* If nonnull, called just before executing a resubmit action.
127 * This is normally null so the client has to set it manually after
128 * calling action_xlate_ctx_init(). */
129 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule *);
131 /* If true, the speciality of 'flow' should be checked before executing
132 * its actions. If special_cb returns false on 'flow' rendered
133 * uninstallable and no actions will be executed. */
136 /* xlate_actions() initializes and uses these members. The client might want
137 * to look at them after it returns. */
139 struct ofpbuf *odp_actions; /* Datapath actions. */
140 tag_type tags; /* Tags associated with OFPP_NORMAL actions. */
141 bool may_set_up_flow; /* True ordinarily; false if the actions must
142 * be reassessed for every packet. */
143 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
145 /* xlate_actions() initializes and uses these members, but the client has no
146 * reason to look at them. */
148 int recurse; /* Recursion level, via xlate_table_action. */
149 int last_pop_priority; /* Offset in 'odp_actions' just past most
150 * recent ODP_ACTION_ATTR_SET_PRIORITY. */
153 static void action_xlate_ctx_init(struct action_xlate_ctx *,
154 struct ofproto *, const struct flow *,
155 const struct ofpbuf *);
156 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
157 const union ofp_action *in, size_t n_in);
159 /* An OpenFlow flow. */
161 long long int used; /* Time last used; time created if not used. */
162 long long int created; /* Creation time. */
166 * - Do include packets and bytes from facets that have been deleted or
167 * whose own statistics have been folded into the rule.
169 * - Do include packets and bytes sent "by hand" that were accounted to
170 * the rule without any facet being involved (this is a rare corner
171 * case in rule_execute()).
173 * - Do not include packet or bytes that can be obtained from any facet's
174 * packet_count or byte_count member or that can be obtained from the
175 * datapath by, e.g., dpif_flow_get() for any facet.
177 uint64_t packet_count; /* Number of packets received. */
178 uint64_t byte_count; /* Number of bytes received. */
180 ovs_be64 flow_cookie; /* Controller-issued identifier. */
182 struct cls_rule cr; /* In owning ofproto's classifier. */
183 uint16_t idle_timeout; /* In seconds from time of last use. */
184 uint16_t hard_timeout; /* In seconds from time of creation. */
185 bool send_flow_removed; /* Send a flow removed message? */
186 int n_actions; /* Number of elements in actions[]. */
187 union ofp_action *actions; /* OpenFlow actions. */
188 struct list facets; /* List of "struct facet"s. */
191 static struct rule *rule_from_cls_rule(const struct cls_rule *);
192 static bool rule_is_hidden(const struct rule *);
194 static struct rule *rule_create(const struct cls_rule *,
195 const union ofp_action *, size_t n_actions,
196 uint16_t idle_timeout, uint16_t hard_timeout,
197 ovs_be64 flow_cookie, bool send_flow_removed);
198 static void rule_destroy(struct ofproto *, struct rule *);
199 static void rule_free(struct rule *);
201 static struct rule *rule_lookup(struct ofproto *, const struct flow *);
202 static void rule_insert(struct ofproto *, struct rule *);
203 static void rule_remove(struct ofproto *, struct rule *);
205 static void rule_send_removed(struct ofproto *, struct rule *, uint8_t reason);
206 static void rule_get_stats(const struct rule *, uint64_t *packets,
209 /* An exact-match instantiation of an OpenFlow flow. */
211 long long int used; /* Time last used; time created if not used. */
215 * - Do include packets and bytes sent "by hand", e.g. with
218 * - Do include packets and bytes that were obtained from the datapath
219 * when a flow was deleted (e.g. dpif_flow_del()) or when its
220 * statistics were reset (e.g. dpif_flow_put() with
221 * DPIF_FP_ZERO_STATS).
223 * - Do not include any packets or bytes that can currently be obtained
224 * from the datapath by, e.g., dpif_flow_get().
226 uint64_t packet_count; /* Number of packets received. */
227 uint64_t byte_count; /* Number of bytes received. */
229 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
230 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
232 uint64_t rs_packet_count; /* Packets pushed to resubmit children. */
233 uint64_t rs_byte_count; /* Bytes pushed to resubmit children. */
234 long long int rs_used; /* Used time pushed to resubmit children. */
236 /* Number of bytes passed to account_cb. This may include bytes that can
237 * currently obtained from the datapath (thus, it can be greater than
239 uint64_t accounted_bytes;
241 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
242 struct list list_node; /* In owning rule's 'facets' list. */
243 struct rule *rule; /* Owning rule. */
244 struct flow flow; /* Exact-match flow. */
245 bool installed; /* Installed in datapath? */
246 bool may_install; /* True ordinarily; false if actions must
247 * be reassessed for every packet. */
248 size_t actions_len; /* Number of bytes in actions[]. */
249 struct nlattr *actions; /* Datapath actions. */
250 tag_type tags; /* Tags (set only by hooks). */
251 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
254 static struct facet *facet_create(struct ofproto *, struct rule *,
256 const struct ofpbuf *packet);
257 static void facet_remove(struct ofproto *, struct facet *);
258 static void facet_free(struct facet *);
260 static struct facet *facet_lookup_valid(struct ofproto *, const struct flow *);
261 static bool facet_revalidate(struct ofproto *, struct facet *);
263 static void facet_install(struct ofproto *, struct facet *, bool zero_stats);
264 static void facet_uninstall(struct ofproto *, struct facet *);
265 static void facet_flush_stats(struct ofproto *, struct facet *);
267 static void facet_make_actions(struct ofproto *, struct facet *,
268 const struct ofpbuf *packet);
269 static void facet_update_stats(struct ofproto *, struct facet *,
270 const struct dpif_flow_stats *);
271 static void facet_push_stats(struct ofproto *, struct facet *);
273 static void send_packet_in(struct ofproto *, struct dpif_upcall *,
274 const struct flow *, bool clone);
278 uint64_t datapath_id; /* Datapath ID. */
279 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
280 char *mfr_desc; /* Manufacturer. */
281 char *hw_desc; /* Hardware. */
282 char *sw_desc; /* Software version. */
283 char *serial_desc; /* Serial number. */
284 char *dp_desc; /* Datapath description. */
288 struct netdev_monitor *netdev_monitor;
289 struct hmap ports; /* Contains "struct ofport"s. */
290 struct shash port_by_name;
294 struct netflow *netflow;
295 struct ofproto_sflow *sflow;
298 struct classifier cls;
299 long long int next_expiration;
303 bool need_revalidate;
304 struct tag_set revalidate_set;
306 /* OpenFlow connections. */
307 struct connmgr *connmgr;
309 /* Hooks for ovs-vswitchd. */
310 const struct ofhooks *ofhooks;
313 /* Used by default ofhooks. */
314 struct mac_learning *ml;
317 /* Map from dpif name to struct ofproto, for use by unixctl commands. */
318 static struct shash all_ofprotos = SHASH_INITIALIZER(&all_ofprotos);
320 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
322 static const struct ofhooks default_ofhooks;
324 static uint64_t pick_datapath_id(const struct ofproto *);
325 static uint64_t pick_fallback_dpid(void);
327 static int ofproto_expire(struct ofproto *);
328 static void flow_push_stats(struct ofproto *, const struct rule *,
329 struct flow *, uint64_t packets, uint64_t bytes,
332 static void handle_upcall(struct ofproto *, struct dpif_upcall *);
334 static void handle_openflow(struct ofconn *, struct ofpbuf *);
336 static struct ofport *get_port(const struct ofproto *, uint16_t odp_port);
337 static void update_port(struct ofproto *, const char *devname);
338 static int init_ports(struct ofproto *);
339 static void reinit_ports(struct ofproto *);
341 static void ofproto_unixctl_init(void);
344 ofproto_create(const char *datapath, const char *datapath_type,
345 const struct ofhooks *ofhooks, void *aux,
346 struct ofproto **ofprotop)
348 char local_name[IF_NAMESIZE];
355 ofproto_unixctl_init();
357 /* Connect to datapath and start listening for messages. */
358 error = dpif_open(datapath, datapath_type, &dpif);
360 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
363 error = dpif_recv_set_mask(dpif,
364 ((1u << DPIF_UC_MISS) |
365 (1u << DPIF_UC_ACTION) |
366 (1u << DPIF_UC_SAMPLE)));
368 VLOG_ERR("failed to listen on datapath %s: %s",
369 datapath, strerror(error));
373 dpif_flow_flush(dpif);
374 dpif_recv_purge(dpif);
376 error = dpif_port_get_name(dpif, ODPP_LOCAL,
377 local_name, sizeof local_name);
379 VLOG_ERR("%s: cannot get name of datapath local port (%s)",
380 datapath, strerror(error));
384 /* Initialize settings. */
385 p = xzalloc(sizeof *p);
386 p->fallback_dpid = pick_fallback_dpid();
387 p->datapath_id = p->fallback_dpid;
388 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
389 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
390 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
391 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
392 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
394 /* Initialize datapath. */
396 p->netdev_monitor = netdev_monitor_create();
397 hmap_init(&p->ports);
398 shash_init(&p->port_by_name);
399 p->max_ports = dpif_get_max_ports(dpif);
401 /* Initialize submodules. */
405 /* Initialize flow table. */
406 classifier_init(&p->cls);
407 p->next_expiration = time_msec() + 1000;
409 /* Initialize facet table. */
410 hmap_init(&p->facets);
411 p->need_revalidate = false;
412 tag_set_init(&p->revalidate_set);
414 /* Initialize hooks. */
416 p->ofhooks = ofhooks;
420 p->ofhooks = &default_ofhooks;
422 p->ml = mac_learning_create();
425 /* Pick final datapath ID. */
426 p->datapath_id = pick_datapath_id(p);
427 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
429 shash_add_once(&all_ofprotos, dpif_name(p->dpif), p);
431 /* Initialize OpenFlow connections. */
432 p->connmgr = connmgr_create(p, datapath, local_name);
439 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
441 uint64_t old_dpid = p->datapath_id;
442 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
443 if (p->datapath_id != old_dpid) {
444 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
446 /* Force all active connections to reconnect, since there is no way to
447 * notify a controller that the datapath ID has changed. */
448 ofproto_reconnect_controllers(p);
453 ofproto_set_controllers(struct ofproto *p,
454 const struct ofproto_controller *controllers,
455 size_t n_controllers)
457 connmgr_set_controllers(p->connmgr, controllers, n_controllers);
461 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
463 connmgr_set_fail_mode(p->connmgr, fail_mode);
466 /* Drops the connections between 'ofproto' and all of its controllers, forcing
467 * them to reconnect. */
469 ofproto_reconnect_controllers(struct ofproto *ofproto)
471 connmgr_reconnect(ofproto->connmgr);
474 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
475 * in-band control should guarantee access, in the same way that in-band
476 * control guarantees access to OpenFlow controllers. */
478 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
479 const struct sockaddr_in *extras, size_t n)
481 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
484 /* Sets the OpenFlow queue used by flows set up by in-band control on
485 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
486 * flows will use the default queue. */
488 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
490 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
494 ofproto_set_desc(struct ofproto *p,
495 const char *mfr_desc, const char *hw_desc,
496 const char *sw_desc, const char *serial_desc,
499 struct ofp_desc_stats *ods;
502 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
503 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
504 sizeof ods->mfr_desc);
507 p->mfr_desc = xstrdup(mfr_desc);
510 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
511 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
512 sizeof ods->hw_desc);
515 p->hw_desc = xstrdup(hw_desc);
518 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
519 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
520 sizeof ods->sw_desc);
523 p->sw_desc = xstrdup(sw_desc);
526 if (strlen(serial_desc) >= sizeof ods->serial_num) {
527 VLOG_WARN("truncating serial_desc, must be less than %zu "
529 sizeof ods->serial_num);
531 free(p->serial_desc);
532 p->serial_desc = xstrdup(serial_desc);
535 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
536 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
537 sizeof ods->dp_desc);
540 p->dp_desc = xstrdup(dp_desc);
545 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
547 return connmgr_set_snoops(ofproto->connmgr, snoops);
551 ofproto_set_netflow(struct ofproto *ofproto,
552 const struct netflow_options *nf_options)
554 if (nf_options && nf_options->collectors.n) {
555 if (!ofproto->netflow) {
556 ofproto->netflow = netflow_create();
558 return netflow_set_options(ofproto->netflow, nf_options);
560 netflow_destroy(ofproto->netflow);
561 ofproto->netflow = NULL;
567 ofproto_set_sflow(struct ofproto *ofproto,
568 const struct ofproto_sflow_options *oso)
570 struct ofproto_sflow *os = ofproto->sflow;
573 struct ofport *ofport;
575 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
576 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
577 ofproto_sflow_add_port(os, ofport->odp_port,
578 netdev_get_name(ofport->netdev));
581 ofproto_sflow_set_options(os, oso);
583 ofproto_sflow_destroy(os);
584 ofproto->sflow = NULL;
588 /* Connectivity Fault Management configuration. */
590 /* Clears the CFM configuration from 'port_no' on 'ofproto'. */
592 ofproto_iface_clear_cfm(struct ofproto *ofproto, uint32_t port_no)
594 struct ofport *ofport = get_port(ofproto, port_no);
595 if (ofport && ofport->cfm){
596 cfm_destroy(ofport->cfm);
601 /* Configures connectivity fault management on 'port_no' in 'ofproto'. Takes
602 * basic configuration from the configuration members in 'cfm', and the set of
603 * remote maintenance points from the 'n_remote_mps' elements in 'remote_mps'.
604 * Ignores the statistics members of 'cfm'.
606 * This function has no effect if 'ofproto' does not have a port 'port_no'. */
608 ofproto_iface_set_cfm(struct ofproto *ofproto, uint32_t port_no,
609 const struct cfm *cfm,
610 const uint16_t *remote_mps, size_t n_remote_mps)
612 struct ofport *ofport;
614 ofport = get_port(ofproto, port_no);
616 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu32,
617 dpif_name(ofproto->dpif), port_no);
622 ofport->cfm = cfm_create();
625 ofport->cfm->mpid = cfm->mpid;
626 ofport->cfm->interval = cfm->interval;
627 memcpy(ofport->cfm->maid, cfm->maid, CCM_MAID_LEN);
629 cfm_update_remote_mps(ofport->cfm, remote_mps, n_remote_mps);
631 if (!cfm_configure(ofport->cfm)) {
632 VLOG_WARN("%s: CFM configuration on port %"PRIu32" (%s) failed",
633 dpif_name(ofproto->dpif), port_no,
634 netdev_get_name(ofport->netdev));
635 cfm_destroy(ofport->cfm);
640 /* Returns the connectivity fault management object associated with 'port_no'
641 * within 'ofproto', or a null pointer if 'ofproto' does not have a port
642 * 'port_no' or if that port does not have CFM configured. The caller must not
643 * modify or destroy the returned object. */
645 ofproto_iface_get_cfm(struct ofproto *ofproto, uint32_t port_no)
647 struct ofport *ofport = get_port(ofproto, port_no);
648 return ofport ? ofport->cfm : NULL;
652 ofproto_get_datapath_id(const struct ofproto *ofproto)
654 return ofproto->datapath_id;
658 ofproto_has_primary_controller(const struct ofproto *ofproto)
660 return connmgr_has_controllers(ofproto->connmgr);
663 enum ofproto_fail_mode
664 ofproto_get_fail_mode(const struct ofproto *p)
666 return connmgr_get_fail_mode(p->connmgr);
670 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
672 connmgr_get_snoops(ofproto->connmgr, snoops);
676 ofproto_destroy(struct ofproto *p)
678 struct ofport *ofport, *next_ofport;
684 shash_find_and_delete(&all_ofprotos, dpif_name(p->dpif));
686 /* Destroy connmgr early, since it touches the classifier. */
687 connmgr_destroy(p->connmgr);
689 ofproto_flush_flows(p);
690 classifier_destroy(&p->cls);
691 hmap_destroy(&p->facets);
694 netdev_monitor_destroy(p->netdev_monitor);
695 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
696 hmap_remove(&p->ports, &ofport->hmap_node);
699 shash_destroy(&p->port_by_name);
701 netflow_destroy(p->netflow);
702 ofproto_sflow_destroy(p->sflow);
704 mac_learning_destroy(p->ml);
709 free(p->serial_desc);
712 hmap_destroy(&p->ports);
718 ofproto_run(struct ofproto *p)
720 int error = ofproto_run1(p);
722 error = ofproto_run2(p, false);
728 process_port_change(struct ofproto *ofproto, int error, char *devname)
730 if (error == ENOBUFS) {
731 reinit_ports(ofproto);
733 update_port(ofproto, devname);
739 ofproto_run1(struct ofproto *p)
741 struct ofport *ofport;
746 if (shash_is_empty(&p->port_by_name)) {
750 for (i = 0; i < 50; i++) {
751 struct dpif_upcall packet;
753 error = dpif_recv(p->dpif, &packet);
755 if (error == ENODEV) {
756 /* Someone destroyed the datapath behind our back. The caller
757 * better destroy us and give up, because we're just going to
758 * spin from here on out. */
759 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
760 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
767 handle_upcall(p, &packet);
770 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
771 process_port_change(p, error, devname);
773 while ((error = netdev_monitor_poll(p->netdev_monitor,
774 &devname)) != EAGAIN) {
775 process_port_change(p, error, devname);
778 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
779 ofport_run(p, ofport);
782 connmgr_run(p->connmgr, handle_openflow);
784 if (time_msec() >= p->next_expiration) {
785 int delay = ofproto_expire(p);
786 p->next_expiration = time_msec() + delay;
787 COVERAGE_INC(ofproto_expiration);
791 netflow_run(p->netflow);
794 ofproto_sflow_run(p->sflow);
801 ofproto_run2(struct ofproto *p, bool revalidate_all)
803 /* Figure out what we need to revalidate now, if anything. */
804 struct tag_set revalidate_set = p->revalidate_set;
805 if (p->need_revalidate) {
806 revalidate_all = true;
809 /* Clear the revalidation flags. */
810 tag_set_init(&p->revalidate_set);
811 p->need_revalidate = false;
813 /* Now revalidate if there's anything to do. */
814 if (revalidate_all || !tag_set_is_empty(&revalidate_set)) {
815 struct facet *facet, *next;
817 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &p->facets) {
819 || tag_set_intersects(&revalidate_set, facet->tags)) {
820 facet_revalidate(p, facet);
829 ofproto_wait(struct ofproto *p)
831 struct ofport *ofport;
833 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
836 dpif_recv_wait(p->dpif);
837 dpif_port_poll_wait(p->dpif);
838 netdev_monitor_poll_wait(p->netdev_monitor);
840 ofproto_sflow_wait(p->sflow);
842 if (!tag_set_is_empty(&p->revalidate_set)) {
843 poll_immediate_wake();
845 if (p->need_revalidate) {
846 /* Shouldn't happen, but if it does just go around again. */
847 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
848 poll_immediate_wake();
849 } else if (p->next_expiration != LLONG_MAX) {
850 poll_timer_wait_until(p->next_expiration);
852 connmgr_wait(p->connmgr);
856 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
858 tag_set_add(&ofproto->revalidate_set, tag);
862 ofproto_get_revalidate_set(struct ofproto *ofproto)
864 return &ofproto->revalidate_set;
868 ofproto_is_alive(const struct ofproto *p)
870 return connmgr_has_controllers(p->connmgr);
874 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
877 connmgr_get_controller_info(ofproto->connmgr, info);
881 ofproto_free_ofproto_controller_info(struct shash *info)
883 struct shash_node *node;
885 SHASH_FOR_EACH (node, info) {
886 struct ofproto_controller_info *cinfo = node->data;
887 while (cinfo->pairs.n) {
888 free((char *) cinfo->pairs.values[--cinfo->pairs.n]);
895 /* Deletes port number 'odp_port' from the datapath for 'ofproto'.
897 * This is almost the same as calling dpif_port_del() directly on the
898 * datapath, but it also makes 'ofproto' close its open netdev for the port
899 * (if any). This makes it possible to create a new netdev of a different
900 * type under the same name, which otherwise the netdev library would refuse
901 * to do because of the conflict. (The netdev would eventually get closed on
902 * the next trip through ofproto_run(), but this interface is more direct.)
904 * Returns 0 if successful, otherwise a positive errno. */
906 ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port)
908 struct ofport *ofport = get_port(ofproto, odp_port);
909 const char *name = ofport ? ofport->opp.name : "<unknown>";
912 error = dpif_port_del(ofproto->dpif, odp_port);
914 VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)",
915 dpif_name(ofproto->dpif), odp_port, name, strerror(error));
917 /* 'name' is ofport->opp.name and update_port() is going to destroy
918 * 'ofport'. Just in case update_port() refers to 'name' after it
919 * destroys 'ofport', make a copy of it around the update_port()
921 char *devname = xstrdup(name);
922 update_port(ofproto, devname);
928 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
929 * true if 'odp_port' exists and should be included, false otherwise. */
931 ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port)
933 struct ofport *ofport = get_port(ofproto, odp_port);
934 return ofport && !(ofport->opp.config & OFPPC_NO_FLOOD);
937 /* Sends 'packet' out of port 'port_no' within 'p'. If 'vlan_tci' is zero the
938 * packet will not have any 802.1Q hader; if it is nonzero, then the packet
939 * will be sent with the VLAN TCI specified by 'vlan_tci & ~VLAN_CFI'.
941 * Returns 0 if successful, otherwise a positive errno value. */
943 ofproto_send_packet(struct ofproto *ofproto,
944 uint32_t port_no, uint16_t vlan_tci,
945 const struct ofpbuf *packet)
947 struct ofpbuf odp_actions;
950 ofpbuf_init(&odp_actions, 32);
952 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
953 ntohs(vlan_tci & ~VLAN_CFI));
955 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, port_no);
956 error = dpif_execute(ofproto->dpif, odp_actions.data, odp_actions.size,
958 ofpbuf_uninit(&odp_actions);
961 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
962 dpif_name(ofproto->dpif), port_no, strerror(error));
967 /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and
968 * performs the 'n_actions' actions in 'actions'. The new flow will not
971 * If cls_rule->priority is in the range of priorities supported by OpenFlow
972 * (0...65535, inclusive) then the flow will be visible to OpenFlow
973 * controllers; otherwise, it will be hidden.
975 * The caller retains ownership of 'cls_rule' and 'actions'. */
977 ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule,
978 const union ofp_action *actions, size_t n_actions)
981 rule = rule_create(cls_rule, actions, n_actions, 0, 0, 0, false);
982 rule_insert(p, rule);
986 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
990 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
993 rule_remove(ofproto, rule);
998 ofproto_flush_flows(struct ofproto *ofproto)
1000 struct facet *facet, *next_facet;
1001 struct rule *rule, *next_rule;
1002 struct cls_cursor cursor;
1004 COVERAGE_INC(ofproto_flush);
1006 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1007 /* Mark the facet as not installed so that facet_remove() doesn't
1008 * bother trying to uninstall it. There is no point in uninstalling it
1009 * individually since we are about to blow away all the facets with
1010 * dpif_flow_flush(). */
1011 facet->installed = false;
1012 facet->dp_packet_count = 0;
1013 facet->dp_byte_count = 0;
1014 facet_remove(ofproto, facet);
1017 cls_cursor_init(&cursor, &ofproto->cls, NULL);
1018 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1019 rule_remove(ofproto, rule);
1022 dpif_flow_flush(ofproto->dpif);
1023 connmgr_flushed(ofproto->connmgr);
1027 reinit_ports(struct ofproto *p)
1029 struct dpif_port_dump dump;
1030 struct shash_node *node;
1031 struct shash devnames;
1032 struct ofport *ofport;
1033 struct dpif_port dpif_port;
1035 COVERAGE_INC(ofproto_reinit_ports);
1037 shash_init(&devnames);
1038 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1039 shash_add_once (&devnames, ofport->opp.name, NULL);
1041 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1042 shash_add_once (&devnames, dpif_port.name, NULL);
1045 SHASH_FOR_EACH (node, &devnames) {
1046 update_port(p, node->name);
1048 shash_destroy(&devnames);
1051 static struct ofport *
1052 make_ofport(const struct dpif_port *dpif_port)
1054 struct netdev_options netdev_options;
1055 enum netdev_flags flags;
1056 struct ofport *ofport;
1057 struct netdev *netdev;
1060 memset(&netdev_options, 0, sizeof netdev_options);
1061 netdev_options.name = dpif_port->name;
1062 netdev_options.type = dpif_port->type;
1063 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1065 error = netdev_open(&netdev_options, &netdev);
1067 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1068 "cannot be opened (%s)",
1069 dpif_port->name, dpif_port->port_no,
1070 dpif_port->name, strerror(error));
1074 ofport = xzalloc(sizeof *ofport);
1075 ofport->netdev = netdev;
1076 ofport->odp_port = dpif_port->port_no;
1077 ofport->opp.port_no = odp_port_to_ofp_port(dpif_port->port_no);
1078 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1079 ovs_strlcpy(ofport->opp.name, dpif_port->name, sizeof ofport->opp.name);
1081 netdev_get_flags(netdev, &flags);
1082 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1084 ofport->opp.state = netdev_get_carrier(netdev) ? 0 : OFPPS_LINK_DOWN;
1086 netdev_get_features(netdev,
1087 &ofport->opp.curr, &ofport->opp.advertised,
1088 &ofport->opp.supported, &ofport->opp.peer);
1093 ofport_conflicts(const struct ofproto *p, const struct dpif_port *dpif_port)
1095 if (get_port(p, dpif_port->port_no)) {
1096 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1097 dpif_port->port_no);
1099 } else if (shash_find(&p->port_by_name, dpif_port->name)) {
1100 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1109 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1111 const struct ofp_phy_port *a = &a_->opp;
1112 const struct ofp_phy_port *b = &b_->opp;
1114 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1115 return (a->port_no == b->port_no
1116 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1117 && !strcmp(a->name, b->name)
1118 && a->state == b->state
1119 && a->config == b->config
1120 && a->curr == b->curr
1121 && a->advertised == b->advertised
1122 && a->supported == b->supported
1123 && a->peer == b->peer);
1127 send_port_status(struct ofproto *p, const struct ofport *ofport,
1130 connmgr_send_port_status(p->connmgr, &ofport->opp, reason);
1134 ofport_install(struct ofproto *p, struct ofport *ofport)
1136 const char *netdev_name = ofport->opp.name;
1138 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1139 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
1140 shash_add(&p->port_by_name, netdev_name, ofport);
1142 ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
1147 ofport_remove(struct ofproto *p, struct ofport *ofport)
1149 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1150 hmap_remove(&p->ports, &ofport->hmap_node);
1151 shash_delete(&p->port_by_name,
1152 shash_find(&p->port_by_name, ofport->opp.name));
1154 ofproto_sflow_del_port(p->sflow, ofport->odp_port);
1159 ofport_run(struct ofproto *ofproto, struct ofport *ofport)
1162 cfm_run(ofport->cfm);
1164 if (cfm_should_send_ccm(ofport->cfm)) {
1165 struct ofpbuf packet;
1168 ofpbuf_init(&packet, 0);
1169 ccm = compose_packet(&packet, eth_addr_ccm, ofport->opp.hw_addr,
1170 ETH_TYPE_CFM, sizeof *ccm);
1171 cfm_compose_ccm(ofport->cfm, ccm);
1172 ofproto_send_packet(ofproto, ofport->odp_port, 0, &packet);
1173 ofpbuf_uninit(&packet);
1179 ofport_wait(struct ofport *ofport)
1182 cfm_wait(ofport->cfm);
1187 ofport_free(struct ofport *ofport)
1190 cfm_destroy(ofport->cfm);
1191 netdev_close(ofport->netdev);
1196 static struct ofport *
1197 get_port(const struct ofproto *ofproto, uint16_t odp_port)
1199 struct ofport *port;
1201 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1202 hash_int(odp_port, 0), &ofproto->ports) {
1203 if (port->odp_port == odp_port) {
1211 update_port(struct ofproto *p, const char *devname)
1213 struct dpif_port dpif_port;
1214 struct ofport *old_ofport;
1215 struct ofport *new_ofport;
1218 COVERAGE_INC(ofproto_update_port);
1220 /* Query the datapath for port information. */
1221 error = dpif_port_query_by_name(p->dpif, devname, &dpif_port);
1223 /* Find the old ofport. */
1224 old_ofport = shash_find_data(&p->port_by_name, devname);
1227 /* There's no port named 'devname' but there might be a port with
1228 * the same port number. This could happen if a port is deleted
1229 * and then a new one added in its place very quickly, or if a port
1230 * is renamed. In the former case we want to send an OFPPR_DELETE
1231 * and an OFPPR_ADD, and in the latter case we want to send a
1232 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1233 * the old port's ifindex against the new port, or perhaps less
1234 * reliably but more portably by comparing the old port's MAC
1235 * against the new port's MAC. However, this code isn't that smart
1236 * and always sends an OFPPR_MODIFY (XXX). */
1237 old_ofport = get_port(p, dpif_port.port_no);
1239 } else if (error != ENOENT && error != ENODEV) {
1240 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1241 "%s", strerror(error));
1245 /* Create a new ofport. */
1246 new_ofport = !error ? make_ofport(&dpif_port) : NULL;
1248 /* Eliminate a few pathological cases. */
1249 if (!old_ofport && !new_ofport) {
1251 } else if (old_ofport && new_ofport) {
1252 /* Most of the 'config' bits are OpenFlow soft state, but
1253 * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the
1254 * OpenFlow bits from old_ofport. (make_ofport() only sets
1255 * OFPPC_PORT_DOWN and leaves the other bits 0.) */
1256 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1258 if (ofport_equal(old_ofport, new_ofport)) {
1259 /* False alarm--no change. */
1260 ofport_free(new_ofport);
1265 /* Now deal with the normal cases. */
1267 ofport_remove(p, old_ofport);
1270 ofport_install(p, new_ofport);
1272 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1273 (!old_ofport ? OFPPR_ADD
1274 : !new_ofport ? OFPPR_DELETE
1276 ofport_free(old_ofport);
1279 dpif_port_destroy(&dpif_port);
1283 init_ports(struct ofproto *p)
1285 struct dpif_port_dump dump;
1286 struct dpif_port dpif_port;
1288 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1289 if (!ofport_conflicts(p, &dpif_port)) {
1290 struct ofport *ofport = make_ofport(&dpif_port);
1292 ofport_install(p, ofport);
1300 /* Returns true if 'rule' should be hidden from the controller.
1302 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1303 * (e.g. by in-band control) and are intentionally hidden from the
1306 rule_is_hidden(const struct rule *rule)
1308 return rule->cr.priority > UINT16_MAX;
1311 /* Creates and returns a new rule initialized as specified.
1313 * The caller is responsible for inserting the rule into the classifier (with
1314 * rule_insert()). */
1315 static struct rule *
1316 rule_create(const struct cls_rule *cls_rule,
1317 const union ofp_action *actions, size_t n_actions,
1318 uint16_t idle_timeout, uint16_t hard_timeout,
1319 ovs_be64 flow_cookie, bool send_flow_removed)
1321 struct rule *rule = xzalloc(sizeof *rule);
1322 rule->cr = *cls_rule;
1323 rule->idle_timeout = idle_timeout;
1324 rule->hard_timeout = hard_timeout;
1325 rule->flow_cookie = flow_cookie;
1326 rule->used = rule->created = time_msec();
1327 rule->send_flow_removed = send_flow_removed;
1328 list_init(&rule->facets);
1329 if (n_actions > 0) {
1330 rule->n_actions = n_actions;
1331 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1337 static struct rule *
1338 rule_from_cls_rule(const struct cls_rule *cls_rule)
1340 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1344 rule_free(struct rule *rule)
1346 free(rule->actions);
1350 /* Destroys 'rule' and iterates through all of its facets and revalidates them,
1351 * destroying any that no longer has a rule (which is probably all of them).
1353 * The caller must have already removed 'rule' from the classifier. */
1355 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1357 struct facet *facet, *next_facet;
1358 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
1359 facet_revalidate(ofproto, facet);
1364 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
1365 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
1368 rule_has_out_port(const struct rule *rule, ovs_be16 out_port)
1370 const union ofp_action *oa;
1371 struct actions_iterator i;
1373 if (out_port == htons(OFPP_NONE)) {
1376 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1377 oa = actions_next(&i)) {
1378 if (action_outputs_to_port(oa, out_port)) {
1385 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
1386 * 'packet', which arrived on 'in_port'.
1388 * Takes ownership of 'packet'. */
1390 execute_odp_actions(struct ofproto *ofproto, const struct flow *flow,
1391 const struct nlattr *odp_actions, size_t actions_len,
1392 struct ofpbuf *packet)
1394 if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t))
1395 && odp_actions->nla_type == ODP_ACTION_ATTR_CONTROLLER) {
1396 /* As an optimization, avoid a round-trip from userspace to kernel to
1397 * userspace. This also avoids possibly filling up kernel packet
1398 * buffers along the way. */
1399 struct dpif_upcall upcall;
1401 upcall.type = DPIF_UC_ACTION;
1402 upcall.packet = packet;
1405 upcall.userdata = nl_attr_get_u64(odp_actions);
1406 upcall.sample_pool = 0;
1407 upcall.actions = NULL;
1408 upcall.actions_len = 0;
1410 send_packet_in(ofproto, &upcall, flow, false);
1416 error = dpif_execute(ofproto->dpif, odp_actions, actions_len, packet);
1417 ofpbuf_delete(packet);
1422 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
1423 * statistics appropriately. 'packet' must have at least sizeof(struct
1424 * ofp_packet_in) bytes of headroom.
1426 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
1427 * applying flow_extract() to 'packet' would yield the same flow as
1430 * 'facet' must have accurately composed ODP actions; that is, it must not be
1431 * in need of revalidation.
1433 * Takes ownership of 'packet'. */
1435 facet_execute(struct ofproto *ofproto, struct facet *facet,
1436 struct ofpbuf *packet)
1438 struct dpif_flow_stats stats;
1440 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1442 flow_extract_stats(&facet->flow, packet, &stats);
1443 stats.used = time_msec();
1444 if (execute_odp_actions(ofproto, &facet->flow,
1445 facet->actions, facet->actions_len, packet)) {
1446 facet_update_stats(ofproto, facet, &stats);
1450 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
1451 * statistics (or the statistics for one of its facets) appropriately.
1452 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom.
1454 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
1455 * with statistics for 'packet' either way.
1457 * Takes ownership of 'packet'. */
1459 rule_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port,
1460 struct ofpbuf *packet)
1462 struct action_xlate_ctx ctx;
1463 struct ofpbuf *odp_actions;
1464 struct facet *facet;
1468 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1470 flow_extract(packet, 0, in_port, &flow);
1472 /* First look for a related facet. If we find one, account it to that. */
1473 facet = facet_lookup_valid(ofproto, &flow);
1474 if (facet && facet->rule == rule) {
1475 facet_execute(ofproto, facet, packet);
1479 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
1480 * create a new facet for it and use that. */
1481 if (rule_lookup(ofproto, &flow) == rule) {
1482 facet = facet_create(ofproto, rule, &flow, packet);
1483 facet_execute(ofproto, facet, packet);
1484 facet_install(ofproto, facet, true);
1488 /* We can't account anything to a facet. If we were to try, then that
1489 * facet would have a non-matching rule, busting our invariants. */
1490 action_xlate_ctx_init(&ctx, ofproto, &flow, packet);
1491 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
1492 size = packet->size;
1493 if (execute_odp_actions(ofproto, &flow, odp_actions->data,
1494 odp_actions->size, packet)) {
1495 rule->used = time_msec();
1496 rule->packet_count++;
1497 rule->byte_count += size;
1498 flow_push_stats(ofproto, rule, &flow, 1, size, rule->used);
1500 ofpbuf_delete(odp_actions);
1503 /* Inserts 'rule' into 'p''s flow table. */
1505 rule_insert(struct ofproto *p, struct rule *rule)
1507 struct rule *displaced_rule;
1509 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1510 if (displaced_rule) {
1511 rule_destroy(p, displaced_rule);
1513 p->need_revalidate = true;
1516 /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a
1517 * 'flow' and an example 'packet' within that flow.
1519 * The caller must already have determined that no facet with an identical
1520 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
1521 * 'ofproto''s classifier table. */
1522 static struct facet *
1523 facet_create(struct ofproto *ofproto, struct rule *rule,
1524 const struct flow *flow, const struct ofpbuf *packet)
1526 struct facet *facet;
1528 facet = xzalloc(sizeof *facet);
1529 facet->used = time_msec();
1530 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
1531 list_push_back(&rule->facets, &facet->list_node);
1533 facet->flow = *flow;
1534 netflow_flow_init(&facet->nf_flow);
1535 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
1537 facet_make_actions(ofproto, facet, packet);
1543 facet_free(struct facet *facet)
1545 free(facet->actions);
1549 /* Remove 'rule' from 'ofproto' and free up the associated memory:
1551 * - Removes 'rule' from the classifier.
1553 * - If 'rule' has facets, revalidates them (and possibly uninstalls and
1554 * destroys them), via rule_destroy().
1557 rule_remove(struct ofproto *ofproto, struct rule *rule)
1559 COVERAGE_INC(ofproto_del_rule);
1560 ofproto->need_revalidate = true;
1561 classifier_remove(&ofproto->cls, &rule->cr);
1562 rule_destroy(ofproto, rule);
1565 /* Remove 'facet' from 'ofproto' and free up the associated memory:
1567 * - If 'facet' was installed in the datapath, uninstalls it and updates its
1568 * rule's statistics, via facet_uninstall().
1570 * - Removes 'facet' from its rule and from ofproto->facets.
1573 facet_remove(struct ofproto *ofproto, struct facet *facet)
1575 facet_uninstall(ofproto, facet);
1576 facet_flush_stats(ofproto, facet);
1577 hmap_remove(&ofproto->facets, &facet->hmap_node);
1578 list_remove(&facet->list_node);
1582 /* Composes the ODP actions for 'facet' based on its rule's actions. */
1584 facet_make_actions(struct ofproto *p, struct facet *facet,
1585 const struct ofpbuf *packet)
1587 const struct rule *rule = facet->rule;
1588 struct ofpbuf *odp_actions;
1589 struct action_xlate_ctx ctx;
1591 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
1592 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
1593 facet->tags = ctx.tags;
1594 facet->may_install = ctx.may_set_up_flow;
1595 facet->nf_flow.output_iface = ctx.nf_output_iface;
1597 if (facet->actions_len != odp_actions->size
1598 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
1599 free(facet->actions);
1600 facet->actions_len = odp_actions->size;
1601 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
1604 ofpbuf_delete(odp_actions);
1608 facet_put__(struct ofproto *ofproto, struct facet *facet,
1609 const struct nlattr *actions, size_t actions_len,
1610 struct dpif_flow_stats *stats)
1612 uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S];
1613 enum dpif_flow_put_flags flags;
1616 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
1618 flags |= DPIF_FP_ZERO_STATS;
1619 facet->dp_packet_count = 0;
1620 facet->dp_byte_count = 0;
1623 ofpbuf_use_stack(&key, keybuf, sizeof keybuf);
1624 odp_flow_key_from_flow(&key, &facet->flow);
1625 assert(key.base == keybuf);
1627 return dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
1628 actions, actions_len, stats);
1631 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
1632 * 'zero_stats' is true, clears any existing statistics from the datapath for
1635 facet_install(struct ofproto *p, struct facet *facet, bool zero_stats)
1637 struct dpif_flow_stats stats;
1639 if (facet->may_install
1640 && !facet_put__(p, facet, facet->actions, facet->actions_len,
1641 zero_stats ? &stats : NULL)) {
1642 facet->installed = true;
1646 /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up
1647 * to the accounting hook function in the ofhooks structure. */
1649 facet_account(struct ofproto *ofproto,
1650 struct facet *facet, uint64_t extra_bytes)
1652 uint64_t total_bytes = facet->byte_count + extra_bytes;
1654 if (ofproto->ofhooks->account_flow_cb
1655 && total_bytes > facet->accounted_bytes)
1657 ofproto->ofhooks->account_flow_cb(
1658 &facet->flow, facet->tags, facet->actions, facet->actions_len,
1659 total_bytes - facet->accounted_bytes, ofproto->aux);
1660 facet->accounted_bytes = total_bytes;
1664 /* If 'rule' is installed in the datapath, uninstalls it. */
1666 facet_uninstall(struct ofproto *p, struct facet *facet)
1668 if (facet->installed) {
1669 uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S];
1670 struct dpif_flow_stats stats;
1673 ofpbuf_use_stack(&key, keybuf, sizeof keybuf);
1674 odp_flow_key_from_flow(&key, &facet->flow);
1675 assert(key.base == keybuf);
1677 if (!dpif_flow_del(p->dpif, key.data, key.size, &stats)) {
1678 facet_update_stats(p, facet, &stats);
1680 facet->installed = false;
1681 facet->dp_packet_count = 0;
1682 facet->dp_byte_count = 0;
1684 assert(facet->dp_packet_count == 0);
1685 assert(facet->dp_byte_count == 0);
1689 /* Returns true if the only action for 'facet' is to send to the controller.
1690 * (We don't report NetFlow expiration messages for such facets because they
1691 * are just part of the control logic for the network, not real traffic). */
1693 facet_is_controller_flow(struct facet *facet)
1696 && facet->rule->n_actions == 1
1697 && action_outputs_to_port(&facet->rule->actions[0],
1698 htons(OFPP_CONTROLLER)));
1701 /* Folds all of 'facet''s statistics into its rule. Also updates the
1702 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
1703 * 'facet''s statistics in the datapath should have been zeroed and folded into
1704 * its packet and byte counts before this function is called. */
1706 facet_flush_stats(struct ofproto *ofproto, struct facet *facet)
1708 assert(!facet->dp_byte_count);
1709 assert(!facet->dp_packet_count);
1711 facet_push_stats(ofproto, facet);
1712 facet_account(ofproto, facet, 0);
1714 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
1715 struct ofexpired expired;
1716 expired.flow = facet->flow;
1717 expired.packet_count = facet->packet_count;
1718 expired.byte_count = facet->byte_count;
1719 expired.used = facet->used;
1720 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
1723 facet->rule->packet_count += facet->packet_count;
1724 facet->rule->byte_count += facet->byte_count;
1726 /* Reset counters to prevent double counting if 'facet' ever gets
1728 facet->packet_count = 0;
1729 facet->byte_count = 0;
1730 facet->rs_packet_count = 0;
1731 facet->rs_byte_count = 0;
1732 facet->accounted_bytes = 0;
1734 netflow_flow_clear(&facet->nf_flow);
1737 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
1738 * Returns it if found, otherwise a null pointer.
1740 * The returned facet might need revalidation; use facet_lookup_valid()
1741 * instead if that is important. */
1742 static struct facet *
1743 facet_find(struct ofproto *ofproto, const struct flow *flow)
1745 struct facet *facet;
1747 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
1749 if (flow_equal(flow, &facet->flow)) {
1757 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
1758 * Returns it if found, otherwise a null pointer.
1760 * The returned facet is guaranteed to be valid. */
1761 static struct facet *
1762 facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow)
1764 struct facet *facet = facet_find(ofproto, flow);
1766 /* The facet we found might not be valid, since we could be in need of
1767 * revalidation. If it is not valid, don't return it. */
1769 && ofproto->need_revalidate
1770 && !facet_revalidate(ofproto, facet)) {
1771 COVERAGE_INC(ofproto_invalidated);
1778 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
1780 * - If the rule found is different from 'facet''s current rule, moves
1781 * 'facet' to the new rule and recompiles its actions.
1783 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
1784 * where it is and recompiles its actions anyway.
1786 * - If there is none, destroys 'facet'.
1788 * Returns true if 'facet' still exists, false if it has been destroyed. */
1790 facet_revalidate(struct ofproto *ofproto, struct facet *facet)
1792 struct action_xlate_ctx ctx;
1793 struct ofpbuf *odp_actions;
1794 struct rule *new_rule;
1795 bool actions_changed;
1797 COVERAGE_INC(facet_revalidate);
1799 /* Determine the new rule. */
1800 new_rule = rule_lookup(ofproto, &facet->flow);
1802 /* No new rule, so delete the facet. */
1803 facet_remove(ofproto, facet);
1807 /* Calculate new ODP actions.
1809 * We do not modify any 'facet' state yet, because we might need to, e.g.,
1810 * emit a NetFlow expiration and, if so, we need to have the old state
1811 * around to properly compose it. */
1812 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
1813 odp_actions = xlate_actions(&ctx, new_rule->actions, new_rule->n_actions);
1814 actions_changed = (facet->actions_len != odp_actions->size
1815 || memcmp(facet->actions, odp_actions->data,
1816 facet->actions_len));
1818 /* If the ODP actions changed or the installability changed, then we need
1819 * to talk to the datapath. */
1820 if (actions_changed || ctx.may_set_up_flow != facet->installed) {
1821 if (ctx.may_set_up_flow) {
1822 struct dpif_flow_stats stats;
1824 facet_put__(ofproto, facet,
1825 odp_actions->data, odp_actions->size, &stats);
1826 facet_update_stats(ofproto, facet, &stats);
1828 facet_uninstall(ofproto, facet);
1831 /* The datapath flow is gone or has zeroed stats, so push stats out of
1832 * 'facet' into 'rule'. */
1833 facet_flush_stats(ofproto, facet);
1836 /* Update 'facet' now that we've taken care of all the old state. */
1837 facet->tags = ctx.tags;
1838 facet->nf_flow.output_iface = ctx.nf_output_iface;
1839 facet->may_install = ctx.may_set_up_flow;
1840 if (actions_changed) {
1841 free(facet->actions);
1842 facet->actions_len = odp_actions->size;
1843 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
1845 if (facet->rule != new_rule) {
1846 COVERAGE_INC(facet_changed_rule);
1847 list_remove(&facet->list_node);
1848 list_push_back(&new_rule->facets, &facet->list_node);
1849 facet->rule = new_rule;
1850 facet->used = new_rule->created;
1851 facet->rs_used = facet->used;
1854 ofpbuf_delete(odp_actions);
1860 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
1863 struct ofpbuf *buf = ofputil_encode_error_msg(error, oh);
1865 COVERAGE_INC(ofproto_error);
1866 ofconn_send_reply(ofconn, buf);
1871 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
1873 ofconn_send_reply(ofconn, make_echo_reply(oh));
1878 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
1880 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1881 struct ofp_switch_features *osf;
1883 struct ofport *port;
1885 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1886 osf->datapath_id = htonll(ofproto->datapath_id);
1887 osf->n_buffers = htonl(pktbuf_capacity());
1889 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1890 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
1891 osf->actions = htonl((1u << OFPAT_OUTPUT) |
1892 (1u << OFPAT_SET_VLAN_VID) |
1893 (1u << OFPAT_SET_VLAN_PCP) |
1894 (1u << OFPAT_STRIP_VLAN) |
1895 (1u << OFPAT_SET_DL_SRC) |
1896 (1u << OFPAT_SET_DL_DST) |
1897 (1u << OFPAT_SET_NW_SRC) |
1898 (1u << OFPAT_SET_NW_DST) |
1899 (1u << OFPAT_SET_NW_TOS) |
1900 (1u << OFPAT_SET_TP_SRC) |
1901 (1u << OFPAT_SET_TP_DST) |
1902 (1u << OFPAT_ENQUEUE));
1904 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
1905 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
1908 ofconn_send_reply(ofconn, buf);
1913 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
1915 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1917 struct ofp_switch_config *osc;
1921 /* Figure out flags. */
1922 dpif_get_drop_frags(ofproto->dpif, &drop_frags);
1923 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1926 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1927 osc->flags = htons(flags);
1928 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
1929 ofconn_send_reply(ofconn, buf);
1935 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
1937 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1938 uint16_t flags = ntohs(osc->flags);
1940 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
1941 && ofconn_get_role(ofconn) != NX_ROLE_SLAVE) {
1942 switch (flags & OFPC_FRAG_MASK) {
1943 case OFPC_FRAG_NORMAL:
1944 dpif_set_drop_frags(ofproto->dpif, false);
1946 case OFPC_FRAG_DROP:
1947 dpif_set_drop_frags(ofproto->dpif, true);
1950 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1956 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
1961 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
1962 struct action_xlate_ctx *ctx);
1965 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
1967 const struct ofport *ofport = get_port(ctx->ofproto, port);
1970 if (ofport->opp.config & OFPPC_NO_FWD) {
1971 /* Forwarding disabled on port. */
1976 * We don't have an ofport record for this port, but it doesn't hurt to
1977 * allow forwarding to it anyhow. Maybe such a port will appear later
1978 * and we're pre-populating the flow table.
1982 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, port);
1983 ctx->nf_output_iface = port;
1986 static struct rule *
1987 rule_lookup(struct ofproto *ofproto, const struct flow *flow)
1989 return rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
1993 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
1995 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
1996 uint16_t old_in_port;
1999 /* Look up a flow with 'in_port' as the input port. Then restore the
2000 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2001 * have surprising behavior). */
2002 old_in_port = ctx->flow.in_port;
2003 ctx->flow.in_port = in_port;
2004 rule = rule_lookup(ctx->ofproto, &ctx->flow);
2005 ctx->flow.in_port = old_in_port;
2007 if (ctx->resubmit_hook) {
2008 ctx->resubmit_hook(ctx, rule);
2013 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2017 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2019 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2020 MAX_RESUBMIT_RECURSION);
2025 flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, uint32_t mask,
2026 uint16_t *nf_output_iface, struct ofpbuf *odp_actions)
2028 struct ofport *ofport;
2030 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
2031 uint16_t odp_port = ofport->odp_port;
2032 if (odp_port != odp_in_port && !(ofport->opp.config & mask)) {
2033 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port);
2036 *nf_output_iface = NF_OUT_FLOOD;
2040 xlate_output_action__(struct action_xlate_ctx *ctx,
2041 uint16_t port, uint16_t max_len)
2044 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2046 ctx->nf_output_iface = NF_OUT_DROP;
2050 add_output_action(ctx, ctx->flow.in_port);
2053 xlate_table_action(ctx, ctx->flow.in_port);
2056 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2057 ctx->odp_actions, &ctx->tags,
2058 &ctx->nf_output_iface,
2059 ctx->ofproto->aux)) {
2060 COVERAGE_INC(ofproto_uninstallable);
2061 ctx->may_set_up_flow = false;
2065 flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD,
2066 &ctx->nf_output_iface, ctx->odp_actions);
2069 flood_packets(ctx->ofproto, ctx->flow.in_port, 0,
2070 &ctx->nf_output_iface, ctx->odp_actions);
2072 case OFPP_CONTROLLER:
2073 nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_CONTROLLER, max_len);
2076 add_output_action(ctx, ODPP_LOCAL);
2079 odp_port = ofp_port_to_odp_port(port);
2080 if (odp_port != ctx->flow.in_port) {
2081 add_output_action(ctx, odp_port);
2086 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2087 ctx->nf_output_iface = NF_OUT_FLOOD;
2088 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2089 ctx->nf_output_iface = prev_nf_output_iface;
2090 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2091 ctx->nf_output_iface != NF_OUT_FLOOD) {
2092 ctx->nf_output_iface = NF_OUT_MULTI;
2097 xlate_output_action(struct action_xlate_ctx *ctx,
2098 const struct ofp_action_output *oao)
2100 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2103 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2104 * optimization, because we're going to add another action that sets the
2105 * priority immediately after, or because there are no actions following the
2108 remove_pop_action(struct action_xlate_ctx *ctx)
2110 if (ctx->odp_actions->size == ctx->last_pop_priority) {
2111 ctx->odp_actions->size -= NLA_ALIGN(NLA_HDRLEN);
2112 ctx->last_pop_priority = -1;
2117 add_pop_action(struct action_xlate_ctx *ctx)
2119 if (ctx->odp_actions->size != ctx->last_pop_priority) {
2120 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_POP_PRIORITY);
2121 ctx->last_pop_priority = ctx->odp_actions->size;
2126 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2127 const struct ofp_action_enqueue *oae)
2129 uint16_t ofp_port, odp_port;
2133 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2136 /* Fall back to ordinary output action. */
2137 xlate_output_action__(ctx, ntohs(oae->port), 0);
2141 /* Figure out ODP output port. */
2142 ofp_port = ntohs(oae->port);
2143 if (ofp_port != OFPP_IN_PORT) {
2144 odp_port = ofp_port_to_odp_port(ofp_port);
2146 odp_port = ctx->flow.in_port;
2149 /* Add ODP actions. */
2150 remove_pop_action(ctx);
2151 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2152 add_output_action(ctx, odp_port);
2153 add_pop_action(ctx);
2155 /* Update NetFlow output port. */
2156 if (ctx->nf_output_iface == NF_OUT_DROP) {
2157 ctx->nf_output_iface = odp_port;
2158 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2159 ctx->nf_output_iface = NF_OUT_MULTI;
2164 xlate_set_queue_action(struct action_xlate_ctx *ctx,
2165 const struct nx_action_set_queue *nasq)
2170 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
2173 /* Couldn't translate queue to a priority, so ignore. A warning
2174 * has already been logged. */
2178 remove_pop_action(ctx);
2179 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2183 xlate_set_dl_tci(struct action_xlate_ctx *ctx)
2185 ovs_be16 tci = ctx->flow.vlan_tci;
2186 if (!(tci & htons(VLAN_CFI))) {
2187 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN);
2189 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
2190 tci & ~htons(VLAN_CFI));
2194 struct xlate_reg_state {
2200 save_reg_state(const struct action_xlate_ctx *ctx,
2201 struct xlate_reg_state *state)
2203 state->vlan_tci = ctx->flow.vlan_tci;
2204 state->tun_id = ctx->flow.tun_id;
2208 update_reg_state(struct action_xlate_ctx *ctx,
2209 const struct xlate_reg_state *state)
2211 if (ctx->flow.vlan_tci != state->vlan_tci) {
2212 xlate_set_dl_tci(ctx);
2214 if (ctx->flow.tun_id != state->tun_id) {
2215 nl_msg_put_be64(ctx->odp_actions,
2216 ODP_ACTION_ATTR_SET_TUNNEL, ctx->flow.tun_id);
2221 xlate_nicira_action(struct action_xlate_ctx *ctx,
2222 const struct nx_action_header *nah)
2224 const struct nx_action_resubmit *nar;
2225 const struct nx_action_set_tunnel *nast;
2226 const struct nx_action_set_queue *nasq;
2227 const struct nx_action_multipath *nam;
2228 enum nx_action_subtype subtype = ntohs(nah->subtype);
2229 struct xlate_reg_state state;
2232 assert(nah->vendor == htonl(NX_VENDOR_ID));
2234 case NXAST_RESUBMIT:
2235 nar = (const struct nx_action_resubmit *) nah;
2236 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2239 case NXAST_SET_TUNNEL:
2240 nast = (const struct nx_action_set_tunnel *) nah;
2241 tun_id = htonll(ntohl(nast->tun_id));
2242 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
2243 ctx->flow.tun_id = tun_id;
2246 case NXAST_DROP_SPOOFED_ARP:
2247 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
2248 nl_msg_put_flag(ctx->odp_actions,
2249 ODP_ACTION_ATTR_DROP_SPOOFED_ARP);
2253 case NXAST_SET_QUEUE:
2254 nasq = (const struct nx_action_set_queue *) nah;
2255 xlate_set_queue_action(ctx, nasq);
2258 case NXAST_POP_QUEUE:
2259 add_pop_action(ctx);
2262 case NXAST_REG_MOVE:
2263 save_reg_state(ctx, &state);
2264 nxm_execute_reg_move((const struct nx_action_reg_move *) nah,
2266 update_reg_state(ctx, &state);
2269 case NXAST_REG_LOAD:
2270 save_reg_state(ctx, &state);
2271 nxm_execute_reg_load((const struct nx_action_reg_load *) nah,
2273 update_reg_state(ctx, &state);
2277 /* Nothing to do. */
2280 case NXAST_SET_TUNNEL64:
2281 tun_id = ((const struct nx_action_set_tunnel64 *) nah)->tun_id;
2282 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
2283 ctx->flow.tun_id = tun_id;
2286 case NXAST_MULTIPATH:
2287 nam = (const struct nx_action_multipath *) nah;
2288 multipath_execute(nam, &ctx->flow);
2291 /* If you add a new action here that modifies flow data, don't forget to
2292 * update the flow key in ctx->flow at the same time. */
2294 case NXAST_SNAT__OBSOLETE:
2296 VLOG_DBG_RL(&rl, "unknown Nicira action type %d", (int) subtype);
2302 do_xlate_actions(const union ofp_action *in, size_t n_in,
2303 struct action_xlate_ctx *ctx)
2305 struct actions_iterator iter;
2306 const union ofp_action *ia;
2307 const struct ofport *port;
2309 port = get_port(ctx->ofproto, ctx->flow.in_port);
2310 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2311 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
2312 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2313 /* Drop this flow. */
2317 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2318 enum ofp_action_type type = ntohs(ia->type);
2319 const struct ofp_action_dl_addr *oada;
2323 xlate_output_action(ctx, &ia->output);
2326 case OFPAT_SET_VLAN_VID:
2327 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
2328 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
2329 xlate_set_dl_tci(ctx);
2332 case OFPAT_SET_VLAN_PCP:
2333 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
2334 ctx->flow.vlan_tci |= htons(
2335 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
2336 xlate_set_dl_tci(ctx);
2339 case OFPAT_STRIP_VLAN:
2340 ctx->flow.vlan_tci = htons(0);
2341 xlate_set_dl_tci(ctx);
2344 case OFPAT_SET_DL_SRC:
2345 oada = ((struct ofp_action_dl_addr *) ia);
2346 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_SRC,
2347 oada->dl_addr, ETH_ADDR_LEN);
2348 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
2351 case OFPAT_SET_DL_DST:
2352 oada = ((struct ofp_action_dl_addr *) ia);
2353 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_DST,
2354 oada->dl_addr, ETH_ADDR_LEN);
2355 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
2358 case OFPAT_SET_NW_SRC:
2359 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_SRC,
2360 ia->nw_addr.nw_addr);
2361 ctx->flow.nw_src = ia->nw_addr.nw_addr;
2364 case OFPAT_SET_NW_DST:
2365 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_DST,
2366 ia->nw_addr.nw_addr);
2367 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
2370 case OFPAT_SET_NW_TOS:
2371 nl_msg_put_u8(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_TOS,
2373 ctx->flow.nw_tos = ia->nw_tos.nw_tos;
2376 case OFPAT_SET_TP_SRC:
2377 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_SRC,
2378 ia->tp_port.tp_port);
2379 ctx->flow.tp_src = ia->tp_port.tp_port;
2382 case OFPAT_SET_TP_DST:
2383 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_DST,
2384 ia->tp_port.tp_port);
2385 ctx->flow.tp_dst = ia->tp_port.tp_port;
2389 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2393 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
2397 VLOG_DBG_RL(&rl, "unknown action type %d", (int) type);
2404 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
2405 struct ofproto *ofproto, const struct flow *flow,
2406 const struct ofpbuf *packet)
2408 ctx->ofproto = ofproto;
2410 ctx->packet = packet;
2411 ctx->resubmit_hook = NULL;
2412 ctx->check_special = true;
2416 ofproto_process_cfm(struct ofproto *ofproto, const struct flow *flow,
2417 const struct ofpbuf *packet)
2419 struct ofport *ofport;
2421 ofport = get_port(ofproto, flow->in_port);
2422 if (ofport && ofport->cfm) {
2423 cfm_process_heartbeat(ofport->cfm, packet);
2427 static struct ofpbuf *
2428 xlate_actions(struct action_xlate_ctx *ctx,
2429 const union ofp_action *in, size_t n_in)
2431 COVERAGE_INC(ofproto_ofp2odp);
2433 ctx->odp_actions = ofpbuf_new(512);
2435 ctx->may_set_up_flow = true;
2436 ctx->nf_output_iface = NF_OUT_DROP;
2438 ctx->last_pop_priority = -1;
2440 if (ctx->check_special && cfm_should_process_flow(&ctx->flow)) {
2442 ofproto_process_cfm(ctx->ofproto, &ctx->flow, ctx->packet);
2444 ctx->may_set_up_flow = false;
2445 } else if (ctx->check_special
2446 && ctx->ofproto->ofhooks->special_cb
2447 && !ctx->ofproto->ofhooks->special_cb(&ctx->flow, ctx->packet,
2448 ctx->ofproto->aux)) {
2449 ctx->may_set_up_flow = false;
2451 do_xlate_actions(in, n_in, ctx);
2454 remove_pop_action(ctx);
2456 /* Check with in-band control to see if we're allowed to set up this
2458 if (!connmgr_may_set_up_flow(ctx->ofproto->connmgr, &ctx->flow,
2459 ctx->odp_actions->data,
2460 ctx->odp_actions->size)) {
2461 ctx->may_set_up_flow = false;
2464 return ctx->odp_actions;
2467 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2468 * error message code (composed with ofp_mkerr()) for the caller to propagate
2469 * upward. Otherwise, returns 0.
2471 * The log message mentions 'msg_type'. */
2473 reject_slave_controller(struct ofconn *ofconn, const const char *msg_type)
2475 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2476 && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
2477 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
2478 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
2481 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2488 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2490 struct ofproto *p = ofconn_get_ofproto(ofconn);
2491 struct ofp_packet_out *opo;
2492 struct ofpbuf payload, *buffer;
2493 union ofp_action *ofp_actions;
2494 struct action_xlate_ctx ctx;
2495 struct ofpbuf *odp_actions;
2496 struct ofpbuf request;
2498 size_t n_ofp_actions;
2502 COVERAGE_INC(ofproto_packet_out);
2504 error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT");
2509 /* Get ofp_packet_out. */
2510 ofpbuf_use_const(&request, oh, ntohs(oh->length));
2511 opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions));
2514 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
2515 &ofp_actions, &n_ofp_actions);
2521 if (opo->buffer_id != htonl(UINT32_MAX)) {
2522 error = ofconn_pktbuf_retrieve(ofconn, ntohl(opo->buffer_id),
2524 if (error || !buffer) {
2533 /* Extract flow, check actions. */
2534 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)),
2536 error = validate_actions(ofp_actions, n_ofp_actions, &flow, p->max_ports);
2542 action_xlate_ctx_init(&ctx, p, &flow, &payload);
2543 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
2544 dpif_execute(p->dpif, odp_actions->data, odp_actions->size, &payload);
2545 ofpbuf_delete(odp_actions);
2548 ofpbuf_delete(buffer);
2553 update_port_config(struct ofproto *p, struct ofport *port,
2554 uint32_t config, uint32_t mask)
2556 mask &= config ^ port->opp.config;
2557 if (mask & OFPPC_PORT_DOWN) {
2558 if (config & OFPPC_PORT_DOWN) {
2559 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2561 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2564 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
2565 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
2566 if (mask & REVALIDATE_BITS) {
2567 COVERAGE_INC(ofproto_costly_flags);
2568 port->opp.config ^= mask & REVALIDATE_BITS;
2569 p->need_revalidate = true;
2571 #undef REVALIDATE_BITS
2572 if (mask & OFPPC_NO_PACKET_IN) {
2573 port->opp.config ^= OFPPC_NO_PACKET_IN;
2578 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2580 struct ofproto *p = ofconn_get_ofproto(ofconn);
2581 const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh;
2582 struct ofport *port;
2585 error = reject_slave_controller(ofconn, "OFPT_PORT_MOD");
2590 port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no)));
2592 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2593 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2594 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2596 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2597 if (opm->advertise) {
2598 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2604 static struct ofpbuf *
2605 make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len)
2607 struct ofp_stats_reply *osr;
2610 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2611 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2613 osr->flags = htons(0);
2617 static struct ofpbuf *
2618 start_ofp_stats_reply(const struct ofp_header *request, size_t body_len)
2620 const struct ofp_stats_request *osr
2621 = (const struct ofp_stats_request *) request;
2622 return make_ofp_stats_reply(osr->header.xid, osr->type, body_len);
2626 append_ofp_stats_reply(size_t nbytes, struct ofconn *ofconn,
2627 struct ofpbuf **msgp)
2629 struct ofpbuf *msg = *msgp;
2630 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2631 if (nbytes + msg->size > UINT16_MAX) {
2632 struct ofp_stats_reply *reply = msg->data;
2633 reply->flags = htons(OFPSF_REPLY_MORE);
2634 *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes);
2635 ofconn_send_reply(ofconn, msg);
2637 return ofpbuf_put_uninit(*msgp, nbytes);
2640 static struct ofpbuf *
2641 make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len)
2643 struct nicira_stats_msg *nsm;
2646 msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX));
2647 nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg);
2648 nsm->type = htons(OFPST_VENDOR);
2649 nsm->flags = htons(0);
2650 nsm->vendor = htonl(NX_VENDOR_ID);
2651 nsm->subtype = subtype;
2655 static struct ofpbuf *
2656 start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len)
2658 return make_nxstats_reply(request->header.xid, request->subtype, body_len);
2662 append_nxstats_reply(size_t nbytes, struct ofconn *ofconn,
2663 struct ofpbuf **msgp)
2665 struct ofpbuf *msg = *msgp;
2666 assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg));
2667 if (nbytes + msg->size > UINT16_MAX) {
2668 struct nicira_stats_msg *reply = msg->data;
2669 reply->flags = htons(OFPSF_REPLY_MORE);
2670 *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes);
2671 ofconn_send_reply(ofconn, msg);
2673 ofpbuf_prealloc_tailroom(*msgp, nbytes);
2677 handle_desc_stats_request(struct ofconn *ofconn,
2678 const struct ofp_header *request)
2680 struct ofproto *p = ofconn_get_ofproto(ofconn);
2681 struct ofp_desc_stats *ods;
2684 msg = start_ofp_stats_reply(request, sizeof *ods);
2685 ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg);
2686 memset(ods, 0, sizeof *ods);
2687 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2688 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2689 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
2690 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
2691 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2692 ofconn_send_reply(ofconn, msg);
2698 handle_table_stats_request(struct ofconn *ofconn,
2699 const struct ofp_header *request)
2701 struct ofproto *p = ofconn_get_ofproto(ofconn);
2702 struct ofp_table_stats *ots;
2705 msg = start_ofp_stats_reply(request, sizeof *ots * 2);
2707 /* Classifier table. */
2708 ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg);
2709 memset(ots, 0, sizeof *ots);
2710 strcpy(ots->name, "classifier");
2711 ots->wildcards = (ofconn_get_flow_format(ofconn) == NXFF_OPENFLOW10
2712 ? htonl(OFPFW_ALL) : htonl(OVSFW_ALL));
2713 ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */
2714 ots->active_count = htonl(classifier_count(&p->cls));
2715 put_32aligned_be64(&ots->lookup_count, htonll(0)); /* XXX */
2716 put_32aligned_be64(&ots->matched_count, htonll(0)); /* XXX */
2718 ofconn_send_reply(ofconn, msg);
2723 append_port_stat(struct ofport *port, struct ofconn *ofconn,
2724 struct ofpbuf **msgp)
2726 struct netdev_stats stats;
2727 struct ofp_port_stats *ops;
2729 /* Intentionally ignore return value, since errors will set
2730 * 'stats' to all-1s, which is correct for OpenFlow, and
2731 * netdev_get_stats() will log errors. */
2732 netdev_get_stats(port->netdev, &stats);
2734 ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp);
2735 ops->port_no = htons(port->opp.port_no);
2736 memset(ops->pad, 0, sizeof ops->pad);
2737 put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
2738 put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
2739 put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
2740 put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
2741 put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
2742 put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
2743 put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
2744 put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
2745 put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
2746 put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
2747 put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
2748 put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
2752 handle_port_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
2754 struct ofproto *p = ofconn_get_ofproto(ofconn);
2755 const struct ofp_port_stats_request *psr = ofputil_stats_body(oh);
2756 struct ofp_port_stats *ops;
2758 struct ofport *port;
2760 msg = start_ofp_stats_reply(oh, sizeof *ops * 16);
2761 if (psr->port_no != htons(OFPP_NONE)) {
2762 port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no)));
2764 append_port_stat(port, ofconn, &msg);
2767 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2768 append_port_stat(port, ofconn, &msg);
2772 ofconn_send_reply(ofconn, msg);
2777 calc_flow_duration__(long long int start, uint32_t *sec, uint32_t *nsec)
2779 long long int msecs = time_msec() - start;
2780 *sec = msecs / 1000;
2781 *nsec = (msecs % 1000) * (1000 * 1000);
2785 calc_flow_duration(long long int start, ovs_be32 *sec_be, ovs_be32 *nsec_be)
2789 calc_flow_duration__(start, &sec, &nsec);
2790 *sec_be = htonl(sec);
2791 *nsec_be = htonl(nsec);
2795 put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule,
2796 ovs_be16 out_port, struct ofpbuf **replyp)
2798 struct ofp_flow_stats *ofs;
2799 uint64_t packet_count, byte_count;
2801 size_t act_len, len;
2803 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
2807 act_len = sizeof *rule->actions * rule->n_actions;
2808 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2810 rule_get_stats(rule, &packet_count, &byte_count);
2812 ofs = append_ofp_stats_reply(len, ofconn, replyp);
2813 ofs->length = htons(len);
2816 ofputil_cls_rule_to_match(&rule->cr, ofconn_get_flow_format(ofconn),
2817 &ofs->match, rule->flow_cookie, &cookie);
2818 put_32aligned_be64(&ofs->cookie, cookie);
2819 calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec);
2820 ofs->priority = htons(rule->cr.priority);
2821 ofs->idle_timeout = htons(rule->idle_timeout);
2822 ofs->hard_timeout = htons(rule->hard_timeout);
2823 memset(ofs->pad2, 0, sizeof ofs->pad2);
2824 put_32aligned_be64(&ofs->packet_count, htonll(packet_count));
2825 put_32aligned_be64(&ofs->byte_count, htonll(byte_count));
2826 if (rule->n_actions > 0) {
2827 memcpy(ofs->actions, rule->actions, act_len);
2832 is_valid_table(uint8_t table_id)
2834 if (table_id == 0 || table_id == 0xff) {
2837 /* It would probably be better to reply with an error but there doesn't
2838 * seem to be any appropriate value, so that might just be
2840 VLOG_WARN_RL(&rl, "controller asked for invalid table %"PRIu8,
2847 handle_flow_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
2849 const struct ofp_flow_stats_request *fsr = ofputil_stats_body(oh);
2850 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2851 struct ofpbuf *reply;
2853 COVERAGE_INC(ofproto_flows_req);
2854 reply = start_ofp_stats_reply(oh, 1024);
2855 if (is_valid_table(fsr->table_id)) {
2856 struct cls_cursor cursor;
2857 struct cls_rule target;
2860 ofputil_cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0,
2862 cls_cursor_init(&cursor, &ofproto->cls, &target);
2863 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2864 put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply);
2867 ofconn_send_reply(ofconn, reply);
2873 put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule,
2874 ovs_be16 out_port, struct ofpbuf **replyp)
2876 struct nx_flow_stats *nfs;
2877 uint64_t packet_count, byte_count;
2878 size_t act_len, start_len;
2879 struct ofpbuf *reply;
2881 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
2885 rule_get_stats(rule, &packet_count, &byte_count);
2887 act_len = sizeof *rule->actions * rule->n_actions;
2889 append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, ofconn, replyp);
2890 start_len = (*replyp)->size;
2893 nfs = ofpbuf_put_uninit(reply, sizeof *nfs);
2896 calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec);
2897 nfs->cookie = rule->flow_cookie;
2898 nfs->priority = htons(rule->cr.priority);
2899 nfs->idle_timeout = htons(rule->idle_timeout);
2900 nfs->hard_timeout = htons(rule->hard_timeout);
2901 nfs->match_len = htons(nx_put_match(reply, &rule->cr));
2902 memset(nfs->pad2, 0, sizeof nfs->pad2);
2903 nfs->packet_count = htonll(packet_count);
2904 nfs->byte_count = htonll(byte_count);
2905 if (rule->n_actions > 0) {
2906 ofpbuf_put(reply, rule->actions, act_len);
2908 nfs->length = htons(reply->size - start_len);
2912 handle_nxst_flow(struct ofconn *ofconn, const struct ofp_header *oh)
2914 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2915 struct nx_flow_stats_request *nfsr;
2916 struct cls_rule target;
2917 struct ofpbuf *reply;
2921 ofpbuf_use_const(&b, oh, ntohs(oh->length));
2923 /* Dissect the message. */
2924 nfsr = ofpbuf_pull(&b, sizeof *nfsr);
2925 error = nx_pull_match(&b, ntohs(nfsr->match_len), 0, &target);
2930 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2933 COVERAGE_INC(ofproto_flows_req);
2934 reply = start_nxstats_reply(&nfsr->nsm, 1024);
2935 if (is_valid_table(nfsr->table_id)) {
2936 struct cls_cursor cursor;
2939 cls_cursor_init(&cursor, &ofproto->cls, &target);
2940 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2941 put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply);
2944 ofconn_send_reply(ofconn, reply);
2950 flow_stats_ds(struct rule *rule, struct ds *results)
2952 uint64_t packet_count, byte_count;
2953 size_t act_len = sizeof *rule->actions * rule->n_actions;
2955 rule_get_stats(rule, &packet_count, &byte_count);
2957 ds_put_format(results, "duration=%llds, ",
2958 (time_msec() - rule->created) / 1000);
2959 ds_put_format(results, "idle=%.3fs, ", (time_msec() - rule->used) / 1000.0);
2960 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2961 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2962 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2963 cls_rule_format(&rule->cr, results);
2964 ds_put_char(results, ',');
2966 ofp_print_actions(results, &rule->actions->header, act_len);
2968 ds_put_cstr(results, "drop");
2970 ds_put_cstr(results, "\n");
2973 /* Adds a pretty-printed description of all flows to 'results', including
2974 * hidden flows (e.g., set up by in-band control). */
2976 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2978 struct cls_cursor cursor;
2981 cls_cursor_init(&cursor, &p->cls, NULL);
2982 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2983 flow_stats_ds(rule, results);
2988 query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target,
2989 ovs_be16 out_port, uint8_t table_id,
2990 struct ofp_aggregate_stats_reply *oasr)
2992 uint64_t total_packets = 0;
2993 uint64_t total_bytes = 0;
2996 COVERAGE_INC(ofproto_agg_request);
2998 if (is_valid_table(table_id)) {
2999 struct cls_cursor cursor;
3002 cls_cursor_init(&cursor, &ofproto->cls, target);
3003 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3004 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
3005 uint64_t packet_count;
3006 uint64_t byte_count;
3008 rule_get_stats(rule, &packet_count, &byte_count);
3010 total_packets += packet_count;
3011 total_bytes += byte_count;
3017 oasr->flow_count = htonl(n_flows);
3018 put_32aligned_be64(&oasr->packet_count, htonll(total_packets));
3019 put_32aligned_be64(&oasr->byte_count, htonll(total_bytes));
3020 memset(oasr->pad, 0, sizeof oasr->pad);
3024 handle_aggregate_stats_request(struct ofconn *ofconn,
3025 const struct ofp_header *oh)
3027 const struct ofp_aggregate_stats_request *request = ofputil_stats_body(oh);
3028 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3029 struct ofp_aggregate_stats_reply *reply;
3030 struct cls_rule target;
3033 ofputil_cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0,
3036 msg = start_ofp_stats_reply(oh, sizeof *reply);
3037 reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg);
3038 query_aggregate_stats(ofproto, &target, request->out_port,
3039 request->table_id, reply);
3040 ofconn_send_reply(ofconn, msg);
3045 handle_nxst_aggregate(struct ofconn *ofconn, const struct ofp_header *oh)
3047 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3048 struct nx_aggregate_stats_request *request;
3049 struct ofp_aggregate_stats_reply *reply;
3050 struct cls_rule target;
3055 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3057 /* Dissect the message. */
3058 request = ofpbuf_pull(&b, sizeof *request);
3059 error = nx_pull_match(&b, ntohs(request->match_len), 0, &target);
3064 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3068 COVERAGE_INC(ofproto_flows_req);
3069 buf = start_nxstats_reply(&request->nsm, sizeof *reply);
3070 reply = ofpbuf_put_uninit(buf, sizeof *reply);
3071 query_aggregate_stats(ofproto, &target, request->out_port,
3072 request->table_id, reply);
3073 ofconn_send_reply(ofconn, buf);
3078 struct queue_stats_cbdata {
3079 struct ofconn *ofconn;
3080 struct ofport *ofport;
3085 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3086 const struct netdev_queue_stats *stats)
3088 struct ofp_queue_stats *reply;
3090 reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3091 reply->port_no = htons(cbdata->ofport->opp.port_no);
3092 memset(reply->pad, 0, sizeof reply->pad);
3093 reply->queue_id = htonl(queue_id);
3094 put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
3095 put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
3096 put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
3100 handle_queue_stats_dump_cb(uint32_t queue_id,
3101 struct netdev_queue_stats *stats,
3104 struct queue_stats_cbdata *cbdata = cbdata_;
3106 put_queue_stats(cbdata, queue_id, stats);
3110 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3111 struct queue_stats_cbdata *cbdata)
3113 cbdata->ofport = port;
3114 if (queue_id == OFPQ_ALL) {
3115 netdev_dump_queue_stats(port->netdev,
3116 handle_queue_stats_dump_cb, cbdata);
3118 struct netdev_queue_stats stats;
3120 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3121 put_queue_stats(cbdata, queue_id, &stats);
3127 handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3129 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3130 const struct ofp_queue_stats_request *qsr;
3131 struct queue_stats_cbdata cbdata;
3132 struct ofport *port;
3133 unsigned int port_no;
3136 qsr = ofputil_stats_body(oh);
3138 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3141 COVERAGE_INC(ofproto_queue_req);
3143 cbdata.ofconn = ofconn;
3144 cbdata.msg = start_ofp_stats_reply(oh, 128);
3146 port_no = ntohs(qsr->port_no);
3147 queue_id = ntohl(qsr->queue_id);
3148 if (port_no == OFPP_ALL) {
3149 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3150 handle_queue_stats_for_port(port, queue_id, &cbdata);
3152 } else if (port_no < ofproto->max_ports) {
3153 port = get_port(ofproto, ofp_port_to_odp_port(port_no));
3155 handle_queue_stats_for_port(port, queue_id, &cbdata);
3158 ofpbuf_delete(cbdata.msg);
3159 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3161 ofconn_send_reply(ofconn, cbdata.msg);
3166 /* Updates 'facet''s used time. Caller is responsible for calling
3167 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3169 facet_update_time(struct ofproto *ofproto, struct facet *facet,
3172 if (used > facet->used) {
3174 if (used > facet->rule->used) {
3175 facet->rule->used = used;
3177 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3181 /* Folds the statistics from 'stats' into the counters in 'facet'.
3183 * Because of the meaning of a facet's counters, it only makes sense to do this
3184 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3185 * packet that was sent by hand or if it represents statistics that have been
3186 * cleared out of the datapath. */
3188 facet_update_stats(struct ofproto *ofproto, struct facet *facet,
3189 const struct dpif_flow_stats *stats)
3191 if (stats->n_packets || stats->used > facet->used) {
3192 facet_update_time(ofproto, facet, stats->used);
3193 facet->packet_count += stats->n_packets;
3194 facet->byte_count += stats->n_bytes;
3195 facet_push_stats(ofproto, facet);
3196 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3201 facet_push_stats(struct ofproto *ofproto, struct facet *facet)
3203 uint64_t rs_packets, rs_bytes;
3205 assert(facet->packet_count >= facet->rs_packet_count);
3206 assert(facet->byte_count >= facet->rs_byte_count);
3207 assert(facet->used >= facet->rs_used);
3209 rs_packets = facet->packet_count - facet->rs_packet_count;
3210 rs_bytes = facet->byte_count - facet->rs_byte_count;
3212 if (rs_packets || rs_bytes || facet->used > facet->rs_used) {
3213 facet->rs_packet_count = facet->packet_count;
3214 facet->rs_byte_count = facet->byte_count;
3215 facet->rs_used = facet->used;
3217 flow_push_stats(ofproto, facet->rule, &facet->flow,
3218 rs_packets, rs_bytes, facet->used);
3222 struct ofproto_push {
3223 struct action_xlate_ctx ctx;
3230 push_resubmit(struct action_xlate_ctx *ctx, struct rule *rule)
3232 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3235 rule->packet_count += push->packets;
3236 rule->byte_count += push->bytes;
3237 rule->used = MAX(push->used, rule->used);
3241 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3242 * 'rule''s actions. */
3244 flow_push_stats(struct ofproto *ofproto, const struct rule *rule,
3245 struct flow *flow, uint64_t packets, uint64_t bytes,
3248 struct ofproto_push push;
3250 push.packets = packets;
3254 action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL);
3255 push.ctx.resubmit_hook = push_resubmit;
3256 ofpbuf_delete(xlate_actions(&push.ctx, rule->actions, rule->n_actions));
3259 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3260 * in which no matching flow already exists in the flow table.
3262 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3263 * ofp_actions, to the ofproto's flow table. Returns 0 on success or an
3264 * OpenFlow error code as encoded by ofp_mkerr() on failure.
3266 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3269 add_flow(struct ofconn *ofconn, struct flow_mod *fm)
3271 struct ofproto *p = ofconn_get_ofproto(ofconn);
3272 struct ofpbuf *packet;
3277 if (fm->flags & OFPFF_CHECK_OVERLAP
3278 && classifier_rule_overlaps(&p->cls, &fm->cr)) {
3279 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3283 if (fm->buffer_id != UINT32_MAX) {
3284 error = ofconn_pktbuf_retrieve(ofconn, fm->buffer_id,
3288 in_port = UINT16_MAX;
3291 rule = rule_create(&fm->cr, fm->actions, fm->n_actions,
3292 fm->idle_timeout, fm->hard_timeout, fm->cookie,
3293 fm->flags & OFPFF_SEND_FLOW_REM);
3294 rule_insert(p, rule);
3296 rule_execute(p, rule, in_port, packet);
3301 static struct rule *
3302 find_flow_strict(struct ofproto *p, const struct flow_mod *fm)
3304 return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr));
3308 send_buffered_packet(struct ofconn *ofconn,
3309 struct rule *rule, uint32_t buffer_id)
3311 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3312 struct ofpbuf *packet;
3316 if (buffer_id == UINT32_MAX) {
3320 error = ofconn_pktbuf_retrieve(ofconn, buffer_id, &packet, &in_port);
3325 rule_execute(ofproto, rule, in_port, packet);
3330 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3332 struct modify_flows_cbdata {
3333 struct ofproto *ofproto;
3334 const struct flow_mod *fm;
3338 static int modify_flow(struct ofproto *, const struct flow_mod *,
3341 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3342 * encoded by ofp_mkerr() on failure.
3344 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3347 modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm)
3349 struct ofproto *p = ofconn_get_ofproto(ofconn);
3350 struct rule *match = NULL;
3351 struct cls_cursor cursor;
3354 cls_cursor_init(&cursor, &p->cls, &fm->cr);
3355 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3356 if (!rule_is_hidden(rule)) {
3358 modify_flow(p, fm, rule);
3363 /* This credits the packet to whichever flow happened to match last.
3364 * That's weird. Maybe we should do a lookup for the flow that
3365 * actually matches the packet? Who knows. */
3366 send_buffered_packet(ofconn, match, fm->buffer_id);
3369 return add_flow(ofconn, fm);
3373 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3374 * code as encoded by ofp_mkerr() on failure.
3376 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3379 modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm)
3381 struct ofproto *p = ofconn_get_ofproto(ofconn);
3382 struct rule *rule = find_flow_strict(p, fm);
3383 if (rule && !rule_is_hidden(rule)) {
3384 modify_flow(p, fm, rule);
3385 return send_buffered_packet(ofconn, rule, fm->buffer_id);
3387 return add_flow(ofconn, fm);
3391 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3392 * been identified as a flow in 'p''s flow table to be modified, by changing
3393 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3394 * ofp_action[] structures). */
3396 modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule)
3398 size_t actions_len = fm->n_actions * sizeof *rule->actions;
3400 rule->flow_cookie = fm->cookie;
3402 /* If the actions are the same, do nothing. */
3403 if (fm->n_actions == rule->n_actions
3405 || !memcmp(fm->actions, rule->actions, actions_len))) {
3409 /* Replace actions. */
3410 free(rule->actions);
3411 rule->actions = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL;
3412 rule->n_actions = fm->n_actions;
3414 p->need_revalidate = true;
3419 /* OFPFC_DELETE implementation. */
3421 static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port);
3423 /* Implements OFPFC_DELETE. */
3425 delete_flows_loose(struct ofproto *p, const struct flow_mod *fm)
3427 struct rule *rule, *next_rule;
3428 struct cls_cursor cursor;
3430 cls_cursor_init(&cursor, &p->cls, &fm->cr);
3431 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
3432 delete_flow(p, rule, htons(fm->out_port));
3436 /* Implements OFPFC_DELETE_STRICT. */
3438 delete_flow_strict(struct ofproto *p, struct flow_mod *fm)
3440 struct rule *rule = find_flow_strict(p, fm);
3442 delete_flow(p, rule, htons(fm->out_port));
3446 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3447 * been identified as a flow to delete from 'p''s flow table, by deleting the
3448 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3451 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3452 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3453 * specified 'out_port'. */
3455 delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port)
3457 if (rule_is_hidden(rule)) {
3461 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
3465 rule_send_removed(p, rule, OFPRR_DELETE);
3466 rule_remove(p, rule);
3470 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3472 struct ofproto *p = ofconn_get_ofproto(ofconn);
3476 error = reject_slave_controller(ofconn, "flow_mod");
3481 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_flow_format(ofconn));
3486 /* We do not support the emergency flow cache. It will hopefully get
3487 * dropped from OpenFlow in the near future. */
3488 if (fm.flags & OFPFF_EMERG) {
3489 /* There isn't a good fit for an error code, so just state that the
3490 * flow table is full. */
3491 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3494 error = validate_actions(fm.actions, fm.n_actions,
3495 &fm.cr.flow, p->max_ports);
3500 switch (fm.command) {
3502 return add_flow(ofconn, &fm);
3505 return modify_flows_loose(ofconn, &fm);
3507 case OFPFC_MODIFY_STRICT:
3508 return modify_flow_strict(ofconn, &fm);
3511 delete_flows_loose(p, &fm);
3514 case OFPFC_DELETE_STRICT:
3515 delete_flow_strict(p, &fm);
3519 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3524 handle_tun_id_from_cookie(struct ofconn *ofconn, const struct ofp_header *oh)
3526 const struct nxt_tun_id_cookie *msg
3527 = (const struct nxt_tun_id_cookie *) oh;
3528 enum nx_flow_format flow_format;
3530 flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10;
3531 ofconn_set_flow_format(ofconn, flow_format);
3537 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
3539 struct nx_role_request *nrr = (struct nx_role_request *) oh;
3540 struct nx_role_request *reply;
3544 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY) {
3545 VLOG_WARN_RL(&rl, "ignoring role request on service connection");
3546 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3549 role = ntohl(nrr->role);
3550 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3551 && role != NX_ROLE_SLAVE) {
3552 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
3554 /* There's no good error code for this. */
3555 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
3558 ofconn_set_role(ofconn, role);
3560 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
3561 reply->role = htonl(role);
3562 ofconn_send_reply(ofconn, buf);
3568 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
3570 const struct nxt_set_flow_format *msg
3571 = (const struct nxt_set_flow_format *) oh;
3574 format = ntohl(msg->format);
3575 if (format == NXFF_OPENFLOW10
3576 || format == NXFF_TUN_ID_FROM_COOKIE
3577 || format == NXFF_NXM) {
3578 ofconn_set_flow_format(ofconn, format);
3581 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3586 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
3588 struct ofp_header *ob;
3591 /* Currently, everything executes synchronously, so we can just
3592 * immediately send the barrier reply. */
3593 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
3594 ofconn_send_reply(ofconn, buf);
3599 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
3601 const struct ofp_header *oh = msg->data;
3602 const struct ofputil_msg_type *type;
3605 error = ofputil_decode_msg_type(oh, &type);
3610 switch (ofputil_msg_type_code(type)) {
3611 /* OpenFlow requests. */
3612 case OFPUTIL_OFPT_ECHO_REQUEST:
3613 return handle_echo_request(ofconn, oh);
3615 case OFPUTIL_OFPT_FEATURES_REQUEST:
3616 return handle_features_request(ofconn, oh);
3618 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
3619 return handle_get_config_request(ofconn, oh);
3621 case OFPUTIL_OFPT_SET_CONFIG:
3622 return handle_set_config(ofconn, msg->data);
3624 case OFPUTIL_OFPT_PACKET_OUT:
3625 return handle_packet_out(ofconn, oh);
3627 case OFPUTIL_OFPT_PORT_MOD:
3628 return handle_port_mod(ofconn, oh);
3630 case OFPUTIL_OFPT_FLOW_MOD:
3631 return handle_flow_mod(ofconn, oh);
3633 case OFPUTIL_OFPT_BARRIER_REQUEST:
3634 return handle_barrier_request(ofconn, oh);
3636 /* OpenFlow replies. */
3637 case OFPUTIL_OFPT_ECHO_REPLY:
3640 /* Nicira extension requests. */
3641 case OFPUTIL_NXT_TUN_ID_FROM_COOKIE:
3642 return handle_tun_id_from_cookie(ofconn, oh);
3644 case OFPUTIL_NXT_ROLE_REQUEST:
3645 return handle_role_request(ofconn, oh);
3647 case OFPUTIL_NXT_SET_FLOW_FORMAT:
3648 return handle_nxt_set_flow_format(ofconn, oh);
3650 case OFPUTIL_NXT_FLOW_MOD:
3651 return handle_flow_mod(ofconn, oh);
3653 /* OpenFlow statistics requests. */
3654 case OFPUTIL_OFPST_DESC_REQUEST:
3655 return handle_desc_stats_request(ofconn, oh);
3657 case OFPUTIL_OFPST_FLOW_REQUEST:
3658 return handle_flow_stats_request(ofconn, oh);
3660 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
3661 return handle_aggregate_stats_request(ofconn, oh);
3663 case OFPUTIL_OFPST_TABLE_REQUEST:
3664 return handle_table_stats_request(ofconn, oh);
3666 case OFPUTIL_OFPST_PORT_REQUEST:
3667 return handle_port_stats_request(ofconn, oh);
3669 case OFPUTIL_OFPST_QUEUE_REQUEST:
3670 return handle_queue_stats_request(ofconn, oh);
3672 /* Nicira extension statistics requests. */
3673 case OFPUTIL_NXST_FLOW_REQUEST:
3674 return handle_nxst_flow(ofconn, oh);
3676 case OFPUTIL_NXST_AGGREGATE_REQUEST:
3677 return handle_nxst_aggregate(ofconn, oh);
3679 case OFPUTIL_INVALID:
3680 case OFPUTIL_OFPT_HELLO:
3681 case OFPUTIL_OFPT_ERROR:
3682 case OFPUTIL_OFPT_FEATURES_REPLY:
3683 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
3684 case OFPUTIL_OFPT_PACKET_IN:
3685 case OFPUTIL_OFPT_FLOW_REMOVED:
3686 case OFPUTIL_OFPT_PORT_STATUS:
3687 case OFPUTIL_OFPT_BARRIER_REPLY:
3688 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
3689 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
3690 case OFPUTIL_OFPST_DESC_REPLY:
3691 case OFPUTIL_OFPST_FLOW_REPLY:
3692 case OFPUTIL_OFPST_QUEUE_REPLY:
3693 case OFPUTIL_OFPST_PORT_REPLY:
3694 case OFPUTIL_OFPST_TABLE_REPLY:
3695 case OFPUTIL_OFPST_AGGREGATE_REPLY:
3696 case OFPUTIL_NXT_ROLE_REPLY:
3697 case OFPUTIL_NXT_FLOW_REMOVED:
3698 case OFPUTIL_NXST_FLOW_REPLY:
3699 case OFPUTIL_NXST_AGGREGATE_REPLY:
3701 if (VLOG_IS_WARN_ENABLED()) {
3702 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3703 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3706 if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) {
3707 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
3709 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3715 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
3717 int error = handle_openflow__(ofconn, ofp_msg);
3719 send_error_oh(ofconn, ofp_msg->data, error);
3721 COVERAGE_INC(ofproto_recv_openflow);
3725 handle_miss_upcall(struct ofproto *p, struct dpif_upcall *upcall)
3727 struct facet *facet;
3730 /* Obtain in_port and tun_id, at least. */
3731 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
3733 /* Set header pointers in 'flow'. */
3734 flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow);
3736 if (cfm_should_process_flow(&flow)) {
3737 ofproto_process_cfm(p, &flow, upcall->packet);
3738 ofpbuf_delete(upcall->packet);
3740 } else if (p->ofhooks->special_cb
3741 && !p->ofhooks->special_cb(&flow, upcall->packet, p->aux)) {
3742 ofpbuf_delete(upcall->packet);
3746 /* Check with in-band control to see if this packet should be sent
3747 * to the local port regardless of the flow table. */
3748 if (connmgr_msg_in_hook(p->connmgr, &flow, upcall->packet)) {
3749 ofproto_send_packet(p, ODPP_LOCAL, 0, upcall->packet);
3752 facet = facet_lookup_valid(p, &flow);
3754 struct rule *rule = rule_lookup(p, &flow);
3756 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3757 struct ofport *port = get_port(p, flow.in_port);
3759 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3760 COVERAGE_INC(ofproto_no_packet_in);
3761 /* XXX install 'drop' flow entry */
3762 ofpbuf_delete(upcall->packet);
3766 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
3770 COVERAGE_INC(ofproto_packet_in);
3771 send_packet_in(p, upcall, &flow, false);
3775 facet = facet_create(p, rule, &flow, upcall->packet);
3776 } else if (!facet->may_install) {
3777 /* The facet is not installable, that is, we need to process every
3778 * packet, so process the current packet's actions into 'facet'. */
3779 facet_make_actions(p, facet, upcall->packet);
3782 if (facet->rule->cr.priority == FAIL_OPEN_PRIORITY) {
3784 * Extra-special case for fail-open mode.
3786 * We are in fail-open mode and the packet matched the fail-open rule,
3787 * but we are connected to a controller too. We should send the packet
3788 * up to the controller in the hope that it will try to set up a flow
3789 * and thereby allow us to exit fail-open.
3791 * See the top-level comment in fail-open.c for more information.
3793 send_packet_in(p, upcall, &flow, true);
3796 facet_execute(p, facet, upcall->packet);
3797 facet_install(p, facet, false);
3801 handle_upcall(struct ofproto *p, struct dpif_upcall *upcall)
3805 switch (upcall->type) {
3806 case DPIF_UC_ACTION:
3807 COVERAGE_INC(ofproto_ctlr_action);
3808 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
3809 send_packet_in(p, upcall, &flow, false);
3812 case DPIF_UC_SAMPLE:
3814 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
3815 ofproto_sflow_received(p->sflow, upcall, &flow);
3817 ofpbuf_delete(upcall->packet);
3821 handle_miss_upcall(p, upcall);
3824 case DPIF_N_UC_TYPES:
3826 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3831 /* Flow expiration. */
3833 static int ofproto_dp_max_idle(const struct ofproto *);
3834 static void ofproto_update_stats(struct ofproto *);
3835 static void rule_expire(struct ofproto *, struct rule *);
3836 static void ofproto_expire_facets(struct ofproto *, int dp_max_idle);
3838 /* This function is called periodically by ofproto_run(). Its job is to
3839 * collect updates for the flows that have been installed into the datapath,
3840 * most importantly when they last were used, and then use that information to
3841 * expire flows that have not been used recently.
3843 * Returns the number of milliseconds after which it should be called again. */
3845 ofproto_expire(struct ofproto *ofproto)
3847 struct rule *rule, *next_rule;
3848 struct cls_cursor cursor;
3851 /* Update stats for each flow in the datapath. */
3852 ofproto_update_stats(ofproto);
3854 /* Expire facets that have been idle too long. */
3855 dp_max_idle = ofproto_dp_max_idle(ofproto);
3856 ofproto_expire_facets(ofproto, dp_max_idle);
3858 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
3859 cls_cursor_init(&cursor, &ofproto->cls, NULL);
3860 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
3861 rule_expire(ofproto, rule);
3864 /* Let the hook know that we're at a stable point: all outstanding data
3865 * in existing flows has been accounted to the account_cb. Thus, the
3866 * hook can now reasonably do operations that depend on having accurate
3867 * flow volume accounting (currently, that's just bond rebalancing). */
3868 if (ofproto->ofhooks->account_checkpoint_cb) {
3869 ofproto->ofhooks->account_checkpoint_cb(ofproto->aux);
3872 return MIN(dp_max_idle, 1000);
3875 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3877 * This function also pushes statistics updates to rules which each facet
3878 * resubmits into. Generally these statistics will be accurate. However, if a
3879 * facet changes the rule it resubmits into at some time in between
3880 * ofproto_update_stats() runs, it is possible that statistics accrued to the
3881 * old rule will be incorrectly attributed to the new rule. This could be
3882 * avoided by calling ofproto_update_stats() whenever rules are created or
3883 * deleted. However, the performance impact of making so many calls to the
3884 * datapath do not justify the benefit of having perfectly accurate statistics.
3887 ofproto_update_stats(struct ofproto *p)
3889 const struct dpif_flow_stats *stats;
3890 struct dpif_flow_dump dump;
3891 const struct nlattr *key;
3894 dpif_flow_dump_start(&dump, p->dpif);
3895 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
3896 struct facet *facet;
3899 if (odp_flow_key_to_flow(key, key_len, &flow)) {
3903 odp_flow_key_format(key, key_len, &s);
3904 VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s",
3910 facet = facet_find(p, &flow);
3912 if (facet && facet->installed) {
3914 if (stats->n_packets >= facet->dp_packet_count) {
3915 facet->packet_count += stats->n_packets - facet->dp_packet_count;
3917 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3920 if (stats->n_bytes >= facet->dp_byte_count) {
3921 facet->byte_count += stats->n_bytes - facet->dp_byte_count;
3923 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3926 facet->dp_packet_count = stats->n_packets;
3927 facet->dp_byte_count = stats->n_bytes;
3929 facet_update_time(p, facet, stats->used);
3930 facet_account(p, facet, stats->n_bytes);
3931 facet_push_stats(p, facet);
3933 /* There's a flow in the datapath that we know nothing about.
3935 COVERAGE_INC(ofproto_unexpected_rule);
3936 dpif_flow_del(p->dpif, key, key_len, NULL);
3939 dpif_flow_dump_done(&dump);
3942 /* Calculates and returns the number of milliseconds of idle time after which
3943 * facets should expire from the datapath and we should fold their statistics
3944 * into their parent rules in userspace. */
3946 ofproto_dp_max_idle(const struct ofproto *ofproto)
3949 * Idle time histogram.
3951 * Most of the time a switch has a relatively small number of facets. When
3952 * this is the case we might as well keep statistics for all of them in
3953 * userspace and to cache them in the kernel datapath for performance as
3956 * As the number of facets increases, the memory required to maintain
3957 * statistics about them in userspace and in the kernel becomes
3958 * significant. However, with a large number of facets it is likely that
3959 * only a few of them are "heavy hitters" that consume a large amount of
3960 * bandwidth. At this point, only heavy hitters are worth caching in the
3961 * kernel and maintaining in userspaces; other facets we can discard.
3963 * The technique used to compute the idle time is to build a histogram with
3964 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
3965 * that is installed in the kernel gets dropped in the appropriate bucket.
3966 * After the histogram has been built, we compute the cutoff so that only
3967 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
3968 * cached. At least the most-recently-used bucket of facets is kept, so
3969 * actually an arbitrary number of facets can be kept in any given
3970 * expiration run (though the next run will delete most of those unless
3971 * they receive additional data).
3973 * This requires a second pass through the facets, in addition to the pass
3974 * made by ofproto_update_stats(), because the former function never looks
3975 * at uninstallable facets.
3977 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3978 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3979 int buckets[N_BUCKETS] = { 0 };
3980 struct facet *facet;
3985 total = hmap_count(&ofproto->facets);
3986 if (total <= 1000) {
3987 return N_BUCKETS * BUCKET_WIDTH;
3990 /* Build histogram. */
3992 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
3993 long long int idle = now - facet->used;
3994 int bucket = (idle <= 0 ? 0
3995 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3996 : (unsigned int) idle / BUCKET_WIDTH);
4000 /* Find the first bucket whose flows should be expired. */
4001 for (bucket = 0; bucket < N_BUCKETS; bucket++) {
4002 if (buckets[bucket]) {
4005 subtotal += buckets[bucket++];
4006 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
4011 if (VLOG_IS_DBG_ENABLED()) {
4015 ds_put_cstr(&s, "keep");
4016 for (i = 0; i < N_BUCKETS; i++) {
4018 ds_put_cstr(&s, ", drop");
4021 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4024 VLOG_INFO("%s: %s (msec:count)",
4025 dpif_name(ofproto->dpif), ds_cstr(&s));
4029 return bucket * BUCKET_WIDTH;
4033 facet_active_timeout(struct ofproto *ofproto, struct facet *facet)
4035 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
4036 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
4037 struct ofexpired expired;
4039 if (facet->installed) {
4040 struct dpif_flow_stats stats;
4042 facet_put__(ofproto, facet, facet->actions, facet->actions_len,
4044 facet_update_stats(ofproto, facet, &stats);
4047 expired.flow = facet->flow;
4048 expired.packet_count = facet->packet_count;
4049 expired.byte_count = facet->byte_count;
4050 expired.used = facet->used;
4051 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4056 ofproto_expire_facets(struct ofproto *ofproto, int dp_max_idle)
4058 long long int cutoff = time_msec() - dp_max_idle;
4059 struct facet *facet, *next_facet;
4061 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
4062 facet_active_timeout(ofproto, facet);
4063 if (facet->used < cutoff) {
4064 facet_remove(ofproto, facet);
4069 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4070 * then delete it entirely. */
4072 rule_expire(struct ofproto *ofproto, struct rule *rule)
4074 struct facet *facet, *next_facet;
4078 /* Has 'rule' expired? */
4080 if (rule->hard_timeout
4081 && now > rule->created + rule->hard_timeout * 1000) {
4082 reason = OFPRR_HARD_TIMEOUT;
4083 } else if (rule->idle_timeout && list_is_empty(&rule->facets)
4084 && now >rule->used + rule->idle_timeout * 1000) {
4085 reason = OFPRR_IDLE_TIMEOUT;
4090 COVERAGE_INC(ofproto_expired);
4092 /* Update stats. (This is a no-op if the rule expired due to an idle
4093 * timeout, because that only happens when the rule has no facets left.) */
4094 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4095 facet_remove(ofproto, facet);
4098 /* Get rid of the rule. */
4099 if (!rule_is_hidden(rule)) {
4100 rule_send_removed(ofproto, rule, reason);
4102 rule_remove(ofproto, rule);
4106 rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason)
4108 struct ofputil_flow_removed fr;
4110 if (!rule->send_flow_removed) {
4115 fr.cookie = rule->flow_cookie;
4117 calc_flow_duration__(rule->created, &fr.duration_sec, &fr.duration_nsec);
4118 fr.idle_timeout = rule->idle_timeout;
4119 fr.packet_count = rule->packet_count;
4120 fr.byte_count = rule->byte_count;
4122 connmgr_send_flow_removed(p->connmgr, &fr);
4125 /* Obtains statistics for 'rule' and stores them in '*packets' and '*bytes'.
4126 * The returned statistics include statistics for all of 'rule''s facets. */
4128 rule_get_stats(const struct rule *rule, uint64_t *packets, uint64_t *bytes)
4131 struct facet *facet;
4133 /* Start from historical data for 'rule' itself that are no longer tracked
4134 * in facets. This counts, for example, facets that have expired. */
4135 p = rule->packet_count;
4136 b = rule->byte_count;
4138 /* Add any statistics that are tracked by facets. This includes
4139 * statistical data recently updated by ofproto_update_stats() as well as
4140 * stats for packets that were executed "by hand" via dpif_execute(). */
4141 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4142 p += facet->packet_count;
4143 b += facet->byte_count;
4150 /* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an
4151 * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to
4152 * their individual configurations.
4154 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
4155 * Otherwise, ownership is transferred to this function. */
4157 send_packet_in(struct ofproto *ofproto, struct dpif_upcall *upcall,
4158 const struct flow *flow, bool clone)
4160 struct ofputil_packet_in pin;
4162 pin.packet = upcall->packet;
4163 pin.in_port = odp_port_to_ofp_port(flow->in_port);
4164 pin.reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION;
4165 pin.buffer_id = 0; /* not yet known */
4166 pin.send_len = upcall->userdata;
4167 connmgr_send_packet_in(ofproto->connmgr, upcall, flow,
4168 clone ? NULL : upcall->packet);
4172 pick_datapath_id(const struct ofproto *ofproto)
4174 const struct ofport *port;
4176 port = get_port(ofproto, ODPP_LOCAL);
4178 uint8_t ea[ETH_ADDR_LEN];
4181 error = netdev_get_etheraddr(port->netdev, ea);
4183 return eth_addr_to_uint64(ea);
4185 VLOG_WARN("could not get MAC address for %s (%s)",
4186 netdev_get_name(port->netdev), strerror(error));
4188 return ofproto->fallback_dpid;
4192 pick_fallback_dpid(void)
4194 uint8_t ea[ETH_ADDR_LEN];
4195 eth_addr_nicira_random(ea);
4196 return eth_addr_to_uint64(ea);
4200 ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED,
4201 void *aux OVS_UNUSED)
4203 const struct shash_node *node;
4207 SHASH_FOR_EACH (node, &all_ofprotos) {
4208 ds_put_format(&results, "%s\n", node->name);
4210 unixctl_command_reply(conn, 200, ds_cstr(&results));
4211 ds_destroy(&results);
4214 struct ofproto_trace {
4215 struct action_xlate_ctx ctx;
4221 trace_format_rule(struct ds *result, int level, const struct rule *rule)
4223 ds_put_char_multiple(result, '\t', level);
4225 ds_put_cstr(result, "No match\n");
4229 ds_put_format(result, "Rule: cookie=%#"PRIx64" ",
4230 ntohll(rule->flow_cookie));
4231 cls_rule_format(&rule->cr, result);
4232 ds_put_char(result, '\n');
4234 ds_put_char_multiple(result, '\t', level);
4235 ds_put_cstr(result, "OpenFlow ");
4236 ofp_print_actions(result, (const struct ofp_action_header *) rule->actions,
4237 rule->n_actions * sizeof *rule->actions);
4238 ds_put_char(result, '\n');
4242 trace_format_flow(struct ds *result, int level, const char *title,
4243 struct ofproto_trace *trace)
4245 ds_put_char_multiple(result, '\t', level);
4246 ds_put_format(result, "%s: ", title);
4247 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
4248 ds_put_cstr(result, "unchanged");
4250 flow_format(result, &trace->ctx.flow);
4251 trace->flow = trace->ctx.flow;
4253 ds_put_char(result, '\n');
4257 trace_resubmit(struct action_xlate_ctx *ctx, struct rule *rule)
4259 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
4260 struct ds *result = trace->result;
4262 ds_put_char(result, '\n');
4263 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
4264 trace_format_rule(result, ctx->recurse + 1, rule);
4268 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
4269 void *aux OVS_UNUSED)
4271 char *dpname, *in_port_s, *tun_id_s, *packet_s;
4272 char *args = xstrdup(args_);
4273 char *save_ptr = NULL;
4274 struct ofproto *ofproto;
4275 struct ofpbuf packet;
4283 ofpbuf_init(&packet, strlen(args) / 2);
4286 dpname = strtok_r(args, " ", &save_ptr);
4287 tun_id_s = strtok_r(NULL, " ", &save_ptr);
4288 in_port_s = strtok_r(NULL, " ", &save_ptr);
4289 packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
4290 if (!dpname || !in_port_s || !packet_s) {
4291 unixctl_command_reply(conn, 501, "Bad command syntax");
4295 ofproto = shash_find_data(&all_ofprotos, dpname);
4297 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
4302 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
4303 in_port = ofp_port_to_odp_port(atoi(in_port_s));
4305 packet_s = ofpbuf_put_hex(&packet, packet_s, NULL);
4306 packet_s += strspn(packet_s, " ");
4307 if (*packet_s != '\0') {
4308 unixctl_command_reply(conn, 501, "Trailing garbage in command");
4311 if (packet.size < ETH_HEADER_LEN) {
4312 unixctl_command_reply(conn, 501, "Packet data too short for Ethernet");
4316 ds_put_cstr(&result, "Packet: ");
4317 s = ofp_packet_to_string(packet.data, packet.size, packet.size);
4318 ds_put_cstr(&result, s);
4321 flow_extract(&packet, tun_id, in_port, &flow);
4322 ds_put_cstr(&result, "Flow: ");
4323 flow_format(&result, &flow);
4324 ds_put_char(&result, '\n');
4326 rule = rule_lookup(ofproto, &flow);
4327 trace_format_rule(&result, 0, rule);
4329 struct ofproto_trace trace;
4330 struct ofpbuf *odp_actions;
4332 trace.result = &result;
4334 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet);
4335 trace.ctx.resubmit_hook = trace_resubmit;
4336 odp_actions = xlate_actions(&trace.ctx,
4337 rule->actions, rule->n_actions);
4339 ds_put_char(&result, '\n');
4340 trace_format_flow(&result, 0, "Final flow", &trace);
4341 ds_put_cstr(&result, "Datapath actions: ");
4342 format_odp_actions(&result, odp_actions->data, odp_actions->size);
4343 ofpbuf_delete(odp_actions);
4346 unixctl_command_reply(conn, 200, ds_cstr(&result));
4349 ds_destroy(&result);
4350 ofpbuf_uninit(&packet);
4355 ofproto_unixctl_init(void)
4357 static bool registered;
4363 unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL);
4364 unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL);
4368 default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
4369 struct ofpbuf *odp_actions, tag_type *tags,
4370 uint16_t *nf_output_iface, void *ofproto_)
4372 struct ofproto *ofproto = ofproto_;
4373 struct mac_entry *dst_mac;
4375 /* Drop frames for reserved multicast addresses. */
4376 if (eth_addr_is_reserved(flow->dl_dst)) {
4380 /* Learn source MAC (but don't try to learn from revalidation). */
4382 && mac_learning_may_learn(ofproto->ml, flow->dl_src, 0)) {
4383 struct mac_entry *src_mac;
4385 src_mac = mac_learning_insert(ofproto->ml, flow->dl_src, 0);
4386 if (mac_entry_is_new(src_mac) || src_mac->port.i != flow->in_port) {
4387 /* The log messages here could actually be useful in debugging,
4388 * so keep the rate limit relatively high. */
4389 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4390 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
4391 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
4393 ofproto_revalidate(ofproto,
4394 mac_learning_changed(ofproto->ml, src_mac));
4395 src_mac->port.i = flow->in_port;
4399 /* Determine output port. */
4400 dst_mac = mac_learning_lookup(ofproto->ml, flow->dl_dst, 0, tags);
4402 flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD,
4403 nf_output_iface, odp_actions);
4405 int out_port = dst_mac->port.i;
4406 if (out_port != flow->in_port) {
4407 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, out_port);
4408 *nf_output_iface = out_port;
4417 static const struct ofhooks default_ofhooks = {
4418 default_normal_ofhook_cb,